Military Communications and Information Technology: A Trusted ...

Military Communications 

and Information Technology: 

A Trusted Cooperation Enabler 

Volume 2 

Warsaw 2012

Reviewers: 

Prof. Milan Šnajder, LOM Praha, Czech Republic 

Prof. Andrzej Dąbrowski, Warsaw University of Technology, Poland 

Editor: 

Marek Amanowicz 

Co-editor: 

Peter Lenk 

© Copyright by Redakcja Wydawnictw Wojskowej Akademii Technicznej. 

Warsaw 2012 

ISBN 978-83-62954-31-5 

ISBN 978-83-62954-52-0 

Publication qualified for printing without editorial alterations made by the MUT 

Publishing House. 

DTP: Martyna Janus 

Cover design: Barbara Chruszczyk 

Publisher: Military University of Technology 

Press: P.P.H. Remigraf Sp. z o.o., ul. Ratuszowa 11, 03-450 Warszawa 

Warsaw 2012

Contents 

Foreword ............................................................................ 5 

Chapter 5 

Tactical Communications and Networks ................................................ 7 

SOA over Disadvantaged Grids Experiment and Demonstrator ............................... 9 

Frank T. Johnsen, Trude H. Bloebaum, Léon Schenkels and team, 

Joanna Śliwa, Przemysław Caban 

GUWMANET – Multicast Routing in Underwater Acoustic Networks ........................ 27 

Michael Goetz, Ivor Nissen 

Network Routing by Artificial Neural Network ............................................ 45 

Michal Turčaník 

An Application of Chord Structure in Tactical Ad-hoc Network .............................. 55 

Jerzy Dołowski, Marek Amanowicz 

Revisiting the DARPA’s Idea of a Programmable Network ................................... 67 

Vladimir Aubrecht, Tomas Koutny 

Selection and Investigation of a Civil Wideband Waveform for Potential Military Use ........... 81 

Ferdinand Liedtke, Matthias Tschauner, Sarvpreet Singh, Marc Adrat, Markus Antweiler 

Experimental Performance Evaluation of the Narrowband VHF Tactical IP Radio 

in Test-Bed Environment .............................................................. 99 

Edward Golan, Adam Kraśniewski, Janusz Romanik, Paweł Skarżyński, Robert Urban 

Hybrid Error Detecting and Correcting System Using Hardware Associative Memories .......... 107 

Ion Tutănescu, Constantin Anton, Laurenţiu Ionescu, Gheorghe Şerban, Alin Mazăre 

Concurrent Error Detection Scheme for HaF Hardware .................................... 117 

Ewa Idzikowska 

Chapter 6 

Spectrum Management and Software Defined Radio Techniques ........................ 133 

A Realistic Roadmap for the Introduction of Dynamic Spectrum Management 

in Military Tactical Radio Communication .............................................. 135 

Bart Scheers, Austin Mahoney, Hans Åkermark 

Dynamic Spectrum Management in Legacy Military Communication Systems ................ 151 

Marek Suchański, Paweł Kaniewski, Robert Matyszkiel, Piotr Gajewski 

Spectrum Issues of NATO Narrowband Waveform ........................................ 161 

Jan Leduc, Markus Antweiler, Torleiv Maseng

4 Military Communications and Information Technology... 

Legacy Waveforms on Software Defined Radio: Can Hierarchical Modulation 

Offer an Added Value to SDR Operators ................................................ 171 

Marc Adrat, Tobias Osten, Jan Leduc, Markus Antweiler, Harald Elders-Boll 

Data Fusion Schemes for Cooperative Spectrum Sensing in Cognitive Radio Networks .......... 187 

Djamel Teguig, Bart Scheers, Vincent Le Nir 

Implementation of an Adaptive OFDMA PHY/MAC on USRP Platforms for 

a Cognitive Tactical Radio Network .................................................... 201 

Vincent Le Nir, Bart Scheers 

Validation of the ITU 1546 Land-Sea Propagation Model for the 900 MHz Band .............. 215 

Krzysztof Bronk, Rafał Niski, Jerzy Żurek, Maciej J. Grzybkowski 

Chapter 7 

Mobile Ad-hoc and Wireless Sensor Networks ......................................... 227 

Algorithms for Channel and Power Allocation in Clustered Ad hoc Networks .................. 229 

Luca Rose, Christophe J. Le Martret, Mérouane Debbah 

High Spatial-Reuse Distributed Slot Assignment Protocol for Wireless Ad-hoc Networks ......... 247 

Muhammad Hafeez Chaudhary, Bart Scheers 

Hybrid Network Synchronization for MANETs ........................................... 265 

Harri Saarnisaari, Teemu Vanninen 

Application of Dezert-Smarandache Theory for Tactical MANET Security Enhancement ........ 277 

Joanna Głowacka, Marek Amanowicz 

Mechanisms of Ad-hoc Networks Supporting Network Centric Warfare ...................... 289 

Rafał Bryś, Jacek Pszczółkowski, Mirosław Ruszkowski 

Using Network Coding in 6LoWPAN WSNs ............................................. 307 

Jarosław Krygier 

Testbed Implementation of Energy Aware Wireless Sensor Network .......................... 319 

Ewa Niewiadomska-Szynkiewicz, Michał Marks, Filip Nabrdalik 

An Energy Aware Self-Configured Wireless Sensor Network ................................ 333 

Marcin Wawryszczuk, Marek Amanowicz 

Chapter 8 

Localization Techniques ............................................................ 347 

Enhanced Location Tracking for Tactical MANETs Based on Particle Filters 

and Additional Information Sources .................................................... 349 

Peter Ebinger, Arjan Kuijper, Stephen D. Wolthusen 

Spatial Localisation of Radio Wave Emission Sources Using SDF Technology .................. 367 

Jan M. Kelner, Piotr Gajewski, Cezary Ziółkowski 

On the Effect of Tuner Phase Noise on TDOA Measurements ............................... 377 

Anders M. Johansson, Patrik Hedström 

Aircraft Tracking Using Mobile Devices ................................................. 387 

Michał Andrzejewski, Radosław Schoeneich 

Index ............................................................................. 397

Foreword 

Modern military operations are conducted in a complex, multidimensional 

and disruptive environment. The challenging political and social environment 

of the operations necessitates establishing coalitions, consisting of many different 

partners of differing levels of trust, e.g. partners from NATO nations, as well 

as non-NATO nations and others such as the local government bodies and local 

forces. Tight collaboration with these partners and the guarantee that the appropriate 

information is shared within the community is vital to the mission efficiency. 

This also requires understanding of these differences and greater trust as well as 

acceptance of the greater risk involved. 

Dynamic environmental changes and limitations of the technical infrastructure 

assets creates additional challenging issues for the effective collaboration 

of the coalition partners. The fragile nature of the communications infrastructure, 

especially at the tactical level, requires robust methods and mechanisms to 

deal with long delays, communication failures or disconnections and available 

bandwidth limitations. 

These all necessitate a better understanding of the environmental conditions 

and appropriate procedural actions, as well as strong technological support, to 

provide the required levels of interoperability, flexibility, security and trusted collaboration 

in connecting heterogeneous systems of all parties involved in the action. 

Many research efforts aimed at the elaboration and implementation of innovative 

communications and information technologies for military systems, 

enabling trusted information exchange and successful collaboration in disadvantaged 

environments, have been undertaken world-wide. The latest selected 

results of such activities that include novel concepts for military communications 

and information systems, as well as innovative technological solutions, are 

presented in this book. 

The book contains the papers originally submitted to the 14 th Military Communications 

and Information Systems Conference (MCC) held on 8–9 October 2012 

in Gdansk, Poland. The MCC is an annual event that brings together experts from 

research establishments, industry and academia, from around the world, as well 

as representatives of the military Communications and Information Systems 

community. The conference provides a useful forum for exchanging ideas on the 

development and implementation of new technologies and military CIS services.

6 

Military Communications and Information Technology... 

It also creates a unique opportunity to discuss these issues from different points of 

view and share experiences amongst European Union and NATO CIS professionals. 

The papers included in this book are split into two volumes, each contains 

selected issues that correspond to the conference topics, and reflect the technology 

advances supporting trusted collaboration of all parties involved in joint 

operations. The first volume is focused on: Concepts and Solutions for Communications 

and Information Systems, Communications and Information Technology 

for Trusted Information Sharing, Information Technology for Interoperability and 

Decision Support Enhancement and Information Assurance & Cyber Defence, while 

the latter on the following: Tactical Communications and Networks, Spectrum 

Management and Software Defined Radio Techniques, Mobile Ad-hoc & Wireless 

Sensor Networks and Localization Techniques. 

The editors would like to take this opportunity to express their thanks to the 

authors and reviewers for their efforts in the preparation of this book. We trust 

that the book will contribute to a better understanding of the challenging issues in 

trusted collaboration in modern operations, scientific achievements and available 

solutions that mitigate the risk and increase the efficiency of information exchange 

in hostile and disruptive environments. We believe that the readers will find the 

content of the book both useful and interesting. 

Marek Amanowicz 

Peter Lenk

Chapter 5 

Tactical Communications 

and Networks

SOA over Disadvantaged Grids Experiment 

and Demonstrator 

Frank T. Johnsen 2 , Trude H. Bloebaum 2 , Léon Schenkels and team 1, 3 , 

Joanna Śliwa 4 , Przemysław Caban 4 

2 FFI, Norway, 

frank-trethan.johnsen@ffi.no, trude-hafsoe.bloebaum@ffi.no 

3 NC3A, Netherlands, Leon.Schenkels@nc3a.nato.int 

4 Military Communication Institute, Zegrze, Poland, 

{j.sliwa, p.caban}@wil.waw.pl 

Abstract: The objective of the IST-090 group is to investigate challenges of SOA applications in disadvantaged 

networks. The group studies possible solutions that can improve the overall efficiency 

of information dissemination when facing different disruptions. This paper presents lessons learned 

from the real-life experiment and demonstration that was carried out by IST-090 group members 

during the MCC 2011 conference. We evaluated several solutions, namely the WS-DDS interface, 

the DSProxy, the Mist protocol, and ESBs that were used as SOA solutions enabling efficient information 

exchange in a disadvantaged environment. The experiment was preceded by separate tests of each 

solution. However, in the combined scenario, the aim was to evaluate the interoperability of these 

solutions and define a long-term plan for either their application in operations or for further functionality 

development. The paper gives an overview of the solutions we investigated, presents a rough 

model of the network environment used, and discusses the results observed and the lessons learned. 

Keywords: component; Web services, publish-subscribe, DDS 

I. Introduction 

NATO is a strategic organization, and, as such, the vast majority of the communications 

infrastructure that is used and deployed by NATO is in the strategic 

realm of static headquarters and Forward Operating Bases (FOB). These tend to be 

served by high speed LAN and WAN communications, with low latency and high 

bandwidth. However, under the NATO Network Enabled Capability (NNEC) vision, 

the stovepipe systems of the past, offering services and applications to a limited, geographically 

co-located group of users will shift to a dynamic federation of systems, 

which will allow services previously only available within the strategic domain to be 

made available in the tactical domain. Likewise, situational awareness information 

1 

The NC3A team members: Rui Fiske, Marc van Selm, Vincenzo de Sortis, and Aad van der Zanden


needs to be fed from the tactical domain into the strategic domain to better inform 

planning and operational decision making. This necessarily involves increased 

transfer of data across tactical links with constrained communications pathways 

in both directions. There is no clear solution yet from industry that will support 

these requirements and so NATO and the member nations themselves need to devise 

suitable mechanisms. The experiment described in this paper is meant to discover 

how solutions may be applied in relation to the anticipated use of services in NNEC. 

Within the NATO and coalition enterprise, there are a wide range of network 

conditions and communications environments. The architects and developers 

of most Functional Area Services (FAS), and often those looking at Service-Oriented 

Architecture (SOA) in particular, tend to consider only high bandwidth, low latency 

local area networks. They find that their applications tend to work well in these 

static environments, with low response times, even for large messages. However, 

thought also needs to be given to the frontline staff; those connected only between 

themselves by Mobile Ad Hoc Networks (MANETs), or by satellite and radio connections 

back to their FOBs. For these users of systems, how does the NNEC SOA 

paradigm deliver the most significant and timely information to them without 

the provision of a “LAN to the man” 

The RTO Group, IST-090, has been tasked with looking at the challenges and 

potential solutions that can be used for SOA over disadvantaged grids [1]. A number 

of NATO member nations were involved in the group, information was shared and 

experiments were conducted together, in order to test the various approaches that 

had been suggested and developed. A number of IST-090 members were invited 

to speak at the Military Communications and Information Systems Conference 

(MCC) in Amsterdam in October 2011. As part of the conference, IST-090 was assigned 

a special area of the conference venue in order to demonstrate the benefits 

of the various solutions that had been presented. As part of this, further experiments 

were conducted over a dynamic network, provided by Norwegian and NC3A-owned 

MANET components. This would provide a real, unreliable network environment 

to assess the actual value of the solutions, and allow further experimentation with 

the MANETs themselves. 

This paper gives a detailed description of the test environment, test plan and test 

results that were conducted and observed at the conference. It describes the network, 

and various SOA-supporting solutions that were deployed by the group members, 

and assesses how each of them provides lessons to consider when designing services 

and systems to be used over poor communications channels. Many different 

potential approaches were identified, many of which could be combined to offer 

a range of network-optimization techniques. 

The remainder of the paper is organized as follows: In Section II we present 

the motivation for pursuing a pervasive SOA in military networks. Related work 

is discussed in Section III. Section IV covers our joint experiment – the setup, 

the execution, and the lessons learned. Finally, Section V concludes the paper.

Chapter 5: Tactical Communications and Networks 

11 

II. SOA motivation 

SOA, realized by Web services technology, has been identified as the crucial 

NNEC enabler [2]. The advantage of SOA is that it provides seamless information 

exchange based on different policies and loose coupling of its components. In a military 

domain it enables making sensitive information resources available in the form 

of services, which can be discovered and used by all mission participants that do 

not need to be aware of these services in advance. 

The most mature technology for implementing SOA, recommended by NATO 

and widely applied in the commercial sector, is Web services. Web services are 

described by a wide range of standards that deal with different aspects of their 

realization, transport, orchestration, semantics, etc. They provide the means to 

build a very flexible environment that is able to dynamically link different system 

components to each other. These standards are based on the eXtensible Markup 

Language (XML) and have been designed to operate in high bandwidth links. 

XML gained wide acceptance and became very popular for the reason that it solves 

many interoperability problems, is human- and machine-readable and facilitates 

the development of frameworks for software integration, independent of the programming 

language. Nevertheless it undoubtedly adds significant overhead, both 

in terms of necessary computation power and consumption of network resources 

while being transported. This means that using SOA in tactical networks is challenging 

and requires optimizations [3], [4]. IST-090 is looking into techniques such 

as compression to mitigate some of the challenges [1]. 

The utilization of Web services and other SOA implementations in a NNEC 

environment (e.g., Enterprise Service Buses (ESBs) and the Data Distribution Service 

(DDS)) has been addressed in many national and international experiments (e.g., Coalition 

Warrior Interoperability Demonstration (CWID) [5], [6], DDS demo [7], 

ESB experiment [8]). These experiments indicate that the technologies improve 

collaboration, interoperation and information sharing in the Federation of Systems 

(FoS). DDS is even perceived as real-time technology that can be tailored for the use 

in low – bandwidth tactical networks. However the idea of IP-based ubiquitous communications 

that is able to feed users with data based on the available communications 

media turns to be difficult to realize at the current stage of available technology 

maturity level. In order to achieve efficient information exchange between FoS users, 

SOA solutions need to work with different types of information and communication 

systems. Service interoperability must be provided among all command levels on 

an end-to-end basis. The challenge is therefore to apply SOA in low bandwidth tactical 

communications systems, which usually cope with high error rates and frequent 

disruptions. Such networks are usually referred to as disadvantaged grids. 

Previous initiatives have focused on information distribution over disadvantaged 

grids. In very low bandwidth environments it is considered to be the best 

solution to use asynchronous replication based middleware that provides static


information distribution between partners that have agreed to use a specific database 

format [9]. This solution is however not very convenient in highly dynamic operational 

scenarios. That is why there has been carried out research on SOA solutions 

that provide flexibility and interoperability and are well suited to work in a FoS. 

III. Related work 

Publish/subscribe, a paradigm for asynchronous communication, has been 

identified as particularly useful in MANETs, where it provides decoupling of provider 

and consumer [10]. In this paper we focus on MANETs, and thus the publish/ 

subscribe paradigm. The NATO Core Enterprise Services (CES) Working Group [11] 

has identified WS-Notification as the standard to use for publish/subscribe in NATO. 

Thus, we employ that standard in our experiment in this paper. 

Net-Centric Tactical Services (NCTS) provide a gateway and software framework 

for tactical users to realize the benefits of information sharing across a SOA environment 

[12]. It is a software framework which resides in the tactical environment and 

supports a set of services and functions to enable communications and messaging 

translation, data publishing, data subscription, and tactical device management. This 

is in key with the NNEC FS [2], which states that in order to apply SOA throughout 

a FoS in various network elements, special devices – so called edge proxies that are to 

support information distribution over disadvantaged grids – must be used. The problem 

is therefore identified and is considered valid for the whole NATO community. 

Those edge functionalities are to adapt service traffic to the capabilities of the tactical 

networks, sometimes in the form of technology gateways. As a consequence, we also 

pursue the gateway and proxy concept in our experiments in this paper. 

In [13] we describe a Web services infrastructure experiment performed during 

the summer of 2011. This experiment is a precursor to the Norwegian and NC3A parts 

of the experiment described in this paper. In 2011 we identified several shortcomings 

of the software we were using then, e.g., that the WS-Notification framework we used 

on the Norwegian side (WSMG from the University of Indiana) was not suitable for 

use in a military environment, since it did not handle disruptions. In the IST-090 

experiment described in this paper we have replaced WSMG with Apache Service- 

Mix in an attempt to leverage a hopefully more mature and stable product. Also, 

we now focus mostly on the MANET aspect of communications, whereas the 2011 

experiment was more geared towards interconnecting infrastructure. 

The SOA over disadvantaged grids experiment and demonstrator discussed 

in this paper was linked to the other IST-090 presentations during the MCC 2011 

in Amsterdam: 

• An overview of the research and experimentation of IST-090 [1] 

• Mediation of network load over disadvantantaged grids using Enterprise 

Service Bus (ESB) technology [8] 

• DDS technology demonstrations related to IST-090 [7]


13 

• Dedicated WS-DDS interface for sharing information between civil and 

military domains [14] 

• An Evaluation of Web Services Discovery Protocols for the Network-Centric 

Battlefield [15] 

• An independent evaluation of Web service reach solutions in disadvantaged 

grids [16] 

• Towards a middleware for tactical military networks – interim solutions 

for improving communication for legacy systems [17] 

• Semantic Description of Web Service QoS Profiles for Context-aware Web 

Service Provision [18] 

These papers give an overview of recent IST-090 related work by the member 

nations. The remainder of this paper focuses on the aspects of the experiment and 

demo performed at MCC 2011. 

IV. Experiment 

The goal of the experiment was to evaluate the possibility of delivering information 

from service producers to service consumers in a disadvantaged environment. 

The networking environment was set up using equipment provided by the NC3A 

and FFI (see Figure 1). Further, technologies studied under the umbrella of IST- 

090 as promising for SOA application on the tactical level were employed: Web 

services, ESBs, and DDS. We built a heterogeneous networking infrastructure that 

modeled cooperation of users on different levels of command. Information was sent 

up and down the echelons of command through the MANET. Thus, we had aspects 

of both infrastructure and disadvantaged networks. We evaluated the interoperability 

of the proposed solutions, as well as their applicability to the scenario that 

was developed by the IST 090. For further information regarding the scenario, 

please refer to [1]. Moreover such real-life experiments always bring us lessons 

learned in terms of implementation correctness and incompatibility of technologies. 

A number of potential approaches have been identified by the participants 

in this experiment, which may improve communications between service providers 

and service consumers. This should lead to a concrete set of recommendations 

for FAS and Communities of Interest (CoI) when commissioning and developing 

their services. Specifically, these approaches are: 

• A DDS domain, which is interfaced to a Web services environment through 

a gateway (MCI’s contribution). 

• DSProxy for delay and disruption tolerant SOAP transport across disadvantaged 

networks (one of FFI’s contributions). 

• Decentralized chat using the experimental Mist protocol, coupled with 

a Mist/XMPP gateway for interoperability with infrastructure networks 

(one of FFI’s contributions). 

• The use of an ESB to optimize the use of bandwidth and manage limited 

connectivity (NC3A’s contribution).


Figure 1. Disadvantaged grid experiment and demonstrator communications network 

Figure 2.SOA infrastructure


15 

The disadvantaged grid networking infrastructure used was based on equipment 

provided by NC3A and Norway: Rajant breadcrumb MANET systems using 

the 2.4 and 5 GHz ISM bands augmented by a simulated satellite link. The NC3A 

provided a complete infrastructure to execute SOA tests and to evaluate the results, 

and functioned as the connecting “hub” between Norway and Poland. This infrastructure 

was built around two ESB instances in charge for the message routing and 

transformation (see Figure 2). Here, the interfaces were used as follows: 

• SIP3 Engine – This is a processing engine for Friendly Force Tracks in NFFI 

V1.3 format that can simulate running tracks, store and filter them. During 

this test campaign it was used as track store, track simulator and SIP3 

interface. The SIP3 Engine is composed of: Track store, Track simulator, 

SIP3 message interface (NFFI over SOAP), IP1 message interface (NFFI 

over TCP/IP), IP2 message interface (NFFI over UDP), and SIP3/IP1/IP2 

interface adapter. 

• WSO2 ESB (version 3.0.1) – This is a COTS ESB based on Apache 

Synapse. During this test campaign it was used for protocol adaptation 

and message routing/conversion. The main services exposed were: a) 

SIP3 to KML converter, which was used to visualize on standard KML 

enabled map viewer the NFFI V1.3 tracks stored in the SIP3 track store; 

b) NVG to KML converter, which was used to visualize on standard 

KML enabled map viewer the NVG layer provided by JocWatch; c) 

WSNotification (notification consumer) to SIP3 (event sink) which 

was used to feed the SIP3 engine with the NFFI tracks published from 

the Norwegian pub/sub broker, and d) WS-DDS to SIP3 (event sink), 

which was used to feed the SIP3 engine with the NFFI tracks published 

from the WS-DDS interface. 

• ServiceMix ESB (version 3.5.0) – This is a COTS ESB that offers a WS- 

Notification v1.3 Pub/Sub interface. During this test campaign it was used 

to publish messages via the WS-Notification protocol. 

• Symbology server – An NC3A service used to graphically represent the units 

given their APP6 symbol. 

• JOCWatch – The Incident Reporting source systems. 

• Google Earth – COTS map viewer used to visualize the tracks produced 

and received from the NC3A systems. 

Figure 1 shows the network diagram that was used in the experiment and 

demonstration. 

The experiment was aimed at connecting Web services and DDS domains, 

which are technologically different realizations of SOA. Web services support both 

publish/subscribe and request/response communications, as opposed to DDS which 

has only publish/subscribe support. On the other hand, DDS supports QoS features 

and real-time systems, which Web services do not. For a comparison, see Table I. 

More details about DDS can be found in [14].


Table I. Web services and DDS comparison 

Property Web services DDS 

Standardization W3C, OASIS, WS-I OMG 

Real-time and QoS support No Yes 

Request/response paradigm Yes No 

Publish/subscribe paradigm Yes Yes 

Current standard suitable for disavantaged grids No No 

Enabling support for disadvantaged grids 

Experimental 

optimizations 

Vendor specific 

tactical extensions 

Figure 3. Mist as a message ferry 

A. Technology components 

There is a combination of parameters at hand that will lead to specific use 

cases where the following conditions can be tested: 

• Different types of information (NFFI, incident reports) 

• Different solutions (DDS and DDS/WS gateway, DSProxy, ESB capabilities) 

• Different network conditions (mostly making use of the MANET but also 

having the possibility to simulate other types of networks). 

We evaluated each tested solution’s usability from an end user perspective, and 

were able to show the viability of the solutions involved for effectively disseminating 

the necessary information. It should be noted that DDS was not actually used 

in the MANET, it was connected to the MANET using a Web services interoperability 

gateway. Further information regarding the separate demo components are 

discussed briefly below. The complete details of the demo are described in [19].


17 

1) Mist: The Mist protocol provides robustness and delay tolerance in dynamic 

wireless ad-hoc networks. We wanted to use two separate 802.11 ad-hoc networks 

with one gateway/relay node connected to 100BASE-TX for the Mist/ 

XMPP experiments. FFI provided the necessary equipment for the mobile 

nodes – a mixture of laptops and Sony Ericsson Xperia mobile phones with 

WiFi enabled were used. One computer, the server in the national services LAN 

where the DSProxy resides, was running the XMPP/Mist gateway software. 

Figure 4. NC3A-NOR interconnection featuring ESBs and the DSProxy 

The XMPP server could be anywhere in the network, as long as it was reachable 

by TCP from the XMPP/Mist gateway node from time to time. In this demo 

the central server was hosted by the NC3A in the NATO LAN. Messages were 

relayed to XMPP when the connection was available and stored for future delivery 

when the link was down, using the ”message ferry” principle shown in Figure 3. 

The message ferry was a single wireless node (mobile phone) which was used 

to carry messages between the two networks. It did not need to be connected 

to both ad-hoc networks at the same time, but it did require the use of 802.11 

to connect in an ad-hoc manner. For in-depth information on using Mist for 

tactical chat, see [20]. 

2) DSProxy: The use of proxy servers for adapting Web services to disadvantaged 

grids is a very promising approach. Through the use of proxies one 

can utilize unmodified Web services at the client and server machines, and 

only the intermediate nodes in the network need to use the proxy software. 

This reduces complexity in the development of applications and servers, and 

therefore also costs. Norway (FFI) has designed and developed the Delay and 

disruption tolerant SOAP Proxy (DSProxy) system. 

The DSProxy is a prototype system implementing a range of concepts, ideas and 

mechanisms which aim to tackle the challenges associated with utilizing web services 

in tactical environments and disadvantaged grids [21]. The DSProxy is a middleware 

component enabling delay and disruption tolerant web services for heterogeneous 

networks. The Java-based software uses acknowledged standards and is designed 

to work with existing COTS Web service clients and services.


Functional tests where NC3A provided NFFI data on a WS-Notification interface 

were shown (see Figure 4). Here, the DSProxy was used to ensure disruption 

and delay tolerant SOAP transport across the unstable MANET and highdelay 

satellite link, whereas the exchange between the NC3A and NOR was conducted 

using ServiceMix to provide WSNotification. 

3) DDS/WS Gateway: DDS is a specification of a publish/subscribe middleware 

for distributed systems created in response to the need to standardize 

a data-centric publish/subscribe programming model for distributed systems. 

It has been suggested for use on the tactical level [7], but such usage requires 

vendor specific extensions. In this paper we focus on interoperability with DDS, 

and do not evaluate any tactical extensions. Poland has developed a gateway 

to bridge the DDS environment to a Web services environment, focusing on 

one specific Web service – blue force tracking using NFFI. This gateway bridging 

the DDS domain with the rest of the infrastructure is shown in Figure 5. 

It enables DDS publishers to deliver NFFI tracks to Web services subscribers 

and, in the opposite direction, NFFI tracks published by Web services can be 

delivered to subscribers on the DDS side. It should be noted that, when using 

the DDS/WS Gateway, for every Web service that needs to be bridged to 

the DDS domain one needs to implement a special purpose message translator 

between the standardized Web services interface and a DDS interface. 

B. Results 

1) FFI infrastructure: ServiceMix was used to exchange NFFI tracks between 

Norway and NC3A. Tracks from Poland were submitted to the NC3A, which 

re-distributed them to Norway. Once the system was up and running everything 

functioned as expected with WS-Notification, but there were some 

initial problems: The version of ServiceMix that was deployed (3.2.3) by FFI 

was not fully standards-compliant, leading to some trouble on NC3A’s side. 

The problems could be handled by a work-around in the subscription message 

and some tuning of the WS-Notification client software. This version 

of ServiceMix supported only IPv4, meaning that notifications could not be 

used over IPv6, even if the communication stack was available. In the future, 

a newer version of ServiceMix, or another application entirely, should 

be considered to provide WS-Notification support. 

Mist functioned as it should by delivering chat messages in a disruption tolerant 

manner across both IPv4 and IPv6. Also, the message ferry principle functioned 

as expected, thereby showcasing one of the strengths of the Mist protocol. 

The DSProxy was used to ensure delay and disruption tolerant communication 

across the „troublesome” networks, i.e., the MANET and the satcom simulator. 

A bridge that had been developed in-house by FFI was used to interconnect 

ServiceMix with the DSProxy, so that notifications could be transported across


19 

the national network to and from the troops. This solution worked as expected 

— internally, instabilities were overcome in the network thanks to the DSProxy, 

while still retaining external compatibility with NC3A through WSNotification on 

the ServiceMix instance. The current implementation of DSProxy has previously 

only been used with IPv4, and at this testing event it was not able to run over IPv6. 

This is an important finding, and support for IPv6 should be sought for in the future. 

Figure 5. WS-DDS is a gateway between the Web services (WS) and Data distribution 

service domains 

2) MCI infrastructure: WS-DDS interface enabled the bidirectional exchange 

of information between the Polish DDS domain and NC3A’s WS domain. 

Additionally, the NFFI tracks from NC3A were sent to the Norwegian domain 

and back (from FFI through NC3A to POL DDS domain). 

Both the WS-DDS interface and the DDS domain emulator operated correctly. 

Initially there was a problem with interoperability between the WS-DDS 

interface and the SIP 3 service. The SIP3 client was created based on the WSDL 

1.1.0, whereas SIP3 service was based on WSDL 1.1.9. After recompilation, and 

some additional modifications in the structure of the NFFI messages exchanged, 

the problem was solved. 

The WS-DDS interface version that was tested currently works in the request/ 

response mode. It would be highly recommended to enhance it with pub-sub functionality 

based on WSNotification. This will improve the functionality of WS-DDS, 

since DDS generally implements a publish/subscribe message exchange pattern. 

Furthermore, this will also decrease the time of information transfer and traffic 

generated in the WS domain. 

3) NC3A infrastructure: The competitive approaches to optimizing traffic using 

the ESB involve two streams of messages being sent to different services 

through the ESB in parallel, a setup further discussed in [8]. One of the services 

is of a higher priority than the other, so it is given more resources in the ESB 

than the other. The number of successful messages to this service is then 

measured. Although the lower priority messages are considered less important,


the number of errors in this background stream (3000 messages over 30 secs) 

is still recorded to give an impression of the negative impact on this stream. 

Two competitive approaches were tested. Prioritization boosts the priority 

of the more important message stream, but does not limit the background 

messages. Throttling, on the other hand, limits the amount of bandwidth that 

a particular service can use, and caps the number of messages that can be 

passed to this service. 

Messages 

Table II. ESB Competetive Test Results 

Background 

errors 

Max Min Average 

Baseline 32 536 57.3 s 7.7 s 32.6 s 

Prioritization 36 601 38.9 s 2.2 s 20.5 s 

Throttling 40 2667 54.2 s 3.3 s 29.5 s 

Prioritization delivered a small but significant (12.5%) improvement on 

the balanced approach, without a massive impact on the background messages. 

Although throttling had a more noticeable effect on the higher priority messages 

(an improvement of 25%), the effect on the background messages was severe. 

The results are shown in Table II. A set of evaluations identifying uncompetitive 

approaches was also performed, these results are shown in [19]. 

It is recommended that a balanced and judicious use of the two approaches 

should be considered for use in production. The relative prioritization and throttling 

parameters will be dependent on the level of service that is offered by the ESB. 

C. Lessons learned 

The use of SOAP over HTTP provides a number of challenges within an environment 

where bandwidth is limited. XML is a verbose data format, and the overhead 

of the SOAP messaging protocol adds additional bytes to the payload. As was seen 

in these experiments, within a MANET environment, where reasonably high 

bandwidth is available between the individual nodes on the MANET, this effect 

can be negligible. 

The use of HTTP can be problematic in case of disruptions due to node 

mobility or interference, though. Also, across radio or satellite communications 

channels, or even a combination of these, then the use of SOAP can cause 

timeouts and limit the number of messages that can be delivered. Therefore, 

mechanisms need to be defined that can provide any level of improvement 

across these network boundaries, and recommendations delivered to the projects 

responsible for the delivery of Functional Area and Core Enterprise Services. 

Even in simple request-response message exchanges between service providers


21 

and consumers, systems can benefit from the judicious use of disabling HTTP 

continue, increasing the number of sockets and lengthening the timeouts on 

the client. The suitability of these approaches should also be considered in addition 

to mediation through any kind of middleware or proxy. However, middleware 

such as ESBs and the DSProxy do help manage poor network communications. 

Further refinements, such as switching to other protocols like DDS or JMS, 

or even prototypes targeted at unreliable networks like Mist, may offer even 

more possibilities. Also, employing compression and other optimization techniques 

must be considered [1]. Any proposed approach should be considered 

on a case-by-case basis, and further research is needed to identify the optimum 

approaches for network optimization. 

1) Data distribution service: DDS is an emerging middleware that is explicitly 

targeted at providing publish/subscribe communications for real-time systems. 

It offers fine and extensive control of QoS parameters, including reliability, 

bandwidth, delivery deadlines, and resource limits that makes it a candidate 

for use in disadvantaged grids. Note, however, that DDS requires vendor 

specific extensions to function in such environments [7]. 

DDS defines a specific API for the messages and subscription handling but 

cannot natively interoperate with SOAP Web services. The WS-DDS gateway 

is filling this gap by building a SOAP web service interface on the top of the DDS 

API so that a standard SOAP data consumer/producer can directly use this interface. 

The experiment with the WS-DDS gateway was very successful, and future 

deployment of DDS could build upon the knowledge gained with this prototype. 

A possible option for future experimentation is to leverage an ESB as frontend 

for SOAP clients and to configure DDS as a transport protocol (as a substitute for 

HTTP). Such a setup could possibly: 

• Make use of the DDS strengths for content-centric message handling and 

QoS definitions. 

• Be a beneficial approach to leveraging COTS middleware. 

• Offer the DDS API to a SOAP service. 

• Hide the complexity of the DDS protocol and the configuration details 

from the SOAP client. 

• Offer the possibility to centrally manage QoS parameters for topics and 

messages. 

• Offer SOAP interfaces that are payload agnostic/independent. 

This approach is similar to a widely adopted solution where a SOAP frontend 

is using the Java Message Service (JMS) as unified transport protocol. The DDS-ESB 

approach must be tested and validated because of the differences between the JMS 

and DDS protocols [22]. A possible drawback of having a SOAP payload agnostic 

interface is that the content-aware DDS is a strong typed protocol, in order to improve 

performance. Extensive tests must be made to guarantee that using generic 

payloads will not negatively impact DDS performance.


2) Publish/subscribe, DSProxy, and Mist: The use of publish/subscribe mechanisms 

further offer a range of possible solutions to unreliable network communications. 

By exchanging messages only in response to clearly-defined 

events, the amount of network traffic can be reduced (through the elimination 

of unnecessary polling), and the effects of intermittent communications 

can be lessened. However, particularly in the coalition environment, with 

multiple heterogeneous systems, it is important to ensure that any solution 

is fully standards-compliant, otherwise there may be interoperability problems, 

as were observed with ServiceMix. 

It is likely in the future that NATO systems will migrate to an IPv6 infrastructure, 

and so it is encouraging that IPv6 functions correctly over the MANET. 

However, it is important that any applications that are developed or procured 

to be used within the NATO enterprise are also able to operate over IPv6. 

The version of ServiceMix used by FFI (3.2.3) supported only IPv4, meaning 

that notifications could not be used over IPv6, even if the communication stack 

was available. It was also an important discovery that the current implementation 

of DSProxy was only able to run over IPv4, and FFI will be looking to support 

IPv6 in the future. 

Mist functioned as expected, and was able to deliver the chat messages in a disruption 

tolerant manner across the networks. Mist supports both IPv4 and IPv6, 

and functioned fully using either protocol. The message ferry principle functioned 

adequately as well, showcasing one of the strengths of the Mist protocol. 

3) Enterprise service buses: The use of ESBs or other middleware offer a number 

of advantages when mediating communications across constrained communications 

channels. The first and most important of these is that it abstracts 

the optimization of the use of bandwidth away from the developers and administrators 

of the services themselves. To the service consumers and providers, 

the network itself should be transparent, and so with the messaging framework 

that is used across this network. By using ESBs to manage traffic optimization, 

the service consumer calls the ESB using SOAP over HTTP, without having to 

consider any other features. Management of the message bus becomes a separate 

concern, and additional optimization features can be deployed without 

the need to redevelop or recompile the applications themselves. In addition 

to this benefit, there are a number of features that are supported by ESBs that 

can greatly improve performance, and these are discussed in Section IV-B3 

above. The use of retries and reliable messaging adds marginal but significant 

improvements over a direct call from service consumer to provider. The most 

obvious benefit that can be delivered by ESBs is the use of compression for 

larger messages, where improvements of over 500 per cent were observed on 

a 111 KB message, using only the standard GZIP compression mechanism. 

Other XML-specific compression suites promise even higher compression 

ratios and therefore even greater improvements.


23 

V. Conclusion 

No single solution stood out as the ”magic bullet” to solve all the requirements 

for high speed connectivity to the edge, but many of them do offer measurable 

improvements in messaging capability. A number of key success factors were 

identified, including the foundation on open standards, ease of management and 

configuration, and transparency to the user. 

We have shown that the standards are important in delivering interoperability 

in heterogeneous FoS environment, but are not sufficient to be used in disadvantaged 

grids. The proposed DSProxy standing upon the standard SOAP communication 

enables compression of SOAP messages and provides store and forward functionalities 

if connectivity is lost. The WS-DDS interface creates unique possibility of connecting 

the WS and DDS architecturally different domains giving the possibility to 

exchange messages vertically between the echelons of command. The Mist protocol 

is a solution for the XMPP deficiencies enabling it to deliver the chat service on 

the tactical level. And finally, the ESBs tuned appropriately give benefits in terms 

of compression and priority of messages. 

In general, the messaging infrastructure should be optimized for the consumers 

of services without the need to incorporate proprietary, ad hoc solutions 

that will ensure tighter coupling between providers and consumers and therefore 

limit the range of potential partners. Where a protocol is not widely understood 

in another domain, then gateways should be used to translate from one standard 

or protocol to another. 

As a suggestion for future experiments, we recommend that the different 

potential solutions should be considered individually and collectively to provide 

optimized communications across the entire enterprise, from static HQs, with 

high speed communications to land forces with intermittent and low quality links 

according to a well defined scenario. Each solution should be considered on a caseby-case 

basis, but once the best solution for the scenario has been identified, then 

it should be quick and easy for an administrator to apply, ensuring the best possible 

experience for front-line users. 

Acknowledgment 

The authors would like to thank the organizers of MCC 2011 for making 

the experiment and demo possible. Furthermore, we would like to acknowledge 

Magnus Skjegstad, who developed the Mist software as part of his ongoing Ph.D. 

thesis work, and supported the chat part of the experiment. He also provided us 

with Figure 3. Finally, we would like to thank Rolf Rasmussen and Johnny Johnsen 

for proofreading.


References 

[1] P.-P. Meiler et al., “An Overview of the Research and Experimentation of IST-090: 

SOA over Disadvantaged Grids,” in proceedings of the MCC 2011, Amsterdam, 

Netherlands, October 2011. 

[2] M. Booth et al., “NATO Network Enabled Feasibility Study vol. II: Detailed Report 

Covering a Strategy and Roadmap for Realizing an NNEC Networking and Information 

Infrastructure, version 2.0,” NATO C3 Agency, 2005. 

[3] D. Dragolov et al., “Service Oriented Architecture – concept of services, 

infrastructure components, implementation practices and military environment 

specifics,” in proceedings of the MCC 2009, Prague, Czech republic, September 2009. 

[4] F.T. Johnsen, “Pervasive web services discovery and invocation in military networks,” 

FFI report 2011/00257, http://rapporter.ffi.no/rapporter/2011/00257.pdf, January 2011. 

[5] J. Busch, “An investigation into deploying web services,” TN 1229, NC3A, December 2007. 

[6] G. Babakhani et al., “Web Trends And Technologies And NNEC Core Enterprise 

Services – Version 2.0,” NC3A Technical Note 1143 (draft), NC3A, The Hague, 

Netherlands, December 2006. 

[7] I.H. Novo et al., “DDS technology demonstrations related to IST-090,” in proceedings 

of MCC 2011, Amsterdam, Netherlands, October 2011. 

[8] R. Fiske et al., “Mediation of network load over disadvantaged grids using Enterprise 

Service Bus (ESB) technology,” in proceedings of the MCC 2011, Amsterdam, 


[9] P. Bartolomasi et al., “NATO Network Enabled Feasibility Study vol. I, ver. 2.0,” 

NATO C3 Agency, 2005. 

[10] P. Costa et al., “Socially-Aware Routing for Publish-Subscribe in Delay- 

Tolerant Mobile Ad Hoc Networks,” IEEE JOURNAL ON SELECTED AREAS 

IN COMMUNICATIONS, VOL. 26, NO. 5, JUNE 2008. 

[11] Consultation, Command and Control Board (C3B), “CORE ENTERPRISE 

SERVICES STANDARDS RECOMMENDATIONS: THE SOA BASELINE PROFILE 

VERSION 1.7,” Enclosure 1 to AC/322-N(2011)0205, NATO Unclassified releasable to 

EAPC/PFP, 11 November 2011. 

[12] S.D. Crane et al., “Bridging the digital divide with net-centric tactical services,” 

AFCEA-GMU C4I CENTER SYMPOSIUM ”CRITICAL ISSUES IN C4I”, George 

Mason University, Fairfax, Virginia Campus, USA, 20-21 May 2008. 

[13] F.T. Johnsen et al., “Towards operational agility using service oriented integration 

of prototype and legacy systems,” in proceedings of the 17th ICCRTS, Fairfax, VA, 

USA, June 19-21 2012. 

[14] P. Caban et al., “Dedicated WS-DDS Interface for Sharing Information Between 

Civil and Military Domains,” In proceedings of MCC 2011, Amsterdam, Netherlands, 

October 2011. 

[15] M. Skjegstad et al., “An Evaluation ofWeb Services Discovery Protocols for 

the Network-Centric Battlefield,” in proceedings of the MCC 2011, Amsterdam, 


[16] A.H. Landa et al., “An Independent Evaluation of Web Service Reach Solutions 

in Disadvantaged Grids,” in proceedings of the MCC 2011, Amsterdam, Netherlands, 

October 2011.


25 

[17] C. Barz et al., “Towards a Middleware for Tactical Military Networks Interim Solutions 

for Improving Communication for Legacy Systems,” in proceedings of the MCC 2011, 

Amsterdam, Netherlands, October 2011. 

[18] J. Sliwa et al., “Semantic Description of Web Service QoS Profiles for Context-aware 

Web Service Provision,” in proceedings of the MCC 2011, Amsterdam, Netherlands, 

October 2011. 

[19] P. Caban et al., “SOA OVER DISADVANTAGED GRIDS EXPERIMENT AND 

DEMONSTRATOR,” NC3A Reference Document 3342, NATO Unclassified, 

The Hague, Netherlands, December 2011. 

[20] M. Skjegstad et al., “Distributed Chat in Dynamic Networks,” in proceedings 

of the 30th IEEE Military Communications Conference (MILCOM2011), Baltimore, 

MA, USA, November 2011. 

[21] K. Lund et al., “Robust web services in heterogeneous military networks,” IEEE 

Communications Magazine, vol. 48, no. 10, pp. 78-83, October 2010. 

[22] Real-Time Innovations Inc., “A Comparison and Mapping of Data Distribution Service 

(DDS) and Java Message Service (JMS),” (on-line), http://www.omgwiki.org/dds/sites/ 

default/files/Comparison of DDS and JMS.pdf, 2006.

GUWMANET – Multicast Routing in Underwater 

Acoustic Networks 

Michael Goetz 1 , Ivor Nissen 2 

1 Fraunhofer Institute for Communication, Information Processing and Ergonomics (FKIE), 

53343 Wachtberg, Germany, michael.goetz@fkie.fraunhofer.de 

2 Research Department for Underwater Acoustics and Marine Geophysics (FWG) – WTD 71, 

24148 Kiel, Germany, ivornissen@Bundeswehr.org 

Abstract: Underwater networks move more and more into the focus of the research community, 

especially for military purposes. They enable the full integration of underwater components like 

submarines or sensor platforms into maritime Network Centric Warfare (NCW). Nevertheless, most 

of the scientific work was done in physical layer methods and medium access protocols. 

In this paper we introduce a new network protocol called Gossiping in Underwater Acoustic Mobile 

Ad-hoc Networks (GUWMANET), which realizes medium access and routing functionality 

in a cross-layer design. In contrast to other routing approaches for underwater networks, we developed 

a network protocol from scratch, fitting the special needs of underwater communication, 

instead of adopting existing terrestrial protocols. 

We use multi-hop strategies to achieve a higher maximum transmission range than that of our 

physical layer method. Additionally, multicast addresses are used which allow an unlimited 

number of nodes. The routes between the nodes are build up automatically and need no preceding 

configuration, which allows a full ad-hoc capability including mobile nodes. In combination 

with the Generic Underwater Application Language (GUWAL), which has a 16 bit header 

with the multicast source and destination address, GUWMANET needs only 10 bits additional 

overhead. This is realized by using gossiping strategies, where each node itself decides whether 

it forwards the data or not. 

Keywords: multicast, multi-hop, gossiping, routing, underwater acoustic networks, mobile ad-hoc 

networks, emission control, implicit acknowledgment, clustering, bursts 


In the last years, underwater networks received more and more attention 

in scientific, industrial and particularly military areas. They enable the full integration 

of underwater components like submarines, Autonomous Underwater Vehicles 

(AUVs), gliders, or bottom sensors into maritime Network Centric Warfare (NCW) 

for example in Mine Counter-Measure (MCM) or Anti Submarine Warfare (ASW) 

operations. Especially, Underwater Wireless Networks (UWNs) moved into the focus 

of the research community to enable full flexibility of the platforms without


the need for any cables, which are not only expensive but also difficult to handle 

and limiting the maneuverability of the moving nodes [1]. 

The ocean is almost impervious to electro-magnetic waves which makes 

them useless for wireless underwater communication over distances greater 

than a hundred meters; wireless communication to submerged nodes can only 

be realized using sound. The network protocol GUWMANET presented in this 

paper, developed by Fraunhofer FKIE and FWG, is designed to establish a robust 

mobile ad-hoc UWN working in all-weather and sea conditions. Within this scope 

GUWMANET is a possible candidate to fulfill delay tolerant scenarios defined by 

the project Robust Acoustic Communications in Underwater Networks (RACUN) 

under the European Defense Agency (EDA) [2], [3]. A detailed description of the scenarios 

follows in Chapter III. 

II. Physical layer restrictions 

In principle, sound waves can propagate several hundreds of kilometers 

in deep waters, nevertheless they underlie natural restrictions which make robust 

underwater communication challenging. Due to the fact that the absorption loss 

of sound waves increases with frequency, the available bandwidth stays in reciprocal 

relation to the maximum transmission range, as shown in Figure 1. In addition, 

the weather conditions influence the maximum transmission range, because breaking 

waves and rain increase the noise level. To overcome these bandwidth limitations, 

we introduce multi-hop strategies in our protocol, which are well-known from 

terrestrial Mobile Ad-hoc Networks (MANETs). 

Figure 1. Upper bound of the User Data Rate (UDR in bps) over transmission range (km) in deep 

and shallow waters and different weather conditions, simplified with homogeneous propagation 

conditions. The worse the weather condition, the lower is the expected transmission range


29 

GUWMANET is designed for a physical layer method based on impulse communication, 

the so called Transient Underwater Acoustic Communication System 

(TUWACS) [4]. It sends out short data bursts with a fixed length of 128 bit in 0.3 s 

plus 10 bits for a network header. This clarifies that there is no room for much protocol 

overhead and an underwater network protocol must be as efficient as possible. 

Another difference to terrestrial communications is the low and varying sound 

propagation speed between 1405 and 1560 meters per second. This does not only 

increase the Doppler compensation complexity significantly, but also the signal propagation 

delay. TUWACS is designed for a maximum internode distance of 10 km. 

This distance leads to an optimal frequency band of 3.5−7.5 kHz. A transmission 

over a distance of 10 km needs 6−7 s, which must be considered in the protocol 

design. A basic assumption of terrestrial protocols is that the transmission time 

is much higher than the propagation time. Therefore, terrestrial network protocols 

cannot be easily adopted. Instead, a completely new one has to be developed from 

scratch to meet the requirements of underwater networks. 

III. Scenario 

With reference to Figure 2, in the scalable RACUN scenario [5] it is assumed 

that an underwater acoustic network is deployed in proximity of a harbor to be 

surveilled. All nodes are bottom-mounted and organized in subsequent barriers. 

These are sets of nodes arranged in a line topology: the first barrier is placed 

in front of the harbor, and is composed of the largest number of nodes in order to 

ensure the largest sensing coverage along the coast. The distance between nearest 

neighbors within a barrier is set to 3 km. The sensing range is assumed to be 2 km. 

Every 8 km comes another barrier which can sense movement as well as relay data 

towards the cooperating fleet at the sea base. Again, the nodes in the barrier are 

arranged in a line topology. While proceeding towards the sea base, the number 

of nodes per barrier is reduced from 5 (in the first barrier) to 2 (in the barrier farthest 

from the shore); in fact, the most important sensing task is carried out near 

the coast, whereas the barriers out at sea are key for relaying packets. Nevertheless, 

their sensed data are useful to confirm the detections of the first barrier and to give 

some further clue about the direction of movement of the boats exiting the harbor. 

The network covers a total area of 16 km × 32 km. We recall that the intended 

maximum transmission range of a node in our networks amounts to about 10 km, 

hence the barriers are typically in range of each other. In this paper, we assume 

that two nodes are deployed close to the last barriers: one buoy on the right 

side and one ship on the left side. These entities are part of the network, and act 

as seaborne sinks. 

Although the number of nodes is reduced after each barrier, our network topology 

features high connectivity, and multiple path alternatives exist. This makes 

the network robust against node failures as well as broken links (for example caused


by the noise of the boat propellers disturbing the communications). Additionally, 

the gateway buoy and the ship stationing at sea are equipped with both acoustic 

and radio communication hardware: therefore, as either receives data over acoustic 

links, such data is immediately relayed to all other sinks and the cooperating fleet 

using radio communications. 

Sea base with 

cooperating fleet 

Last barrier 

Ship 

Buoy 

8km 

3km 

First Barrier 

Surveilled harbor 

Figure 2. RACUN harbor surveillance network scenario, with 14 bottom nodes organized in 4 parallel 

barriers. Additionally, a ship and a gateway buoy act as sea-borne sinks, as they gather detection information 

from the network and relay it to the cooperating fleet stationing in the sea base area 

The traffic generation pattern in this scenario is inherently event-based, for 

example the detection of an intruder by a sensor node. Most of the time there is no 

communication in the network. Also the network protocol should work reactively 

instead of generating periodically control messages. This saves energy and keeps 

the network covert. 

IV. Application data 

Besides the physical layer restrictions, it is important to know which kind 

of data will be transmitted in the network in order to design an appropriate network 

protocol. Due to the low bandwidth the application data format must be as short and 

efficient as possible. For this propose we specified the so called Generic Underwater


31 

Application Language (GUWAL) [6] which is suitable for any kind of underwater 

application. It is based on the following four parcel types: 

1. Data Request 

2. Data (sensor data, status, GPS position, ...) 

3. Command and Control (sleep, move, change mode, ...) 

4. Text Message (SMS) 

In GUWAL the basic parcel size is 128 bits; but it is also possible to use 

variable length if needed. All parcels include an operational header, a checksum 

of 16 bits each, and 96 bits payload with variable fields depending on the parcel 

type, as shown in Table 1. Among other fields, the header contains a source and 

a destination address used for a cross-layer approach. An operational address 

is 6 bits long, whereby the first two bits define the type of the node. The following 

groups are defined: 

1. Gateway Node (buoy/ship with acoustic and radio link) 

2. Bottom Node (environment sensor or relay node) 

3. Mobile Node (submerged unit: diver, AUV, submarine) 

4. Surface and Air Nodes (node without acoustic link like ship, airplane or 

operation center via satellite). 

Table 1. Format of a GUWAL parcel with source, destination 

and priority field as header and a checksum 

Position Length Field 

1-2 2 Parcel Type 

3 1 End-to-End Acknowledgment 

4 1 Priority Flag 

5-10 6 Operational Source Address 

11-16 6 Operational Destination Address 

17-112 96 Payload 

113-128 16 Checksum 

The latter four bits are a node identifier to distinguish nodes of the same type, 

whereby zero is defined as broadcast to all nodes of the same type. The address 

with all bits set to one is reserved for broadcasting to all nodes in the network 

regardless of their type. As a result, there are 15 network addresses in groups 

1-3 and 14 in group 4, hence 59 in total. In order to allow more than 59 participants 

in the network, it is envisaged that multiple nodes can share the same 

address. For example, it is not mandatory from an operational point of view to 

distinguish bottom nodes in the same area, especially if they are only relay nodes. 

As a consequence, the network protocol may handle nodes with the same address 

as multicast groups. How this is achieved by GUWMANET is explained in detail 

in the following chapter.


V. Protocol design 

In this chapter we introduce our new network protocol GUWMANET. In the first 

part, we introduce the Medium Access Control (MAC) and the Automatic Repeat re- 

Quest (ARQ) mechanisms which are used in GUWMANET. Furthermore, the details 

of the addressing problem resulting from shared multicast addresses are discussed. 

The introduction of a local nickname additional to the GUWAL network address 

is motivated. At the end we describe the initialization steps and the routing algorithm, 

followed by an extension for improvements based on network-coding strategies. 

A. Medium Access Control 

The MAC layer manages the access to the acoustic communication channel. 

This can be done in a contention-free manner like code, space, frequency or time 

division multiple access or randomized, accepting possible collisions in the shared 

medium water. We decided to apply a time-based random access method, which 

is much more efficient than fixed divisions in networks with event based and bursty 

traffic and a physical layer using short impulses, as in underwater scenarios. A detailed 

survey of MAC mechanisms with regard to underwater acoustic networks 

can be found in [7]. 

In random access methods every node can access the sound channel whenever 

it has data to send while regarding specific rules. The methods can be subdivided 

into two types, with previous channel reservation or direct data transmission. 

In our case channel reservation is inefficient, because the data packets consisting 

of 128 bits are already very short and a reservation would take a long time due to 

Round Trip Times (RTT) of up to 15 s having internode distances of 10 km. Moreover, 

the transmission times are with 0.3 s short compared to the long propagation delay. 

Most of the time, nodes wait for incoming data packets instead of transmitting. 

This is an important distinction to terrestrial networks. 

Another consideration is to apply carrier-sensing before transmitting data. 

However, sensing the medium prior to a transmission does not allow drawing 

conclusions about the channel state at the receiver side several seconds later when 

the signal arrives. Hence, we use a simple ALOHA [8] method without carriersensing 

in GUWMANET. Nevertheless, we point out that underwater acoustic 

communication is half-duplex. It is not possible to transmit data during a reception 

or vice versa. 

B. Automatic Repeat reQuest 

Using a random access method like ALOHA means that packet collisions may 

occur, even if the transmissions itself are very short. Additionally, bit errors due 

to high noise or low signal strength can result in packet losses. Therefore, ARQ


33 

methods have to be used to guarantee a successful reception. Typically, the receiver 

node sends an acknowledgment packet back to the transmitter to indicate a correct 

packet reception. If an acknowledgment stays out, the source node will automatically 

repeat the packet after a predefined period of time. 

The application language GUWAL already supports end-to-end acknowledgments, 

which are sent by the final destination node as operational notice of receipt. 

This acknowledgment is optional and is requested by setting the acknowledgment 

bit in the header. 

In multi-hop networks it is reasonable to add additional link layer packet 

repetition mechanisms, which will detect packet losses at each hop. In wired networks 

explicit packets have to be sent to inform the transmitter about a successful 

reception. In MANETs this can be done in an implicit way without additional 

transmissions due to the shared medium in wireless multi-hop networks. The source 

node is able to overhear if the next hop forwards the packet and therefore gets a so 

called implicit acknowledgment. If the packet will not be forwarded, the source 

node will automatically repeat the packet until it overhears a packet forwarding attempt. 

In consequence, the destination node must also repeat the packet to inform 

the previous hop about the successful reception. 

In GUWMANET this implicit acknowledgments are used with exponential 

backoff timers for packet error control as described later in Section G. Additionally, 

our network protocol is delay tolerant and supports so called Emission Control 

(EMCON), which means that nodes can decide to stay silent for an arbitrary time 

period. This is for example important for submarines, which do not want to get 

detected due to transmissions. Therefore, it is possible that acknowledgments stay 

out, even the transmission was successful. A packet is repeated with exponential 

backoff to attempt to deliver the packet successfully, but it cannot be guaranteed 

that the packet was received correctly. It is also not possible to make assumptions 

like a node is broken even if it did not reply for a longer period. Nevertheless, for 

this period where a node stays in EMCON state it is not available for packet forwarding 

and related routes have to be updated. 

C. Addressing 

As mentioned earlier, each network address can be used as multicast group 

including multiple nodes of the same type. From operational view, it is not mandatory 

to distinguish nodes inside this group, but for the network protocol it is 

mandatory to facilitate forwarding mechanisms. Therefore, we introduce a second 

local address of 5 bits in addition to the global 6 bit operational network address, 

which is independent from the network address and the node type. In the following 

this local address is called nickname. 

To allow full ad-hoc capability, the local nicknames are not predefined and 

have to be chosen automatically after network deployment. In our MANET exists


no master node coordinating the address allocation, therefore each node has to 

choose its nickname by itself. These should be unique in the local 2-hop neighborhood 

to allow all nodes to distinguish between its neighbors. Typically, this 

is done making a neighborhood discovery, where the new node requests lists with 

neighborhood information from all its neighbors. We propose another algorithm to 

reduce the number of transmissions and thereby the energy consumption, because 

battery power is a limited resource in underwater scenarios. 

The idea is to overhear at first the network traffic for a listen period T L . If during 

this period communication takes place in the neighborhood, the node becomes 

acquainted with its neighbors and learns their nicknames passively. Additionally, 

it gets information about its 2-hop neighborhood, because the network header 

of GUWMANET includes beside the transmitter’s nickname also the nickname 

of the last hop, as described later. After T L the node knows a subset of its 2-hop 

neighborhood and chooses randomly a presumably free local nickname. Now 

it transmits to its local neighborhood a nickname notification (NN) parcel to indicate 

its presence and nickname choice. The chosen nickname is included in the network 

header as source address. All other information is included inside a special GUWAL 

parcel. GUWAL has reserved a separate data type for such network control parcels 

where each network protocol can define up to eight individual control parcels. 

The network control parcel NN of GUWMANET includes the predefined 

multicast GUWAL address of the node, its unique 16 bit MAC address and a list 

with up to 10 local nicknames of its by now known 1-hop neighbors, as shown 

in Table 2. Typically, underwater networks are sparse; anyway, multiple NN packets 

are sent if there are more than 10 neighbors. The NN parcels are also used for 

network initialization described in the next section. 

Table 2. Format of a Nickname Notification (NN) parcel included in a GUWAL Data frame 

Position Length Field 

1-2 2 Parcel Type (Data) 

3 1 End-to-End Acknowledgment (no) 

4 1 Priority Flag (low) 

5-10 6 Source Address (own address) 

11-16 6 Destination Address (broadcast = 111111 2 ) 

17-20 4 Data Type (Network Control) 

21-23 3 Network Control Type (NN) 

24-42 19 Timestamp 

43-92 50 Nicknames of 10x 1-hop Neighbors 

93-108 16 MAC (own address) 

109-111 3 unused 

113-128 16 Checksum


35 

If any of the neighbors receives a NN and has an objection, because it already 

has a neighbor with this nickname, it replies with a nickname collision notification 

(NCN) parcel which has the same fields as the NN parcel. But instead of using 

the own 16 bit MAC address the one of the discovering node is used which 

was included in the corresponding NN parcel. This allows the discovering node to 

extend its 2-hop neighborhood list and to choose another free nickname. Before 

transmitting a new updated NN parcel the node waits a period T C to collect further 

NCN parcels if such were send. In the case a node receives an NN parcel in which 

it is not included in the 1-hop neighborhood list, it automatically sends its own 

NN parcel to inform the other node about its presence, if it has not already sent 

an NCN parcel for that nickname notification. 

NCN parcels are repeated with a binary exponential backoff algorithm until 

a new NN parcel is received or a limit of L rep is reached. An exponential backoff 

is used to allow fast repetitions to correct simple packet errors at the beginning 

and late repetitions to handle (temporally) asymmetric or broken links. NN parcels 

are not repeated automatically, only after the incoming of NN or NCN parcels 

of neighbor nodes. 

Although the probability is low, it is possible that nickname collisions stay 

undetected. This may occur due to packet losses, asymmetric or temporally broken 

links during the nickname allocation, or mobile nodes moving to other areas. 

Therefore, the network protocol is designed to tolerate double nickname occupancies. 

This only leads to redundancies during packet forwarding as will be shown 

later. If a nickname collision is detected later, the detecting node will try to fix this 

by sending an NCN parcel, followed by the same procedure as described above. 

D. Initialization 

In our network protocol each node among the nickname needs some initial 

network control information. This information is exchanged automatically with 

an initialization parcel after a node deployment to allow full ad-hoc capability. 

This parcel includes for example a 32 bit UNIX timestamp. The timestamp is used 

as reference time for a shortened 19 bit timestamp which is used in GUWAL, see 

Table 2. It allows date stamping inside a three month operating period with an accuracy 

of 15 s. 

Due to the limited battery power resulting in an operation period of three 

month, inaccurate clocks with a clock drift of one second per week and a lack of synchronization 

it is not necessary to have a longer timestamp with higher accuracy. 

Before a node can use this shortened GUWAL timestamp it must know the starting 

point regarding to the UNIX timestamp where the GUWAL timestamp is zero. 

The initialization parcel is send out as reply to parcels with a timestamp set 

to zero, here the NN parcel. After the node received the initialization parcel it is 

ready for use.


E. Routing 

In this section we describe the routing strategies of GUWMANET in detail. 

As mentioned before, multi-hop strategies and multicast addresses are used to 

fulfill the operational needs of an underwater network. Besides this, an important 

design aspect was to have as less overhead as possible due to fixed burst length 

of the underlying impulse communication. 

The basic idea behind GUWMANET is to leave the decision whether to forward 

a parcel to the nodes themselves, comparable to the behavior when gossiping. 

A node hears a parcel/gossip and decides if there are nodes in the neighborhood 

which may be interested in this information. If any node repeats everything we have 

simple flooding, but this is very inefficient and wastes a lot of energy. Therefore, 

the nodes in our network learn passively if they are on the direct route to the destination. 

As already mentioned, we need only 10 bits additional network protocol 

overhead for this purpose. The 5 bit local nickname of the transmitter and during 

route establishments the nickname of the last or next hop. 

In the beginning, the nodes have only information about the topology in the local 

two-hop neighborhood. To get global topology information, each node stores 

all incoming packets: 

P D × H × T × S 

whereby D represents the 128 bits GUWAL data packets, H the network headers, 

T the local receive times and S the estimated Signal-to-Noise Ratio (SNR). 

After a listen period T L plus an additional random backoff time T Backoff each 

node forwards the data packet d, even if it is addressed itself as destination, because 

there might exist more nodes in the network with the same operational network 

address (multicast). The forwarding node puts its own nickname into the source 

address field of the network header. For the last hop field all received parcels with 

data content d in M are considered: 

P d : ={(d', h', t', s') P | d' = d} 

In this subset all parcels with a SNR lower than a threshold SNR min are filtered 

out: 

P : {( d ', h', t ', s') P | s' SNR 

} 

d 

As last hop field the own nickname is chosen if the node is the first transmitter 

or the nickname of the transmitter of the first received parcel m in P * d or, 

if P * d is empty, the parcel with the highest SNR is chosen: 

 

* * * 

p' Pd : p" Pd : t' t", if Pd 

 

P 

 

p' P : p" P : s' s", else 

d 

d 

If a node overhears that it was elected as last hop inside a forwarded message 

of one of its neighbors, it generates a temporary routing entry with the two 6 bits 

d 

min


37 

source and destination addresses included in the data part and the physical addresses 

of the last and next hop. 

All addressed destination nodes receiving the packet have to reply with 

an acknowledgment parcel. With the help of the temporary routing entries 

the acknowledgment packet can be routed back to the source node. Each node 

that was elected as last hop knows that it is on the direct way back to the source 

node and responsible for data forwarding. All nodes being involved in the back 

routing process convert the temporary routing entries into permanent ones and 

will now forward all data packets with the same GUWAL source and destination 

address tuple. We emphasize that after this process only the routes from 

the source node to the destination nodes are learned but not vice versa, because 

there might be more nodes with the same address like the source node which 

are not considered yet. 

After a route is established, all following data transmissions are sent with the last 

hop field set to zero. This indicates that there is a known route and all other nodes 

without permanent routing entries should not flood this message again. If the route 

gets broken due to node or link failures, the last hop field is reused again as before, 

which initiates a new route discovery with flooding. 

Mobile nodes like submarines and AUVs possess a special role during the route 

discovery process. Due to their mobility, they are bad candidates for static routes. 

Therefore, they are only elected as last hops if there exist no alternatives after 

an additional waiting period of T mob . Also, the established routes to or from mobile 

nodes have a limited lifetime. If a node has not forwarded a message during 

a predefined time period T life the routing entries are discarded and a new routing 

process is started, because it is unlikely that this route still works. Mobile nodes 

can be easily identified by their operational address; they have the group number 3 

as defined in Chapter IV. 

We explain in the next section with a simple example how the above described 

route establishment works. 

F. Route establishment example 

Figure 3 shows an example topology whereby each dot represents a node 

and each line a connection between two nodes. The number above each node 

represents its local nickname, whereby the apostrophes are only for distinction 

in this description. In reality the nicknames 1 and 1’ are equal. The local nicknames 

should only be unique in the two-hop neighborhood if possible. In the following, 

we explain how the route establishment works; if node 1 with the GUWAL address 

A wants to transmit data to node 4’ with the GUWAL address B.


Figure 3. Example network topology with local nicknames 

First of all, node 1 broadcasts its data packet. Thereby, the source nickname 

and the last hop field inside the network header are set to 1: 

1 → [1,1] + [A → B: DATA] 

In the beginning, no nodes have routing information. Therefore, all neighbors 

of node 1 which receive this message forward the data. The parcel itself stays steady, 

whereby the source nickname of the network header is replaced by the nickname 

of the forwarding node. As last hop the nickname of node 1 is chosen, as shown 

in Figure 4. 

2 → [2,1] + [A → B: DATA] 

In this way, the message is flooded through the network. 

Figure 4. Node 2 rebroadcasts the data of node 1. Node 1 overhears that it was chosen as last hop 

and creates a temporally routing entry 

Figure 5. Temporally routes (see Table 3) used to send back the acknowledgment. 

On the way back, the routing entries get persistent 

During this flooding process each node overhearing a message with a last hop 

nickname equal to its own creates a routing entry in a temporally routing table, 

as shown in Table 3. 

Also the addressed node 4’ with the GUWAL address rebroadcasts the data parcel, 

because there might be more nodes with the GUWAL address B in the network: 

4’ → [4,3] + [A → B: DATA]


39 

After this, node 4’ sends out an acknowledgment parcel which is forwarded 

back to node 1 using the temporally routing tables. During the complete back 

forwarding the last hop field is set to zero: 

4’ → [4,0] + [B → A: ACK] 

Table 3. Temporally routing entries of all nodes 

Node From To Next Node From To Next 

1 A B 2 7 A B 8 

1 A B 3 8 A B 6’ 

1 A B 4 3’ A B 4’ 

1 A B 5 5’ A B 7’ 

1 A B 6 6’ A B 2’ 

1 A B 7 7’ A B 1’ 

2 A B 3’ 4’ A B 8’ 

3 A B 5’ 

Due to the empty last hop field, the message is not flooded back. Instead, 

only node 3’ forwards the acknowledgment back, because it has a temporally 

routing entry after overhearing his election as last hop from node 4’ in the previous 

transmission. This is repeated until the acknowledgement reaches node 

1 and a complete persistent route is established. Even, if there is more than one 

destination node, all routes are learned simultaneously. Additionally, we point out 

that it is necessary to store the data parcel itself in each temporally routing entry, 

to distinguish parallel route discoveries. The acknowledgment parcel includes 

the 16 bit checksum of the data parcel which allows an association of the ACK 

with the temporary routing entry. 

As mentioned before, this route is valid for transmissions from A to B only 

and not vice versa, because the GUWAL address of A is not necessarily used 

as unicast address. 

G. Packet loss and broken link handling 

In general, the probability of collisions in the underwater channel is very low; 

even though flooding creates a lot of redundancy during the route establishment 

phase. That is because the transmission time is very low compared to the propagation 

time; in consequence the channel is idle most of the time and not occupied 

with an ongoing transmission. In our underlying impulse communication we have 

a transmission time of 0.3 s for a GUWAL parcel whereas the propagation time 

can be up to 6−7 s. This is a fundamental difference to terrestrial networks where 

the data rate is much higher and the signals propagate with light speed.


On the other side, the bit error rate of an underwater channel can be very 

high, depending on weather conditions and noise. Therefore, packet losses must be 

handled. As mentioned in the previous chapter we use implicit acknowledgments 

to indicate successful transmissions from hop to hop. If a packet forwarding from 

a neighbor stays out, the message will be repeated. In GUWMANET a message 

is retransmitted up to 5 times with an exponential backoff. After this, the link 

is declared as broken and the neighbor is removed from the neighborhood list. 

If a link is broken during a forwarding process on an established route, the route 

has to be updated. As already mentioned, this is done by using the last hop field. 

The neighbor nodes will flood the message again, if the last hop field is nonzero. 

This results in a new route discovery process to renew the broken route. 

Nevertheless, these single routes without redundancy are very vulnerable to 

temporally link failures or asymmetric links. To overcome this problem, we introduce 

in the next section an enhancement of GUWMANET, which adds additional 

redundancy along a route if necessary. 

H. Corridor as route enhancement 

In this section we introduce a modification of GUWMANET to enhance 

the packet delivery ratio during bad weather conditions. Rain and breaking waves 

on the surface increase the channel noise significantly which leads to a much 

higher bit error rate. To overcome this problem, additional redundancy is generated 

by involving the neighbors along the direct route. We call this enlarged route 

corridor, which is illustrated in Figure 6. 

Figure 6. Corridor as route enhancement for a higher packet delivery ratio 

For this objective, the route discovery process is modified in the following 

way. If a data parcel reaches one of its addressed destination nodes it will send 

out an acknowledgement as usual. But this time, the last hop is not left empty 

anymore. Instead, it is used in acknowledgment parcels as next hop field, wherein 

the transmitter puts its elected previous hop. This node is intended to forward 

the acknowledgment back to the source node. Now, not only this elected next 

hop will forward the packet, but also all nodes having this elected node as direct 

neighbor. This is like gossiping in the real world; gossip is circulated if you know 

your neighbor is interested in it too.


41 

Figure 7 shows an example. Node 4’ has elected node 3’ as next hop, whereby 

node 6’ and node 7’ knowing node 3’ and 4’ participate as corridor nodes in the forwarding 

process. Multiple receptions at node 3’ allow utilization of network-coding 

strategies, as described later in Chapter VII. This technique allows restoring messages, 

even if all single transmissions are error-prone. 

Figure 7. Adding next hop during back forwarding of an acknowledgment 

To avoid unnecessary transmissions, the additional corridor nodes only forward 

the data if the node on the direct route did not. For this purpose, we introduce 

an additional backoff T corridor to the normal backoff T backoff combined with the link 

quality L quality which is between 0 and 1: 

T forward = T backoff + T corridor . (2 − L quality ) 

The timer is canceled if the node overhears a transmission of the node on 

the direct route or reset if the message was repeated first by another neighbor. 

During the back forwarding of the acknowledgment the nodes on direct node 

make their persistent routing entries as before. But this time, also the nodes on 

the corridor make persistent routing entries, which indicate that they are jointly 

responsible to forward data parcels. Note, during data forwarding the next/last 

hop is still not used and left empty as before. The packet forwarding will be only 

decided with the usage of the persistent routing entries. 

VI. Evaluation 

Sea trials are costly due to the need of expensive equipment and personal. Therefore, 

we developed an underwater acoustic emulation system to test GUWMANET 

in an environment as close as possible to real hardware. This emulation test bed 

uses a real acoustic channel with the same physical layer methods as underwater. 

The only difference is, that the communication takes place in air instead of water. 

Insulation material is used to model the absorption losses of sound waves, which 

allows the modulation of a scenario of 8 km × 30 km. Only the propagation delay 

is artificially created by delaying incoming transmissions. 

A cluster of 10 computers equipped with microphones and loudspeakers are 

used to emulate the network nodes. They are arranged in a topology similar to 

the RACUN scenario described in Chapter III with 4 barriers consisting of 9 bottom 

nodes and an additional mobile node.


We implemented GUWMANET inside this emulations system and made 

a concept study of the routing mechanisms and first performance analyses. The route 

establishment algorithms were tested with 3 hops and multicast addresses. These 

emulations are the first step of the evaluation and will be enlarged to additional 

simulations and sea trials. 

VII. Network-coding 

In underwater acoustic networks high bit error rates are common. Therefore, 

error detection and correction techniques have to be used to enable reliable communications. 

Common ways are the use in the form of so called checksums in combination 

with ARQ and Forward Error Correction (FEC) mechanisms in the physical 

layer methods. If a packet error is detected, the simplest way is to drop the packet 

and wait/ask for a retransmission. But retransmissions waste valuable energy, occupy 

the channel and increase the packet delay. 

A much more efficient way is to use a posteriori error correction techniques 

like clustering of all incoming short 128 bit parcels. The idea behind this is to merge 

multiple corrupted parcels and merge them into correct one. In real life gossiping 

messages contains lies; the gossip average results in true facts. The network-coding 

approach with clustering techniques was proposed in the RACUN project [9] and 

is an essential part of gossiping. 

VIII. Conclusion 

In this paper we introduced a new network called Gossiping in Underwater 

Mobile Ad-Hoc Networks (GUWMANET). This protocol is designed from scratch 

fitting to the special needs of underwater communication. It is designed to fulfill 

the requirements of the scenarios defined in the RACUN project; an European project 

for robust underwater communication, supporting stationary as well as bottom nodes. 

GUWMANET is based on impulse communication as physical layer method, 

which sends out short data bursts with a fixed length of 128 bit. This makes it necessary, 

that our protocol gets along with only 10 bit additional protocol overhead. 

These 10 bits are used for two 5 bit local nicknames, identifying the transmitter 

node and the last hop. With the help of these two fields a persistent route can be 

found, even if there are multiple destinations. We also defined the Generic Underwater 

Application Language (GUWAL) to overcome the restrictions of 128 bits for 

application data. 

At least, we introduced a protocol enhancement using corridors for packet 

forwarding. These corridors generate additional redundancy if necessary and enable 

network-coding strategies to restore error-prone messages. This is done by using 

clustering algorithms, which can be used to the low message size of 128 bit and 

the low number of packet transmissions in underwater networks.


43 

IX. Future work 

As mentioned in the evaluation chapter, we already made first concept studies 

and analyses of GUWMANET. The next steps are simulations with the simulation 

framework DESERT Underwater, an NS-Miracle-based [11], [13] framework to 

DEsign, Simulate, Emulate and Realize Test-beds for underwater network protocols [10] 

developed by the University of Padova. This framework allows us to enlarge our 

studies to higher node numbers as were foreseen in the RACUN scenario. Beside 

this, GUWMANET can be compared with other network protocols which are 

already implemented in DESERT underwater. 

After detailed simulation and emulation studies sea trials are planned to 

demonstrate the capability of our network protocol to fulfill the requirements 

of the RACUN scenario. 

X. Acknowledgment 

We gratefully acknowledge the partners of the project Robust Acoustic Communications 

in Underwater Networks (RACUN) for their helpful discussions and 

advices. The RACUN project is part of the EDA UMS program (European Unmanned 

Maritime Systems for MCM and other naval applications), and is funded 

by the Ministries of Defence of the five participating nations Germany, Italy, Netherlands, 

Norway, Sweden. Partners of this project are: Atlas Elektronik (Germany), 

WTD71-FWG (Germany), L-3 Communications ELAC Nautik (Germany), TNO 

(The Netherlands), Kongsberg Maritime (Norway), FFI (Norway), FOI (Sweden), 

SAAB (Sweden), WASS (Italy) and CETENA (Italy). 

References 

[1] I.F. Akyildiz, D. Pompili, T. Melodia, “Underwater Acoustic Sensor Networks: 

Research Challenges”, Ad Hoc Networks, vol. 3, no. 3, pp. 257-279, May 2005. 

[2] European Defense Agency (EDA) Project Arrangement No. B0386, http://www. 

racun-project.eu 

[3] J. Kalwa, “The RACUN-project: Robust acoustic communications in underwater 

networks – An overview”, OCEANS, 2011 IEEE – Spain, pp. 1-6, 6-9 June 2011, DOI: 

10.1109/Oceans-Spain.2011.6003495. 

[4] I. Nissen, “Alternativ Ansatz zur verratsarmen Unterwasser-kommunikaton durch 

Verwendung eines Transienten im Kontext von IFS und JUWEL”, FWG Research Report 

IB 2009-3; Research Department for Underwater Acoustics and Marine Geophysics 

(FWG) – WTD 71, Kiel, 2009. 

[5] M. Goetz, S. Azad, P. Casari, I. Nissen, M. Zorzi, “Jamming-Resistant Multipath 

Routing for Reliable Intruder Detection in Underwater Networks”, Proceedings


of the Sixth ACM Inernational Workshop on Underwater Networks, WUWNet’11, 

Seattle, Washington, USA, Dec. 1-2, 2011. DOI: 10.1145/2076569.2076579. 

[6] I. Nissen, M. Goetz, “Generic Underwater Application Language (GUWAL) – 

A Specification Approach”, Research Report WTD71 – 0161/2010 FB, Kiel, Germany, 

20.12.2010. 

[7] R. Otnes, A. Asterjadhi, P. Casari, M. Goetz, T. Husøy, I. Nissen, K. Rimstad, 

P. van Walree, M. Zorzi, “Underwater Acoustic Networking Techniques”, SpringerBriefs 

in Electrical and Computer Engineering, DOI: 10.1007/978-3-642-25224-2_1, 2012. 

[8] N. Abramson, “Development of the ALOHANET”. IEEE Transactions on Information 

Theory, 31(2):119-123, 1985. 

[9] I. Nissen, M. Goetz, T. Wiegmann, T. Schäl, “Clustering of tactical underwater 

messages”, Research Report WTD71-0065/2012, Kiel 2012. In Preparation. 

[10] R. Masiero, S. Azad, F. Favaro, M. Petrani, G. Toso, F. Guerr, P. Casari, M. Zorzi, 

“DESERT Underwater: an NS-Miracle-based framework to DEsign, Simulate, Emulate 

and Realize Test-beds for Underwater network protocols”, Oceans 2012, Yeosou, Republic 

of Korea, May 2012. 

[11] The Network Simulator – ns-2, Last time accessed: March 2012. Available: 

http://nsnam.isi.edu/nsnam/index.php/User_Information 

[12] The DESERT Underwater libraries – DESERT, Last time accessed: March 2012. 

Available: http://nautilus.dei.unipd.it/desert-underwater 

[13] The Network Simulator – NS-Miracle, Last time accessed: March 2012. Available: 

http://dgt.dei.unipd.it/download

Network Routing by Artificial Neural Network 

Michal Turčaník 

Department of Informatics, Armed Forces Academy, Liptovský Mikuláš, Slovakia, 

michal.turcanik@aos.sk 

Abstract: Author presents a design of the artificial neural network for routing in the sensor network 

in this paper. The routing table is replaced by artificial neural network. The main aim is to realize this 

operation as fast as possible. Optimized neural network is implemented in the reconfigurable hardware. 

The paper concludes with possible future research areas. 

Keywords: artificial neural network, FPGA, network routing 


Sensor networks consist of small nodes with sensing, computation, and wireless 

communications capabilities. Many routing, power management, and data 

dissemination protocols have been specially designed for sensor network where 

saving energy is an essential design issue. The focus, however, has been given to 

the routing protocols which might differ depending on the application and network 

architecture. 

A new approach has been proposed for the routing problem in the sensor 

network in this paper. The routing table is replaced by artificial neural network 

in this approach. The main aim is to realize this operation as fast as possible. 

In this paper author used offline learning method. Offline training regards to 

learning procedure on a general-purpose computing platform before the trained 

system is implemented in hardware. The software that was chosen for offline training 

is MATLAB. Also was used Xilinx ISE to synthesize of VHDL code and simulate. 

This paper is organized as follows: part II introduces the sensor network topology. 

Part III presents artificial neural network (ANN). Using ANN for network 

routing and ANN optimization is presented in part IV. Part V presents the results 

and discussion about results, and finally part VI concludes the paper. 

II. Sensor network topology 

Networking unattended sensor nodes may have profound effect on the efficiency 

of many military and civil applications such as target field imaging, intru-


sion detection, weather monitoring, security and tactical surveillance, distributed 

computing, detecting ambient conditions such as temperature, movement, sound, 

light, or the presence of certain objects, inventory control, and disaster management. 

Deployment of a sensor network in these applications can be in random fashion 

(e.g., dropped from an airplane) or can be planted manually (e.g., fire alarm sensors 

in a facility). For example, in a disaster management application, a large number 

of sensors can be dropped from a helicopter. Networking these sensors can assist 

rescue operations by locating survivors, identifying risky areas, and making the rescue 

team more aware of the overall situation in the disaster area [1]. 

Figure 1. Sensor network (R – router, S – sensor) 

In the past few years, an intensive research that addresses the potential of collaboration 

among sensors in data gathering and processing and in the coordination 

and management of the sensing activity were conducted. However, sensor nodes 

are constrained in energy supply and bandwidth. Thus, innovative techniques that 

eliminate energy inefficiencies that would shorten the lifetime of the network are 

highly required. Such constraints combined with a typical deployment of large 

number of sensor nodes pose many challenges to the design and management sensor 

networks and necessitate energy-awareness at all layers of the networking protocol 

stack. For example, at the network layer, it is highly desirable to find methods for 

energy-efficient route discovery and relaying of data from the sensor nodes so that 

the lifetime of the network is maximized [2]. 

III. Artificial neural network 

Artificial neural networks have been trained to perform complex functions 

in various fields, including pattern recognition, identification, classification, speach,


47 

vision, and control systems [9, 10, 11]. Neural networks are composed of simple 

elements operating in parallel. These elements are inspired by biological nervous 

systems. As in the nature, the connections between elements largely determine 

the network function. A neural network can be trained to perform a particular 

function by adjusting the values of the connections (weights) between elements. 

Typically, neural networks are adjusted, or trained, so that a particular input leads 

to a specific target output. There, the network is adjusted, based on a comparison 

of the output and the target, until the network output matches the target. Typically, 

many such input/target pairs are needed to train a network [3]. 

Many variations of the perceptron were created by Rosenblatt [4]. One of the simplest 

was a single-layer network whose weights and biases could be trained to produce 

a correct target vector when presented with the corresponding input vector. 

The training technique used is called the perceptron learning rule. The perceptron 

generated great interest due to its ability to generalize from its training vectors and 

learn from initially randomly distributed connections. Perceptrons are especially suited 

for simple problems in pattern classification. They are fast and reliable networks for 

the problems they can solve. Feedforward networks often have one or more hidden 

layers of sigmoid neurons followed by an output layer of linear neurons. Multiple layers 

of neurons with nonlinear transfer functions allow the network to learn nonlinear 

relationships between input and output vectors. The linear output layer is most often 

used for function fitting (or nonlinear regression) problems [7]. 

A multi-layer perceptron (Fig. 2) consists of neurons and synapsies (connections). 

Each neuron has an input, activation and output function. Each synapsy between 

two neurons has a weight. The units are organized in layers. Three different types 

of neurons are distinguished: input neurons, hidden neurons and output neurons. 

The input units receive the input data, and the output units provide the output [8]. 

Figure 2. Multilayer perceptron


The calculation of the final output values proceeds layer by layer. First, the input 

signals are applied to the input layer, and each neuron of the input layer calculates 

its output value. Next, these values are propagated to the next layer; until the output 

layer is reached. 

IV. Using the artificial neural network for network routing 

A. Clasical table look-up aproach in the sensor network 

The primary role of routers is to forward data towards their final destination. 

A router must decide for each incoming packet where to send it next. More exactly, 

the forwarding decision consists in finding the address of the next-hop router as well 

as the egress interface through which the packet should be sent. This forwarding 

information is stored in a forwarding table that the router computes based on the information 

gathered by routing protocols. To consult the forwarding table, the router 

uses the packet’s destination address as a key; this operation is called address lookup. 

Specifically, the address lookup operation is a major bottleneck in the forwarding 

performance of today’s routers. This paper presents a proposal a new method 

to realize algorithm for efficient address lookup. 

B. Artificial neural network topology 

Neural networks map between a data set of numeric inputs and a set of numeric 

targets. To solve a problem of routing in the mash topology of the sensor 

network (Fig. 1) was used artificial neural network. A three layer feed-forward 

network with sigmoid hidden neurons and linear output neurons are used to solve 

multi-dimensional problems. 

Figure 3. The structure of the multilayer perceptron for network routing


49 

The structure of the multilayer perceptron is shown in Fig. 3. The input layer 

of the neural network receives input data for routing in the sensor network. Input 

data represent address (node ID) of the data packet that should be transferred 

through sensor network (Fig. 1). The interface status represents information 

about status of all interfaces for single router. All active interfaces can be used 

to route data. Data communications using inactive interfaces are rerouted to active 

interfaces. Some sensor nodes may fail or be blocked due to lack of power, 

physical damage, or environmental interference. The failure of sensor nodes 

should not affect the overall task of the sensor network. If many nodes fail, routing 

protocols must accommodate formation of new links and routes to the data 

collection base stations. 

The number of the input layer neurons corresponds to the input information 

for multilayer perceptron. The number of the neurons of the hidden layer is a target 

of the optimization. The number of the neurons of the output layer corresponds to 

the number of the interfaces of the router of the sensor network. 

The value of the output layer´s neurons represents index of the interface that 

will be used to send data packet to the destination. If the value of the output neuron 

is equal to 1, interface will be used. Otherwise, this interface will not be used. All 

layers are fully interconnected and there are not feedback connections between 

the hidden and input layers. 

C. ANN optimization 

To optimize neural network structure must be done analysis of the number 

of neurons in hidden layer from point of view of correct classification [5, 6]. 

To solve this problem three training sets and several models of neural network 

were created. 

Number of neurons for input and output layer is equal to each other for all 

models of neural network. Number of neurons in hidden layer is variable and 

it changes in interval from 10 to 100 neurons for training, validation and testing 

sets. 

Training, validation and testing sets were created on the base of topology 

of the sensor network shown in Fig. 1. The number of samples for neural network 

training depends on router, for which is ANN created. Each sample is represented 

by one rule of the routing table. There are two main routing rule groups. One group 

of rules represents situation when all interfaces are in active state. The second group 

of rules are applied when some interface is inactive and another interface must be 

used to forward data packet. 

Training set was used during network training to adjust error and it consists 

of 70% of samples. Validation set was used to measure network generalization 

and it consists of 15% of samples. Training is halted when generalization stops 

improving. Testing set have no effect on training phase and provide independent


measure of network performance during and after training. Testing set consists 

of 15% of samples. 

The results of Matlab simulation are shown in following tables and graphs. 

Number 

of hidden neurons 

Table I. Mean squared error for analyzed ann 

Mean Squared Error 

Training Validation Testing 

10 8.487e-3 2.453e-2 1.526e-1 

15 6.954e-2 7.843e-2 9.085e-2 

20 7.743e-17 1.081e-16 1.099e-16 

Mean squared error (MSE) is the average squared difference between outputs 

and targets. Lower values are better. Zero means no error. The lowest value of MSE 

is for ANN with 20 neurons in the hidden layer. 

Number 

of hidden neurons 

Table II. Regression for analyzed ann 

Regression 

Training Validation Testing 

10 9.690e-1 9.074e-2 4.225e-1 

15 7.458e-1 6.926e-1 6.661e-1 

20 9.999e-1 9.999e-1 9.999e-1 

Regression R values measure the correlation between outputs and target. 

An R value of 1 means a close relationship, 0 a random relationship. R values are 

again the best for ANN with 20 neurons in the hidden layer. 

ANNs with higher number of neurons in the hidden are not shown in the tables, 

because the value of MSE and R value are worse than ANN with 20 neurons. 

The following regression plots display the network outputs with respect to targets 

for training, validation, and test sets (Fig. 1, Fig. 2 and Fig. 3). For a perfect fit, 

the data should fall along a 45 degree line, where the network outputs are equal 

to the targets. For this problem, the fit is reasonably good for all data sets, with 

R values in each case of 0.93 or above. The ANN with 20 neurons in the hidden 

layer has the best results.


51 

Figure 4. Regression R for ANN with 10 neurons in the hidden layer 

Figure 5. Regression R for ANN with 15 neurons in the hidden layer


Figure 6. Regression R for ANN with 20 neurons in the hidden layer 

V. Simulation and synthesis results 

VHDL (VHSIC hardware description language) is a hardware description 

language used in electronic design automation to describe digital and 

mixed-signal systems such as field-programmable gate arrays and integrated 

circuits [7]. ANN was implemented by VHDL language and synthesized using 

Xilinx ISE 13.3. 

Weights and biases that are obtained from Matlab simulation after training 

phase are floating point. To be synthesis possible, weights and biases of the artificial 

neural network are expressed using fixed-point numbers (range scale 

was multiplied by 1000) in the FPGA-based implementation. There is no difference 

in results between FPGA-based implementation and simulation. 

The Virtex®-5 LXT ML505 is a general purpose FPGA and RocketIO 

GTP development board. Provides feature rich general purpose evaluation 

and development platform. Includes on-board memory and industry standard 

connectivity interfaces. It delivers a versatile development platform for embedded 

applications. Virtex®-5 LXT ML505 was used as a platform for realization 

of the analyzed ANN.


53 

ANN 

Table III. Analysis of ann fpga realisation 

FPGA 

Delay Number of LUT Slices 

10 142 ns 1098 524 

15 149 ns 1527 714 

20 152 ns 1822 865 

The results of the practical realisation of the ANN are summarized in previous 

table. Main parameters to compare were delay, number of slices and number 

of 4-inputs LUTs. FPGA (Field Programmable Gate Array) is an integrated circuit 

containing a matrix of user-programmable logic cells, being able to implement 

complex digital circuitry. The elementary programmable logic block in Xilinx FPGAs 

is called slice. LUTs (Look-Up Tables) can implement any 4-input boolean function, 

used as combinational function generators. They are usually the number of LUTs 

and the number of slices (and not the number of registers) that are used to compare 

the size of FPGA designs. 

Delay for all ANN has almost the same value. The highest number of slices 

and 4-inputs LUTs has ANN with neurons in the hidden layer. The lowest requirement 

has ANN with 10 neurons in the hidden layer. The modern FPGA circuits 

(thanks to their size) can accommodate easily requirements of the proposed ANNs. 

VI. Conclusion 

In this paper, was presented a new method for sensor routing table realisation 

using neural networks. It was proposed artificial neural network topology 

and optimization criterions to solve a problem of routing in the mash topology 

of the sensor network. The main advantage of the using ANN for table look-up 

process is speed. The size of the look-up table in this approach could not influence 

the speed of decision where to send packet next as it is in other methods. Time to get 

decision from ANN is always the same. ANN could be trained to change the routes 

in case of topology changes that could be predicted in the training phase. Thanks 

to this ANN does not have to be retrained in the real environment after topology 

changes. Analyzed ANNs were practically realised in hardware platform. Execution 

of the training process in the hardware can improve this proposal. Future work 

can be oriented to realisation embedded system based on MicroBlaze processor 

to train ANN in FPGA architecture. This method can be executed in parallel on 

an FPGA chip. 

Another interesting issue for routing protocols is the consideration of node 

mobility. Most of the current protocols assume that the sensor nodes and the sink 

are stationary. However, there might be situations such as battle environments 

where the sink and possibly the sensors need to be mobile.


In such cases, the frequent update of the position of the command node and 

the sensor nodes and the propagation of that information through the network 

may excessively drain the energy of nodes. New routing algorithms are needed 

in order to handle the overhead of mobility and topology changes in such energy 

constrained environment. 

Other possible future research for routing protocols includes the integration 

of sensor networks with wired networks (i.e. Internet). Most of the applications 

in security and environmental monitoring require the data collected from the sensor 

nodes to be transmitted to a server so that further analysis can be done. On 

the other hand, the requests from the user should be made to the sink through 

Internet. Since the routing requirements of each environment are different, further 

research is necessary for handling these kinds of situations. 

This work has been supported by the MOD of Slovak Republic grants No. 4/2011 

“Partial dynamic reconfiguration of the digital systems in military applications”. 

References 

[1] J.N. Al-Karaki and A.E. Kamal, “Routing Techniques in Wireless Sensor Networks: 

A Survey”, IEEE Wireless Communications, vol. 11, no. 6, Dec. 2004, pp. 6-28. 

[2] O. Al-Kofahi and A.E. Kamal, “Survivability Strategies in Multihop Wireless 

Networks”, IEEE Wireless Communications, vol. 17, no. 5, Oct. 2010, pp. 71-80. 

[3] J.L. Elman,”Finding structure in time, Cognitive Science,” vol. 14, 1990, pp. 179-211. 

[4] F. Rosenblatt, “Principles of Neurodynamics,” Washington D.C., Spartan Press, 

1961. 

[5] I. Mokriš and M. Turčaník, ”Contribution to the analysis of multilayer perceptrons 

for pattern recognition,” Neural Network World. vol. 10, no. 6 (2000), ISSN 1210-0552, 

p. 969-982. 

[6] M. Turčaník, I. Mokriš, Possible Approach to Optimization of Neural Network 

Structure. Proc. of Conf. SECON 97, Warsaw, pp. 326-335. 

[7] D. Pattreson, Artificial Neural networks – Theory and Applications, Prentice Hall, 

1996. 

[8] S. Melacci, L. Sarti, M. Maggini, M. Bianchini, “A Neural Network Approach to 

Similarity Learning,” Lecture Notes in Computer Science, Artificial Neural Networks 

in Pattern Recognition, June 30, 2008, pp. 133-136. 

[9] S. Ezzati, H.R. Naji, A. Chegini, P. Habibimehr, “Intelligent Firewall on 

Reconfigurable Hardware,“ European Journal of Scientific Research. ISSN 1450- 

-216X vol. 47 no. 4 (2010), pp. 509-516. 

[10] M. Harakaľ, J. Chmúrny, The Tesseral Processor for Image Processing Based on 

Hierarchical Data Structures, Radioengineering, vol. 6, no. 4, 1997, pp. 1-5. ISSN 1210- 

-2512. 

[11] M. Kuffová, Simulation of fatigue process. In: Mechanics. – ISSN 1734-8927. – vol. 27, 

no. 3 (2008), p. 110-112.

An Application of Chord Structure 

in Tactical Ad-hoc Network 

Jerzy Dołowski, Marek Amanowicz 

Institute of Telecommunications, Faculty of Electronics, 

Military University of Technology, Warsaw, Poland, 

{jerzy.dolowski,marek.amanowicz}@wat.edu.pl 

Abstract: Mobile ad hoc network (MANET) is expected to become a prominent standard in a tactical 

environment. However, a tactical network requires a reliable mechanism of discovering resources. 

In this paper, we propose using a Peer-to-Peer structure to achieve this goal. The crucial aspects 

of the Chord structure’s implementing over MANET are pointed out. A cross-layer communication 

is proposed to improve the joining process. The results of simulations show that our solution is able to 

work in a fully autonomous way. 

Keywords: Peer-to-Peer, Chord, MANET 


Contemporary battlefield requires a deployable command and control system. 

A tactical system should cope with an increasing number of services and sensors. 

SOA (Services Oriented Architecture) is considered a method of tactical services’ 

consolidation. An implementation of the SOA system requires a distributed register 

which would not be prone to damage in a tactical environment. We noticed that 

Peer-to-Peer system could be used for this purpose. 

The Peer-to-Peer system is a set of equal entities (nodes) that communicate 

with each other to share resources. The nodes form a self-organized overlay network. 

Among many Peer-to-Peer systems we have chosen the Chord structure. Attractive 

features of Chord include its simplicity, provable correctness and provable performance. 

However, the routing table of Chord is not based on topological feature. 

Thus, we propose modification of Chord in order to improve its distance awareness. 

A mobile ad hoc network (MANET) is a multi-hop wireless network operating 

without infrastructure. All nodes in the network have to cooperate in order to 

perform routing. An application of Chord in MANET requires a robust mechanism 

of bootstrap. We propose the use of discovering features of MANET. 

The rest of the paper is organized as follows: Section 2 presents the related 

work, section 3 provides a brief overview of the Chord, whereas in section 4


we describe, in more details, the crucial aspects of the Chord on top of a MANET 

as well as the proposed solutions. We demonstrate the effectiveness of our system 

in Section 5 and the paper is concluded in Section 6. 

II. Related work 

Various Peer-to-Peer systems have been proposed for MANET. Most of them 

are unstructured (e.g. [6]). Usually unstructured systems use some forms of flooding 

to find resources. Flooding-based look-up is rather inefficient and does not scale 

well. This makes them impractical in a tactical network where bandwidth is limited. 

Nevertheless, there are approaches to overcome this limitation, for instance [7]. 

In the structured Peer-to-Peer systems, the topology is strictly controlled. They 

use a Distributed Hash Table (DHT) which allows an efficient and deterministic 

search for resources. Therefore, in our opinion only structured Peer-to-Peer systems 

should be taken into account. 

Chord overlay which matches to the underlay network is presented in [8]. 

This feature is achieved using the minimum-spanning tree in each node. Additionally, 

the identifier space is divided among the nodes in a non-contiguous way. 

Another improvement of Chord over MANET, which is proposed in [5], consists 

in forwarding lookup queries via physically adjacent nodes. In order to make it possible, 

a node has to maintain its physically adjacent neighbors. As a consequence, 

it results in generating redundant routing traffic. 

A different approach is an integration of the DHT with ad hoc network routing. 

MAD Pastry [10] is an example of that manner. In this idea, the overlay is built 

on top of the AODV protocol and both protocols cooperate strictly. 

III. The Chord structure 

In the Chord [9], each node has a unique identifier ranging from 0 to 2 m −1. 

The nodes form a one-dimensional ring according to their identifiers. Each node 

maintains a pointer to its successor and predecessor node. A successor of the node 

is the first node whose identifier is higher than its own identifier. Moreover, each 

node maintains information about (at most) m other neighbors, called Fingers. They 

are collected in a Finger Table. The i-th row of this table indicates the successor for 

interval [n+2 i , n+2 i+1 ), where n is the identifier of the current node. 

The resources are mapped to nodes based on their keys (identifiers). In Chord, 

a resource is mapped to the first node whose identifier is equal to or follows its key. 

An example of a Chord overlay network formed by nine nodes arranged on a circle 

is shown in Figure 1.We present also an example of the Finger Table of node n = 24, 

and location of two resources.


57 

Figure 1. Example of a Chord ring (m = 5) 

A message is forwarded toward a closer node in the Finger Table with the highest 

identifier value less than or equal to the identifier of the destination node. 

As a consequence, the key lookup mechanism in Chord takes O (log N) hops, where 

N is the total number of nodes [9]. 

IV. Adaptation of Chord for MANET 

A. The key issues 

Considering the introduction of Chord in the MANET, we have to take into 

account several key issues. Firstly, a mechanism of bootstrapping is requested. 

A new node which intends to join an existing Peer-to-Peer structure has to communicate 

with at least one node of this structure. Therefore, a discovery method 

is mandatory. Secondly, the state machine of the Chord node has to be adjusted 

to the ad hoc environment. And thirdly, the routing in Chord is not based on 

an underlay network topology. If the node was aware of real distances to others, 

it could choose a closer node during a routing process. We suggest implementing 

an additional mechanism which will improve nodes’ awareness of distance. 

B. A cross-layer mechanism 

The MANET manager discovers and maintains routes to other nodes. We propose 

using these routes by the overlay layer during bootstrapping. The nodes that 

could be potentially used for bootstrap are stored in a Bootstrap List.


We assume that the MANET manager adds every discovered route to the Bootstrap 

List using a cross-layer communication – Figure 2. 

When the node intends to join a structure, it selects a random node from 

the Bootstrap List. Next, the node sends a request towards it. In case of no response 

this entry will be removed from the list. If the list is empty, the node will form 

a one-member ring. 

Figure 2. The node 

The process of adding entries by the MANET manager is independent from 

the Chord procedure. 

C. A state machine 

In Chord, the node maintains an internal state machine. When the node 

has successfully joined a ring or has formed a new one-member ring, it goes 

into the READY state. In this state, the node waits for incoming requests of joining, 

but it does not send these requests itself. The abovementioned behavior 

arises from the assumption that an accessibility of the node is invariable during 

its operation. However, the use of MANET as an underlay network results 

in dynamic changes of the topology. Thus, the internal state machine should 

be adapted to the nature of the ad hoc network. The proposed procedure is depicted 

in Figure 3. 

We propose to introduce a new state called READY_ALONE. The node goes 

into this state when it fails to join any member of the Chord ring. The node permanently 

tracks changes of its Bootstrap List in this state. Besides this, the node 

is ready to receive a request of joining from another node. Furthermore, we also 

propose extending the READY state. The node staying in this state should try


59 

to join the newly discovered nodes. It is essential for detecting nodes that are 

members of another ring (partition). 

Figure 3. The modified Chord procedure 

D. An improvement of the distance awareness 

The topology of an overlay network does not reflect the underlying (physical) 

topology. This may result in deterioration of the system effectiveness. In particular, 

the latency of a request may decrease. Therefore, we propose to use Vivaldi algorithm 

that is one of the Internet Coordinate Systems. 

Vivaldi [4] is a distributed technique to estimate the latency between nodes 

in the Internet. In Vivaldi, a new node computes its coordinates after collecting 

latency information from only a few other nodes. Each time a node makes a request 

to another node, it measures the network latency to the node. A response 

includes the answering node’s current coordinates. The requesting node refines its 

coordinates based on the latency measurement and the responding node’s informa-


tion. As Vivaldi is a decentralized procedure, no fixed infrastructure needs to be 

deployed. Furthermore, the node does not require any initial data. These features 

align well with the requirements of the Peer-to-Peer system. 

Figure 4. The pseudocode to find the successor node of an identifier k 

However, in order to take advantage of that algorithm it is necessary to modify 

the Chord procedures by extending the Finger Table. 

The main idea of this proposal is to prepare a set of nodes which potentially 

could be predecessors of an identifier k. Then, the node whose distance is the smallest 

one is selected. 

In the original Chord, a row of the Finger Table contains only one successor. 

In our modification, several successors are placed in each row. Moreover, the distance 

obtained owing to Vivaldi is associated with each successor. The procedure 

of finding the successor has to be modified too. The appropriate pseudocode 

is shown in Figure 4.


61 

V. Evaluation 

We conducted experiments using a computer simulation technique in order 

to evaluate the proposed solution. We selected the OMNeT++ [11] simulation 

environment and OverSim model [2]. 

The aim of the evaluation was to determine whether it is possible to effectively 

search for an identifier in the Chord structure built on top of MANET as well 

as what is the volume of the maintenance traffic. 

A. Model 

A model of mobile ad hoc network was prepared. We choose the DYMO 

(Dynamic MANET On-demand) protocol [3]. The network consisted of a fixed 

number of nodes, ranged from 10 to 60. Each node acted as a Chord node. Moreover, 

the node was able to use the Vivaldi mechanism and extending Finger Table. 

This feature was controlled during experiments. 

We prepared three scenarios of our simulation investigations. In the first 

scenario, the nodes were placed in the simulation area and stayed in their position 

motionless during operation (a stationary network). In the second and third scenario, 

the nodes moved according to the Random Way Point mobility model [1]. 

In this model, the node chooses a random destination point, and next moves toward 

it at a constant speed. After reaching its destination, the node stays in it for 

a pause time. Then, it begins to move toward the next destination. In our experiments, 

the current value of the speed was chosen from the ranges: 1…5 m/s, and 

1…10 m/s in the second and third scenario, respectively. 

B. Tests and performance metrics 

Each node that had become a member of a Chord ring started the test application. 

This application performed the following tests: One-way lookup, KBR lookup, 

and DHT lookup. 

During the One-way lookup test, the test application sends a message toward 

a random identifier. In the KBR lookup test, the application asks the Chord to look 

for a node that is responsible for the given identifier. 

In order to evaluate an impact of the Vivaldi mechanism on the process 

of obtaining a resource we prepared the DHT lookup test. We assumed that there 

are copies of the same resource shared in the Chord structure. In this test, the application 

that looks for the resource is eligible to select one of the nodes which 

possess that resource. 

We use the following metrics to analyze the protocol performance: 

– One-way Delivery Ratio – The total fraction of requests that reach the destination 

successfully.


– One-way Latency – The delay between reaching a destination by the message 

and sending it. 

– Lookup Success Ratio – The total fraction of requests for which the originator 

receives an answer successfully. 

– Lookup Success Latency – The delay between receiving answer and sending 

the request. 

– Get Latency – The cumulative delay between receiving a resource and 

sending the request in DHT lookup test. 

Additionally, the total volume of the sent and received traffic essential for maintenance 

(i.e. refreshing successor, and predecessor; refreshing entries in the Finger 

Table) was written. 

C. Results 

The results of our evaluation are shown in Figures 5-11. The notation “ps = 1” 

means that the Vivaldi mechanism and extending Finger Table were active, whereas 

“ps = 0” denotes the original Chord structure. 

The delivery ratio for One-way lookup test (Figure 5) reflects the general 

efficiency of a node to communicate in the overlay network. Increasing the number 

of nodes diminishes the ability to successfully communicate in the overlay. 

We found out that the performance of the wireless network is the primary cause 

of that observation. The phenomena that are peculiar to mobile ad hoc network 

(especially the hidden station) hinder transmission. The improvement of the underlay 

network was not our goal. 

Figure 5. The one-way delivery ratio


63 

Nonetheless, the obtained results prove that our application of Chord enables 

fully autonomous work on top of MANET. Moreover, our modifications bring a slight 

increase in the delivery ratios (Figure 5 and Figure 7) as well as some improvement 

of the latencies (Figure 6 and Figure 8). 

Figure 6. The one-way latency 

Figure 7. The lookup success ratio


Figure 8. The lookup latency 

The knowledge about real distances between nodes may also be exploited 

to optimize a process of getting resources. Owing to our modification, the node 

is given a chance to choose the nearest node which possesses the requesting resource. 

As shown in the Figure 9, the latency is decreased. 

Figure 10 and Figure 11 depict the total sent and received maintenance traffic, 

respectively. The underlay network is not overloaded by this kind of traffic. However, 

it should be stressed that the frequency of refreshing has considerable impact on 

the value of the above-mentioned traffic. 

Figure 9. The Get latency


65 

Figure 10. Sent maintenance traffic 

Figure 11. Received maintenance traffic 

VI. Conclusion 

In this paper, we proved that the Chord structure is able to operate on top 

of MANET in a fully autonomous manner. The adaptations we proposed resolve 

the joining process as well as improve distance awareness of the node. Although 

the effectiveness of work is not perfect, the network is not degraded due to the maintenance 

traffic. 

During future work we intend to test our solution in a real ad hoc network. 

Therefore, we have started practical implementation of the modified Chord 

structure.


References 

[1] N. Aschenbruck, E. Gerhards-Padilla, and P. Martini, “A survey on mobility 

models for performance analysis in tactical mobile networks,” in: Proceedings 

of Military Communications and Information Systems Conference (MCC 2007), 

Sep. 2007, Bonn, Germany, pp. 25-26. 

[2] I. Baumgart, B. Heep, and S. Krause, “OverSim: A scalable and flexible overlay 

framework for simulation and real network applications,” in: Proceedings 

of the International Conference on Peer-to-Peer Computing, Seattle, WA, USA, 

Sep. 2009, pp. 87-88. 

[3] I. Chakeres, and C. Perkins, “Dynamic MANET On-demand (DYMO) Routing,” 

IETF Internet-Draft, draft-ietf-manet-dymo-21, 2011. 

[4] F. Dabek, R. Cox, F. Kaashoek, and R. Morris, “Vivaldi: A decentralized network 

coordinate system,” in: Proceedings of the ACM SIGCOMM, Portland, Aug. 2004, 

pp. 15-26. 

[5] R. Kummer, P. Kropf, and P. Felber, “Distributed lookup in structured peer-topeer 

ad-hoc networks,” in: Proceeding of the on On the Move to Meaningful Internet 

Systems, 2006, pp. 1541-1554. 

[6] L.B. Oliveira, I.G. Siqueira, and A.A.F. Loureiro, “On the performance of ad hoc 

routing protocols under a peer-to-peer application,” Journal of Parallel and Distributed 

Computing, vol. 65, Issue 11, Nov. 2005. 

[7] N. Shah, and D. Qian, “An Efficient Unstructured P2P Overlay over MANET Using 

Underlying Proactive Routing,” in: Proceedings of Seventh International Conference 

on the Mobile Ad-hoc and Sensor Networks, 2011, pp. 248-255. 

[8] N. Shah, D. Qian, and Rui Wang, “MANET adaptive structured P2P overlay,” Peerto-Peer 

Networking and Applications, vol. 5, Number 2, 2012, pp. 143-160. 

[9] I. Stoica, R. Morris, D. Liben-Nowell, D.R. Karger, M.F. Kaashoek, F. Dabek, 

and H. Balakrishnan, “Chord: A Scalable Peer-to-Peer Lookup Protocol for 

Internet Applications,” IEEE/ACM Transactions on Networking, vol. 11(1), Feb. 2003, 

pp. 17-32. 

[10] T. Zahn, and J.H. Schiller, “DHT-based unicast for mobile ad hoc networks,” 

in: Proceedings of Fourth Annual IEEE International Conference on Pervasive 

Computing and Communications Workshops, 2006, pp. 179-183. 

[11] OMNeT++ discrete event simulation system, http://www.omnetpp.org

Revisiting the DARPA’s Idea 

of a Programmable Network 

Vladimir Aubrecht 1 , Tomas Koutny 2 

1 Faculty of Applied Sciences, University of West Bohemia, Pilsen, Czech Republic, 

aubrecht@kiv.zcu.cz 

2 Faculty of Applied Sciences, University of West Bohemia, Pilsen, Czech Republic, 

txkoutny@kiv.zcu.cz 

Abstract: In 1995, DARPA recognized the shortcomings of IP networks. As a response, the experts 

initiated a project that should become the cornerstone of new military communication architecture. 

The goal was a network that would be resistant to attack and to operate despite successful attacks, 

while allowing a secure sharing of data at all levels of command. The project was named Active 

Networks. However, it failed to meet performance and security criteria. We revisited this problem, 

identified performance and security bottlenecks so that we present an advance to fast, secure and 

scalable active network server – Smart Active Node. 

Keywords: programmable network, active networking, smart active node 


In 1995, DARPA recognized the shortcomings of IP networks. As a response 

the experts initiated a project that should become the cornerstone of new military 

communication architecture. The goal was a network that would be resistant to 

attack and to operate despite successful attacks, while allowing a secure sharing 

of data at all levels of command. The project was named Active Networks. 

Initially, DARPA funded the research on active networks. A number of wellaccepted 

papers appeared to demonstrate the capability of the active-networking 

architecture to fulfill the desired goals. With a time, it has shown that these papers 

presented rather a proof of the concept than a real-world usable implementation 

of an active networking server. The existing active networking servers failed to 

widespread beyond the research activities. As DARPA ceased funding this research, 

the research stopped gradually. Active Networks failed to meet performance and 

security criteria. 

The history of computer networks shows a periodic attempt to benefit from 

network programmability. For example, it is the Wormhole approach to loadsharing 

[1], the paradigm of mobile agents [2] and finally, the active networking.


Therefore, we decided to revisit the idea of active networking. We identified performance 

and security bottlenecks so that we present an advance to fast, secure 

and scalable active network server – Smart Active Node. 

Our decision was based on a development of load-redistribution method 

that runs on a simple active networking server – Grade32 [3, 4]. Grade32 runs 

processor-native code, but provides no security. In a short, we added a security to 

the Grade32 approach. 

The paper is organized as follows. The second section describes the related 

work. The third section presents Smart Active Node’s (SAN) architecture and 

server’s scalability. The fourth section describes the programming environment 

for the capsules and active applications. The fifth section presents the use of Ada’s 

rendez-vous to implement process isolation and making calls to the applicationprogramming 

interface (API). The sixth section describes the implementation 

of access rights. The seventh section describes capsule transmission. The eighth 

section presents results. The ninth section concludes the paper. 


A. Active networks 

In active networks, a capsule supersedes the packet. The capsule is a packet 

associated with a program code. The capsule’s header contains a hash digest 

of the program code. This program code can handle the data being transmitted. 

A network node can download the program code as needed. 

In general, a network protocol is associated with a program code that handles 

its packets. In an IP network, the program code has to already run at the network 

node, prior the packets are received. Therefore, the services provided are rigid and 

has to be standardized first. In an active network, the network node can download 

and execute the program code as needed. Therefore, only the program code notation 

and execution environment have to be standardized. As a result, the network 

can learn new protocols and functionality immediately. 

In IP networks, the destination node has to already run a program that 

can handle the incoming packets of a given protocol. Active Networks do not have 

this restriction. 

B. Preceding concept 

Let us discuss the concept of active networking that precedes SAN. In the bottom, 

there is a so-called NodeOS. It is either a standard operating system such as Linux or 

Windows, or it is a specifically written operating system such as JanOS [5]. 

On the top of the NodeOS, there are so-called execution environments. Each 

environment handles a particular notation of program code. ANTS [6, 7], BEES [8]


69 

and Magician [9] use Java bytecode. Proof of the concept for load redistribution 

in Active Networks, called Grade32 [3, 4], uses native x86 code. 

A process running in the execution environment is either an active application, 

or a capsule. The application injects the capsule into the network. If network’s 

implementation allows it, the capsule can replicate itself by injecting another capsule 

into the network. Grade32 allow the capsule to run an active application at 

the node. This approach was necessary for Grade32 to implement process migration 

in a distributed environment. It is not found in preceding implementations such 

as ANTS and Magician. 

Initially, the capsule program code was either pre-installed at the node, or 

it was transmitted with the capsule. In later implementations of active networking 

servers, each capsule contains a hash digest of capsule’s program code [7]. Using 

code distribution protocol, node uses this digest to obtain the program code from 

other nodes. Only the capsules of a basic code distribution protocol have to use 

standardized code identifiers, in place of the hash digest, to identify particular 

requests. Otherwise, it would be impossible for any two nodes to exchange a program 

code. Active node caches recently used program code at a local repository. 

Projects like PLAN [10], RCANE [11], SANE [12] and SNAP [13] focused 

on security and performance. In contrast to the general-purpose Java bytecode, 

specialized languages appeared. Their goal was simplicity that would limit resource 

consumption and shorten a capsule’s execution time. 

Reference [14] gives an overview on CSANE active network concept, which 

aims for security and scalability. CSANE is based on cluster processing. It builds 

on ANTS, JanOS and Linux. 

C. Related work 

Taking a closer look on the concept of the Google Chrome [15] operating 

system, we see a resemblance with the active networks concept. There is a simple, 

underlying operating system that provides hard application isolation – sandboxing 

[16]. API and the definition of web services define the Execution Environment. 

Also, it features a security manager that prevents running a malware. 

Although the Native Client [17] is not primarily designed for a use in active 

networks, there are ideas valuable to a high performance execution environment. 

III. Architecture 

A. Smart Active Node 

For the Smart Active Node, we highlight this: 

• SAN runs on top of a standard operating system. 

• SAN uses standard Java Development Kit for programming the capsules 

and to achieve interoperability across different platforms.


• SAN does not execute capsules in a single process. Instead it takes advantage 

of process isolation that is provided by the underlying operating system. 

• SAN uses Ada’s rendez-vous and Remote Procedure Call to further isolate 

a capsule program code from accessing SAN’s internal data structures. 

• SAN runs without a virtual machine such as JVM. The capsule program 

code is compiled just once to the native instruction set for the target 

processor. 

• SAN applies security measures when compiling the bytecode to the native 

instruction set, thus eliminating this overhead from the capsule runtime. 

• To increase performance, SAN benefits from supporting a distributed 

memory model. 

• For SAN, we adopted a concept of roles for a fine-grain control over execution 

of different program codes of capsules. 

B. Farmer – worker model 

The SAN server is a farmer-worker model that allows network-node scalability 

using the distributed memory model. Furthermore, the client-server model let us 

to use the means of process isolation and resource-limits which are provided by 

the underlying operating system. The resources are the processor time, the memory 

and network bandwidth. 

The farmer manages incoming capsules and assigns them to particular workers 

for further processing. The SAN server can run on a cluster, i.e. the workers 

can be distributed physically. SAN server and SAN worker communicate using 

the remote procedure call (RPC) mechanism. Therefore, SAN uses RPC to assign 

a capsule to a particular worker. 

Existing implementations of RPC, e.g. Open Networking Computing RPC, 

rely on the TCP/IP stack. As active networks were supposed to supersede IP, SAN 

has a custom RPC implementation. The SAN RPC implementation uses SAN’s 

internal networking interfaces instead of relying on the presence of TCP/IP stack. 

SAN networking interface can encapsulate an ISO/OSI layer 2 protocol such 

as the Ethernet. Furthermore, SAN utilizes a proprietary data serialization to avoid 

performance penalties by eliminating unnecessary data conversions. 

C. Farmer 

In the current implementation, the farmer provides the main subsystems: 

• Network subsystem – a communication with other nodes 

• Security subsystem – monitoring the networking traffic and capsule execution, 

while enforcing the security measures 

• Repository – caching compiled program codes of capsules 

• API subsystem – providing an API to capsules via RPC


71 

As the SAN server development will progress, part of the subsystems will be 

moved to the workers. This will reduce an average time a capsule/worker needs to 

access a farmer’s subsystem. 

Figure 1. SAN’s Famer-Worker Model 

D. Worker 

In the current implementation, the worker is a sandbox for an executing capsule. 

It takes an advantage of operating-system provided mechanisms for process 

isolation and enforcing limits on processor–time consumption, memory usage, 

network bandwidth usage, etc. To the running capsule, the worker provides the API. 

IV. Programming environment 

A. The server programming language 

Initially, we followed some of the well-accepted papers on active networking. 

The programming language was Java. Both, server and capsule codes were written 

in Java and compiled to bytecode. 

While the Java Virtual Machine (JVM) executed the server program code, 

we needed to execute the capsule program code separately in order to enforce 

the security measures. If we would let the JVM to execute the capsule program 

code, we would have no control over its execution. 

We needed to control processor-time consumption, memory usage, network 

bandwidth usage and method calls to prevent malicious actions as such calling 

System.exit. 

Java was designed to be architecture neutral. To follow this principle, we 

were left with one choice – to interpret the capsule bytecode in a JVM. Because 

of the incurred speed penalty, we decided to use C++ for the server development.


While the developments costs are higher with C++, we got a robust and efficient 

platform to execute the capsule program code, with security measures applied. 

B. The capsule programming language 

While Java is performance-inadequate for the server development, it has benefits 

for programming the capsules. Using a Java bytecode to C++ cross-compiler, 

and a C++ compiler, we compile the bytecode into the target-processor optimized 

machine code. In addition, the Java’s portability allows the SAN servers to run 

across several processor architectures. 

The Java’s simplicity provides an opportunity to analyze the bytecode for a possible 

security breach just once, to save the processor time later on. For example, 

the absence of pointers in Java greatly improves the security. 

Per a class, the bytecode to C++ cross-compiler generates the header file with 

class’ declaration and C++ file with class’ implementation. The header file includes 

header files of other required classes. The Java Runtime Environment is replaced 

with a C++ equivalent. In addition to standard packages, the SAN Java development 

kit (SAN JDK) provides classes to call SAN API. Finally, the capsule program 

is linked using the pre-compiled SAN JDK and compiled capsule’s cross-compiled 

C++ code. The linker generates a dynamically linked library that is saved for a later 

reuse, once another capsule of the same protocol arrives. 

Several JDK methods are security-unsafe. For example, letting a capsule to use 

reflection would be a security risk. In SAN JDK, such methods are not available at 

all. Such a malicious capsule will not even compile. Classes such as java.io.File, 

which access the server’s file system, have custom implementation which forbids 

the capsule to access the file system of the underlying operating system. The new 

operator is overloaded with a custom implementation that forbids memory allocation, 

if the capsule overcomes a given limit. 

To monitor processor-time consumption, a separate thread monitors execution 

times of capsule. It terminates a capsule, which exceeds the limit. 

V. Programming interface 

A. SAN programming interface 

The capsule program code is compiled into a dynamically linked library. 

The library exports a routine. SAN server calls this routine to execute the program 

code. The routine takes one parameter. This parameter is an object implementing 

a pre-defined interface to access SAN API. Capsule accesses SAN API in an objectoriented 

manner. The API offers these services: 

• Network service – capsule transmission 

• Routing service – accessing and updating the routing table


73 

• Storage service – an inter-process communication using the blackboard 

concept [18] 

• Diagnostic service – for debugging 

B. Calling SAN API 

Standard Java compiler compiles the capsule program code. At this point, the SAN 

API interface is not implemented. Later on, the SAN server compiles the bytecode 

to C++ source code. This compiled source code is linked with an object that makes 

the actual calls to the SAN API interface. This object is a wrapper for Ada’s rendezvous 

synchronization mechanism. 

The Java synchronization model relies on a monitor construction. The monitor 

has its working data and synchronized methods. For an exclusive access, a thread calls 

a synchronized method to manipulate with the monitor’s data. The thread continues to 

receive processor-time quanta based on its priority. Also, its access rights are not elevated. 

With Ada’s rendez-vous, a thread makes an entry call to the server. Then, 

the calling thread is suspended until the server’s thread finishes execution of the call. 

Only the server thread accesses its data. Furthermore, the server’s thread priority 

and access rights are independent on the calling thread. 

The SAN server runs in a user-address space. The compiled capsule program 

code cannot issue an equivalent of x86 int and syscall instructions to isolate SAN’s 

kernel from the running capsule. Therefore, we use Ada’s rendez-vous to implement 

secure API calls. 

To a capsule programmer, the Java capsule program code makes a call to 

a Java interface. 

VI. Access rights 

A particular networking application may require additional access rights to 

complete its task. For example, a routing service requires an access right to modify 

the routing table. Similarly, time-synchronization service requires an access right to 

set the system time. In SAN, we implement the access rights with a concept of users 

and roles. In a principle, we follow the Windows NT security model of users, 

access rights and privileges. 

A role is a set of access rights. For example, the routing-service role includes 

all rights required to modify the routing table, but not to e.g. set the system time. 

In addition to the access rights, a role has resource-consumption limits defined. 

A user has authorization credentials and a set of roles. For the development, 

we have defined four roles: 

• Administrator – no restriction 

• Power User – There is no restriction on capsule’s code execution time. 

This is used for the routing service.


• Users – The execution time limit is increased against the anonymous users. 

• Anonymous – This applies to unknown capsule codes and unauthorized 

users. The code being executed has lowest priority and no access right 

is granted. 

The capsule code can elevate its default access rights. By using different credentials, 

it can re-authorize to the server. 

VII. Transmitting a capsule 

A. All-Nodes-Execute flag 

In the original concept, the capsule’s code executed on every node visited. 

This implied an additional overhead. For example, let us consider a frequently used 

protocol such as TCP. The intermediate nodes have no need to modify packets’ 

payload. The packets need to be forwarded only. Therefore, we eliminate overhead 

of the original active-networking concept by introducing All-Nodes-Execution flag. 

When this flag is not set, the capsules’ code executes at the destination node only. 

As a result, the intermediate nodes do not need to create sandboxes, thus reducing 

the associated overhead to increase the node throughput. 

With this behavior, we maintain low costs of the TCP protocol, while offering 

an additional feature. When the capsule executes at the destination node, it attempts 

to deliver its payload. Instead of TCP/UDP ports, we will identify a listening process 

with a globally unique ID (GUID). This concept was used successfully in the loadredistribution 

method [3, 4]. If the listening process migrated to a different node, 

it left its new address using the black-board mechanism. Thus, the capsule’s code 

could lookup the new address and set it as the new destination address. This concept 

reduces the overhead of handling the error when the listening process is not 

present at the destination node. 

SAN provides the black-board concept with the Storage service. It is a keyvalue 

dictionary. The key is a GUID and the value is a sequence of bytes. Its size 

and persistency is given by respective resource limits. With the storage service, we 

implement an inter-process communication, when two communicating processes 

do not need to execute concurrently. 

With the All-Nodes-Execute flag set, the capsule’s code executes at every visited 

node. We use this flag for a routing protocol. To be specific, we use the AntNet 

routing protocol [19] to fill the routing table. 

B. Quality of Service and Type of Service 

To implement Quality of Service and Type of Service, we experiment with 

a concept of alternative routing table. SAN’s capsule has Time-Sensitive flag. With 

this flag set, the capsule is routed using the better routes in the sense of AntNet


75 

protocol routing metric. If this flag is not set, the capsule is routed to a worse 

route if the better route’s link is saturated. As a result, we prioritize the capsule 

transmission while attempting to reduce a capsule elimination due to an insufficient 

bandwidth. 

C. Delivering a payload 

When a capsule delivers its payload to a process, the SAN would invoke 

a registered callback to handle the delivery. The callback executes as a separate 

process and asynchronously to the main process of the networking application. 

As a result, the main process does not need to be executing. The callback can handle 

the capsule, while requiring fewer resources. Later, the main process can process 

the cumulative results, which were produced by the callback on receiving capsules’ 

payloads. This way, we attempt to support mobile devices, where the power and 

memory consumption is critical. 

When a capsule successfully delivers its payload, it would either forward itself 

to a new destination, or finish its execution. 

VIII. Results 

A. Testing software 

To test the performance of the SAN server, we used a custom implementation 

of the ping command. First, we implemented the ping command as the SAN 

active-networking application. Then, we re-implemented the same application using 

WinAPI and TCP/IP protocol. This guaranteed that we tested principally the same 

applications – the first one using IP, the second one using the active network. 

The testing ping application transmitted data for two minutes. We counted 

the total number of transmitted bytes in packet/capsule payloads. The payload size 

increased gradually. 

The operating system used was 64-bit Windows 7, version 6.1.7601. The server 

and cross-compiled bytecode were compiled with the Microsoft Optimizing 

Compiler, version 17.00.40825.2. The bytecode was cross-compiled to C++ using 

a proprietary compiler [20]. The generated images were 64-bit executables and 

dynamically loaded libraries. 

To measure the time, we used QueryPerformanceCounter and QueryPerformanceFrequency 

WinAPI functions. 

B. Testing network 

The network nodes used were connected with 100 Mb/s link speed in a full 

duplex, using Ethernet. In the current implementation, SAN uses TCP/IP to trans-


mit the capsules. This implies that the IP-ping should have less overhead than the active-networking 

pings. 

Two servers ran on Intel64, family 6, model A, stepping 7. We tested the ping 

on the localhost and over the network. The localhost machine had 8 GB RAM and 

3.4 GHz processor frequency. The other machine had 4 GB RAM and 2.3 GHz 

processor frequency. 

C. Discussion 

Table 1 gives achieved bit rate and latency for a ping over a network, i.e. when 

packets and capsules processed through the entire networking stack of the operating 

system. Table 2 gives the same, but for the localhost when the packets and 

capsules are not processed by the lowest level of the networking stack of the operating 

system. 

Size 

[B] 

Bit rate 

[Mb/s] 

Table I. Network Ping 

SAN 

Latency 

[µs] 

Bit rate 

[Mb/s] 

IP 

Latency 

[µs] 

64 17.1 643 0.5 945 

128 18.3 734 1.0 942 

256 24.2 811 2.1 942 

512 15.2 923 16.3 239 

1024 18.0 1097 18.3 428 

2048 20.7 1395 22.6 693 

4096 20.1 1804 27.9 1121 

8192 21.9 2201 35.2 1777 

Size 

[B] 

Table II. Localhost Ping 

Bit rate 

[Mb/s] 

SAN 

Latency 

[µs] 

Bit rate 

[Mb/s] 

IP 

Latency 

[µs] 

64 5.1 218 17.4 28 

128 5.7 223 33.7 29 

256 18.3 232 65.1 30 

512 41.9 236 114.9 34 

1024 77.1 241 211.2 37 

2048 136.8 259 363.4 43 

4096 179.2 276 726.7 43 

8192 193.5 297 1275.5 49


77 

The tests were not designed to achieve the maximum bit rate possible, but 

to give the difference between the IP processing and the programmable network 

processing. 

Table 3 gives memory usage for the SAN ping over the network. The memory 

usage measured was the Peak Working Set, which is the largest working set that 

has been observed for the process. 

Table III. Memory Usage 

Size [B] Server [MB] Worker [MB] 

64 8.1 5.4 

128 8.2 5.4 

256 8.9 5.7 

512 10.5 6.1 

1024 12.0 6.9 

2048 16.8 8.5 

4096 25.0 11.8 

8192 39.3 14.3 

Looking at the measured quantities of the IP ping, we can see the effect 

of Nagle’s algorithm and that the operating system used a reduced processing for 

the localhost communication. 

Looking at the latencies of the SAN ping, we see an overhead implied by 

the sandbox usage. Looking at the bit rates over the network, we see that SAN capsule 

suppresses the Nagle’s algorithm due to the capsule’s size overhead. With a payload 

size greater than 4096 B, we see that capsule serialization has to be improved. 

The localhost results confirm this conclusion. Particularly, Table II shows results 

above 200 Mb/s for IP, but less than 200 Mb/s for SAN. As the ping-mechanism 

is essentially the same for IP and SAN, the reason for the achieved bitrate difference 

must be related to the present implementation of capsule processing. 

Looking at the memory usage, we see that it increased with capsule’s payload 

size. However, the memory usage was stable. If it was not, the memory usage 

would increase at least to 328 MB for the 8192 B payload and at least to 77 MB 

for the 64 B payload. 

IX. Conclusion 

In this paper, we present an advance on the DARPA’s idea of a programmable 

network. We show that the implementation presented could provide a sufficient 

performance, while enforcing the security measures on the code being executed. 

We achieved these results by using a different approach than the initial implementations 

did.


With the All-Nodes-Execution flag not set, data flows are routed in same manner 

as in IP networks and no sandboxes are created at intermediate nodes. If we 

would consider rigid protocol hashes for frequently used protocols such as TCP 

and UDP, it would not be even necessary to create sandboxes at all. SAN network 

would behave as IP network in such a case, only the address:port-like data structure 

would have greater size than with IP protocol. In addition, the SAN network would 

still maintain its programmability for other protocols. 

If we would consider protocols at higher ISO/OSI level than TCP and UDP, 

programmable protocols could possibly provide an increased efficiency over IPbased 

protocols. For example, reference [22] presents an improved efficiency with 

a multi-cast algorithm. 

The node throughput depends on the packet buffer size. The node has to 

buffer the packet prior processing it. Regarding larger payloads and the All-Nodes- 

Execution flag set, SAN could provide a blocking access to the payload. The capsule’s 

execution would be suspended, if the capsule requires such a part of payload that 

has not been received yet. Such an optimization technique would further increase 

SAN’s throughput, as it processes a capsule ahead of receiving its complete payload. 

On the other hand, we still pay with processor time and memory consumption for 

the network programmability. Therefore, an improved efficiency of networking 

protocols at higher ISO/OSI levels should be given by their design. As demonstrated 

with e.g. [22], we consider this as possible. 

However, the SAN server is still under development [21]. There are a number 

of optimization opportunities, which should improve the node’s throughput 

furthermore. 

References 

[1] J.F. Shoch and J.A. Hupp, ”The ‘Worm’ Programs – Early Experience with a Distributed 

Computation”, Communications of the ACM, 1982. 

[2] C.G. Harrison, D.M. Chess and A. Kershenbaum, “Mobile Agents: Are they a good 

idea”, Technical Report, IBM Research Division, T.J. Watson Research Center, March 

1995. 

[3] T. Koutny and J. Safarik, “Load Redistribution in Heterogeneous Systems”, 

Proceedings of the Third International Conference on Autonomic and Autonomous 

Systems, Athens, Greece, 2007. 

[4] T. Koutny and J. Safarik, “Simulating Distributed Applications in an Active Network”, 

Proceedings of 6th Eurosim Congress, Ljubljana, Slovenia, 2007. 

[5] P. Tullmann, M. Hibler and J. Lepreau, “Janos: A Java-oriented OS for Active Networks”, 

IEEE Journal on Selected Areas of Communication, vol. 19, no. 3, March 2001. 

[6] D.J. Wetherall, “Developing Protocols with the ANTS Toolkit”, http://www.cs.utah.edu/ 

flux/janos/ants-manual-2.0.2/papers/programming.ps Last Accessed on April 11, 2012.


79 

[7] D.J. Wetherall, J. Guttag and D. Tennenhouse, “ANTS: A Toolkit for Building 

and Dynamically Deploying Network Protocols”, IEEE Open Architectures and Network 

Programming 1998, San Francisco, CA USA, 1998 

[8] T. Stack, E. Eide and J. Lepreau, “BEES: A Secure, Re-source-Controlled, Java Based 

Execution Environment”, IEEE Open Architectures and Network Programming 2003, 

San Francisco, CA, USA, 2003. 

[9] S.F. Bush and A.B. Kulkarni, “Active Networks and Active Network Management – 

A Proactive Manage-ment Framework”, Kluwer Academic/Plenum Publishers, 2001. 

[10] M. Hicks, J.T. Moore, D.S. Alexander, C.A. Gunter and S.M. Nettles, “PLANet: 

an active internetwork”, Proceedings of Eighteenth Annual Joint Conference of the IEEE 

Computer and Communications Societies, New York, NY, USA, 1999. 

[11] P. Menage, “RCANE: A Resource Controlled Framework for Active Network Services”, 

Proceedings of the First International Working Conference on Active Networks, 

Berlin, Germany, 1999. 

[12] D.S. Alexander, P.B. Menage, A.D. Keromytis, W.A. Arbaugh, K.G. Anagnostakis 

and J.M. Smith, “The Price of Safety in an Active Network”, Journal of Communications 

and Networks, Special Issue on Programmable Switches and Routers, vol. 3, Number. 1, 

March 2001. 

[13] W. Eaves, L. Cheng, A. Galis, T. Becker, T. Suzuki, S. Denazis, C. Kitahara, 

“SNAP Based Resource Control for Active Networks”, Proceedings of IEEE Global 

Telecommunications Conference, Taipei, Taiwan, 2002. 

[14] C. Xiao-lin, Z. Jing-yang, D. Han, L. Sang-lu and C. Gui-hai, “A Cluster-Based 

Secure Active Network Environment”, In Wuhan University Journal of Natural Sciences, 

vol. 10, Number 1, 2005, pp. 142-146, doi: 10.1007/BF02828636. 

[15] J. Gray, “Google Chrome: The Making of a Cross-Platform Browser”, In Linux Journal, 

vol. 2009, 2009. 

[16] Ch. Reis, A. Barth and Ch. Pizano, “Browser Security: Lessons from Google Chrome”, 

In Communications of the ACM, vol. 52, 2009, pp. 45-49, doi: 10.1145/1536616.1536634. 

[17] B. Yee, D. Sehr, G. Dardyk, J.B. Chen, R. Muth, T. Ormandy, S. Okasaka, 

N. Narula, and N. Fullagar, “Native Client: A Sandbox for Portable, Untrusted 

x86 Native Code”, Proceedings of 2009 IEEE Symposium on Security and Privacy, 

Oakland, California, USA, 2009. 

[18] R.S. Engelmore and A. Morgan, editors Blackboard Systems, Addison-Wesley, 1988. 

[19] G. di Caro and M. Dorigo, “An Adaptive Multi-Agent Routing Algorithm Inspired 

by Ants Behavior”, Proceedings of PART98 – Fifth Annual Australasian Conference 

on Parallel and Real-Time Systems, 1998. 

[20] T. Koutny, “Static Cross-Compilation of Java Bytecode”, Submitted to IEEE Transactions 

on Computers, March 2012. 

[21] T. Koutny and V. Aubrecht et al., “Smart Active Node”, http://www.san.zcu.cz/ 

Last Accessed on April 11, 2012. 

[22] V. Ramakrishna, M. Robinson, K. Eustice and P. Reiher, “An Active Self-Optimizing 

Multiplayer Gaming Architecture”, Cluster Computing, vol. 9, Issue 2, 2006.

Selection and Investigation of a Civil Wideband 

Waveform for Potential Military Use 

Ferdinand Liedtke, Matthias Tschauner, Sarvpreet Singh, 

Marc Adrat, Markus Antweiler 

Communication Systems, Fraunhofer-FKIE, Wachtberg, Germany, 

ferdinand.liedtke@fkie.fraunhofer.de 

Abstract: This contribution presents the results of an ongoing study 1 evaluating civil wireless 

communication systems or Waveforms (WFs) for their potential military use. It is a continuation 

of the work presented in [1] where several civilian communication systems with their main strengths 

and weaknesses have been evaluated. The aim of the past and present activities is to examine which 

of these communication systems or WFs have valuable characteristics. In addition, it is analyzed which 

WFs can probably be modified and supplemented to fulfill military requirements. The focus is on 

Wide Band Waveforms (WBWFs) which could help to fill the capacity gap of the “last tactical mile”. 

The studies will contribute to the international activities of finding IP and network capable WBWFs 

for military Software Defined Radios. 

In this paper several fundamentals of tactical communications are sketched, followed by some remarks 

on modern civil WFs already in use or tested by military forces. After agreeing on several significant 

assessment criteria, three interesting civil systems are short listed before one of these systems is finally 

selected for further investigation. For the selected system, WLAN IEEE 802.11n, the first critical 

modules are identified and some ideas for their modification and/or enhancement are presented. 

Keywords: Wideband Waveform; Software Defined Radio; Wireless Military Communications; 

Tactical Communications 


The international defense forces stride ahead towards further development 

of the network enabled capabilities for the network centric warfare. Important elements 

of this research are the communication capabilities, especially the Software 

Defined Radio (SDR) technology. For this technology, appropriate hard- and software 

modules and new Waveforms (WFs) are needed. In addition to narrow band WFs, 

the particular focus is on the development of new, so-called Wide band WFs (WBWFs) 

with spectral channel bandwidths ≥ 1 MHz for the transmission of payload data 

1 

This research project was performed under contract with the Federal Office of the Bundeswehr for Information 

Management and Information Technology, Germany.


rates > 1 Mbps (Megabits per second) per channel. Several military driven efforts 

are pursued in this area. Beside international activities, e.g. from NATO and from 

the consortium on Coalition Wideband Networking Waveform (COALWNW) [2], 

several nations are thinking about designing additional WBWFs for national use. Keeping 

this background in mind, it will be interesting to look at civil used WBWFs, too. 

In the preceding work [1], the main strengths and weaknesses of four categories 

of modern communication systems have been examined: broadcast, cellular, wireless 

access networks and trunked radio systems. The results show that several of those 

systems have some military suitable characteristics, but further efforts will be required 

to get military usable WBWFs. Thus, it is appropriate to select at first a few or only one 

suitable modern communication system for further investigation. For the selection 

procedure, many military requirements, both of operational and of technical manner, 

can be collected, weighted and evaluated in a detailed way. But, it is recognized 

that only a few evaluation criteria are significant enough for the ongoing efforts that 

a comparatively pragmatic and short selection procedure is possible. 

This is accomplished in the second phase of our investigations which are discussed 

in this contribution. In this context, modern civil communication systems 

which are already in use or tested by military forces are searched. The recognition 

of the state of military deployment of civil WFs has to be kept in mind while deciding 

for a suited communication system for further investigation. In the following 

text “communication(s)” will be mostly abbreviated by COM(s). 

To get a better understanding of the different military COM requirements 

concerning bandwidth, data rate, range and mobility, a glimpse on tactical COMs 

and the necessity to fill the capacity gap of the “last tactical mile” is presented in Section 

II. Section III contains some remarks on and examples of modern civil COM 

systems already in use or tested by military forces. In Section IV, key requirements 

of a WBWF for military use are summarized. In Section V, at first, the extraction 

of three potentially suited civil COM systems is described. After having outlined 

several important strengths and weaknesses of these systems, finally, the WLAN 

system IEEE 802.11n is selected for further investigation (WLAN: Wireless Local 

Area Network). In Section VI, the first critical modules of this system are identified 

and ideas for their modification or enhancement are presented. Section VII 

gives a short overview of the initial steps for implementing the new WF with SCA 

(Software Communications Architecture) conformity on a simulation platform. 

II. Tactical communications and difficulties 

for the “last tactical mile” 

Generally, the military operations planned and executed at the divisional 

level or below are the tactical ones. In modern scenarios, planning and execution 

of actions are increasingly accomplished by military subordinated smaller groups, 

whereby they have to strive for the superior aims. To solve all the different tasks,


83 

powerful COM equipment is necessary. Important requirements for tactical COM 

systems are: IP and networking capability, sufficient transfer capacity, adequate range 

and fair mobility, also for relevant parts of the infrastructure. The aim is to fulfill 

the necessity for Command, Control (C2) and Communication/Coordination (C3) 

on the move. The main subsystems of the tactical COM system are pointed out 

in Fig. 1 with blue framed boxes. These are: the trunk network, the Combat Net 

Radio (CNR) and the data distribution subsystems [3]. 

Figure 1. Main elements of a tactical COM system, derived from [3] 

The arrangement of these subsystems concerning their range-capacity-mobility 

trade-offs is depicted in Fig. 2. 

Figure 2. Range-capacity-mobility trade-off, similar to [3] 

The three blue framed circles designate the areas related to the trunk network 

(bottom left), the CNR (top) and the data distribution (bottom right) subsystems. 

The trunk network transfers large amounts of information with high data rates 

between the relevant headquarters and between the staff personnel and the control 

officers, traditionally down to brigade level. The connections are by satellite or 

other longer range wireless or lined connections. In general, the range is large and 

the mobility is minor. 

The CNR in its traditional form is a complement of the trunk network; it has high 

mobility, small values for spectral bandwidth and data rate – traditionally 25 kHz


and around 20 kbps – and a medium range. The personnel communication system 

(PCS) is smaller and, in its traditional versions, less powerful than the CNR. 

The main purpose of both these radio types is to transfer voice and low rate data 

information, e.g. sensor-to-shooter information. Modern tactical radios offer higher 

throughput rates, e.g. 112.5 kbps like the SDR-7200 from Elbit/Tadiran [4]. 

But, for filling the capacity gap between the trunk network and the CNRs or PCSs 

for the transfer of more capacity demanding information, e.g. situational awareness 

data, an additional tactical subsystem is necessary, which is the data distribution 

subsystem. This system has to transfer more information to and from the acting 

teams, and nowadays it is the “bottle neck” of the tactical COM system. This problem 

of the “last tactical mile” is similar to that of the last mile in civil environments. 

Although there exist several special solutions for filling this gap, e.g. the U.S. systems 

EPLRS (Enhanced Position Locating and Reporting System) and NTDR (Near Term 

Digital Radio), important enhancements can be gained by modern SDRs with different 

WFs. Modern narrow band WFs will be used for CNR purposes and WBWFs 

will serve for (bidirectional) data distribution. For satisfying the different military 

requirements, it is necessary that the new WBWFs are scalable. 

III. Modern civil wireless COM systems already in use or tested 

by military forces 

Before going on to the evaluation of civil COM systems concerning their potential 

military use, a review of the present situation is necessary, i.e. a check, if and 

where civil wireless COM technologies are already in use or tested by military forces. 

A. Broadcast systems 

An interesting wideband system is the terrestrial Digital Video Broadcast system 

DVB-T/T2. It uses the modern and flexible modulation/multiplex type OFDM 

(Orthogonal Frequency Division Modulation/Multiplex). But, in its original form, 

it has no backward channel. One known modification is the Universal Multi-Media 

Link system from Thales Defense [5]. It is provided with a backward channel so 

that bidirectional COM is possible. The system is intended for use by safety forces 

in difficult scenarios like disasters. 

B. Cellular systems 

The well-known commercial cellular systems are eagerly used or tested by 

military forces though they seem to have some drawbacks like single points of failure 

(base stations) and limited mobility of the infrastructure. The cellular technologies 

of the third generation (3G) and beyond, i.e. UMTS, CDMA2000, HSDPA, HSPA 

and the 4G/LTE (fourth generation of civil cellular systems/Long Term Evolution)


85 

are in the focus. The literature about the military use mainly concerns activities 

from U.K. and U.S. [6]. On one hand, cellular systems are used in less dangerous 

scenarios, e.g. in command posts, logistical establishments, airports and separated 

areas. On the other hand, they are also used for more robust operations. For those 

cases, small and transportable base stations, also airborne ones are available with 

the aim to have one’s own infrastructure with non-traditional radio frequencies and 

extended range and mobility. The currently used or tested handhelds are commercial 

smartphones or small tablet PCs, enlarged by ciphering and personnel identification 

circuits. In less peaceful scenarios, these hand helds are connected to the CNRs by 

wire or wireless, and the information transfer takes place through the CNRs. For 

future use, so-called sleeve modules are developed which can pick up the handhelds 

and provide the power supply, the crypto module and interfaces. For peacekeeping 

missions, the hand helds can also use the civil cellular infrastructure, if available. 

So, the military used cellular technology can fulfill particular tasks of the trunk 

network, the data distribution subsystem and the CNR, too. 

Examples of initiatives from U.K. and U.S. are: the Training and Doctrine Command 

– Brigade Combat Team initiative, where smartphones are tested together with the JTRSs 

(Joint Tactical Radio Systems) AN/PRC-154 (so-called Rifleman Radio) and 155. 

In a DARPA (Defense Advanced Research Project Agency) project, the Harris radio 

AN/PRC-117G SDR is tested together with tablet PCs. Connections by WLAN are 

provided for the near surrounding. A third example is the Monax system from Lockheed 

Martin. The used RF is non-traditional, modules for ciphering are developed and IP 

connections up to the end-users are available. The developed mobile base stations can be 

integrated into land and aerial vehicles. Some of the used WFs are tolerant against larger 

latencies, e.g. for GEO-SATCOM connections. A further U.S. development is FASTCOM, 

a pico cell system from Textron/Overwatch for robust missions. The base stations are 

mobile. The COM types are voice, data and streaming videos, e.g. from smartphone to 

an unmanned aerial vehicle and reverse. A high grade ciphering module will be developed. 

Up to 100 subscribers can be provided. In the U.K., Roke Manor Research is discussing 

ideas and concepts concerning the use of civil cellular technology, i.e. 3G/UMTS, 

4G/LTE and WiMAX, for military purposes (WiMAX: Wireless interoperability 

for Microwave Access). Another U.K. project is Roke’s Battlefield Connect, a femto 

cell system for a range up to 40 km with up to 7.2 Mbps for non-demanding IP 

connections, VoIP and near real-time video transfer. The handhelds would be usable 

up to 120 km/hr. The used technology is UMTS/HSDPA, HSPA, or, for the future, 

4G/LTE or WiMAX. In the final version, 24 subscribers will be provided in one cell, 

each with up to 14.4 Mbps for the downlink and up to 1.9 Mbps for the uplink. 

C. Wireless access networks 

The relevant access networks dealt with in [1] are the WLAN family IEEE 

802.11 and WiMAX IEEE 802.16. WiMAX in its mobile form, i.e. IEEE 802.16e,


is mainly a cellular system with base stations and the appropriate infrastructure. 

But, to preserve the same classification as in [1], it is dealt with in this section 

as access network. The ranges of WLAN and WiMAX in their standardized forms 

are very different. WLAN can provide connections in the near surrounding, while 

the range of the mobile WiMAX is up to 15 km. 

In the safety or military range of application, WiMAX is proposed for high data 

rate linking over medium ranges. In Norway, its use for rescue operations is proposed. 

WLAN is used for short range connections as it is known from civil applications. 

For example it is used in headquarters and command posts. A particular proposal 

is its supplemental use in command posts during the set-up and set-down periods, 

if the line connections are not yet or no longer established. A further WLAN application 

is the remote control of platforms like sensors, weapon systems or vehicle 

based COM nodes. Examples for remotely controlled COM nodes are: Tactical 

Cross Domain Solution from U.S. and C2UK C-Lite from U.K. [6]. 

D. Trunked radios 

The well-known trunked radios TETRAPOL and TETRA are well proven and 

broadly used for safety and military forces. The systems have a comparatively good 

physical and electronic robustness. But, they are designed only with comparably 

small channel bandwidths for low or moderate data rates. 

IV. Requirements for a wideband waveform in wireless military 

tactical use 

The term Waveform (WF) comprises of all components of a COM system 

in the seven ISO/OSI layers. The focus of the currently accomplished studies is on 

the lower layers, PHY and MAC (Physical and Medium Access Control layers). 

The requirements of a WBWF are orientated towards the necessity to fulfill mainly 

the tasks of a modern bidirectional data distribution subsystem, i.e. to find a WBWF 

with IP and flexible networking capability and with appropriate capacity, range 

and mobility. The new WF should be flexible in use, i.e. it should be scalable and 

adaptable to different applications and transmission conditions. Certain robustness 

against disturbing influences, natural and intentional, is aspired. 

A. General requirements 

General requirements, which are resulting from the facts mentioned in the sections 

above, are compiled as follows: 

• command, control and coordination (C3), also on the move, 

• single points of failure should be avoided as far as possible, hence an adhoc 

capable system/WF would be advantageous,


87 

• shared situational awareness, thus a payload data rate > 1 Mbps seems 

advisable, 

• with a range of up to 40 km (as Roke’s Battlefield Connect System) for land 

based platforms, 

• with multi hop and multi cast capabilities, 

• for use within mobile platforms; also with participation of a slow airborne 

vehicle; hence, land based: up to 100 km/hr and airborne: up to 400 km/hr, 

• secure and IP capable information transport, 

• Quality of Service (QoS), from low latency to best effort, 

• WF with inherent flexibility and scalability, thus a OFDM WF seems adequate, 

• providing a tactical mobile extension of the deployed Protocol Core Network, 

• for use within SDRs taking into account SCA conformity, 

• with resilience against noise and interference, 

• operation within the usual military spectral opportunities (RF bands) and 

constraints, and 

• COM capability between military and civil forces. 

Further require ments like a certain protection against hostile detection, interception 

and jamming and the provision for the WF usage in high speed airplanes 

or for Radio Based Combat Identification will be desirable. But, the last mentioned 

requirements are secondary ones and can be fulfilled later. 

These requirements for military use have to be kept in mind while assessing 

civil WBWFs. But, it cannot be expected that the civil WFs in their original forms 

can fulfill many of the requirements because they have been developed with different, 

commercially driven goals. 

B. Further aspects 

Some factors proposed from our side are: 

• For the present, half-duplex traffic will be sufficient. 

• Multiple Input Multiple Output (MIMO) is not yet necessary, i.e. Single 

Input Single Output (SISO) will be sufficient. 

• The new WBWF shall be demonstrated on particular simulation and development 

platforms. 

If appropriate transceiver modules will be available later on, the possible 

WF characteristics concerning RF, transmitter power, range and mobility will be 

exploited. 

V. System evaluation and selection of one civil WBWF 

for further investigation 

As a first step, a pre-election of a few appropriate civil WBWF systems is accomplished. 

The result is a selection of three modern wireless COM systems with


large channel bandwidths. Later on, as a second selection step, one of these systems 

is finally selected for further investigation. For the pre-election and the final selection 

comparable pragmatic evaluation criteria are used. 

A. Evaluation and pre-election 

For the pre-election, only three significant assessment criteria are employed: 

(1) modern wireless WBWF system with 

(2) bidirectional information transfer and 

(3) easy scalability. 

All the newer systems mentioned in Section III are modern WBWF systems, 

except the trunked radios. The newer trunk radio versions (TETRA2/3) allow 

a maximum channel bandwidth of only 150 kHz. This fact, together with the comparatively 

low development speed of such radios, will result in the exclusion of this 

technology from further analysis for a suitable WBWF. 

In the second evaluation criterion, the requirement for bidirectional information 

transfer is not fulfilled from the broadcast system DVB-T/T2. Additionally, 

it has a structure with comparably long frames. Both facts would demand a remarkable 

work load for modification or replacement of the modules in question. 

The consequence is the exclusion of the broadcast systems from further investigations 

as well. 

In the third evaluation criterion, the scalability can be easily fulfilled of OFDM 

WFs. Additionally, OFDM has significant advantages relating to robustness within 

multipath propagation and the flexibility of using its particular frequency channels. 

So, from the cellular and the wireless access network systems those WBWF 

systems with this modulation/multiplex type are the most favored ones for our 

further investigations: the newest cellular standard 4G/LTE, the WiMAX IEEE 

802.16e and the WLAN IEEE 802.11n. 

B. Final selection of one system/WF for further investigation 

For accomplishing the final selection, the relevant system characteristics 

of the three pre-elected systems are gathered in Table I, whereby the focus is essentially 

on those system characteristics which are different for the systems. The principally 

equal characteristics, like the OFDM WBWFs, the bidirectional information 

transfer, the use of IP protocols, the multicast capability and the variability of packet 

lengths are not picked up in the comparison because they do not contribute to discrimination. 

The table contents written in red or green colored text will be explained 

below. In addition to the technical characteristics noted in Table I, several other 

facts concerning the information gathering and the possibilities for modifications 

or enhancements are quite relevant for our final system selection. These facts are 

collected in Table II below.


89 

Table I. Relevant characteristics of pre-elected COM systems (with focus on those 

characteristics which are different between the three systems under consideration) 

Characteristics 

4G/LTE 

WiMAX 

802.16e 

Max. channel 20 MHz 20 MHz 20 MHz 

bandwidth BW chmax (40 MHz) 

Max. link data rate 

in BW chmax 

for SISO operat. 

RF bands 

Bandwidth (data 

rate) scalability 

Modulation types 

DL: 75 Mbps 

UL: 18.75 Mbps 

0.8 and 2.6 GHz 

(Europe) 

Yes; 

1.25 … 20 MHz 

QPSK, 16-QAM 

and 64-QAM 

Multiplex method DL: TDMA / 

OFDMA; 

UL: TDMA / 

SC-FDMA 

Ad-hoc net work 

capabil. 

Multi hop relay 

capability 

Transmission of 

speech and data 

with QoS (priority, 

latency, etc.) 

Necessity of base 

stations 

Real time 

capability 

Range 

Mobility 

Remarks 

DL: 128 Mbps, 

UL: 56 Mbps, 

both with FDD 

2.3, 2.5 and 3.5 GHz 

for mobile operat. 

Yes; 

1.25 … 20 MHz 

QPSK, 16-QAM 


TDMA/ 

Scalable OFDMA 

(S-OFDMA) 

No No Yes 

No, but handover 

capability 

Not yet; the necessary 

IP multimedia 

subsystem is not yet 

imple mented; 

Yes 

Yes 

WLAN 

802.11n 

75 Mbps (150 Mbps) 

2.4 and 5 GHz 

(in U.S. also < 1 GHz) 

Yes; frequency raster 

5, 10 and 20 MHz 

BPSK, QPSK, 16-QAM 


CSMA/CA 

Yes 

In principle: 

Yes; the necessary wireless 

multi me dia module is still 

included; 

Yes Yes No (for the ad-hoc mode); 

Yes; 

smallest latency 

time 5 ms; 

Several kilometers; 

under spe cial 

condi tions up to 

100 km 

Handhelds 

usable up tp 

350 km/hr 

comparatively 

autonomous base 

stations; 

very expen sive 

infra structure 

Yes; 

smallest latency 

time 1 ms 

Several hundred 

meters up to 50 km 

be tween fixed stations; 

up to 15 km 

between fixed 

and mobile stations 

Handhelds 

usable up to 

120 km/hr 

802.16e standard 

for mobile 

operation; 

expensive infrastructure 

With restrictions because 

of CSMA/CA and relativly 

large frame and packet 

lengths; 

Typically: < 100 m indoor 

and ≤ 250 m outside; 

with more transmitter 

power: several kilometers; 

Restricted to walking pace 

Eco/sleep mode; 

frame aggregation 

and block acknoledgment; 

mature and broadly used 

technology; 

cheap in-frastructure


Table II. Information gathering and possibilities for modifications and enhancements 

Inform. gathering, 

possib. for modific. 

and enhancements 

Access of source code 

and simulation models 

Possibilities for code generation 

from simulation tools 

like MATLAB/SIMULINK 

Possibilities for incremental 

modifications and enhancements 

Expenditure for modifications 

and enhancements 

4G/LTE WiMAX 802.16e WLAN 802.11n 

Strongly restricted 

because of property 

rights 

Not available 

Difficult because 

of concetanated 

PHY and MAC 

Restricted 

because 

of property rights 

Only simple PHY 

model available 

Difficult because 

of concetanated 


PHY: Yes, but 

without time sync; 

MAC: No 

Complete PHY 

model available 

Promising because 


are separable 

Very high High Moderate 

While the access of system specifications is good for all three systems, the availability 

of simulation tools and the possibilities for adaptations and enhancements are different. 

As can be acknowledged from Table II, there are clear advan tages for WLAN 802.11n, 

because of (partial) availability of source code and simulation tools and the possibility 

to separate the PHY and MAC layers. These advantages alleviate modifications and 

replacements of critical modules and are the crucial factors in our decision to select 

the WLAN WF for further investigation. For the other two WFs, much work ought 

to be invested to reach comparable circumstances. In Table II, the advantageous facts 

of WLAN 802.11n are emphasized by using green colored text. 

As can be gathered from Table I, WLAN 802.11n has several further advantages 

(emphasized with green colored text): the ad-hoc network and multi-hop relay capabilities, 

the included multi-media module, the independence from base stations, 

the eco/sleep mode, the possibility for frame aggregation and block acknowledgment 

and the availability of cheap components. The eco/sleep mode can perhaps help 

to realize the military required “radio silence”. The frame aggregation and block 

acknowledgment reduce the overhead caused by control information. 

However, the WLAN WF not only has advantages but also disadvantages which 

are emphasized by using red colored text. The main disadvantage is the CSMA/ 

CA (Carrier Sense Multiple Access/Collision Avoidance) multiplex method which 

impedes the required real time information transfer if more than only a few users 

are active. Furthermore, CSMA/CA does not efficiently exploit the possible channel 

capacity. But, this multiplex method can be replaced as it is discussed in the next 

section. The other two emphasized disadvantages, limited range and only pedestrian 

mobility, are not independent from the multiplex method and can also be 

improved concerning the military requirements. Furthermore, the topic of insufficient 

security has to be worked on in a later study phase.


91 

VI. Ideas for modifications and enhancements of WLAN IEEE 

802.11n modules 

In this section some critical modules of WLAN 802.11n are discussed and 

evaluated for necessary modifications and enhancements. Due to the military 

requirements on a data distribution subsystem, WLAN 802.11n (as described 

in the specifications) has some disadvantages, e.g. in multiplex method, range and 

mobility. In the following subsections, initial ideas for the elimination of these 

disadvantages will be presented. 

A. Multiplex method 

1) Selection of an alternative multiplex method 

In WLAN 802.11n, the CSMA/CA multiplex method does not guarantee real 

time behaviour and cannot exploit the possible channel capacity due to the fact 

that a random based channel access is used. Therefore, multiplex methods with 

systematic channel access like Time Division Multiple Access (TDMA), Frequency 

Division Multiple Access (FDMA) or Code Division Multiple Access (CDMA) 

are better suited. However, CDMA needs an expensive power controller to realize 

non-discriminatory data transmission between all users, especially mobile 

ones. A realisation without fixed base stations would be very demanding. FDMA 

is normally reserved for the separation of different user groups. Therefore, at first, 

the focus is on TDMA. When replacing the existing multiplex method in WLAN 

802.11n, the following requirements need to be preserved or newly fulfilled: 

• support of many network nodes, 

• IP support and ad-hoc network capability, 

• QoS capability, i.e. appropriate latency and priority, based transmission 

between nodes, and 

• bandwidth efficiency. 

As a possible solution, the use of the Unified Slot Assignment Protocol 

(USAP) [7, 8] is proposed. USAP is a TDMA based multiplex method with multichannel 

support. The channel access is managed due to distributed slot assignments, 

so that random based packet collisions are avoided. With a collision-free 

transmission the reliability rises and QoS constraints can be fulfilled more easily. 

2) Evaluation of the USAP multiplex method 

With the exchange of the multiplex method, an evaluation of the new WBWF 

based on the PHY layer scheme of WLAN 802.11n combined with the MAC layer 

scheme of USAP is necessary. Therefore, a few equations for estimating the data 

rate above the MAC layer are derived. 

To estimate the slot time duration T slot (B) dependent on the packet size B 

bytes in TDMA systems, several relevant parameters have to be known. Firstly,


an upper bound of the COM distance D max that should not be exceeded by any two 

node pairs in the existing network is needed. This strict constraint is important to 

prevent crossover collisions between neighbouring slots that have been assigned 

by the USAP for transmission. Secondly, an additional time period T add needs to 

be considered which includes time uncertainties on the half-duplex switching 

duration, the hard- and software processing delays and for having an additional 

spare time. Further, the transmission time for a PHY layer packet of size B needs 

to be known. From the WLAN 802.11n specification [9, 10], it is extracted that for 

a given channel bandwidth BW, the number of bytes B in the packet and the chosen 

modulation and coding schemes expressed in the data bits per OFDM symbol 

N dbps,k is sufficient to calculate the slot duration. The values for N dbps,k can be 

taken from Table III for different modulation and coding schemes, assigned by 

the index k. The slot duration T slot,k (B) for the USAP multiplex method can then 

be calculated by: 

Dmax 

80 22+8B 

 

T 

slot,k 

B =T 

add 

+ 6 

. 

c0 BW Ndbps,k 

 

The parameter c 0 is the velocity of light. The first two terms of (1) stem from 

USAP part and the third term is the transmission time for a PHY layer packet 

with B bytes corresponding to the WLAN 802.11n specification. T slot,k is valid 

either for a payload or a control time slot. Equation (1) can be transformed such 

that the maximum number of transmittable bytes B k in a single data slot can be 

calculated: 

 

1 BW 

D 

max 

B 

k 

= Tdataslot Tadd 6 Ndbps,k 

22 . 

8 

 

80 c 

(2) 

 

 

 

0 

The next step is the estimation of T dataslot . USAP has a strict separation of payload 

and control data. A frame with duration T frame is divided by mini slots for USAP 

control information and by data slots for the payload. The number of available mini 

slots in a frame is given by N bootstrap,slot . Similarly, the number of data slots is given 

by the sum of broadcast slots N broadcast,slot and reservation/standby slots N res/stby,slot . 

Thus, the duration of a single data slot T dataslot can be computed by: 

(1) 

Tframe N bootstrap,slot 

T slot,k0 

(B20) 

T 

dataslot 

= . 

N N 

broadcast,slot 

res/stby,slot 

(3) 

For USAP in each mini slot, B = 20 bytes with 16 bytes for USAP control 

and 4 bytes for the code redundancy check needs to be transmitted. T slot,k=0 can be 

calculated from (1).


93 

Table III. Number of data bits per symbol dependend on the WLAN 802.11n 

modulation and coding schemes [9, 10] 

Index k Modulation Coding N dbps,k 

0 BPSK 1/2 26 

1 BPSK 3/4 39 

2 QPSK 1/2 52 

3 QPSK 3/4 78 

4 16-QAM 1/2 104 

5 16-QAM 3/4 156 

6 64-QAM 2/3 208 

7 64-QAM 3/4 234 

8 64-QAM 5/6 260 

The data rate R above the MAC layer can then be computed by: 

Nbroadcast,slot 

Nres/stby,slot 

R 1 B 

mean, 

(4) 

T 

frame 

whereas ρ is the average slot utilization of the data slot assignment of USAP, 

ε is the packet error rate and B mean is the assumed mean number of bytes in a packet. 

As an example, the following parameter values are chosen for an initial estimation 

of the data rate: BW = 5 MHz, D max = 50 km, T add = 1 ms, ε = 0.05, ρ = 0.6 

and B mean = B k=1 . Corresponding to [7, 8] the further parameter values are chosen 

as: T frame = 0.125 sec, N bootstrap,slot = 13, N broadcast,slot = 2 and N res/stby,slot = 8. With 

the given parameter values T dataslot can be calculated from (3) together with (1) 

to 10.71 ms. From (2) and Table III B mean = B k=1 is calculated to 2873 bytes, and, 

together with (4), the data rate R is gained to 1.0235 Mbps. This data rate (above 

the MAC layer) can be seen as an initial value which could be adequate for a military 

WBWF. It should be emphasized that the corresponding payload data rate will be 

smaller and the corresponding link data rate will be remarkably higher. 

B. Range 

The original WLAN 802.11n system is not specified to reach high ranges. 

This disadvantage is part of restriction to the transmission power of at most 

100 mW in the 2.4 GHz RF Band, 200 mW in the 5.1 GHz to 5.3 GHz RF band and 

1000 mW in the 5.4 GHz to 5.7 GHz RF band (European specification). However, 

this disadvantage could be eliminated by use of higher transmit power and by 

operating in the military RF bands with lower center frequencies. Due to the fact 

that higher transmit power enlarges the total energy consumption of the system, 

further improvements of critical modules of the PHY layer, e.g. those for detection,


synchronization or modulation and coding will be helpful to compensate somewhat 

the higher energy consumption. 

As a first step, the well-known preamble coding and detection algorithm 

from Schmidl & Cox [11] is implemented and analyzed. This algorithm is applicable 

for WLAN 802.11n. But, in the meantime Minn-Zeng-Bhargava published 

a more powerful preamble coding scheme [12], with which better symbol timing 

estimations at the receiver side can be achieved. At present, this proposed method 

is adapted for use together with the new WBWF. 

As a second step, an alternative header coding scheme is analyzed. The WLAN 

802.11n uses a BPSK modulation and a convolutional code rate of 1/2 with generator 

polynomial G = [171 133] for header coding. With this configuration, 

the 42 header bits are mapped to two symbols with 48 subcarriers for each symbol. 

As an alternative, the use of a QPSK modulation and convolutional coding rate 

of 1/4 with generator polynomial G = [171 153 135 127] is considered. Both configurations 

produce the same total coding rate of 42/96. In an AWGN environment 

the new configuration outperforms the classical configuration by around 0.4 dB 

as shown in Fig. 3. The use of iteratively based channel coding schemes like turbo 

or low density parity check schemes would not result in performance improvements 

due to the very short block length of a header with only 42 bits. 

Figure 3. Bit Error Rate (BER) and Header Error Rate (HER) dependend on E b /N 0 

for two schemes for header modulation and coding; coding rate 42/96


95 

C. Mobility 

Another important goal in designing a military WBWF is the mobility capability. 

WLAN 802.11n is developed only for walking pace. Therefore, some modifications 

and enhancements, and, also for the simulations, the use of an appropriate 

channel model are required. It is decided to use the ITU models pedestrian A and 

vehicular B [13] which were officially employed together with the WiMAX development. 

For taking in account high platform mobility, a channel response with 

remarkable values for delay spread, Doppler spread and Doppler shift has to be 

provided. Another designing aspect is the choice of the OFDM guard interval 

length. It has to be chosen in accordance to the largest expected path delay. WLAN 

802.11n uses short guard intervals in terms of 1/4 or 1/8 of a symbol length. For 

the case of 1/4 symbol length duration, the resulting time durations are 3.2 µs, 1.6 µs 

or 0.8 µs for transmission with those symbol rates which are related to the channel 

bandwidths of 5, 10 or 20 MHz. But, since here ITU vehicular model B is used, 

in which a maximum delay of 20 µs is pre-determined, the guard interval length 

of the new WF has to be enlarged to that value. 

VII. SCA based implementation of WLAN IEEE 802.11n 

In parallel to reviewing the WF, its implementation on a simulation platform, 

according to the SCA standards, has to be prepared. Various SCA development 

tools are available in the market with which such implementations can be accomplished. 

For preparing the implementation of a newly designed WBWF, as a first 

step, the general SCA based implementation of WLAN 802.11n is accomplished. 

A. Modeling – building blocks 

The model we propose is depicted in Fig. 4 and contents the following modules. 

The four green framed boxes on the top are those of the WF in which signal 

processing functionality can be added. They are the SCA based resources. The blue 

framed boxes below represent the services offered by the SDR platform which a WF 

can use. They emulate all the SCA based devices, e.g. the serial port, the sound 

card (not shown in Fig. 4) etc. The red framed box in the bottom center represents 

the crypto module, which is bypassed in our current simulation environment. 

The boxes are individually described as: 

• DataReadService is responsible for reading and writing data to and from 

the SerialPort Device. It also contains the functionality of using push-totalk, 

i.e. the communication starts when a particular key is pressed. 

• RedPayloadProcessing (RPP) handles the payload and control data on the red 

security side. 

• BlackPayloadProcessing (BPP) handles the payload and control data on 

the black security side.


• Signal-in-Space (SIS) contains the actual PHY and MAC signal processing 

code of the waveform. The C++ code of the waveform which is compiled 

as a static library (e.g.: WLAN signal processing) is integrated within 

the C++ code of this SCA resource. 

• SCA Devices Platform (blue framed boxes at bottom) is a combination 

of all the SCA Devices used, see Fig. 4. It includes also further devices like 

the sound card etc. 

We focus on PHY and MAC which correspond to SIS Resource and BPP 

Resource, respectively. But for simplicity, we keep it together in the SIS Resource 

in the current implementation. In future, a separation between PHY and MAC will 

be accomplished. Also, a subdivision of SIS will be required to run parts of it on 

GPP, DSP and FPGA. But currently, everything is kept working on a GPP. 

B. Payload workflow 

Fig. 4 shows the workflow of the payload data going through the building 

blocks of the SCA based model of the WF. In the first step, the data to be 

transmitted is given to the arrangement using the SerialPort Device. Then it is 

read by the DataReadService Resource using the DataRead functionality implemented 

in it. Then the data is given to the RPP Resource from where it is given 

to the SecurityAdapterRed Device. The data from there is given to the crypto 

module. From the crypto module it is committed to the SecurityAdapterBlack 

Device. The BPP Resource then takes the data from the adapter and passes it on 

to the SIS Resource. The actual signal processing of the data takes place in this 

component. For the present, the WLAN signal processing functionality is added 

into this resource. The processed data are then given to the Transceiver (Trx) 

Device which transfers it out to the antenna. In the case of receiving information, 

the data flow in the reverse direction. 

Figure 4. Building blocks for SCA implementation


97 

C. Separation of red and black security sides and between payload 

and control data 

the implementation of the WF is done keeping in view the concept of red and 

black security separation. This means that the system can be divided into two parts: 

• Red side: It handles unencrypted information usually from devices like 

keyboard, audio card etc. 

• Black side: It handles the information in encrypted form and uses it for 

signal processing. 

In addition to this concept it is also recommended that there should be a clear 

separation between the payload and control information. 

• Payload is also called user data. It is that part of the transmitted data which 

is the fundamental purpose of the transmission. 

• Control information provides the control data the network needs to process 

and deliver the user data. 

Keeping in view to the above mentioned separations, all the building blocks 

on the red side of the system are implemented with a clear payload and control 

separation. 

D. Defining the interfaces 

The interfaces between the SCA based resources and devices, which handles 

the payload and control data, have to be defined beforehand. 

VIII. Conclusions 

After a short glimpse on tactical communications, followed by remarks on 

modern civil WFs already in use or tested by military forces, several requirements 

on WBWFs for military use are outlined. With the help of only a few significant 

assessment criteria, three modern civil WBWF systems are short listed as candidates 

for further investigations. In a second assessment step, WLAN IEEE 802.11n is finally 

selected, and first critical modules are pointed out for modifications or enhancements. 

These modules are: the CSMA/CA multiplex method, the preamble coding 

and the header coding. The most critical part, the CSMA/CA multiplex method, 

could be exchanged by a TDMA based method. Here, USAP is observed closely, 

and a first estimate for the expected data rate above the MAC layer is presented. 

Furthermore, an enhanced preamble coding with the aim to improve the detectability 

is considered. Finally, a modulation and coding configuration scheme to 

improve the header coding and therefore decrease the header error rate is proposed. 

In parallel to reviewing the new WF, its implementation on an appropriate simulation 

platform with SCA conformity is prepared.


In future, more critical modules will be investigated, and, if necessary, modified 

or enhanced. All modules, those taken unchanged from WLAN IEEE 802.11n 

and the modified or enhanced ones shall be integrated in appropriate simulation 

and development platforms to develop a new military usable WBWF. 

References 

[1] S. Couturier et al., “Evaluation of wireless civilian communication systems for 

military applications”, MCC 2011, Amsterdam, The Netherlands, October 2011. 

[2] R. Pengelley, “UK strives for joined-up communic ations networks”, Jane’s International 

Defense Review, April 2011. 

[3] M.J. Ryan, M.R. Frater, “Tactical communications for the digitized battlefield”, 

Artech House, 2002. 

[4] “Armada compendium, tactical radios 2010”, Supplement to Armada INTERNATIONAL, 

issue no. 3 (volume 34), June/July 2010. 

[5] Universal Multi-Media Link, Data Sheet, Thales Defense Solutions and Services, 2009. 

[6] R. Pengelley, “Digital delights: Commercial wireless commu nications on 

the battlefield”, Jane’s International Defense Review, September 2011. 

[7] C.D. Young, “USAP: A Unifying dynamic distributed multichannel TDMA Slot 

Assignment Protocol”, Proc. of IEEE Milcom, vol. 1, October 1996. 

[8] C.D. Young, “USAP multiple access: Dynamic resource allocation for mobile multihop 

multichannel wireless networking”, Proc. of IEEE Milcom, November 1999. 

[9] IEEE Standard for information technology – Telecommunications and information 

exchange between systems – Local and metropolitan area networks – Specific 

requirements, Part 11: “Wireless LAN Medium Access Control (MAC) and Physical 

layer (PHY) specifications”, 2007 (WLAN 802.11a/b/g). 

[10] IEEE Standard for information technology – Telecommunications and information 

exchange between systems – Local and metropolitan area networks – Specific 

requirements, Part 11: “Wireless LAN Medium Access Control (MAC) and Physical 

layer (PHY) specifications”, Amendment 5: “Enhancements for higher throughput”, 

2009 (WLAN 802.11n). 

[11] T.M. Schmidl, D.C. Cox, “Robust frequency and timing synchro nization for OFDM”, 

IEEE Trans. on Com., vol. 45, no. 12, pp. 1613-1621, December 1997. 

[12] H. Minn, M. Zeng, V.K. Bhargava, “On timing offset estimation for OFDM systems”, 

IEEE Com. Let., vol. 4, no. 7, pp. 242-244, July 2000. 

[13] ITU-R M.1225, “Guidelines for evaluations of radio transmission technologies for 

IMT-2000”, 1997.

Experimental Performance Evaluation 

of the Narrowband VHF Tactical IP Radio 

in Test-Bed Environment 

Edward Golan, Adam Kraśniewski, Janusz Romanik, 

Paweł Skarżyński, Robert Urban 

Radiocommunications Department, Military Communication Institute, Zegrze, Poland, 

{e.golan, a.krasniewski, j.romanik, p.skarzynski, r.urban}@wil.waw.pl 

Abstract: This paper evaluates the efficiency of IP packets transmission in narrowband networks 

based on the RRC 9210 tactical radios. Based on measurements performed in the laboratory test-bed 

environment, we examined the user throughput for selected length of UDP datagrams under AWGN 

channel. We determined BER characteristics vs. channel attenuation. We also found the thresholds 

of BER, when the data rate of the radio interface drops automatically as a result of the increased channel 

attenuation. We mapped the BER to the data rate of the radio interface. We found that the maximum 

throughput offered to the user amounts to 9,1 kbit/s. We also noticed, that the optimum size of UDP 

datagrams is equal to 512 bytes. Based on these results, we assessed the efficiency of the RRC 9210 

tactical radios in the IP-mode. 

Keywords: Tactical radio, tactical VHF communication, BER characteristics, user throughput 


The rapid growth of needs of the C4I systems for the information exchange 

enforced specific requirements for the tactical communications systems. As a result, 

military solutions within the scope of communications employ broadband technologies, 

mostly based on the IP protocol. This trend also applies to mobile wireless 

systems, which are essential mean of communications in the changing and unpredictable 

battlefield environment. The effort of researchers and military engineers 

is heavily oriented towards IP radio at the tactical level (Tactical Internet) [1,2,3] and 

also providing new services, e.g., BFT (Blue Force Tracking) and RFT (Red Force 

Tracking). Broadband technology enables to get an increased network throughput, 

better reliability and to offer wide range of services. If the network efficiency is high 

enough, the set of provided services is almost the same as in the wire network. 

Finally, the end user can get full access to the network resources from any place, 

using cable connection or wirelessly. This leads to the unification of the interfaces 

and applications and thus enables to get high level of interoperability.


On the contrary, there are narrowband radios that can operate in IP mode. 

For example, Polish Land Forces are equipped with tactical VHF radio RRC 9210, 

which offers throughput reaching up to 19,2 kbit/s under favorable propagation 

conditions [4,5,6,8]. RRC 9210 radios have Ethernet 10/100BaseTX interface. 

The latest firmware enables to operate in IP mode. 

The advantage of narrowband radios is to get long-range links and stable connections. 

However, the question about the efficiency of the narrowband IP packets 

transmission seems to be justifiable. Until now, this problem was not discussed 

widely in the literature. This may result from the fact, that manufacturers do not 

provide detailed information of the implemented mechanisms and network performance. 

Usually, the information is limited to the simple statement, that the Ethernet 

interface and IP protocol stack are built-in the radio. From the system designing 

point of view it is important to assess properly the network efficiency [7] and to 

plan deployment of stations as well as the set of provided services. 

In the literature, there are many papers dedicated to the efficiency of the broadband 

wireless systems. A lot of important issues have been recognized, which may 

have strong influence on the efficiency of the IP transmission in narrowband wireless 

networks [9,10,11]. They include, e.g., the informational overhead resulting from 

the transfer of the data through the complete protocol stack. While transporting 

data from the application layer to the physical layer, headers coming from successive 

layers are added and consequently the size of the transmitted data increases 

significantly [8]. In the physical layer the management and control frames such 

as RTS, CTS or ACK are exchanged, which also limits the available bandwidth. 

Moreover, apart from the user data, additional information is transmitted, e.g., synchronization, 

link test. 

If the TCP transmission is considered, then the size of the packets amounts to 

several hundred of bytes. In the physical layer packets are fragmented and typically 

are send as a few frames. This process has a significant impact on the transmission 

delay, especially in case of errors and retransmissions. In addition, one of the critical 

parameters of the TCP protocol is Timeout. The value of the Timeout should be 

adjusted to the narrowband channel characteristics. 

Taking into account issues mentioned above, the detailed test of VHF radio 

are strongly required. Such tests will help to assess the efficiency of IP packets 

transmission in destructive military environment. The test results will provide 

the necessary information about the rules of the radio configuration, the network 

capacity, recommended deployment of nodes and set of offered services. Moreover, 

it will be possible to determine the typical throughput offered to the user. 

The paper is organized as follows. In Section II the narrowband VHF communication 

was characterized. Section III describes the IP modes of RRC 9210 

radio. In Section IV the test scenario and assumptions were presented, while 

in Section V test results are discussed. Section VI contains conclusions and Section 

VII future work.


101 

II. Narrowband VHF communication 

Fast development of technologies to support reliable data transmission through 

radio channels was caused by increasing demands on services quality offered by 

wireless communication systems. It is not easy to realize such expectations because 

of transmission channel, which is changeable in a random way. Signal sent from 

transmitter can reach receiver by multiple propagation paths at different delays 

as a result of reflection from terrain obstructions. It causes random fluctuations 

in the received signal’s amplitude and phase briefly called multipath fading, which 

results in burst errors in received signal [9]. This phenomenon depends on type 

of environment and radio mobility, thus location can have important influence on 

signal reception. 

Transmission channel of tactical system based on VHF radio is narrowband 

(usually 25 kHz) and is characterized by flat fading. Such fading occurs when coherence 

bandwidth is greater than signal bandwidth and causes signal suppression [10]. 

VHF transmission is very suitable for short distance terrestrial communication, 

only slightly farther than the line of site from the transmitter to the receiver. 

The requirements that must be fulfilled by modern military communications 

systems indicate, that the development of such systems is focused on broadband 

technologies, mostly based on IP protocol. Classical tactical radio networks are 

organized with the use of narrowband transceivers operating in the VHF and HF 

frequency range. In such case radio communications between multiple correspondents 

is provided in omni-directional mode and integration with wired systems 

is realized by means of Radio Access Points (RAPs). 

Implementation of TCP/IP protocol stack in tactical radios caused positive 

changes in operation of radio networks. Transmission reliability increases, because 

IP radio network can operate despite partial destruction or jamming and does not 

need integrating devices such as RAPs. 

III. IP modes of RRC 9210 tactical radio 

Polish Land Forces are equipped with tactical VHF F@stnet family radios, 

which includes the RRC 9210 manpack radios with maximum power of 10W and 

vehicle version called RRC 9310AP with output power increased to 50W. These 

radios have Ethernet 10/100BaseTX interface and may operate IP mode if the latest 

firmware is used. From these reasons RRC 9210 manpack radios were selected to 

evaluate the performance of IP packets transmission in VHF channel. 

RRC 9210 tactical radio offers two IP modes: 

• IP-MUX (Simultaneous Voice & Data over IP mode). 

• IP PAS (data packet over IP mode). 

IP-MUX mode replaced older MUX mode, which was designed to transmit/ 

broadcast voice and IP data on the same radio channel. It offers simplex data


transmission or transmission in TDMA triggering mode with the data rate up 

to 4,8 kbit/s. Synchronization of hopping is made by station, which play NCS 

role in the network. Though, core synchronization type assumes using GPS. 

Switching of the station to IP-MUX mode is initialized by an operator. Addition 

as well as removal of SUB stations can be commanded at any time via NCS station. 

The automatic selection of routing tracks is provided in the latest version 

of the firmware. To provide interoperability with previously delivered radios, 

a firmware update is necessary. 

IP PAS mode was designed only for data transmission with the data rate up to 

19,2 kbit/s. The hopping synchronization is triggered in distracted mode without 

NCS station (there is no primary station in the network). Retransmission of data 

can be done with maximum 5 hops, Fig. 1. In that case, one radio serves also as relaying 

node. Such functionalities provide significant extension of network coverage, 

although, it costs notable increase of the transmission delay. Therefore, this mode 

is dedicated to non-real time services only. 

Figure 1. RRC 9210 in IP PAS mode 

Figure 2. Test-bed configuration


103 

Network topology to which radio station belongs is updated automatically. 

As a result each station “knows” its surroundings. If there is lack of direct connection, 

then the most optimum neighboring node is selected automatically to play 

a role of a retransmitting station. 

IV. Test scenario and assumptions 

The aim of the tests was to assess the efficiency of IP packet transmission of RRC 

9210 radio in IP mode. The test-bed consisted of two radio terminals configured 

in a point-to-point connection, which is shown in Fig. 2. The first step in the test 

was to select one of two IP modes of the radio. After preliminary test the IP-PAS 

mode was chosen, as it allocates all of the available bandwidth for data transmission 

only. Therefore, this mode allows to measure the available bandwidth in a simple 

point-to-point connection, Fig. 2. 

Figure 3. BER vs. channel attenuation 

All tests were performed in laboratory environment, which guarantee repeatability 

and stability of experiments. AWGN channel was used as a simulator 

of transmission channel. It does not generate burst errors, typical for real VHF 

channel, but only errors distributed at a uniform rate. However, to determine BER 

characteristics, such test-bed configuration was acceptable. In the next step of tests, 

the model of VHF channel with multipath fading will be used. 

The channel attenuation was changed from 130 dB to 140 dB with the step 

of 1 dB. The output power was set to +27 dBm in each radio. Thus, the signal level 

on the receiver site was equal to: 

• –103 dBm – maximum signal level. 

• –113 dBm – the level of sensitivity.


Figure 4. User throughput 

Radio was configured in the IP-PAS mode with adaptive data rate and correction 

code. End-points were the terminals with specialized software to generate 

and analyze an IP traffic. 

Bit error rate (BER) parameter was chosen as a channel quality metric. Therefore, 

BER was measured for each value of attenuation level. Fig. 3 presents BER 

characteristics vs. channel attenuation. 

Next step of experiments was to determine the available bandwidth vs. the channel 

quality. To measure the maximum user throughput, UDP protocol was used, 

as it does not need acknowledgements and retransmissions [11]. If TCP protocol 

were chosen, measured throughput would not be easy to analyze, because of additional 

acknowledgements in transport layer. 

To sum up the scenario, the quality of VHF channel was changed by selection 

of the attenuation level. The source terminal generated UDP datagrams of 512 Bytes 

size to achieve the traffic of 12 kbit/s. 

V. Test results 

Fig. 4 shows the user throughput measured for different BER levels. For each 

period of 20 seconds the throughput was measured and then the average value 

was calculated. After each 10-minute period of measurement, the level of BER 

was changed by selection of the channel attenuation level. In Fig. 4 'n' denotes 

the number of consecutive 20-second periods. The maximum user data throughput 

amounts to 9,10 kbit/s on average. In that case, RRC 9210 operates with maximum 

data rate. Because the channel is of high quality (BER from 4,02*10 -7 to 8,45*10 -6 )


105 

the redundancy of correction codes is low. When the BER achieve the level of 8,18*10 -5 , 

then the user data throughput drops to 3,59 kbit/s. This is a result of applying 

of correction codes with higher redundancy. When the channel quality is worsen 

again and BER achieves the level of 1,88 *10 -2 , then the user throughput drops to 

1,49 kbit/s. In such a situation, radios detect high level of errors and apply stronger 

correction code. Further increase of the channel attenuation makes that the transmission 

is impossible and finally connection between two radios is simply lost. 

VI. Summary 

In this paper the efficiency of IP packets transmission in narrowband networks 

based on the RRC 9210 tactical radios was evaluated. Based on measurements 

performed in the laboratory test-bed environment, the user throughput was determined 

under AWGN channel. The maximum throughput offered to the user 

amounts to 9,10 kbit/s in point-to-point connection. The minimum throughput 

amounts roughly 1,5 kbit/s. The thresholds of BER when the data rate of the radio 

interface drops automatically were determined. These BER levels allow to estimate 

the network throughput in a given propagation conditions. 

These results will be helpful to assess the channel quality in a real environment 

and to predict user throughput. Finally it will enable to define the set of offered 

services. Results of experiments presented in this paper allow to presume, that 

the network will not be efficient for real time services. Only non-real time services 

will be offered, like chat, e-mail, short message transfer or FTP. However, the range 

of provided services in a real environment needs to be confirmed by additional 

experiments. 

VII. Future work 

For future work it would be interesting to examine the efficiency of the IP 

transmission in real environment. Based on the test results, we are going to investigate 

the network capacity, the set of offered services and also the rules of device 

configuration for IP-mode. 

Furthermore, we would like to devote attention to the aspect of the relaying 

and test the efficiency of the network, when nodes operate on large area.


References 

[1] Collective work edited by M. Amanowicz, “Zaawansowane metody i techniki 

tworzenia świadomości sytuacyjnej w działaniach sieciocentrycznych” (Advanced 

methods and techniques for creating situational awareness in the network centric 

activities). 

[2] B. Marczuk, “Radiostacje szerokopasmowe” (Broadband radio stations), The Land 

Forces Review 03/2006. 

[3] L. Stypik, “Cyfrowe wsparcie pola walki – przyszłość czy rzeczywistość” (Digital 

battlefield support – the future or reality), The Land Forces Review 10/2010. 

[4] M. Gruszka, “Taktyczny Internet w praktyce” (Tactical Internet in practice), The Land 

Forces Review 01/2011. 

[5] E. Golan, A. Kraśniewski, J. Romanik and P. Skarżyński, “Ocena potrzeb 

i możliwości wykorzystania szerokopasmowych radiostacji osobistych i pokładowych 

na szczeblu taktycznym” (Assessment of needs and possibilities of using the broadband 

on-board and personal radio stations at the tactical level), Journal of KONBIN, Safety 

and reliability systems, Warsaw 2011. 

[6] D. Pawłocki, “ W stronę transmisji danych z protokołem IP” (Towards a data 

transmissions including the IP protocol) The New Military Technology 09/2011. 

[7] J. Dudczyk, “Optymalny dobór parametrów radiostacji szerokopasmowej żołnierza” 

(Optimal selection of parameters of the soldier’s broadband radio station) NTV 

no. 9/2011. 

[8] J. Romanik, P. Gajewski and J. Jarmakiewicz, A Resource Management Strategy 

to Support VoIP across Ad hoc IEEE 802.11 Networks, ThinkMind Digital Library, 

Proceedings of The Fourth International Conference on Communication Theory, 

Reliability and Quality of Service, April 17-22, 2011, Budapest, Hungary, pp. 15-21, 

ISBN 978-1-61208-005-5. 

[9] R. Urban, “Metoda określania średniej długości paczek błędów w kanale UKF w oparciu 

o analizę zmian wartości BER” (The method of determining mean length of error 

bursts in VHF channel based on fluctuation analysis), Biuletyn WAT, Warszawa 2007, 

vol. LVI, s. 433-440. 

[10] B. Sklar, “Digital Communications: Fundamentals and Applications” (2nd Edition), 

Prentice-Hall, Upper Saddle River, New Jersey 2004. 

[11] W. Wysota and J. Wytrębowicz, “End to End QoS Measurements of TCP 

Connections”, PPAM’07 Proceedings of the 7th international conference on Parallel 

processing and applied mathematics, Springer-Verlag Berlin, Heidelberg 2008.

Hybrid Error Detecting and Correcting System 

Using Hardware Associative Memories 

Ion Tutănescu, Constantin Anton, Laurenţiu Ionescu, 

Gheorghe Şerban, Alin Mazăre 

Faculty of Electronics, Communications and Computers – University of Pitesti, Romania, 

{gheorghe.serban, constantin.anton, laurentiu.ionescu, ion.tutanescu, alin.mazare}@upit.ro 

Abstract: This paper presents a solution of design and implementation of a hardware error 

correction and detection system using associative memories. This type of memory allows search 

of a stored binary value, having as an input data a partial (or modified) amount of this value. 

This property can be used in communication, for detection and correction of errors. In our 

experiments the encoder just associate message with corresponded word code, which is sent 

to communication channel. The decoder can associate back the word code received from communication 

channel with the message, even if the received word code has errors. This is due 

to associative memory recognition ability. Experimental results obtained were compared with 

performances of other hardware systems. 

Keywords: associative memories, error detecting and correcting codes, Field Programmable Gates 

Array (FPGA), Hybrid Automatic Request (H-ARQ) 


Errors’ correction and detection is a very important feature in modern 

communications. There are several methods for error correction and detection. 

BCH codes (Bose, Chaudhuri, and Hocquenghem) are widely used in communication 

networks, computer networks, satellite communication, magnetic and 

optic storage systems. 

In this paper we present a hybrid ARQ (Automatic Request) and FEC (Forward 

Error Correction) solution using hardware associative memory. 

BCH codes operate over finite fields or Galois fields. BCH codes can be defined 

by two parameters that are: length of code words, n, and the number of errors to 

be corrected, t. 

The BCH codes are a class of cyclic codes whose generator polynomial is a product 

of distinct minimal polynomials corresponding to 

α, α 2 , …, α 2t , 

GF 2 m is a root of the primitive polynomial g(x)[1]. 

where


An irreducible polynomial g(x) of degree m is said to be a primitive if only 

if it divides polynomial form of degree n, for no n less than 2 m 1. In fact, 

m 

2 1 

every binary primitive polynomial g(x) of degree m is a factor of x 1 [2]. 

In our application we present two Hybrid ARQ solutions. First use a polynomial 

of 10 degree, which can be used to correct 2 erroneous bits and detect 5, 

in a word code of 20 bits size for 10 bits size message: 

10 7 6 4 2 

gx ( ) x xxxx 1 

(1) 

This polynomial is used to generate word codes with 5 Hamming distance. 

Secondary, we use a polynomial which generates word codes with 10 Hamming 

distance, which can correct 4 erroneous bits from 5-bits size messages and 

20 bits size word code. This solution has a very high correction capacity (from 

5 bits message can correct 4 erroneous bits) and can detect all errors from message: 

15 14 13 12 10 8 5 4 

g( x) x x x x x x x x 

1 

(2) 

Field-Programmable Gate Arrays (FPGAs) have become one of the key digital 

circuit implementation media over the last decade [3]. One bit patterns will 

produce operational circuits and can be used in many areas like the communication 

systems. Our hardware scheme is based on polynomial generator for errors 

detection and correction. 

FPGA circuits represent a compromise between circuits with microprocessor 

and ASIC circuits (Application Specific Integrated Circuits) [4]. First, they 

present flexibility in programming, called here reconfiguration, which is a feature 

for microprocessors. 

Even if FPGA cannot be programmable while operation, they can be configured 

anytime is needed, having a structure based on RAM programmable machines, as we 

see in Figure 1. On the other hand, they allow parallel structures implementation, 

with response time less than a system with microprocessor. 

Figure 1. Communication system with hardware errors’ detection and correction 

The system proposed in this paper is based on the use of reconfigurable 

FPGA circuits for hardware implementation of error detection and correction 

algorithms.


109 

Our new design system can be integrated inside a multilayer protocol communication. 

Thus, hardware module takes the tasks from others components 

of communication system (for example from the computers). 

In the next section we present the hardware circuits for encoder and decoder. 

The first presented solution is a simple binary decoder structure while the second 

is based on a hardware linear associative memory. 

Section 3 present experimental results obtained after implementations for 

both H-ARQ solutions. The implementation is done on a FPGA circuit. 

II. ENCoder and decoder 

We designed the encoder and decoder using dedicated binary circuits. Thus, 

the encoder will be attached as a physical device to any system which transmits 

dates to a communication channel while the decoder will be attached to the system 

which receives the dates. 

In this section we describe the operation of the two systems and our design 

method proposed for them. 

A. Encoder 

The operation of encoder is illustrated in Figure 2. 

Figure 2. Encoder operation flowchart 

According to the first solution, the message is received from the emitter system. 

The message can be received serially (Ethernet or USB) or in parallel if the emitter 

use a protocol defined by itself. The experiments were performed using 5-bits 

size message words (thus can be encoded letters from Latin alphabets) and 10-bits 

message words (which can contain extended ASCII charset).


Each message has associated a word code. To 10 bits-size message word we 

have a Hamming distance 5, which means that we can correct 2 erroneous bits and 

can detect 5 erroneous bits. We generate word codes for all 10-bits size combinations. 

Thus we have 2 10 locations in memories to encoder and decoder. 

In fact, the communication encoder is designed as a simple binary decoder. 

The messages are generated in a linear increasing manner and word code is obtained 

by multiplying message polynomial with generator matrix (see Figure 3). 

The word codes are equidistant in Hamming space (distance is 5). This orthogonally 

property is required in order to have the same error correction and 

detection capacity to each received word code. 

Figure 3. Association between message and BCH word code to the encoder 

The second solution uses 5-bits size message, which are encoded with 20-bits 

size word code. The Hamming distance between word codes is 10 so we have a correction 

capacity of 4 erroneous bits. 

The encoder is designed in the same way as in previous solution and the number 

of location is 2 5 . Thus, in both cases, we have a very reduced hardware structure 

for encoder. 

B. Decoder 

Decoder, on the other hand, has a more complex structure. This is because 

it also has the error detection and correction function. The decoder can match 

received words with expected word codes but, in some cases, mistakenly received 

word is not found in any of the stored ones. 

Error correction would actually identify the correct word code. This involves 

finding the closest word in memory. Once identified, it will determine what message 

is received. 

Tasks to be performed at the receiver are presented in the following chart 

(see Figure 4).


111 

Figure 4. Decoder operation flowchart 

First, the word code is taken from the communication channel. In our 

experiments the word code is of 20 bits size, for both solutions. This word code 

is then compared with all stored words. It is a binary comparison. Finally, we 

find out which are the erroneous bits. Each memory cell counts the erroneous 

bits and we can determine minimal value from all locations. The location 

which contains an word code with minimum erroneous bits is associated with 

the correct message. 

The associative memory has a more complex structure that the encoder (see 

Figure 5). Thus, each location consists in a binary comparing circuit, as it is illustrated 

in Figure 4, and in a bits counter (compressor). This compressor count 

“1” bits from the comparator output. There’s also a register in which are stored 

word code and message.


Figure 5. Structure of location in associative memory to the decoder 

The message with erroneous bits will be transmitted as a 10 or 15 bits size 

data pair (5 or 10 bits message and 5 bits number of erroneous bits – from 20 bits 

of received word code). All these data pairs, at each location separately, will be 

compared to determine the minimum. For this operation we use a combinational 

network which gets the minimal value of number of erroneous bits, as is illustrated 

in Figure 6. 

Figure 6. Minimum computation circuit 

The data pairs from the associative memory are applied to left side (x axis). 

To y axis we find the data pairs with a minimal value of erroneous bits, which travel 

to bottom side of circuit.


113 

The minimum cell (Min), which enters in the structure of the circuit, compares 

the number of erroneous bits from two data pairs and selects only the data pair 

with the minimum number of erroneous bits. 

III. Experminental results 

Since it’s a hardware system used to detect and correct errors in communication, 

items that are considered for performance analysis of this system are given 

by response time, area occupied on silicon surface and communication speed. 

In our experiments, we tested two different solutions. The first solution encodes 

large message package (10 bits) and can correct a small number of erroneous bits 

(2 errors). The second solution encodes a smaller message and has a greater error 

correction capacity. Implementation was performed to a FPGA Xilinx Spartan 3 

family, a low cost family circuit (3-20$ per chip inside the family). A complete 

communication system will integrate both encoder and decoder on the same chip. 

We will analyze them separately. 

In our analysis we take in account different members of Spartan 3 FPGA family. 

The logical cell (Look Up Table, LUT) and the logical-routing cell (Slice) are 

the same structure for all members inside the family. The differences come from 

the number of logical and routing cells integrated on chip. 

For example, XC3S50 FPGA has 1728 logic cells and 768 logical-routing cells, 

while XC3S1000 has 17280 logic cells and 7680 logic – routing cells. Of course, 

the price reflects this difference. For a minimum cost we implement the systems 

to the smallest and cheapest circuits inside family. 

Thus, the communication encoder consists in a simple binary decoder. Its 

structure is very simple, in terms of hardware, occupying approximately 1% of all 

chip resources for a XC3S50 chip. So we use, for encoder of both solution XC3S50 

circuit (3 $/chip). 

Table I. Results from encoder and decoder implementation – synthesis report 

Method Circuit Area on silicon (XC3S) Response time 

H-ARQ 

w = 20, 

m = 10 

H-ARQ 

w = 20, 

m = 5 

Encoder 

Decoder 

Encoder 

Decoder 

a) for 10 bits of message 

b) for 5 bits message 

Used slices 10 – 1.3% capacity of XC3S50 

Used LUTs 17 – 1.0% capacity of XC3S50 

Used slices 6550 – 84% capacity of XC3S1000 

Used LUTs 11494 – 66% capacity of XC3S1000 


Used LUTs 15 – 0.87% capacity of XC3S50 


Used LUTs 812 – 47% capacity of XC3S50 

10.926 ns a 

24.048 ns a 

9.280 ns b 

21.573 ns b


The circuit simplicity implies a very small response time of 9.2 ns for each 

5-bits size message or 10.9 ns for each 10-bits size message (see Table I). So, when we 

use 5 bits communication means communication at 1 Gbps for encoder. The same 

speed is for 10 bits message. 

The decoder, more complex than the encoder, still has a simple structure 

consisting only of combinational circuits arranged in an array. Comparison operation 

between the input data into the associative memory (word code received from 

the communication channel) and the word code stored is performed in parallel 

for all locations. 

Therefore, circuit’s complexity is increased by the number of locations. To 

10 bits message solution we have a number of 1024 different location. This number 

is reflected in capacity of 84% allocated from XC3S1000 logical resources. The decoder 

implemented for second solution, with 5 bits per message but highest error 

correction capacity can be implemented in the smallest XC3S50. 

On the other hand, the response time for both decoders is approximately 

the same (24 ns – 10 bits, 22 ns – 5 bits). This happens because all the cells respond 

in parallel (this value doesn’t take in account the response time of minimum circuit). 

We have a communication speed for 10 bits message packets (encoded in 20 bits 

word code) of 416 Mbps and for 5 bits message (encoded in 20 bits word code) 

the speed is 227 Mbps. First solution makes correction for 2 erroneous bits and 

the second solution makes correction for 4 erroneous bits. 

Because of coding it is a reduction in speed of communication with 1/4 (5-bits 

message is encoded with 15-bits word code) or ½ (5-bit size message is encoded 

with 20-bits word code). For example, if communication is to 1 Gbps the real communication 

speed is 333 Mbps. However, in our approach, this reduction is not 

presented because of parallel computation of entire message word. 

Our system can be integrated in Ethernet communications, as additional 

services added to physical layer (see Figure 7). 

Figure 7. Hardware correction layer services by using our BCH encoder-decoder 

Interposition of this circuit in Ethernet communication system will automatically 

corrected 2-4 error bit locally. Thus, it provides error correction services for


115 

protocols in the highest layer. The TCP services will be significantly relieved of task 

that, in other circumstances, would have their back. To increase the communication 

speed, we can use a “secured” UDP communication, because we have already, 

from physical layer, error correction services. 

Compared with other structures, based on hardware implementation of error 

correction circuit, performance has improved in this version (see Table II, where 

SR-ARQ means Selective Repeat Automatic Repeat Request). 

Table II. Comparison with other hardware implemented error correction circuits 

Method 

Parallel computing, classic 

Hardware BCH SR-ARQ 

Associative memory ARQ 

(w = 20, m = 10) 

Associative memory A RQ 

(w = 20, m = 5) 

Correction (t) and detection 

(f) capability 

t = 1 erroneous bit 

f = 0 

t = 3 erroneous bits 

f = 7 erroneous bits 





Communi-cation speed 

149 Mbps 

87 Mbps 

416 Mbps 

227 Mbps 

Thus, we have a parallel computing circuit which corrects 1 erroneous bit, with 

no detection, presented in [5]. Because of parallel computing structure we have 

here a 149 Mbps communication speed. Another circuit uses a hardware hybrid 

BCH SR-ARQ error correction and detection algorithms [6]. 

IV. Conclusions 

Our solution presents a modern method concerning implementation of error 

correcting codes with associative memories. Associative memories allow a very easy 

design for error correcting codes and obtain good performances, both in detection 

and correction. Also, they provide a very high response speed because of parallel 

processing. 

Our method, presented in this paper, consist in storing a set of word codes 

and the message associated with them. The correction is achieved by comparing 

the received word code with all stored word codes and by selecting the nearest word 

code, in terms of Hamming distance. Also this method allows us detection of errors 

while the correction is performed, increasing computing speed. Our method 

improves the capacity of communication channel in Ethernet communication. 

In future works we will implement in FPGA error correcting codes with 

a higher power correction and detection. We will test these new error correcting 

systems in order to measure their throughput and to decide their capacity to work 

in real conditions.


References 

[1] M.Y. Rhee, “Error Correcting Coding Theory”, McGraw-Hill, Singapore, 1989. 

[2] S. Lin, and D.J. Costello Jr., “Error Control Coding”, Prentice-Hall, New Jersey, 1983. 

[3] J. Rose, S.D. Brown, R.J. Francis, “Field Programmable Gate Arrays”, Kluwer Academic 

Publishers, 1992. 

[4] T. Fogarty, J. Miller, and P. Thompson, “Evolving digital logic circuits on Xilinx 

6000 family FPGAs,” in Soft Computing in Engineering Design and Manufacturing, 

P. Chawdhry, R. Roy, and R. Pant (eds.), Springer: Berlin, 1998, pp. 299-305. 

[5] L. Ionescu, C. Anton, I. Tutănescu, A. Mazăre, G. Şerban, “Hardware 

implementation of BCH Error Correcting Codes on FPGA”, International Journal 

of Intelligent Computing Research (IJICR), vol. 1, Issue 3, ISSN 2042-4655, Published 

by Infonomics Society, p.148-153, September 2010. 

[6] G. Şerban, C. Anton, L. Ionescu, I. Tutănescu, A. Mazăre, “Implementation 

of a 64-bit hybrid SR-ARQ algorithm on FPGA”, Proceeding of International Conference 

on Applied Electronics, ISBN 978-80-7043-987-6, ISSN 1803-7232, IEEE Catalog 

Number CFP1169A-PRT, p. 337-340, Pilsen, 7-8 September 2011.

Concurrent Error Detection Scheme 

for HaF Hardware 

Ewa Idzikowska 

Poznań University of Technology, 

pl. M. Skłodowskiej-Curie 5, 60-965 Poznań, Poland, 

ewa.idzikowska@put.poznan.pl 

Abstract: HaF (Hash Function) is a dedicated cryptographic hash function considered for verification 

of the integrity of data. It is suitable for both software and hardware implementation. HaF has an iterative 

structure. This implies that even a single transient error at any stage of the computation 

of hash value results in a large number of errors in the final hash value. Hence, detection of errors 

becomes a key design issue. In the hardware design of cryptographic algorithms, concurrent error 

detection (CED) techniques have been proposed not only to protect the encryption and decryption 

process from random faults but also from the intentionally injected faults by some attackers. In this 

paper, we show the propagation of errors in the VHDL model of HaF-256 and then we propose and 

analyse some error detection schemes. In proposed CED scheme all the components are protected 

and all single and multiple, transient and permanent bit flip faults will be detected. 

Keywords: hash function, HaF, S-box, concurrent error detection, hardware redundancy, time redundancy, 

DWC 


A hash function H is a transformation that takes an input m and returns 

a fixed-size string, which is called the hash value h (that is, h = H(m)). Hash functions 

with just this property have a variety of general computational uses, but 

when employed in cryptography, the hash functions are usually chosen to have 

some additional properties, e.g. H(m) must be relatively easy to compute for any 

given m, one-way and collision-free [11]. Very important role of a cryptographic 

hash function is in the provision of message integrity checks, digital signatures, 

password storage and verification etc. 

Hash functions are computationally complex, and in order to satisfy the high 

throughput requirements of many applications, they are often implemented by 

means of VLSI (Very Large Scale Integration) devices. The high complexity of such 

imple mentations raises concerns regarding their reliability. There is a need to develop 

methodologies and techniques for designing robust cryptographic systems, 

and to protect them against both accidental faults and intentional intrusions and


attacks, in particular those based on the malicious injection of faults into the device 

for the purpose of extracting the secret information [3] and [4]. If an attacker 

deliberately generates a glitch attack, causing a flip-flop state to change or corrupt 

data values when they are transferred from one digest operation to another, even 

a single fault can result in multiple errors in the hash value computed. The severity 

of the problem necessitates detection of errors a key design issue. A digest round 

consists of several operations. Errors can creep in at any of these operations and 

can affect one or several bits at any of the operations in a digest round. As HaF 

is considered for use in security services, concurrent error detection is very important. 

This necessitates an analysis on the propagation of error from the point 

of origin to the output. 

CED has certain associated penalties such as hardware cost and the performance 

degradation due to interaction between the circuit and the detection logic, 

which need to be considered while designing the error detection circuit. The design 

goal of the CED is to achieve 100% error detection with minimal penalty. 

CED techniques involve redundancy in the form of hardware, time or information. 

A CED circuit based on hardware redundancy can for example duplicate 

the complete circuit. It means that hardware overhead is more than 100%. In time 

redundancy, the same hardware is used to perform both the normal computation 

and re-computation using the same input data [9]. The advantage of this technique 

is that it uses minimum hardware. The drawbacks of this technique are that it entails 

≥100% time overhead and it can only detect transient faults. In information 

redundancy technique, data are appended with additional bits and a coding scheme 

is used to detect errors. Coding techniques marginally increase the hardware as well 

as performance overhead. Combinations of the above techniques are also employed 

to minimize the overhead for CED [1] and [5]. 

In this paper the propagation of errors in HaF-256 is studied. We take into 

consideration single, transient as well as permanent faults injected at different 

stages of hash value computation. It is found that even a single error injected resulted 

in half the bits of hash value being in error and the errors are spread across 

the computed hash value. Next we focused on CED techniques and proposed error 

detection schemes to protect basic operations such as multiplication mod (2 n +1), 

addition modulo 2, addition modulo 2 n There is also presented schemes to protect 

S function, step function and round function of HaF-256. The proposed approaches 

are tested and the results are presented. 

This paper is organized as follows. Sec. II presents family HaF of hash functions. 

There is shown a method of one block processing, round function, operations 

of step function. In Sec. III error analysis is carried out to understand the effect 

of an error injected into the hash computation circuit. Error detection schemes, 

possible faults and faults models are described in Sec. IV. Simulation results are 

presented in Sec. V. An error detection scheme for the HaF circuit is proposed 

in Sec. VI. Concluding remarks are in Sec. VII.


119 

II. Family HaF of hash functions 

The family HaF is formed of three hash functions: HaF-256, HaF-512 and 

HaF-1024, producing hash values of the length equal to 256, 512 and 1024 bits, 

respectively. The general model for HaF is presented on Fig. 1 [2]. 

The original message m has to be formatted before hash value computation 

begins. After formatting the message m we have the message M. This massage 

is divided on blocks M 0 , M 1 , …, M k–1 . Each block M i is processed with the salt s by 

the iterative compression function φ. After processing all blocks we receive the hash 

value h(m) = H k as the result. 

The length of formatted message M should be a multiple of 16n bits. It means, 

that the length of the input block equals 16n bits, where n is a parameter depending 

of the hash value we want to obtain. The parameter n equals 16, 32 and 64 bits 

for HaF-256, HaF-512 and HaF-1024 respectively. We consider HaF-256 function 

in this article. It means, that n = 16. 

Figure 1. Model for hash function HaF [2] 

The block M i is processed in two rounds. The method of one block processing 

is depicted on Fig. 2. M i , H i and s are inputs for compression function. The parameter 

n indicates the length of the working variable A r , rÎ{0, 1, …, 15}. 

The round function (Fig. 3) has two inputs N i , H i , (H 0 = IV is an initial value) 

and two outputs N i * , H i * . Before processing in round #1, the block M i is modified,


N i = M i Å s, where s is a salt, |s| = 16n (Fig. 2). In the round #l (lÎ{1, 2}) four least 

significant bits of N i indicate the number of bits the string N i is rotated to the left 

(


121 

Figure 4. Step function F j [2] 

In the Fig. 4. the following notations are used: 

a ⊙ b – multiplication mod (2 n +1) of n-bit non-zero integers a and b, 

v


III. Error propagation 

Error propagation analysis is carried out to understand the effect of an error 

injected into the hash computation circuit. Error was injected in: 

• the inputs of a round function, 

• the inputs of step function. 

Experiments were conducted by injecting a single bit flip error in one of the inputs 

of round function or step function and obtaining the number of erroneous 

bits at the output of the same block. The errors were introduced at different bits 

randomly in every step and the number of bits that were in error was computed. 

As an example we show the propagation of error injected in first step of round #1 

in subblock A 13 . Instead value F9DD is the faulty value F95D. Output values A 0 , 

A 1 , …, A 15 after processing each of 16 steps of round function, are shown in Table I. 

The faulty values are depicted by bold, italic font. 

One faulty bit in the step #0 of round function causes about 43% faulty bits 

in the last, #15 step. Faulty bit in the input of round #1 of round function causes 

49% faulty bits in the output of round #2. 

This analysis helps us in choosing suitable error detection schemes. 

Table I. Error propagation in round function of HaF-256 

Faulty free input value A0, A1, …, A15; A13 = F9DD 

076A3663D2541F389E94CA4E1A759C92ED3282E55F31FC1514A6F9DDBA029ABE 

Faulty input value A0, A1, …, A15; A13 = F95D 

076A3663D2541F389E94CA4E1A759C92ED3282E55F31FC1514A6F95DBA029ABE 

Step 

Faulty outputs 

0 3663D2541F389E94CA4E1A759C92ED3282E598AFFC1514A6F95DBA029ABEDFA8 

1 D2541F389E94CA4E1A759C92ED3282E598AF0AFE14A6F95DBA029ABEDFA8C321 

2 1F389E94CA4E1A759C92ED3282E598AF0AFE530AF95DBA029ABEDFA8C3210741 

3 9E94CA4E1A759C92ED3282E598AF0AFE530AAEFCBA029ABEDFA8C3210741A47E 

4 CA4E1A759C92ED3282E598AF0AFE530AAEFC015D9ABEDFA8C3210741A47E1059 

5 1A759C92ED3282E598AF0AFE530AAEFC015D5F4DDFA8C3210741A47E105943F0 

6 9C92ED3282E598AF0AFE530AAEFC015D5F4DD46FC3210741A47E105943F0D7D2 

7 ED3282E598AF0AFE530AAEFC015D5F4DD46F90E10741A47E105943F0D7D2C084 

8 82E598AF0AFE530AAEFC015D5F4DD46F90E1A083A47E105943F0D7D2C084A3FF 

9 98AF0AFE530AAEFC015D5F4DD46F90E1A0833F52105943F0D7D2C084A3FF9977 

10 0AFE530AAEFC015D5F4DD46F90E1A0833F522C8843F0D7D2C084A3FF9977CA78 

11 530AAEFC015D5F4DD46F90E1A0833F522C88F821D7D2C084A3FF9977CA785AAE 

12 AEFC015D5F4DD46F90E1A0833F522C88F821E96BC084A3FF9977CA785AAE875B 

13 015D5F4DD46F90E1A0833F522C88F821E96B4260A3FF9977CA785AAE875B8AF7 

14 5F4DD46F90E1A0833F522C88F821E96B4260FFD19977CA785AAE875B8AF7D530 

15 D46F90E1A0833F522C88F821E96B4260FFD1BBCCCA785AAE875B8AF7D5307F1C


123 

IV. Error detection schemes 

Fault detection is achieved in a circuit by including redundancy in one form 

or the other. The selection of a suitable scheme depends on the type of errors to 

be covered, error models and the performance degradation acceptable in terms 

of hardware overhead and delay. The schemes based only on time redundancy are not 

capable of handling permanent faults. The simplest form of error detection scheme 

which can detect transient as well as permanent faults with hardware redundancy 

is the DWC scheme. The hardware overhead in this case is little >100% with very 

little additional delay, these can handle unrestricted error models. 

The earliest error detecting scheme is the parity bit scheme. It is a well-known 

fact that parity codes can detect all single bit errors and all errors with multiple 

odd erroneous bits. To improve the error detection capability and detect also even 

erroneous bits, multiple parity bits are required [7] and [8]. In HaF-256 first of all 

DWC scheme will be used. 

A. Faults models 

In our considerations we use a fault model wherein either transient or permanent 

faults are induced randomly into the device. We consider single and multiple 

faults. Faults are modelled as an 16-bit error vector E = {e 15 ,...,e i ,...,e 1 ,e 0 }, where 

e i Î{0,1} and e i = 1 indicates that bit i is faulty. The number of ones in this vector 

is equal to the number of inserted faults. Fault simulations were performed for two 

kinds of fault models. In one model a fault flips the bit, and in the other model 

(stuck-at-0/1) the bit takes a constant value 0 or 1. 

Let X = {x 15 ,...,x 1 ,x 0 } be an error-free vector of bits. 

Vector Xe = {xe 15 ,...,xe 1 ,xe 0 } is an erroneous vector [6] and [7]: 

• xe i = x i Å e i – if the fault flips the bit, 

• xe i = x i + e i – for stuck-at-1 fault, 

• xe i = x i × ē i – for stuck-at-0 fault, 

where: Å – xor, + – or, × – and. 

B. Error detection in basic operations 

To protect basic operations such as multiplication mod (2 n +1) (⊙), addition 

modulo 2 (Å) and addition modulo 2 n (⊞) in step function of HaF-256 (Fig. 3) we 

used DWC (Duplication With Comparison) scheme as it shown in Fig. 5. 

Step function F j with DWC scheme elements for one of addition modulo 

2 n (⊞) operation is presented in Fig. 6. In this figure gray boxes are an extra elements 

to detect errors in A 1 ⊞A 6 operation. The outputs of the addition block and 

the duplicated addition block are compared (box errcheck) and if are different, 

an error signal is generated.


Figure 5. Duplication With Comparison scheme 

Figure 6. Step function F j with DWC for one of addition modulo 2 n (⊞) operation 

Experiments were conducted by injection errors into the inputs of an operation 

block and observe if the errors are detected. Capability of single and multiple, 

transient and permanent fault detection using this scheme of fault detection 

is presented in Sec. V. 

C. Error detection in S-boxes 

S-box is a substitution function and is used to obscure the relationship between 

the plaintext and the ciphertext. It is an important element of cryptographic


125 

algorithm and it should posses some properties, which make linear and differential 

cryptanalysis as difficult as possible. 

In each step of round function of HaF a substitution S j (n) depending on 

n and j is used. It consists of four S-boxes S 0 , S 1 , S 2 and S 3 each of dimension 16×16 

working in such a way that: 

• for n = 16, S j (16) = S (j) mod 4 

• for n = 32, S j (32) = S (j) mod 4 || S (j+1) mod 4 , 

• for n = 64, S j (64) = S (j) mod 4 || S (j+1) mod 4 || S (j+2) mod 4 (x) || S (j+3) mod 4 . 

HaF S-box has been generated using the multiplicative inverse procedure 

similar to AES (Advanced Encryption Standard) with a randomly chosen primitive 

polynomial defining the Galois field. Nonlinearity of this S-box is 32510 and its 

nonlinear degree is 15. Sixteen Boolean functions that constitute this S-box have 

nonlinearities equal to 32510 or 32512 and are all of degree 15 [2]. 

A 16×16 S-box can be stored as a table containing 65536 values indexed by 

an input of the S-box function, i.e., x 1 , x 2 , …, x 16 . The table stores S-box outputs 

(16 bits: f 1 (x 1 , x 2 , …, x 16 ), f 2 (x 1 , x 2 , …, x 16 ), …, f 16 (x 1 , x 2 , …, x 16 )). 

Concurrent Error Detection schemes for S-box are presented in [9] and [10]. 

There are three different approaches to error detection. Two of these methods 

– parity based Concurrent Error Detection approach and DWC scheme – are 

the methods with hardware redundancy [10], the third one is time redundancy 

method and use involutional time redundancy CED to protect the S-boxes core 

of function HaF [9]. 

Capabilities of detection single and multiple, transient and permanent fault 

using these schemes are presented in Sec. V. 

D. Error detection in step function 

The round function (Fig. 3.) consists of 16 steps. The step #j is indicated by 

the integer j Î {0, 1, …, 15} and is shown in Fig. 4. To protect this function we used 

DWC (Duplication With Comparison) scheme as it shown in Fig. 5. 

Round function with DWC scheme elements for step function F j is presented 

in Fig. 7. In this figure, gray box is an extra step function block. The outputs 

of these two blocks are compared (box errcheck) and if are different, an error 

signal is generated. 

Experiments were conducted by injection errors inside or into inputs of the step 

function block and observation if the errors are detected. Capability of single and 

multiple, transient and permanent faults detection using this scheme of fault detection 

is presented in the Sec. V.


Figure 7. Round function with CED elements for step function F j protection 

E. Error detection in round function 

The message block M i is processed in two rounds. The method of one block 

processing is depicted in Fig. 2. Round function with DWC is presented in Fig. 8. 

The gray box is an extra Round function block. The outputs of Round #1 and Round 

#1b are compared (box errcheck) and if are different, an error signal is generated. 

Capability of single and multiple, transient and permanent faults detection 

using this DWC scheme is presented in Sec. V. 

V. Simulation results 

In order to measure the detection capability of the proposed in Sec. IV error 

detection schemas we used VHDL (Very High Speed Integrated Circuits Hardware 

Description Language) hardware description language and the VHDL simulator, 

Active-HDL by Aldec. In this section we provide simulation results related to 

the fault coverage of the proposed approaches. We present simulation results on


127 

the vulnerability of these schemes for fault models from Section IV.A. The faults 

were injected into inputs of an basic operation blocks or also inside the other blocks, 

and observed if the error is detected. We consider random faults, in the sense that 

the faulty value is assumed to be random and uniformly distributed. The VHDL 

model of the HaF-256 function has been modified by injected faults. The output 

signals have been compared to correct signals. In this way, the obtained fault coverage 

gives a measure of the error detection capability. 

Figure 8. DWC for round function 

A. CED in basic operations 

In this experiment we focused on transient and permanent, single and multiple 

stuck-at faults and bit flips faults. Errors were injected in the input of an operation block 

and observe if the error is detected. The obtained faults coverage for addition modulo 

2 n operation (⊞) is shown in Fig. 9 for permanent faults and in Fig. 10 for transient 

faults. Single permanent stuck-at-0/1 faults are detected by proposed CED in 57.9%, 

transient faults in 57.2%. All single bit flip faults, permanent and transient, are detected. 

Fig. 9. and Fig. 10. show also dependence of error detection probability on 

the number of injected faults. Not only single, but also multiple bit flip faults are 

detected always.


Percentage of permanent stuck-0/1 error detection for other operations is presented 

in Table II. DWC scheme detects all bit flip errors. 

Figure 9. Probability of permanent error detection using DWC for addition modulo 2 n operation (⊞) 

Figure 10. Probability of transient error detection using DWC for addition modulo 2 n operation (⊞)


129 

Table II. Probability of permanent error detection for operations of step function of HaF-256 

Stuck-at-0/1 

Number of errors 1 2 3 4 5 

a ⊙ b 57.2 74.9 88.1 93.1 96.6 

v Å w 61.1 82.1 85.4 93.3 96.5 

v ⊞ w 57.9 80.3 90.3 95.3 96.8 

p 1 (x) Ä p 2 (x) 60.9 78.2 85.8 93.2 96.1 

B. CED in S-boxes 

In this paper Concurrent Error Detection schemes for S-boxes of the HaF 

algotithm was tested for transient and permanent, single and multiple stuck-at error 

and bit flips errors. The obtained detection percentage for differnet CED schemes 

for transient faults is shown in Table III. and for permanent faults in Table IV. 

The best detection percentage of errors is for DWC scheme, first for all for 

single stuck-at-0/1 errors. A difference between DWC and parity bit scheme reaches 

more than 25% however this difference is much smaller for multiple errors. There 

is no difference between detection percentage for these three error detection schemas 

for bit flip errors — all are detected. 

A study of concurrent error detection possibilities of these three CED schemas 

shows, that involutional time redundancy CED and also parity bit based scheme 

is a good choice. The hardware overhead is very small and detection percentage for 

bit flip errors is practically the same as for DWC. 

Table III. Probability of transient error detection in S-box 

Fault type Stuck-at-0/1 Bit flip fault 

No. of errors 1 2 3 4 5 1 2 3 4 5 

Parity bits 54.5 79.4 83.5 93.1 95.2 94.5 100 100 100 100 

DWC 72.6 86.6 92.9 96.7 99.4 100 100 100 100 100 

Involution 65.1 81.8 91.5 91.9 95.1 99.9 100 100 100 100 

Table IV. Probability of permanent error detection in S-box 


No. of errors 1 2 3 4 5 1 2 3 4 5 

Parity bits 56.3 82.3 85.5 94.9 97.0 100 100 100 100 100 

DWC 75.8 87.7 95.1 98.3 99.7 100 100 100 100 100 

Involution 69.2 85.9 95.7 95.9 98.9 100 100 100 100 100


CED in step function 

In this experiment we focused on transient and permanent, single and 

multiple stuck-at faults and bit flip faults. Errors were injected in the input 

of step function block or inside the block and observed if the error is detected. 

Detection percentage of stuck-at-0/1 and bit flip errors, is shown in Table V. for 

transient errors and in Table VI. for permanent errors. Using this CED method 

we can detect all bit flip faults. The worse case is for single stuck-at faults. Only 

63.6% transient faults and 69.4% permanent faults is detected. Multiple stuck-at 

faults are detected in 80-95%. 

Table V. Probability of transient error detection in step function 


No. of errors 1 2 3 4 5 1 2 3 4 5 

DWC 63.6 79.6 84.3 90.3 95.6 100 100 100 100 100 

Table VI. Probability of permanent error detection in step function 


No. of errors 1 2 3 4 5 1 2 3 4 5 

DWC 69.4 82.8 85.3 93.1 95.9 100 100 100 100 100 

C. CED in round function 

CED in round function was tested also for stuck-at faults and for bit flip faults. 

There was considerate single and multiple faults, transient and permanent faults. 

The faults was inserted inside the function block or into the inputs. Probability 

of errors detection is presented in Table VII (transient error) and in Table VIII 

(permanent error). The obtained detection percentage for bit flip faults, single and 

multiple, is 100%. 

Table VII. Probability of transient error detection in round function 


No. of errors 1 2 3 4 5 1 2 3 4 5 

DWC 66.5 82.6 85.3 92.1 96.5 100 100 100 100 100 

Table VIII. Probability of Permanent error detection in round function 


No. of errors 1 2 3 4 5 1 2 3 4 5 

DWC 70.6 83.9 86.7 94.2 96.7 100 100 100 100 100


131 

VI. CED for hash function HaF 

After analyzing the propagation of errors in HaF and the probability of error 

detection for elementary function of HaF using different CED schemes, to protect 

HaF circuit we used DWC scheme for step function protection and DWC for two 

basic operations – addition modulo 2 (Å) and addition modulo 2 n (⊞). Hardware 

redundancy for step function DWC protection is smaller than redundancy for 

round function protection, and probabilities of error detection are comparable. 

We take into consideration transient and permanent, single and multiple 

stuck-at faults and bit flip faults. Errors were injected into the input of HaF or inside 

this function and than we observed if the errors are detected. Detection percentage 

is presented in Table IX. and Table X. All, transient and permanent, bit flip faults 

are detected. Multiple stuck-at faults are detected in 79-95%. Single stuck-at fault 

are detected at approximately 60-65%. 

Table IX. Probability of transient error detection in HaF 


No. of errors 1 2 3 4 5 1 2 3 4 5 

DWC 58.6 78.7 84.1 89.4 94.1 100 100 100 100 100 

Table X. Probability of permanent error detection in HaF 


No. of errors 1 2 3 4 5 1 2 3 4 5 

DWC 65.4 81.9 85.0 91.9 95.5 100 100 100 100 100 

VII. Concluding remarks 

Fault attacks are becoming a serious threat to hardware implementations 

of cryptographic systems. Proper countermeasures must be adopted to foil them. 

In this paper first the propagation of errors in HaF-256 is studied. We take 

into consideration single, transient as well as permanent faults injected at different 

stages of hash value computation. It is found that even a single error injected 

resulted in half the bits of hash value being in error and the errors are spread across 

the computed hash value. 

We proposed also error detection schemes for basic operations and elementary 

functions of hash function HaF and finally for HaF-256 circuit. In order to 

measure the fault detection capability of these schemas we used VHDL models and 

the VHDL simulator, Active-HDL by Aldec. We presented simulation results on 

the vulnerability of these schemes for single and multiple faults and for transient 

and permanent faults. These methods can provide high coverage for multiplebit 

errors, which are the most common fault attacks. The coverage depends heavily


also on the fault model. Simulation experiments conducted on a large number 

of test cases show that our schemes have 100% fault coverage in the case of bit flip 

errors. These solution can be useful for concurrent checking cryptographic chips, 

especially designed for platforms with limited resources 


This research was partially supported by the Polish Ministry of Education and 

Science as a 2010-2013 research project. 

References 

[1] G. Bertoni, L. Breveglieri, I. Koren, P. Maistri, V. Piuri, “Error Analysis and 

Detection Procedures for a Hardware Implementation of the Advanced Encryption 

Standard”, IEEE Transactions on Computers, vol. 52, no. 4, pp. 492-505, 2003. 

[2] T. Bilski, K. Bucholc, A. Grocholewska-Czuryło, J. Stokłosa, „Hash Function 

Concept”, Report no. 596, Poznań University of Technology, Poznań, 2011. 

[3] D. Boneh, R. DeMillo, R. Lipton, “On the Importance of Eliminating Errors 

in Cryptographic Computations”, Journal of Cryptology, vol. 14, pp. 101-119, 2001. 

[4] D. Boneh, R. DeMillo, R. Lipton, On the importance of checking cryptographic 

protocols for faults. Proceedings of Eurocrypt, pp. 37-51. Springer-Verlag LNCS 1233, 

1997. 

[5] L. Breveglieri, I. Koren, P. Maistri, “An Operation-Centered Approach to Fault 

Detection in Symmetric Cryptography Ciphers”, IEEE Transactions on Computers, 

vol. 56, no. 5, 635-649, 2007. 

[6] E. Idzikowska, K. Bucholc, “Concurrent Error Detection in S-boxes”, International 

Journal of Computer Science & Applications, vol. 4, no. 1, pp. 27-32, 2007. 

[7] E. Idzikowska, K. Bucholc, “Error detection schemes for CED in block ciphers”, 

Proceedings of the 5th IEEE/IFIP International Conference on Embedded and 

Ubiquitous Computing, pp. 22-27. Shanghai 2008. 

[8] E. Idzikowska, “CED for S-boxes of symmetric block ciphers”, Pomiary, Automatyka, 

Kontrola, vol. 56, no. 10, pp. 1179-1183 2010. 

[9] E. Idzikowska, “CED for involutional functions of PP-1 cipher”, Proceedings of the 5th 

International Conference on Future Information Technology. Busan 2010. 

[10] E. Idzikowska, “Errors detection in S-boxes of hash function HaF-256”, Borzemski L., 

Grzech A., Świątek J., Wilimowska Z. (eds.), Information Systems Architecture and 

Technology – Web Information Systems Engineering, Knowledge Discovery and Hybrid 

Computing, Wroclaw University of Technology Press, Wrocław, 2011, 231-240. 

[11] J. Stokłosa, T. Bilski, T. Pankowski, “Bezpieczeństwo danych w systemach informatycznych”, 

PWN, Warszawa-Poznań 2001.

Chapter 6 

Spectrum Management 

and Software Defined Radio Techniques

A Realistic Roadmap for the Introduction 

of Dynamic Spectrum Management in Military 

Tactical Radio Communication 

Bart Scheers 1 , Austin Mahoney 2 , Hans Åkermark 2 

1 CISS Department, Royal Military Academy (RMA), Brussels, Belgium, 

bart.scheers@rma.ac.be 

2 Communications Development, Security and Defense Solutions, Saab AB, Linköping, Sweden, 

{austin.mahoney, hans.akermark}@saabgroup.com 

Abstract: Cognitive radio (CR) technology has to date not been adopted by the military even though 

more than a decade has passed since its inception. This paper describes how cognitive radio and 

associated dynamic spectrum management (DSM) procedures can be introduced in military tactical 

radio communication. CR and DSM address key challenges that face future military tactical 

radio communication and their successful introduction can reduce or even overcome current 

spectrum scarcity and deployment difficulty. A DSM roadmap is introduced where military users 

develop trust in, and experience with, these novel technologies in manageable steps. The first step 

in this DSM roadmap involves the introduction of a military band dedicated for CRs and subsequent 

steps gradually increase the spectrum that may be utilized for military cognitive operation. 

A high-level vision of how existing military spectrum management procedures will change in the future 

with the introduction of DSM is also presented resulting in a significant reduction in the workload 

of spectrum management personnel. 

Keywords: cognitive radio; dynamic spectrum management; military tactical communication; 

roadmap 


Military tactical networks are being required to support a greater number 

of services than ever before. In addition, the bandwidth requirements associated 

with many of the new services are also rapidly increasing. The combination 

of these two factors means that we are nearing a time when there will be insufficient 

bandwidth to support the services required for future military operations. Today’s 

military operations are also typically undertaken by multiple nations cooperating 

in a coalition force. The spectrum and frequency planning activities associated with 

This research work was carried out in the frame of the CORASMA – EDA Project B-0781-IAP4-GC.


the deployment of a large multi-national coalition force are extremely complex and 

unacceptably long, and can even delay the start of an operation [1]. 

Both of the aforementioned problems, spectrum scarcity and deployment burden, 

are to a large degree consequences of the centralized and static nature of current 

spectrum management [1]. Dynamic spectrum management and cognitive radios 

seek out and use a part of the electromagnetic spectrum in ways that are not predictable, 

so that it is not generally known which set of frequencies that a radio will use at 

any given time”. The DSM process may be seen as a harmonization of, and dynamic 

interaction between, both a human element in the form of spectrum regulators and 

spectrum planners or managers and an autonomous element in the form of one or 

more cognitive radio networks. DSM represents a fundamental change from existing 

spectrum management procedures in the way that spectrum is allocated and used for 

both civilian and military domains. 

This paper aims to address two particular aspects of DSM, first: 

• how a new and unproven technology such as DSM can be introduced 

in military tactical radio communication, and second 

• how current spectrum management procedures would be affected by such 

a change. 

To address these two aspects, this paper starts in section 2 with an overview 

of current spectrum management procedures in the military tactical domain and 

their inadequacy for future military operations. Section 3 highlights the key challenges 

for a proposed roadmap describing the gradual and systematic extension 

of current spectrum management procedures that is presented in section 4. Section 

5 relates the proposed roadmap to future spectrum management procedures 

to clarify possible DSM practices in a future setting where cognitive enabled radios 

will be the norm in military tactical radio communication. This paper ends in section 

6 with major conclusions. 

II. Current NATO spectrum management procedures 

at the operational level 

The worldwide use of the electromagnetic spectrum is regulated by the International 

Telecommunication Union (ITU). The ITU sets out the global radio 

regulations in the form of a global Frequency Allocation Table (FAT) from which 

each nation defines its own FAT covering many aspects of radio regulation, such 

as the generic type of radio service that is permitted within each frequency band 

and whether the band is allocated for military, civil or shared use. In what follows, 

existing non-cognitive military spectrum management procedures are summarized 

for coalition operations as defined in ACP 190(C) [2]. ACP 190(C) defines three 

main phases to each operation: Planning, Deployment (also known as Implementation) 

and Recovery. The description provided is structured according to these 

three phases involving the elements introduced in Figure 1.

Chapter 6: Spectrum Management and Software Defined Radio Techniques 

137 

Combined Task 

Force Commander 

(CTFC) 

Host Nation 

Administration 

Combined Spectrum 

Management Cell 

(CSMC) 

Component Spectrum 

Manager (CSM) 

Lead Nation 

National Spectrum 

Manager 


Manager 


Manager 

National 

Element 

- Networks 

- Radios 

National 

Element 

- Networks 

- Radios 

National 

Element 

- Networks 

- Radios 

Figure 1. Elements of a coalition deployment and their relationships 

A. Planning phase 

The primary purpose of the planning phase is to produce the Battlespace 

Spectrum Management Plan (BSMP). The BSMP is used to inform the coalition 

force entities of issues relating to the management and planned usage of the spectrum 

during the operation by providing a mapping between all radio and network 

systems and frequencies. 

In the planning phase, the Combined Task Force Commander (CTFC) assumes 

overall command of the forthcoming operation as part of a mandate from 

a higher authority specifying the conditions under which the coalition force will 

operate. The CTFC establishes a Combined Spectrum Management Cell (CSMC) 

responsible for coordinating the spectrum requirement of the force, acquiring 

the necessary spectrum and assigning frequencies for the systems used by the forces 

within the operational area. The CSMC will normally delegate the authority to manage 

frequency allotments to one or more Component Spectrum Managers (CSMs), 

where maritime, land, air, logistics and special forces are typical components. Each 

component may be composed of multiple National Elements (NEs). 

When a coalition is formed, a single nation is given the responsibility to act 

as the Lead Nation. The Lead Nation is responsible for providing and sustaining 

frequencies for the force through the spectrum management process and providing 

technical support to the CSMC. Each nation within the coalition force is obliged to 

set up its own spectrum management cell with a National Spectrum Manager (NSM) 

to coordinate spectrum processes for its national forces. Each NSM will identify 

all radio equipment to be deployed by the NEs, including equipment parameters


and operational (geographical) areas, and provide this information to the CSMC. 

The CSMC combines the spectrum requirements, operational areas and radio 

equipment characteristics provided by the individual NSMs into the Electronic 

Order of Battle (EOB). The EOB expresses the overall spectrum requirement for 

the operation. 

Depending on the tools available to the CSMC, it should be possible to model 

the overall spectrum requirement for the operation that is detailed in the EOB (i.e. 

frequency planning). This modeling activity makes use of the topographical data 

associated with the operational area to identify where frequency re-use is possible. 

It also takes into account any equipment frequency constraints, frequency hopping 

systems, airborne emitters, preferred frequency allotments and any protected 

frequencies. The CSMC uses the output of the modeling activity and the EOB to 

acquire the necessary spectrum to satisfy the operation requirement. Depending 

on whether or not the operation is conducted with the support of the host nation, 

the CSMC either establishes contact with the host nation spectrum administrators 

and submits a frequency request detailing the spectrum requirements or uses 

electronic surveillance to select the most favorable spectrum without regard for 

existing local users. 

When the available spectrum has either been allocated by the host nation or 

identified through observation, the CSMC (and CSMs) will produce the frequency 

assignment tables for all radio equipment associated with the coalition force and 

incorporate them into the BSMP. These tables include any constraints on the use 

the frequencies including transmit power, antenna height and location. 

B. Deployment phase 

In the deployment phase, each NE implements the BSMP as received from 

the CSMC via their CSM prior to deployment. Interference may be encountered at 

any time due to host nation emitters, conflicting allied systems or enemy jamming. 

Each NE is responsible for investigating the interference that is encountered to try to 

determine the source and, if the source is local, endeavor to reduce the interference 

or eliminate it using appropriate action. If local action is impractical or unsuccessful, 

the interference and resultant loss of capability is reported to the CSM using 

a specified interference report format and procedure. Each CSM is responsible for 

the real-time control and management of the spectrum within its area of operation 

(i.e. the spectrum used by its NEs). The CSM responsibilities include the resolution 

of any frequency conflicts between its NEs and other interference issues by making 

appropriate frequency assignment changes or modifying allotments if other efforts 

to alleviate the interference are ineffective. The CSM will send any spectrum changes 

to the appropriate NEs where they are implemented. 

Whilst the CSMC may delegate responsibility for real-time spectrum management 

to the CSMs, it retains overall control and will become involved where


139 

coordination between components is required. The CSMC’s responsibilities include 

maintaining a close relationship with the host nation so that all interference reports 

associated with host nation emitters that are received from CSMs are sent to 

the host nation administration to be resolved. The CSMC will also act to resolve 

spectrum conflicts between components and other interference issues by making 

appropriate frequency assignment changes or modifying allotments if other efforts 

to alleviate the interference are ineffective. 

C. Recovery phase 

The recovery phase is a period of transition where each individual NE is responsible 

for informing the CSM of the time and date when the element will stop 

using their frequency assignments and handing back frequencies to the CSM for 

reassignment to another unit. Each CSM is responsible for reviewing and consolidating 

the spectrum in use by the force component, identifying any NE changes 

that need to be incorporated into a new BSMP. The CSMC is responsible for reviewing 

and consolidating the spectrum in use by the force as a whole, incorporate 

any changes into a new BSMP to meet the requirements of the new force which 

is passed on to an incoming force or the (newly established) civil administration. 

III. Key challenges for the introduction of DSM in military 

tactical environments 

The previous section shows that today’s military spectrum management procedures 

are, in the main, centralized and static. For example, the frequency bands 

allocated for military use within the relevant FAT are typically not modified on 

a regular basis. The frequency planning procedures that are applied within these 

military bands are also performed to meet the particular operational requirements 

of an operation including factors such as size, scope and composition of the deployed 

force. To a large extent, frequency management activities are typically performed 

prior to the operation in the planning phase and are often time consuming and 

complex especially for large coalition operations. Hence, once the assignments are 

made, they are generally fixed for the duration of the operation. Cognitive radio 

and DSM technologies offer the potential to significantly enhance the operational 

effectiveness and management of military tactical radio communications. 

However, these new technologies are relatively complex and represent a paradigm 

shift in capability and operational procedure for military spectrum managers 

and end-users. Several hurdles must be overcome if cognitive radio technology is to 

be adopted by the military as the technology is currently not in use even though more 

than a decade has passed since its inception. Firstly, as with any novel technology, 

there is general mistrust in its capability and military users tend to be conservative 

and risk averse towards new and unproven technology. Secondly, if the cognitive


radio concept is adopted, it will be introduced into service in a gradual fashion 

where CRs must be able to safely coexist with existing legacy radio equipment and 

existing procedures for the foreseeable future. These issues call for a carefully managed 

and incremental introduction of CR technology over time. 

Summarizing, the extension of current spectrum management procedures 

requires that: 

• DSM is introduced in a step-wise fashion gradually increasing military users’ 

experience with and understanding of CR technology and equipment, 

and where 

• DSM procedures co-exist with legacy equipment and current spectrum 

management procedures taking on a gradually increasing role as more CR 

technology is coming into service. 

Following these two principles, the introduction of CR and DSM in military 

tactical radio communication should be regarded as a evolution and not 

a revolution enabling a more flexible and efficient use of the spectrum in future 

military operations, and to ease the burden of the pre-mission frequency planning 

procedures. 

IV. A relastic roadmap for the introduction of dynamic 

spectrum management 

This section proposes a realistic DSM roadmap that identifies how cognitive 

radio technology may be introduced into the military communications 

domain in discrete steps although no specific timeframe is proposed. This DSM 

roadmap starts from the present day, with no cognitive radios, and with each increment 

takes a step further into the future with an increasing exposure to DSM 

in terms of number of devices, spectrum access complexity and freedom in operating 

spectrum. The DSM roadmap is designed such that military users may develop trust 

in and experience with the technology in an incremental fashion, and limit risk to 

existing operational capabilities. 

A. The first step – a dedicated band for CR systems 

As mentioned previously, it is imperative that this first step involves minimal 

risk to existing legacy operations and allows military end-users the opportunity to 

build trust in and evaluate CR technology. We propose the introduction of a band 

exclusively dedicated to CRs within the military allocation. All cognitive devices 

would operate within this band and all existing legacy systems would operate outside 

this dedicated band that is governed by current spectrum management procedures. 

Initially, we would expect only a limited number of CRs within a coalition force and 

the bandwidth requirement for this CR-only band would be small in comparison 

to that required to support the legacy systems of the coalition.


141 

Civilian 

Bands 

Military 

Bands 

Today's spectrum 

management procedures 

Step 1: A dedicated band 

for CR cystems 

Step 2: Opportunistic military 

use of limited civilian bands 

Time 

Military bands 

(legacy radio only) 

Civilian bands 

(no military users) 

CR as secondary users 

(military bands) 

Step 3: Coexistence of CR 

with military legacy systems 

in military bands 

Step 4: Flexible use of military 

bands and wide scale opportunistic 

use of civilian bands 

Dedicated CRonly 

band 

CR as secondary users 

(civilian bands) 

Mostly CR (military 

bands, flexible use) 

1) General considerations 

Figure 2. DSM roadmap representation 

We propose that the CR-only band would be managed under a managed 

commons DSA model 1 where all CR users would be considered equal and no 

traditional spectrum management procedures, such as channel assignment, would 

be required leading to a reduced pre-operation preparation time. Separate CRNs 2 

would be able to coexist within the same band by autonomously and cooperatively 

or non-cooperatively selecting different operating channels within the band 

(i.e. channels with the least interference). In the managed commons model case, 

all CR users/networks share spectrum using an agreed management protocol 

that encapsulates technology agnostic spectrum access rules. This CR-only band 

may be utilized by military users at all levels of the military hierarchy. However, 

it may be best to initially target the technology at the lower priority echelons and 

for particular applications (see below). If CRs demonstrate success at this lower 

level, they are more likely to gain acceptance and be more widely adopted across 

all layers of the military hierarchy. 

1 

2 

A DSA model is here used as classification of a high-level Dynamic Spectrum Access (DSA) technique following 

the terms introduced in [3]-[5]. 

A CRN is a network where a number of cognitive radios interoperate.


The choice of a dedicated band for CR systems as a first step in the road-map 

has some important advantages. 

a) No risk for interference with legacy systems 

Military end-users are generally conservative and risk averse. The use of a cognitive 

system in a shared-use spectrum access mode that could degrade a legacy 

communication system would not be acceptable. The use of a dedicated band for 

cognitive systems eliminates this risk. 

b) Relaxation of system requirements 

From a technical point of view, a dedicated band based on the managed commons 

model will relax system requirements for CR systems. The primary reason for 

this statement is that the cognitive band will lack primary users, placing less severe 

constraints on CR sensing and decision making functions to gather and process 

spectrum information to adapt its behavior. Cognitive systems would thus only 

need to address interference from other cognitive systems and jammers, which 

can be expected to be less time-critical compared to more challenging channel 

evacuation requirements. 

c) Familiarity 

The concept of a license free band such as the current 2.4 GHz ISM band, 

is not new for most military end-users and the introduction of a license free military 

band for CR is likely to be regarded not as a revolution, but as a case where military 

procedures catch up to the less restrictive procedures used in the civilian world. 

d) Compatibility with current NATO spectrum management procedures 

In NATO, the civil/military spectrum Capability Panel 3 (CAP3) is responsible 

for the harmonization of radio frequency use among NATO allies. CAP3 is placed 

under the new C3B sub-structure (Consultation, Command and Control Board). 

The military part of the CAP3 panel manages the harmonized military spectrum 

by allocating bands to applications or services, such as wide-band land systems or 

satellite broadcast services. As such, the proposed first step would be compatible 

with established administrative spectrum management procedures where CAP3 

would allocate one or multiple bands exclusively for CR systems. However, NATO 

currently considers CR as a technology rather than as a service or application. 

CAP3 has therefore refrained from making the proposed allocation necessitating 

further negotiations and lobbying to overcome this hurdle. 

e) Incentives for research and development 

One of the main obstacles in the technological development of military CR 

is the deadlock created by the general lack of confidence in the technology. As long 

as there is no clear sign from the military end-user showing an interest in the tech-


143 

nology and thus establishing a potential market, the military communications 

industry will be hesitant to invest in the development of military CR systems. On 

the other hand, as long as there are no military off-the-shelf CR products, which 

can prove the concept and clearly demonstrate key CR benefits, the military enduser 

will remain skeptical towards the technology. The creation of a dedicated 

band for CR systems can end this deadlock and be an incentive for the industry to 

start investing and developing products, comparable to what happened in the civil 

2.4 GHz ISM band. 

f) Extensibility 

As mentioned previously, an important motivation for this first step is gaining 

trust. Once the concept of CR proves to work in this dedicated band, the band 

can be easily extended and integrate more complex spectrum access models with 

civilian and/or military primary users. 

In summary, in light of the advantages described above, the introduction and 

use of a dedicated band for CR is by far the most appropriate first step in the roadmap 

towards the introduction of cognitive radio in the military. If this first step 

proves to be successful, the following steps can be implemented to overcome 

spectrum scarcity and alleviate current deployment burdens. 

2) Spectrum suitability, rules and limitations 

Spectrum is a scarce resource, both civilian and military, and the introduction 

of a CR-only band will require sacrifices. In our opinion, the most 

suitable option is to define multiple smaller bands within different spectrum 

regions and use them as dedicated CR-bands. It is obvious that the NATOharmonized 

bands are the most appropriate. From a technical point of view, 

and taking into account the possible applications, we think that at least a band 

of 5 to 10 MHz should be defined in the NATO harmonized UHF band I 

(225-400 MHz). Other possible candidate frequency bands are the military UHF band II 

(790-960 MHz), the military UHF band III (1350-2690 MHz) and the SHF band 

(4400-5000 MHz), where cognitive systems for short range wireless networks 

could be envisaged. 

An in-depth study of the rules and limitations for the use of this dedicated 

band for CR systems is out of the scope of the paper. We will therefore only comment 

on three aspects of such rules and limitations: 

• The types of applications that are allowed and are forbidden to use this 

band together should be defined. Some example of systems that could be 

allowed are wideband land systems, line-of-sight microwave links and tactical 

wireless local area networks whereas pulsed systems like surveillance 

radars and broadcast stations are likely to be excluded. 

• The technical elements of a set of restrictions need to be studied and 

standardized in a future STANAG. Some possible restrictions are the use


of a spectrum mask, a radiated power bound per application and the duty 

cycle of the systems. 

• The need for a pre-defined channelization in the dedicated band should 

also be studied and if necessary the choice of the channel bandwidth. 

B. The second step – opportunistic military use of limited civilian bands 

A long-term driving factor is the increasing pressure felt by governments and 

regulatory authorities to reallocate spectrum traditionally reserved for military 

use to support civilian wireless services. In the event that spectrum allocated for 

military operations diminishes on this basis, allowing the military to operate in civilian 

bands as non-interfering secondary users offers a way to increase the total 

military spectrum pool. 3 

In this second step, we propose to extend the military CR footprint by allowing 

military CR systems to operate as secondary users within limited civilian bands 

(taking advantage of white spaces in time and space). The terrestrial television 

and radio UHF/VHF bands are likely first candidates due to the ideal propagation 

characteristics for tactical radio systems, the relatively stable channel availability 

and the maturity of the commercial IEEE 802.22 standard which may be leveraged 

for military implementations. 

Within these television and radio bands, the military CR systems would operate 

on a non-interfering secondary user basis under a spectrum overlay model. 

Military CRNs may or may not cooperate with each other and/or with the primary 

civilian users regarding spectrum access. As noted in step 1, the necessary avoidance 

of primary user transmissions under the overlay model represents an additional complexity 

over and above operations under a managed commons model (as in step 1). 

From a military perspective, a significant advantage of this step in the DSM roadmap 

is that whilst the military users will have acquired more operating spectrum, no interference 

risk is introduced for existing military operations using legacy systems. 4 

Military CRs would be operating within both the dedicated CR-only band (as peers) 

and within limited civilian bands (as secondary users). All legacy systems would 

be operating in remaining military allocations as today. This isolation in frequency 

maintains the CR and legacy system coexistence assurance provided in step 1. This 

step also allows military users to evaluate the more complex overlay spectrum access 

technique with no additional risk to military legacy operations. 

3 

4 

Here it is assumed that the military agree to be subservient to civil users (i.e. act as secondary users). This 

may be the case in training or humanitarian relief operations. In more aggressive operations (such as Defense 

of National Territory), military forces are unlikely to operate under this premise. They will more likely secure 

spectrum on an exclusive-use basis either through negotiation with the host nation or by force. 

There may be an interference risk to civilian primary users. Of course, the level of acceptability of risk depends 

on the importance of civilian services being supported in the specific band and the context of the military 

operation. The risk may be deemed acceptable for TV and broadcast radio bands but not so for civilian 

air traffic control bands.


145 

C. The third step – coexistence of CR with military legacy systems 

in military bands 

In this third step we propose to build on the experience developed in the second 

step where military CRs were allowed to operate as secondary users within limited 

civilian bands. We propose that military CR systems would also be allowed to coexist 

with legacy systems in limited military bands. The legacy systems would be 

designated as primary users. The CRs would operate on a non-interfering secondary 

user basis. This will allow military CR systems to take advantage of white spaces 

in time and space left fallow by military legacy systems. 

The proposed third step differs from the previous two steps in the proposed 

DSM roadmap in that CRs would be allowed to coexist in the same spectrum 

as legacy systems and thus represents the first step where there is an increase in risk 

of interference by CRs to existing military operations. However, with the experience 

of the previous steps in the roadmap and only allowing a limited spectrum 

overlap (between military legacy and CR systems), this risk may be easily managed. 

D. The fourth step – flexible use of military bands and wide scale 

opportunistic use of civil bands 5 

The previous steps have identified how CR and DSM technology may be 

evaluated and implemented in an incremental process. A fundamental component 

of these steps is the coexistence with existing civilian and military legacy systems 

as described in steps 2 and 3, respectively. For this final step we assume that most 

military legacy systems have been retired from service and that cognitive radios are 

the norm (i.e. a far future setting). 6 We also assume that military acquisition agencies 

and end-users have accepted, and are comfortable with, the new technology. 

These assumptions enable the exciting possibility for a significantly more flexible 

and effective use of spectrum within the military domain. 

In this final step we propose that military spectrum will be managed under 

a highly flexible and dynamically variable combination of the uncontrolled commons, 

managed commons, dynamic exclusive-use and overlay DSA model types 

controlled using DSM policies. We also propose the continual and extensive use 

of civil spectrum bands on a secondary use basis. A CRN will be provided DSM 

policies which identify where in spectrum it will be allowed to operate and the associated 

operating conditions, which it must adhere to. Such allocations may include 

multiple bands each managed under a different DSA model type. The DSM strategy 

implemented by the CRN management system will autonomously determine 

5 

6 

The first three steps may not be accompanied by the proposed fourth step as it is entirely possible that 

the third step is sufficient to overcome challenges with spectrum scarcity and deployment difficulty. 

Some portions of the spectrum may always need to be reserved for fixed frequency operations (e.g. GPS and 

some safety critical applications) and for any remaining legacy radios. However, this may still be managed 

under the new DSM paradigm with the exclusive-use DSA model type.


the most favorable operating location within the allowed spectrum given the local 

time-variable environmental conditions. 

CRNs operating within bands designated under either of the commons types 

or the overlay model type (as secondary users) will not require complex traditional 

frequency planning and should realize more efficient and effective use of the spectrum. 

However, systems operating under these conditions will not benefit from 

the preferential (and often sole) spectrum access rights enjoyed under existing 

military processes, which may have a detrimental impact on existing levels of quality 

of service (QoS) that need to be managed by planning efforts as well as more dynamic 

cognitive monitoring and control loops to better reflect the needs of an ongoing 

operation. These future spectrum management procedures are clarified in the following 

section. 

V. Future dynamic spectrum management procedures 

in the military tactical domain 

The introduction of CR technology and DSM will involve the modification 

of existing military spectrum management practices. It is anticipated that these 

changes will simplify and shorten the spectrum planning activity required prior to 

an operation. This section addresses these procedural changes using the same three 

phases of operation as in the current procedure with an emphasis on the planning 

phase. 7 

A. The DSM planning phase 

In a CRN future, CRs and CRNs will not, in general, need to be assigned 

specific operating frequencies. CRs will instead be allowed to dynamically and 

autonomously select the best available operating spectrum within specific bands 

according to certain rules. This means that future DSM planning processes will 

change from current spectrum management procedures as described in the following 

starting with an overview followed by a detailed example. 

1) Overview of DSM planning considerations 

CRs will not, in general, be assigned specific operating frequencies which 

is fundamental difference from current spectrum planning that result in the compilation 

of the BSMP that includes the (static) frequency assignments for all radio and 

network equipment. CRs will instead be allowed to dynamically and autonomously 

select the best available operating frequency channel within specific bands accord- 

7 

In the proposed roadmap, CR systems and legacy radio systems may coexist. It is proposed that legacy radio 

systems will continue to be managed using the principles described in section 2 whereas future CR systems 

managed as described in this section. The overall management procedure will therefore be a blend between 

static and dynamic processes.


147 

ing to certain rules formalized by the DSM policies. 8 This means that the contents 

of the BSMP will fundamentally change in the future and include DSM polices 

rather than today’s mapping between all radio and network systems to frequencies. 

The EOB can also be expected to undergo changes making DSM policy generation, 

refinement and distribution the key activities to be undertaken during planning 

phase of future operations. 

A future DSM hierarchy would in the planning phase define and refine the DSM 

policies for the underlying CRNs in preparation for deployment. Each decision 

making entity in the hierarchy is expected to use a computer-based DSM policy 

generation and management tool for this purpose. The host nation administration 

will define the highest level DSM policies. These mainly specify the frequency 

allocations and conditions of their use, which may be used by the CR systems 

of the coalition force in the forthcoming operation. These DSM policies may be 

updated and refined by each lower layer in the SM hierarchy prior to being downloaded 

into the policy repositories in the individual CRN management. DSM policy 

refinements made by the individual NESMs and CSMs are reported back up the SM 

hierarchy to the CSMC. These feedback paths ensure that the CSMC is completely 

aware of how the spectrum will be utilized during the mission. The CSMC would 

here not be permitted to change the DSA model type of a band unless the band 

has been allocated to it by the host nation on an exclusive-use basis. This is true 

for all decision making entities in the SM hierarchy, which must respect the DSA 

model decisions made by the hierarchical level immediately above it. 

The CSMC will update the DSM policies provided to it by the host nation to 

reflect the decisions made. Where host nation allocations are split into sub-bands, 

the DSM policy associated with each sub-band inherits the access conditions 

of the parent allocation, together with any additional or replacement conditions 

specified by the CSMC. The relevant DSM policies are then distributed to the CSMs. 

Note that the particular decisions made by the CSMC depend on the requirements 

of the individual national elements supplied to it by the NSMs. 

2) A DSM planning example 

As an example to illustrate the DSM planning phase, we consider a military 

operation involving two battle groups, BG A and BG B, from two different nations. 

The host nation administration provides the military force with two frequency 

bands, one in the VHF band from 47-50 MHz and one in the UHF I band from 

318-328 MHz, both on an exclusive use basis. The administration also allows the military 

force to make use of the civilian terrestrial television band (470-830 MHz) on 

a secondary use basis. Figure 3 illustrates the allocations defined by the spectrum 

management hierarchy (line 1). 

8 

DSM policies are here used as a mechanism to guide, control and bound the autonomous behavior of CRs.


The CSMC decides how the host nation allocation should be used by the military 

force on a high level (line 2 in Figure 3). Whilst the focus here is for the BG A operation, 

the CSMC must also consider the spectrum needs for BG B, that is operating 

in another region of the same area of responsibility. 

47 

MHz 

VHF 

50 

MHz 

318 

MHz 

UHF I 

328 

MHz 

470 

MHz 

UHF II 

830 

MHz 

1 

Host Nation 

Military Use 

Military Use 

Terrestrial TV 

2 

CSMC 


3 

NESM A 


Company Waveforms 

Coalition Waveform 

Band Allocation Key 

Dynamic 

Exclusive Use 

Civil Overlay 

(Military as SU) 

Uncontrolled 

Commons 

Military Overlay 

(PU) 

Military Overlay 

(SU) 

Figure 3. Spectrum plan as output from DSM planning processes at different levels in the spectrum 

management hierarchy. Primary and secondary users are designated as PU and SU, respectively 

The CSMC assigns primary status to NESM A for BG A communications 

for the 48-49 MHz and 318-320.5 MHz bands. These preferential assignments are 

intended for the higher priority waveforms associated with BG A. The CSMC also 

assigns primary status to NESM B for BG B communications for the 49-50 MHz 

and 320.5-323 MHz bands. We assume that a coalition waveform is used for communication 

between elements in the two battle groups does not support DSM 

and is assigned exclusive-use access for the 47-48 MHz band. The CSMC assigns 

the 323-328 MHz VHF band as uncontrolled commons spectrum to be shared by 

all BG A and BG B cognitive systems (intended for squad or platoon level waveforms). 

The terrestrial television band is assigned to be used equally by all systems 

in the military force on a secondary basis. 

Line 3 in Figure 3 illustrates the planning decisions made by NESM A as the next 

level below the CSMC in the SM hierarchy (the decisions made by NESM B are 

not shown as they are of no further interest in this example). NESM A decides that 

it has three different company waveforms that are high priority and are thus given 

primary status in separate bands as shown. Underlying squad and platoon level 

waveforms are not provided dedicated allocations – they are required to find their 

own operating spectrum. 

B. The DSM deployment phase 

At the beginning of the deployment phase, all CRs should have downloaded 

the appropriate DSM policies and be able to initiate communication activities 

within their possible allocations, under the specific conditions defined in the planning 

process. During deployment all CRNs will regularly send performance and


149 

spectrum status reports back up to the relevant NESM provided resources are allocated 

for this purpose. The NESMs will send aggregate status reports to the relevant 

CSM, status reports that are further aggregated to the CSMC. These status reports 

allow spectrum managers at all levels of the SM hierarchy to continuously evaluate 

spectrum usage and network performance within their allocations throughout 

an operation. Spectrum managers at all levels will have the ability to update and 

disseminate DSM policies at any time to ensure the coalition force communication 

needs are continuously met. 

Each CRN will be responsible for dynamically and autonomously selecting 

the best available operating frequency within its specific allocations defined by 

the active DSM policies and be transparent to the end-user. With this capability, 

many interference issues may be handled directly by the CRNs themselves through 

the execution of the selected DSM strategy and spectrum mobility, with no involvement 

of any spectrum management personnel. Each NESM, in conjunction with 

a network manager, will be responsible for monitoring the status reports provided 

by their CRN management systems. During deployment, an NESM may make act 

to optimize the performance of the underlying CRNs by updating the active DSM 

policies associated with any allocations provided to it on an exclusive-use basis 

by the relevant CSM. 

C. The DSM recovery phase 

During the recovery phase, spectrum managers at all levels of the SM hierarchy 

will be responsible for reporting to their immediate superior manager that their 

units no longer require their spectrum allocations. The associated DSM polices 

may be modified to better reflect the needs of the remainder of the coalition force. 

The CSMC will be responsible for consolidating the master DSM policy database 

such that it may be used by an incoming / replacement force. 

VI. Conclusions 

Military DSM is and requires a fundamental shift in capability and spectrum 

management procedures. This paper has attempted to clarify the need for an incremental 

approach for implementing DSM into the military domain. We have outlined 

a DSM roadmap for this purpose. Through this roadmap we have described how 

novel CR technology with non-cognitive legacy radio systems can co-exist over 

the transitionary period towards an all CR future. This roadmap has also addressed 

how military users may develop trust in, and experience with, this novel technology 

in manageable steps. The first step in this DSM roadmap involves the introduction 

of a military band dedicated for CRs. It has been proposed that this band is managed 

under the managed commons models using multiple smaller bands within different 

NATO-harmonized spectrum regions.


We have also presented a high-level vision of how existing military spectrum 

management procedures will change in the future with the introduction of DSM 

using ACP 190(C) as a baseline reference. With the application of secondary use 

and uncontrolled use of spectrum, we foresee a significant reduction in the workload 

of spectrum management personnel. We have focused on both the important 

planning and deployment phases of military operations. The decisions made by 

the spectrum management entities, both before and during an operation, will 

principally revolve around dynamically creating, updating, activating, deactivating 

and deleting DSM policies. 

References 

[1] XG Working Group, The XG Vision Request for Comments, v. 2.0, available through 

http://www.ir.bbn.com/~ramanath/pdf/rfc_vision.pdf 

[2] Combined Communications and Electronics Board, Guide to Spectrum Management 

in Military Operations, ACP 190(C), september 2007, available through http://jcs. 

dtic.mil/j6/cceb/acps/acp190/ACP190C.pdf 

[3] M. Buddhikot, “Dynamic spectrum access: models, taxonomy and challenges”, IEEE 

International Symposium on New Frontiers in Dynamic Spectrum Access Networks, 

2007, pp. 649-663. 

[4] Q. Zhao, A. Swami, “A survey of dynamic spectrum access: signal processing and 

networking perspectives”, Proceedings from the IEEE International Conference on 

Acoustics, Speech and Signal Processing, 2007, pp. 1349-1352. 

[5] Q. Zhao, B.M. Sadler, “A survey of dynamic spectrum access: signal processing, 

networking and regulatory policy”, IEEE Signal Processing Magazine, vol. 24, May 2007, 

pp. 79-89.

Dynamic Spectrum Management 

in Legacy Military Communication Systems 

Marek Suchański 1 , Paweł Kaniewski 1 , Robert Matyszkiel 1 , 

Piotr Gajewski 2 

1 Military Communication Institute, Zegrze, Poland, 

{m.suchanski, p.kaniewski, r.matyszkiel}@wil.waw.pl 

2 Military University of Technology, Warsaw, Poland 

piotr.gajewski@wat.edu.pl 

Abstract: Increasing demands are being placed on dynamic spectrum access as a result of overload 

of the frequency spectrum. In this paper we present a concept of frequency broker as a crucial element 

of coordinated spectrum management system for battlefield communication network. 

Critical problem in such system is automation of the collision-free frequency planning and management. 

In this paper, the solution of this problem basing on the graph theory is described. The new 

model of electromagnetic waves attenuation prediction is also presented. 

Keywords: dynamic spectrum management, frequency broker, radio wave attenuation, frequency 

assignment, frequency planning 


The rapid development of the technical systems that use wireless technologies 

causes increasing of the spectrum shortage problem. The devices that are “dependent” 

on the spectrum appear – apart from the classic uses such as e.g. radio broadcasting 

– also as more and more complex systems such as the cellular telephony, WLAN 

802.11, 802.16 (WiMAX) wireless networks, the GPS systems as well as simple 

radio-devices e.g. wireless telephones and devices for home RTV equipment control, 

garage gates control etc. The saturation of such type of devices grows causing overload 

of the frequency spectrum and thus increase of the level of interference. An illustration 

of this phenomenon can be a forecast of use of the spectrum fragment intended for 

broadband mobile systems that has been constructed by the USA national regulator 

(FCC – Federal Communication Commission) that foresees occurrence of significant 

spectrum deficiency for these systems already in 2014 year (see Figure 1) [1]. 

The work is supported by Polish Ministry of Science and Higher Education funds under the project 

No. OR 00018712.


The spectrum static management and methods that have been used so far, 

lead to drastically low effectiveness of spectrum usage. It was proved in many 

measurements that were made in various parts of the world [2]. It was found, on 

the basis of such experiments, that within the frequency below 3 GHz, an average 

level of the spectrum use does not exceed 10% [3]. 

Such observations and experiments have caused growing of common awareness 

of a need of the spectrum use rationalization, among others, by application 

of more efficient management methods. 

Figure 1. Forecast of increasing use of spectrum intended for broadband mobile systems 

That is why at the end of XX century appeared a concept of a dynamic access 

to the spectrum. The most advanced form of which is so called opportunistic access. 

This concept will find its embodiment in future solutions of the cognitive radio. 

A spectrum coordinated management system can be used to rationalize 

the spectrum efficiency of the communication systems that are built on the basis 

of currently operated radio equipment. In some cases to solve this problem, it is 

possible to use the frequency broker that operates in a quasi-real time. Such a possibility 

exists in the communication systems that are based on radio stations susceptible 

to radio data remote control. In polish army, VHF radio stations of the family 

of PR4G and HF type Harris RF-5800H have such possibilities. 

This article describes a concept of the spectrum coordinated management 

system in which the critical problem is to find an effective algorithm to provide 

and to modify the collision-free frequency plans. The presented below algorithm 

is based on the graph theory using a specific models to predict an electromagnetic 

waves attenuation of the path between a transmitter and a receiver. 

II. Frequency broker concept 

The main element of the spectrum coordinated management system is the frequency 

broker. The tasks of the frequency broker are generation and distribution


153 

of radio data for radio networks that are covered by such broker and that ensure 

a collision-free operation. 

To correctly generate the radio data that ensures a collision-free operation 

of the radio networks, the frequency broker must obtain the data that defines a set 

of accessible frequencies. Under Polish conditions the unit responsible for the frequency 

management in armed forces is the Military Office of Frequency Management 

– NARFA PL. To define the organizational structure of the radio communication 

system, the frequency broker should cooperate with automated command 

systems from which it receives information about location and type of all radio 

equipment that is grouped in the radio network. Additionally, to currently verify 

the received frequency set, a close cooperation with the electronic warfare systems 

is recommended. Such cooperation enables an assessment of the electromagnetic 

environment real state in defined nodes of the radio communication system. 

Figure 2 shows a concept of the frequency broker usage that takes into account 

a hierarchic command system from an operational level to a subunit level. 

Figure 2. Concept of broker use in military forces 

Starting from the tactical level, the frequency broker is connected with the command 

system section that is responsible for the communication (G6, J6). It exchanges 

data to execute frequency assessment procedures. The radio data from the broker 

is sent to a properly level of the radio networks that confirm their acceptance or 

generate demand for spectrum resources in a reverse channel. The intermediate 

level broker is connected with a higher or lower level broker to distribute data 

about allocations of the spectrum resources to a lower level and data collection 

about electromagnetic situation in the area of the lower level broker responsibility.


Figure 3 shows a system diagram that takes into account structural connections 

between the NAFRA PL and local brokers and a block diagram of the frequency 

broker that illustrates its modular structure. 

The basic broker element is a planning module that enables generation of radio 

data on the basis of jamming measures and criteria defined earlier. The accepted 

measures and criteria of the jamming result directly from the radio equipment type 

that is used in the radio communication system. For correct assessment of a interfering 

signal and the level of a usable signal it is necessary to use reliable waves 

attenuation models and hence in the further part of the article is presented a radio 

waves attenuation model in an urban area (GUPL) complete with test results that 

verify this model. 

Figure 3. Broker architecture and its functional connections 

The collision-free radio plans is generated by broker taking into account a reliable 

waves attenuation model and some earlier defined data such as: a set of accessible 

frequencies, a structure of radio communication system, jamming measures 

and criteria, radio stations operating mode and parameters. To solve this problem 

an optimization method is used. An example of the greedy algorithm used in our 

broker is presented below. 

III. Concept of frequency planning 

The problem of frequency assignment, even in a simplest model is a NP-hard 

problem (it is a generalized problem of a graph colouring) and that is why a working 

out of a polynomial algorithm that finds an optimal solution is impossible at the present 

state of knowledge. The best known algorithms that bring optimal solutions 

of the NP-hard problems work in an exponential time what means that in practice 

they are suitably only to solve not too complex cases of very limited input data [5].


155 

The data sizes in practical use exceed considerably the ones suitable for solution 

in an exponential time and therefore the worked out algorithms should find 

a possibly best solution in the time assumed in advance what means a necessity 

of selection of one of suboptimal methods used in NP-hard problem solutions. Hence 

for the considered applications, the following types of algorithms can be helpful: 

• approximation and heuristic algorithms (e.g. a greedy algorithm); 

• branch & cut type algorithms that reduce the exponential complexity of indepth 

searching by suitably early elimination of ineffective searching paths; 

• artificial intelligence algorithms: genetic algorithms, ant colony algorithms, 

tabu search, simulated annealing. 

It seems that the simple and small time-consuming method to solve frequency 

assignment question is the greedy algorithm. The greedy algorithm executes always 

operation that is the most advantageous operation at a given moment. Thus, it selects 

a locally optimal solution expecting that it leads to a globally optimal solution. 

A separation matrix is an input element for the frequency assignment algorithm. 

The separation matrix elements determine a smallest possible distance expressed 

in the frequency field that ensures a collision-free operation of two radio networks. 

For co-site locations (the distance between two correspondents of two different 

radio networks is within 1,5-10 m), the radio station producers define 

separation as the percentages that determine the distance between the upper frequency 

of the lower band and the lower frequency of the upper band that ensures 

the networks compatibility. 

The formalized expression of such a state when two radio networks operating 

in the subbands F min1 – F max1 and F min2 – F max2 (F min2 > F max1 ) do not interfere 

themselves has the form: 

B* F F F 

 

min 2 min 2 max1 

Where “B” is a coefficient of a co-site separation and in case of the above 

mentioned radio station of the PR4G family is 0,09. 

To determine the separation coefficients for other networks that are not cosite 

networks, it is assumed that the usable signal “S” reduced by the protection 

coefficient “O” must be greater than the interfering signal “Z”. 

SZO 

On the basis of literature and experimental tests the separation of 10 dB 

has been assumed. 

After correct construction of the separation matrix it is possible to approach 

frequency assignment for the defined radio networks. The input data for such 

an algorithm is: 

• a set of accessible frequencies for use in the frequency assignment process; 

• defined networks complete with the type of used radio stations (an operation 

frequency range, a transmitter power, a receiver selectivity characteristics);


• defined criteria of the co-site separation; 

• operation modes for particular radio networks (a FH frequency hopping 

mode, an operation mode at the DFF fixed frequency); 

• a defined protective coefficient; 

• a separation matrix; 

Moreover, if a given radio network is operating using of the FH mode, the minimum 

and maximum number of frequency subbands should be defined for a given 

radio network as well as hops in the defined frequency subband. 

Figure 4. Algorithm frequency assignment used in frequency broker 

Figure 4 presents a functional diagram of the frequencies assignment algorithm 

that uses the greedy algorithm. The defined structure of the radio communication 

system can be treated as a full graph in which the nodes are the vertexes 

whereas the graph edges are determined on the basis of the values contained 

in a separation matrix. The task of the algorithm of the frequency assignment 

is to solve the travelling salesman problem, it means finding the Hamilton path 

of minimum weights sum. The algorithm sums all edge weights in every node and 

sorts the nodes depending on the obtained sum of the edge weights. In the first


157 

step, a frequency is assigned to the node of the highest sum of the edge weights that 

is treated as the graph starting node. Searching of next “unvisited” graph vertexes 

is done by selection of a lowest weight edge. 

After all vertexes are visited the total sum of the edge weights of the visited 

vertexes is saved. 

In the described algorithm every node is the starting node from which searching 

the Hamilton path is started. 

The result of the algorithm operation is a path of the edge weights smallest 

sum of the visited vertexes. 

In every case it is checked if the number of used frequencies does not exceed 

the set of the accessible frequencies. In a special case, when the algorithm can not 

determine any path that meets the above mentioned criterion, the assignment 

of radio frequencies is not possible. 

After the solution is selected that is the best solution (it requires the smallest 

number of accessible frequencies to allocate them to defined radio networks), there 

is carried out a selection of the radio frequencies that exist in the set of accessible 

frequencies and that are not used and a trial to assign successive frequency subbands 

to co-site networks. 

IV. Prediction of electromagnetic waves attenuation 

To determine an usable signal level and an interfering signal level, waves 

attenuation prediction models are used in the planning module. A detailed 

analysis of such models leads to the conclusion that there is lack of suitable models 

that serve for prediction of radio waves attenuation in urban environments 

in which more and more military operations are carried out for the frequencies 

within 30-88 MHz and 225-400 MHz most often used by army and for the geometry 

of the systems that are characterized by low antennas hanging above ground 

(1-3 m) (see Table I). [1] 

Among the items specified in the Table I, the first three items (Okumura, 

Hata, COTS 231) are the most popular empiric models in the engineering practice 

that are used for prediction of waves attenuation in urban environments. Unfortunately, 

these models are useful only for highly hanged antennas because they are 

constructed for designing of conventional cellular systems in which the base station 

antennas are mounted at high altitudes. Other items presented in the Table 1 show 

the analysis results for frequencies other than for above mentioned Okumura, Hata 

and COST 231 models, even when they refer to parts of the military frequency 

ranges also always regards the situation when one of antennas is mounted at high 

altitude. A confirmation of such an observation are also results of the problem 

thorough analysis included in [2]. Wishing to fill the existing gap, the authors 

of this publication proposed an adaptation of the model given by Rappaport [3] 

that was constructed for the frequencies of 900 and 1800 MHz. Thus, they made


a linear extrapolation of the model coefficients and carried out its initial experimental 

verification. According to that model called the General Urban Path Loss (GUPL) 

the attenuation of waves for the frequency range of 30-88 MHz that propagate 

in an urban environment is expressed by the following formula: 

d 

L[ dB] 10lg( ) 10nlg( ) d FAF 

d 

d 

4 

o o 

Where: 

 

d o – reference distance in a close zone (m) d o 

(when the antenna 

2 

largest dimension < λ, do @ 30 meters); 

λ – wave length (m); 

d – distance between transmitter and receiver (m); 

β – power component which indicates that received power decays with distance 

at a rate of 10β dB/decade; 

n – path loss exponent; 

α – attenuation constant (dB/m); 

FAF – floors attenuation coefficient (dB) 

Author 

Frequency 

(MHz) 

Table I 

Distance 

(km) 

HT 

(m) 

Y. Okumura 15-1920 1-100 30-1000 

HR 

(m) 

M. Hata 150-1500 ≥ 1 30-200 1-10 

COST 231 800-2000 0.02-5 4-50 1-3 

H. Xia 900, 1900 0.001-2 3.2, 8.7, 13.4 1.6 

V. Erceg 1956 0.01-0.5 3.3, 6.6 1.5 

D. Har 900, 1900 0.06-2 3.2, 8.7, 13.4 1.6 

A. Kanatas 1890 0.02-0.18 4 1.7 

H. Masui 3350, 8450, 15750 0.02-0.5 4 2.7 

Y. Oda 457-15450 ≥ 20 

T. Rao 200, 400, 450 0.5-10.5 ≥ 20 3 

N. Blaunstein 902-928 7 2-3 

W. Young 150, 450, 800, 3700 0.108-16.3 138 2 

In Table II are presented values of β, n and α coefficients for the following 

scenarios: 

Scenario 1: outdoor RF propagation in an urban canyon; 

Scenario 2: indoor propagation (same building, same/multiple floor(s)); 

Scenario 3: indoor-to-indoor propagation (between two different buildings, same/ 

multiple floor(s)); 

(1)


159 

Scenario 4: indoor-to-outdoor propagation; 

Scenario 5: outdoor-in-indoor propagation. 

Scenario 

Power component 

β 

Table II 

Path loss 

exponent 

n 

Attenuation 

constant 

α (dB/m) 

1 2,2 1,8 0,06 

2 2,63 1,5 0,65 

3 

5 (if number of penetrated floors = 0), 

4 (if number of penetrated floors > 0) 

2 0,65 

4 3,6 4 0 

5 3,6 4 0 

The analysis of the GUPL model carried out by us shown a necessity of making 

corrects of FAF values given in [2] – in particular for higher than 4 number 

of floors. That is why experiments were carried out. The results of these experiments 

are discussed in [4] and the obtained corrected values of the FAF coefficient are 

presented in the Table III and Table IV. 

Table III 

Recommended FAF 

Number of floors 30 MHz 49 MHz 87,5 MHz 

1 1,9 3,44 1,76 

2 5,85 7,67 5,5 

3 9,1 15,7 10,26 

4 15,46 18,75 14,15 

5 21,25 25,15 19,8 

Table IV 

Recommended FAF 

Number of floors 230 MHz 320 MHz 

1 6,4 7,95 

2 11,8 10,75 

3 14,5 15 

4 21,9 18,15 

5 26,6 20,1 

These experiments enabled at the same time verify positively the GUPL model 

usability for signal attenuation prediction in urban areas – both inside buildings 

and in street “canyons”.


V. Summary 

The problems related to the frequency assignment were subject to wide analyses 

as regards use in the cellular telephony network planning. However, the results 

of these analyses are not useful for the frequency assignment in military radio 

networks what results from a different specificity of these systems. 

In commercial systems, the aim of frequency assignment is to minimize 

the number of used frequencies, what is motivated by the concession costs. In the battlefield 

radio networks it is desired to use maximum number of frequencies, what 

is motivated by the system resistance to intentional interference. 

In the presented article a way of solution of the problem of frequency assignment 

to radio networks has been discussed using the greedy algorithm. The greedy algorithm 

selects locally optimal solution assuming that such a strategy will lead to a globally 

optimal solution. It should be noted that it is one of possible solutions of the travelling 

salesman problem. Another way of optimal frequency assignment is use of the SAT 

solvers. The SAT-solver is a program that for a given formula written in a form CNF 

(Conjunction Normal Form) finds if there exists such a sentence variables substitution 

for which the formula is true. If such a substitution exists the program returns 

it as a result. At the moment, in the Military Communication Institute is conducted 

work on use of the SAT solvers in the frequency assignment. 

References 

[1] http://blogs.uco.edu/graduate/files/2012/02/chart_wireless_data_2.gif 

[2] J. Andrusenko, R.J. Miller, J.A. Abrahamson, N.M. Merheb Emanuelli, R.S. 

Pattay and R.M. Shuford, “VHF General Urban Path Loss Model for Short Range 

Ground-to-Ground Communications” IEEE Transaction on Antennas and Propagation, 

vol. 56, No. 10, 2008. 

[3] T.S. Rappaport, Wireless Communication: Principle and Practice, Upper Saddle 

River, NJ: Prentice Hall PTR, 2002 (second edition). 

[4] P. Gajewski, M. Suchański, P. Kaniewski, R. Matyszkiel, „Prediction of VHF and 

UHF Wave Attenuation In Urban Environment” 19th International Conference on 

Microwave,Radar and Wireless Communications MIKON-2012, May 21-23. 

[5] T.C. Cormen, C.E. Leiserson, R. Rivest, “Introduction to Algorithms”, Massachusets 

Institute of Technology, Thirteenth priinting, 1994.

Spectrum Issues of NATO Narrowband Waveform 

Jan Leduc 1 , Markus Antweiler 1 , Torleiv Maseng 2 

1 Fraunhofer Institute for Communication, Information Processing and Ergonomics FKIE, 

Wachtberg, Germany, jan.leduc@fkie.fraunhofer.de 

2 Forsvarets Forskningsinstitutt FFI, Kjeller, Norway, Torleiv.Maseng@ffi.no 

Abstract: In order to fulfill most of the military operational requirements, two kinds of waveforms 

are needed, a wideband networking waveform, e.g. the Coalition Wideband Networking 

waveform (COALWNW) with high data rates, enabling network centric warfare and a Narrow 

Band Waveform (NBWF) covering long ranges and enabling the parallel transmission of data and 

voice. As a complement to COALWNW, the NATO Line of Sight Communications Capability 

Team is developing a new NBWF. Legacy versions of the NBWF exist as national waveforms only, 

operating historically in the VHF tactical communications band ranging from 30 MHz to 88 MHz 

occupying a channel bandwidth of 25 kHz. The new NBWF is expected to operate in the same 

frequency range, employing the same channelization; furthermore, the NBWF should allow operation 

in the tactical UHF band as well. The current waveform proposal is based on Continuous 

Phase Modulation (CPM), which is widely used in mobile communications, due to the constant 

envelope property of the modulation scheme. 

In this paper, we first introduce principles of the newly defined NBWF. Furthermore, we want to 

point out some unapparent spectrum usage issues of the current proposal. 

Keywords: Narrowband Waveform; Bandwidth; Continuous Phase Modulation; Spectral Efficiency 


In the NATO Technical Note 1246 [1], it was identified that two types of waveforms 

are needed to fulfill all operational requirements. The first kind of waveform 

is a wideband networking waveform enabling high data rates for advanced network 

enabled capabilities. This kind of waveform is addressed in the COALWNW project 

in order to achieve interoperability. The second kind of waveform, a narrow 

band waveform, enabling long range transmission employing the VHF 25 kHz 

channelization is currently defined at NATO within the NATO Line of Sight Communications 

Capability Team. The focus of this paper is on the NBWF. The NBWF 

will be mainly used in Combat Net Radios (CNR) for interoperability on the lower 

command levels. Currently there is no secure and networking capable CNR 

This research project was performed under contract with the Federal Office of the Bundeswehr for Information 

Management and Information Technology, Germany.


waveform available as a NATO Standard, since legacy versions of the NBWF are 

available as national waveforms only. 

There exist several solutions for the NBWF at present, but the one described 

in [2] is a CPM employing a scheme with joint iterative demodulation and decoding 

of convolutional codes. We appreciate, that the lead development was done 

at the Communications Research Centre (CRC), Canada [3], [4]. 

The waveform is described by a variety of modes for a 25 kHz channelization. 

These modes are mainly distinguished by the provided data rates. The three lower 

rate modes, namely NR (R refers to robust), N1 and N2, offering 10 kbps, 20 kbps 

and 31.5 kbps respectively, can be considered as long range modes, while the modes 

N3 and N4, offering 64 kbps and 96 kbps respectively can be considered as high 

throughput modes. It is worth to mention that this CPM-waveform uses a very interesting 

approach to realize higher data rates, which will be described subsequently. 

There are two possibilities for increasing the data rate. The first one is to increase 

the amount of information within one symbol duration by transmitting e.g. 2 Bit/ 

Symbol instead of 1 Bit/Symbol, which is the common way of increasing the data 

rate. The CPM waveform follows another, the second possible approach. While keeping 

the amount of information constant during one symbol duration, the symbol 

duration itself gets reduced, thus the symbol rate increases. When the symbol rate 

increases, it is known that the bandwidth occupation of the system increases as well, 

which is not desired, since the use of a 25 kHz channel is a mandatory requirement 

(Requirement 2.07.03, 2.07.04 and 2.07.08 in [5]). Thus, the maximum phase 

change within a symbol, h the modulation index, needs to be reduced to compensate 

the spectral growth. This technique is very ambitiously applied in the modes N3 

and N4, because the modulation indices become very small. 

In this paper, the modulation parameters of this particular waveform are 

investigated to see, whether the chosen approach for generating higher data rates 

is appropriate for CNR NBWF or not. 

The paper is organized as follows: In Section II, we give some more insights 

on the physical layer of the CPM NBWF as defined in [2]. In Section III, we describe 

the simulation environment. Simulations results are presented in Section IV. 

In Section V, the results and possible consequences are summarized. 

II. CPM NBWF 

CPM is a digital phase modulation of the same family as the Continuous-Phase 

Frequency-Shift Keying (CPFSK) used in many legacy VHF tactical and civilian 

waveforms. The most significant properties are, a constant envelope in the timedomain 

signal, a fundamental robustness to amplitude modulation distortion and 

an efficient utilization of power in the transmit amplifiers since the power amplifier 

can work in saturation. A complete overview of CPM can be found in [7]. 

The complex baseband representation of the CPM waveform without filtering is


163 

2Es 

jt, 

 

st , e 

(1) 

T 

with E s denoting the symbol energy and T s representing one symbol period. The parameter 

denotes the bipolar represented information input values. The phase 

of the n th symbol of this CPM modulation can be computed by 

n 

, 2 , 

t h qt iT 

s 

i 

s 

(2) 

i 

with nT t ( n 1) T . The accumulated phase is defined by 

s 

s 

nL 

i 

(3) 

i 

2h 

n 

K 

1 

where K is a given normalization constant (for the NBWF) and indicates 

2 

the kind of CPM modulation. In case of L = 1 a full response system is considered, 

if L > 1 there is a memory of length L in the system and adjacent symbols overlap. 

The phase shaping pulse is 

t 

qt gd 

(4) 

 

and g represents the – REC (Rectangular) partial response phase shaping 

pulse. 

As already mentioned the CPM-NBWF is described in modes, by employing 

different physical layer parameters, but for all modes there is no transmission 

filter defined [2] and the phase shaping is done with a rectangular pulse. Thus, two 

adjacent modulations symbols are connected by equally spaced samples. The list 

of important waveform parameters are given below in Table I. 

CPM-Mode 

Table I. CPM-Waveform-Modes and specific parameters 

User Data Rate 

[kbps] 

L M h Code Rate 

Symbol Rate 

[ksps] 

NR 10 2 2 1/2 1/3 30 

N1 20 2 2 1/2 2/3 30 

N2 31.5 2 2 1/4 3/4 42 

N3 64 3 2 1/6 4/5 80 

N4 96 3 2 1/11 3/4 128 

In the first columns of Table I, the five different modes are indicated with 

their respective user data rate in the second column. Furthermore, L is denoting 

the ratio of width of the phase shaping pulse and the symbol duration; thus


all specified modes use partial response modulation. The parameter M indicates 

the number of signal states, which is binary for all the different modes; h is representing 

the modulation index (max. phase change in a symbol duration), which 

the developers of the waveform are using to compensate the spectral growth due 

to the increasing symbol rates (and data rate). 

The Forward Error Correction (FEC) is realized with a convolutional mother 

code with code rate R = 1/3 and constraint length 4. The mother code is given by 

the octal generator polynomials G 13,15,17 8 

and is defined to be in a non-recursive 

form. For all modes except NR, the mother code is punctured by the puncturing 

matrices given in [2]. Convolutional codes are vulnerable to burst errors, where 

a contiguous sequence of bits or symbols gets corrupted. In order to break these burst 

errors in single bit errors; an interleaving scheme is employed between FEC and 

modulator. The interleavers have been designed following the “S-Random/Dithered 

Relative Prime (DRP)” approach [6] which ensures a minimum distance between 

adjacent input bits. Since, a joint iterative demodulation and decoding 

scheme is suggested by the inventors in the receiver, the interleaver has another 

important function besides enhancing the robustness against burst errors. 

The interleaver guarantees that the sources of reliability information in the generation 

of extrinsic information on the receiver side are independent, namely 

the independence between the inner code (the CPM modulator) and the outer 

convolutional code. 

A conclusive block diagram with the transmitter of the CPM-NBWF is given 

in Fig. 1. 

Figure 1. Block diagram of CPM-NBWF transmitter 

III. Simulation environment 

From a spectrum perspective, the modes given in Table 1 are following 

the design constraint, that 99% of the power is within 25 kHz bandwidth. As shown 

in [8], even this constraint is not completely fulfilled. The 99% power bandwidth 

is given by: 

• N1 = 25.89844 kHz 

• N2 = 23.78906 kHz 

• N3 = 27.10938 kHz 

• N4 = 28.50000 kHz


165 

However, another evident constraint tells, that all information of the waveform 

is embedded within 25 kHz. We have to bear in mind that the increase of data rate 

is achieved with an increase of the symbol rate and the symbol rate is somehow reciprocal 

to the bandwidth. Hence, we want to analyze, if the decrease of the modulation 

index h is able to compensate the spectral growth due to the increase of the symbol 

rate. Therefore, only the modulation scheme is under analysis, since the FEC shall 

only be used to increase the robustness to interferences and not to compensate 

shortcomings of the application of the modulation scheme. Furthermore, the FEC 

does not contribute to the spectral shaping of this waveform. The robustness to 

interferences is not considered in this paper. 

A block diagram of the simulation is given in Fig. 2. First, uniformly distributed 

random bits are generated and then CPM modulated using the parameters 

of Table I (excluding FEC). The system employs oversampling with factor 16. This 

means that after the modulator each bit (M = 2) for all modes) is represented by 

16 samples. The signal stream is then filtered by an equiripple Finite Impulse Response 

(FIR) low pass filter. The pass frequency f 

pass 

is variable and fstop 

1.2 fpass 

, 

is indicated in Fig. 2. Furthermore, the pass band ripple is limited 0.1 dB and 

the stop band attenuation is given by A 60 dB . 

stop 

Figure 2. Block diagram of simulation chain 

The filtered signal is then demodulated by the respective CPM demodulator 

and the Bit Error Rate (BER) is measured. 

IV. Simulation results 

The previous described setup permits to measure the BER dependent on the (by 

f 

the FIR filter limited) one-sided bandwidth W 1.1fpass 

fpass 

and with 

2 

this, a possibility to verify, whether all information is contained within the 12.5 kHz 

one-sided (25 kHz two-sided) channel.


First, the generated results are presented in Fig. 3, where the BER is drawn 

over the one-sided bandwidth W in kHz. 

Figure 3. One-sided bandwidth W versus BER 

In Fig. 3, it is easy to observe that the mode N1 needs clearly less bandwidth 

than the maximum of 12.5 kHz. Also, the mode N2 can transmit all information 

within a bandwidth smaller than the 12.5 kHz one-sided (25 kHz two-sided) channel. 

One could assume that the symbol rate for those modes could be further increased 

in order to maximize the data rate, but we have to bear the 99% power bandwidth 

in mind. Thus, the long range modes are compliant to the spectral requirements. 

(NR is disregarded here, since NR and N1 have the same modulation parameters 

and only differ in the puncturing). 

The behavior under filtering is very different for the high throughput modes N3 

and N4, since they have a higher bandwidth demand to transmit all the containing 

information. If the maximum bandwidth W=12.5 kHz (one-sided) is granted to 

the system, the mode N3 has a BER of approximately 6% and the mode N4 of approximately 

35%. Those are not tolerable values and to be able to transmit the bits 

at a BER of 10 -7 , the mode N3 would need two channels, thus 50 kHz (two sided) 

and the mode N4 three channels, thus 75 kHz (two sided), in order to be compliant 

to the current NATO channelization. This observation reduces the spectral 

efficiency of the system dramatically, if we assume that NATO will not change 

the 25 kHz channelization for this waveform. The results of the above section are 

summarized in Table II. 

All modes are advertised as occupying 25 kHz bandwidth; the respective 

spectral efficiencies are indicated in Table II. Simulations results show (Fig. 3)


167 

that the necessary bandwidth occupation for the modes N3 and N4 is higher; 

the realistic, corrected spectral efficiency of the high throughput modes N3 and 

N4 is approximately as high as the spectral efficiency of the long range mode N2, 

as can be seen in Table II. Furthermore, the high throughput, low-h, modes are 

more sensitive to noise, as can be seen in Table III. 

CPM-Mode 

Table II. CPM-Waveform with realistic spectral efficiences 

Spectral Efficiency 

a) 

25 

Necessary bandwidth 

in NATO channels 

Spectral Efficiency 

b) 

real 

NR 0.4 1 channels (25 kHz) 0.4 

N1 0.8 1 channels (25 kHz) 0.8 

N2 1.26 1 channels (25 kHz) 1.26 

N3 2.65 2 channels (50 kHz) 1.28 

N4 3.84 3 channels (75 kHz) 1.28 

a) is defined as the user throughput divided by 25 kHz band width 

b) is defined as the user throughput divided by necessary band width 

Table III. Nosie sensitivity of low-h CPM (Oversampling 16) 

CPM-Scheme 

(Modulation parameters) 

M = 2; L = 2; h = 1/2 

(N1 without FEC) 

M = 2; L = 2; h = 1/4 


M = 2; L = 3; h = 1/6 


M = 2; L = 3; h = 1/11 


Necessary E b / N 0 

to reach a BER of 10 -3 Necessary E b / N 0 

to reach BER of 10 -6 

7.5 dB 11.1 dB 

12.8 dB 16.5 dB 

17.7 dB 21.4 dB 

23 dB 26.4 dB 

Thus, combining two or respectively three channels to provide the necessary 

bandwidth to N3 respectively N4 is disadvantageous. It would be the same 

from a spectral occupation (see Table II) perspective and significantly better 

from a robustness (see Table III) perspective, if e.g. the N2 waveform would be 

employed with double or triple data rate and occupying the respective two or 

three frequency channels. 

However, we further want to verify, if the modulation index can be used to 

compensate the spectral growth of the waveform, due to the increase of the symbol 

rate. To do so, the results are normalized to the symbol rate and in Fig. 4 

B . T s (T s is the symbol duration; B = 2W) versus the BER is plotted.


Figure 4. B . T s (T s is the symbol duration; B = 2W) versus BER 

It can be observed that all modes have a different normalized bandwidth 

occupation. The higher the throughput, the lower the bandwidth needs. The valid 

question at this point is, if the normalized bandwidth gains of N3 and N4 are caused 

by the reduction of h, because the value of L (ratio phase pulse length / symbol time) 

also has an impact on the bandwidth [7] and L = 3 for the modes N3 and N4 and 

L = 2 for the modes N1 and N2. To verify this, further simulations are conducted, 

with the modulation settings of N3 and N4 but employing a value of L = 2. These 

results are plotted using dashed lines in Fig. 4. 

As it can be seen, these “verification modes” perform normalized to the symbol 

rate very similar to the mode N2. Thus, if L = 2 is used for the low h modes 

(N3 and N4), they are almost as bandwidth efficient as N2. Thus, the observed 

bandwidth gains between the mode N2 and N3/N4 are due to the higher values 

of L. Hence, there are only negligible bandwidth gains from reducing h beyond ¼. 

V. Conclusions 

In this paper the upcoming NATO NBWF has been analyzed, regarding 

the necessary bandwidth occupation of the system. This particular waveform 

is keeping the amount of information constant during one symbol duration and


169 

to increase the data rate, the symbol duration itself gets reduced and the symbol 

rate increases. Thus the bandwidth occupation of the system increases as well, 

which is not desired, since the use of a 25 kHz channel (two-sided) is a mandatory 

requirement [5]. This increase of bandwidth is compensated with the reduction 

of the modulation in h of the CPM modulation scheme, which is the maximum 

phase change within a symbol duration. Therefore, while making the symbols 

duration shorter, the phase change within a symbol is also reduced. It is shown 

that the spectral growth cannot be compensated by a reduction of h. The modes 

N3 and N4 occupy (to function properly) two and respectively three 25 kHz 

channels, which does not meet [5], especially because military spectrum is a very 

scarce resource. 

On the other hand the modes N1 and N2 are very robust and well-defined 

waveforms fulfilling all requirements regarding spectral occupation [5]. Therefore, 

we propose the following two steps: 

1. Move on with N1 and N2 for a fast ratification and make these new waveforms 

available to the user 

2. Revisit higher data rate modes in the future for a possible annex and ensure 

usable modes fulfilling requirements of all nations. 

There are certain possible solutions for overcoming the issue pointed out 

in the paper within the second proposed step. 

The other waveform candidates, which were submitted to the LOS Comms 

CaT could be taken into consideration for the high throughput modes or modifications 

would had to be made to the current waveform, 

If the CPM-waveform is kept also for the high throughput modes there are 

also certain modification possibilities. 

First, the modulation parameters of the mode N2 are kept, but the number 

of bits per symbol M = 1 would have to be increased to M = 2 respectively M = 3, 

which would double/triple the data rate (63 kbps resp. 94.5 kbps) on the expense 

of the demodulator complexity [7]. 

Second, the symbol rate of N2 gets increased (doubled/tripled offering (63 kbps 

resp. 94.5 kbps); occupying two resp. three 25 kHz channels, while being fairly 

resistant to interference compared to the current modes N3 and N4. 

References 

[1] Technical Note 1246, ‘Wireless communication architecture (land tactical): scenarios, 

requirements and operational view’, NC3A, The Hague, Netherlands, 2007. 

[2] LOS Comms CaT (formaly known as: SC/6 AHWG/2), “Technical Standards for Narrow 

Band Physical Layer of the NATO Network enabled Communications Waveform and 

VHF Propagation Models (STANAG Draft 4),” 2010.


[3] C. Brown and P. Vigneron, “Spectrally Efficient CPM Waveforms for Narrowband 

Tactical Communications in Frequency Hopped Networks”, Military Communications 

Conference, 2006. IEEE MILCOM 2006, 23-25 Oct. 2006. 

[4] C. Brown and P. Vigneron, “Equalisation and Iterative Reception for Spectrally 

Efficient CPM in Multipath Environments”, Military Communications Conference, 

2010. IEEE MILCOM 2010, Oct. 31-Nov. 3 2010. 

[5] NATO C3B, “Requirements for a Coalition Tactical Radio Waveform,” (AC/322-N 

(2011)0010-REV1), January 2011. 

[6] D. Divsalar, G. Montorsi, S. Benedetto and F. Pollara, “Serial concatenation 

of interleaved codes : Design and performance analysis,” in IEEE Trans. Inform. 

Theory (1998), April, pp. 409-429. 

[7] J.G. Proakis, “Digital Communications,” Mcgraw-Hill Publ.Comp., Edition: 4., August 

2000. 

[8] J. Nieto, “NBWF Investigations”, LOS Comms Cat Workshop, Rome, Italy, March 2011.

Legacy Waveforms on Software Defined Radio: 

Can Hierarchical Modulation Offer 

an Added Value to SDR Operators 

Marc Adrat 1 , Tobias Osten 1 , Jan Leduc 1 , Markus Antweiler 1 , 

Harald Elders-Boll 2 

1 Fraunhofer Institute for Communication, Information Processing and Ergonomics FKIE, 

Wachtberg, Germany, marc.adrat@fkie.fraunhofer.de 

2 Cologne University of Applied Sciences, Cologne, Germany, harald.elders-boll@fh-koeln.de 

Abstract: Military tactical communications is taking the next step in its evolution. Many nations 

spend considerable efforts to bring the novel Software Defined Radio (SDR) technology into service. 

SDRs allow military radio operators to change waveforms on-the-fly according to the mission needs. 

New capabilities can be loaded as so-called Waveform Application (WFA). Before novel waveforms 

with wideband networking capabilities will be available in the future, most nations have launched 

projects to port legacy waveforms to SDRs. These WFAs on the modern SDRs shall ensure interoperable 

communication to legacy radios in situations where both types of radio equipment are deployed 

at the same time in the same mission. 

In this paper, we present first results of our analysis if an added value can be provided to the operators 

at SDRs, e.g., in terms of a higher data throughput or robustness. As an example, we apply the concept 

of hierarchical modulations to a legacy waveform. The modulation scheme of the legacy waveform 

acts as the base-layer which ensures the over-the-air interoperability to legacy radios. Additional 

information can be transmitted between SDRs only utilizing some extra enhancement-layers. 

Keywords: Software Defined Radio, Waveform Application, Hierarchical Modulations, Bit Interleaved 

Coded Modulation with Iterative Decoding 


It is expected that the modern Software Defined Radio (SDR) technology will 

considerably extend the capabilities of future military tactical communications. 

Among many other advantages of SDR technology, one key benefit is that waveforms 

can be loaded onto the SDRs as so-called Waveform Applications (WFA) 

according to the mission needs. Another key benefit is that SDRs are expected to 

be powerful enough to host modern WFAs with wideband networking capabilities. 

This research project was performed under contract with the Federal Office of the Bundeswehr for Information Management 

and Information Technology, Germany.


Such Wideband Networking Waveforms (WNW) typically set the most demanding 

requirements for the design of high capable heterogeneous SDR platforms. 

However, if WFAs with lower computational demands are deployed on SDRs, like 

legacy waveforms, some of the computational resources remain unused. These 

spare resources might probably be beneficially utilized by applying advanced signal 

processing algorithms. An example will be discussed later on in this paper. 

Moreover, it can be anticipated that SDR technology will be introduced to the 

military forces in a step-by-step process. There will be a significant period of time 

where legacy radios and modern SDRs hosting the corresponding legacy WFA will 

be deployed at the same time in the same mission. Guaranteeing interoperability 

between both types of radio equipment, legacy radios and modern SDR, is the 

utmost requirement in the Porting process. Porting in this context means that the 

waveform functionality which is known from the legacy radio is implemented 

as a piece of software which runs on the SDR platform. This piece of software is called 

Waveform Application. Usually, interoperability is guaranteed by reproducing the 

known functionality of the legacy radios one-to-one in the WFA for SDRs. 

However, as we have already proposed in [1], it might be of value to apply 

modern advancements in digital signal process ing at the receiving end of a communication 

link. While keeping interoperability on the air interface, the operator 

at the receiving SDR can experience a benefit if compared to the operator at the 

legacy radio thanks to the more sophisticated receiver signal processing. Such 

benefit can be, e.g. a reduced bit error rate or an extended communication range. 

In the example given in [1] the concept of Bit Interleaved Coded Modulation 

with Iterative Decoding (BICM-ID) [2] [3] has been applied at the receiver using 

some MIL-STD188-110B-like waveform modes [4]. It has been shown that some 

gains are achievable, but unfortunately that these gains have to be considered 

as negligibly small if the concept of BICM-ID is directly applied to the standardized 

configuration settings. It has also been shown that substantial gains can be 

realized if a change of a single line of software code at the transmitter is tolerable. 

Of course, this would violate the key requirement to guarantee interoperability to 

legacy radios. 

In this paper, we analyze if the concept of Hierarchical Modulation can beneficially 

be applied. Again we use some MIL-STD188-110B-like waveform modes 

as example. Our objective is to provide some extended services to the operators 

at the SDRs. Using the known configuration settings from the legacy waveform 

as a base-layer in the hierarchical modulation scheme, allows ensuring interoperability 

between legacy radios and SDRs. Thanks to some enhancement-layers in the 

hierarchical modulation scheme additional data becomes transferable within an 

SDR-to-SDR communication link. 

Important note: In this paper, we focus on presenting the basic idea, discussing 

some useful design guidelines as well as showing first simulation results. These 

simulation results focus on the analysis if an enhanced data throughput can be


173 

provided as added value to the operator at an SDR. In a second complementing 

paper, see [8], we discuss if a higher robust ness (and with this a higher range) can 

be realized. 

This paper is structured as follows. In Section II we explain the idea of hierarchical 

modulation in the context of legacy waveforms on SDRs in more detail. 

In Section III we describe the simulation environment. Simulations results are 

presented in Section IV. After having given a short outlook on the results in [8] and 

having described some future steps of our analysis in Section V, we finally conclude 

our findings in Section VI. 

II. Hierarchical modulations in the context of legacy 

waveforms on SDRs 

Hierarchical modulation allows to multiplex and to modulate multiple streams 

of user data into a single stream of modulation symbols. It is sometimes also referred 

to as Layered Modulation because one of the input data streams determines 

the base-layer symbols and the other input data streams determine the extra 

enhancement-layer symbols. 

One prominent practical implementation of hierarchical modulation can 

be found in the area of digital video broadcasting where the base-layer carries 

the information of a robust, but typically low-resolution video stream. The extra 

enhancement-layers, which might only be decodable by receivers under very good 

channel conditions, allow to increase the resolution and therewith the quality 

of the video. 

Thus, our basic idea is to apply a similar technique in the porting process 

of legacy waveforms to modern software defined radios. In the ported counterpart 

of the legacy waveform, the information of the legacy waveform represents the 

base-layer guaranteeing interoperability. On top of that the operators at SDRs can 

transmit additional data using the extra enhancement-layers. 

A. Example base-layer and enhancement-layers 

To simplify matters, in the following we restrict our considerations to a comprehensible 

example using an 8-PSK signal constellation as base-layer. Such a digital 

modulation scheme is widely used in a couple of present military tactical communication 

schemes like in NATO STANAG 4285 [5] or MIL-STD188-110B 

Appendix C [4]. The extension of our considerations to other digital modulation 

schemes is straight forward. 

The left part of Figure 1 shows an 8-PSK signal constellation set with a Gray 

labeling for the individual symbols. This set shall serve as the base-layer for our 

hierarchical modulation schemes. Note, in this simple example the base-layer carries 

information solely in its phase.


Figure 1. Different Signal Constellation Sets with exemplary symbol labels; left: 8-PSK base-layer, 

center: 8-PSK base-layer plus 4-ASK enhancement layer (star); 

right: 8-PSK base-layer plus 4-QAM enhancement layer (diamonds) 

In order to provide an add-on to the operators at SDRs we are aiming for 

transmitting some additional data in so-called enhancement-layers. Two examples 

are shown in the center and in the right part of Figure 1. 

In the first example, we apply a 4-ASK enhancement-layer as an overlay to 

the 8-PSK base-layer. This results in an overall 32-QAM scheme which would 

allow the operator at the SDR to transmit two additional bits. Simply speaking, 

the base-layer selects one of the eight sectors in the I/Q-plane (inphase/ 

quadrature) while the enhancement-layer selects one out of four magnitudes 

in each sector. Later on, this scheme will also be referred to as “star-shaped” 

32-QAM scheme. 

In the second example we again apply an enhancement-layer with 4 signal 

constellation points to the 8-PSK base-layer. However, inspired by [6] the slightly 

different placing of signal constellation points allows exploiting the I/Q-plane more 

effectively. In the rest of this paper, this scheme will be called “diamond-shaped” 

32-QAM scheme. 

Two categories of questions arise immediately: 

• How to design the enhancement-layer in detail How to place the signal 

constellation points How to label them 

• What are the beneficial/adverse consequences with respect to the performance, 

e.g. in terms of bit error rate behavior What are the consequences 

for the operators at the legacy radios resp. software defined radios 

Answers to the design aspects (first category) will be given in the next subsection. 

The consequences (second category) will be analyzed in Sections III and IV. 

B. Designing the enhancement-layer 

When designing the enhancement-layer we have to consider a couple of design 

rules. Among others, these are:


175 

1. The mean energy E S required per modulation symbol shall be the same for the 

original 8-PSK scheme and the extended star- resp. diamond-shaped 32-QAM. 

2. The Euclidean distance between any pair of signal constellation points shall 

be such that not a single pair exists which reveals a dedicated bottleneck 

with respect to the bit error rate performance. 

3. The symbol labels shall be optimized in view of a receiver using the 

concept of Bit-Interleaved Coded Modulation with Iterative Decoding 

(BICM-ID) [2][3]. 

The first two design rules are related to the placing of signal constellation 

points in the I/Q-plane while the third design rule is dedicated to the labeling. 

Both aspects will be optimized separately from each other. 

1) Placing of signal constellation points for the star-shaped 32-QAM scheme: Figure 

2 shows an extract of the star-shaped 32-QAM scheme. 

Figure 2. Design of star-shaped 32-QAM scheme (extract) 

Let us assume that the four signal constellation points representing the 4-ASK 

enhancement-layer are equidistantly separated by the distance α. The 4-ASK scheme 

shall be centered around β. In order to make sure that the mean energy E s of the 

resulting star-shaped 32-QAM scheme is the same as for the original 8-PSK scheme 

(see first design rule), it can be shown that α and β have to satisfy the condition 

5 2 

. 

E S 

 

4 

(1) 

If the spacing between the four signal constellation points representing the 

4-ASK enhancement-layer tends towards zero, i.e. 0, the term . 

E S


Thus, the star-shaped 32-QAM scheme degenerates to the original 8-PSK scheme 

(this fact can be considered as a consistency check). If we increase α, the term β starts 

to decrease. While doing so, we have to take into consideration that the Euclidean 

distance between the inner signal constellation points on radius r * 1 becomes smaller 

the larger α resp. the smaller β is. In order to satisfy the second design rule, we enforce 

* 

that the Euclidean distance between the inner signal constellation points on radius r 1 

is also α. It can be shown that in this case α and β have to satisfy the condition 

13sin 22.5 

 

. 

(2) 

2 sin 22.5 

Solving Eqs. (1) and (2) for E S = 1 yields α = 0.331 and β = 0.928999. From 

that we can easily determine the radii of the star-shaped 32-QAM scheme shown 

in the center of Figure 1 

r 

∗ 

1 

r 

∗ 

2 

r 

∗ 

3 

r 

∗ 

4 

= 0. 

432499 

= 0. 

763499 

= 1. 

094499 

= 1. 

425499 

with * 0.331. 

(3) 

2) Placing of signal constellation points for the diamond-shaped 32-QAM scheme: 

Figure 3 shows an extract of the diamond-shaped 32-QAM scheme. 

Figure 3. Design of diamond-shaped 32-QAM scheme (extract) 

In this scheme we assume that two of the signal constel lation points are placed 

with a fixed phase offset of 22.5° on a circle with radius r r r 

. 2 

 

3 

A third point 

is located at a larger radius r ◊ 4 , while the Euclidean distance α to the first two points


177 

on radius r shall be the same as in between the first mentioned two points. The 

fourth point is located at a smaller radius r 1 ◊ . Again, in order to make sure that the 

mean energy E s of the resulting diamond-shaped 32-QAM scheme is the same as for 

the original 8-PSK scheme (see first design rule), it can be shown that r and r 1 

◊ 

have to satisfy the condition 

1 

 

 

2 

2 

4 

S 

2 cos11.25 3sin11.25 . 

r E r 

(4) 

Note, while doing the consistency check, whether the diamond-shaped 32-QAM 

scheme degenerates to the original 8-PSK scheme, e.g. by letting r 

1 

r 0 , we 

have to keep in mind that at the same time the phase offset of 11.25° disappears. 

With this in mind, the term in the inner parenthesis in (4) becomes 1. 

Again, in order to ensure that the Euclidean distance between the inner signal 

constellation points on radius r ◊ 1 is at least α, the terms r and r ◊ 1 have also to satisfy 

the condition 

sin11.25 

r1 

r . 

(5) 

sin 22.5 

From (4) and (5) follows 

 

r 0.5098 

 

r2 r3 

1.0001 with 0.3902. 

 

r 1.31878 

4 

The minimum Euclidean distance between any combination of two signal constellation 

points in the diamond-shaped 32-QAM scheme is α ◊ = 0.3902. This is slightly 

higher than for the star-shaped 32-QAM scheme α* = 0.331. This offers the potential 

for a higher performance as it will be discussed in more detail in Section IV. 

3) Finding the optimal labels for the star-shaped resp. diamond-shaped 32-QAM 

scheme: The symbol labels for both 32-QAM schemes shall be chosen such 

that a receiver design based on the BICM-ID concept is optimally supported. 

The concept of BICM-ID was first described in [2] and is based on a serial 

concatenation of a Forward Error Correction (FEC) component, a bit-level interleaver 

and a digital modulation scheme. On the receiving end, there is a feedback loop 

between the BCJR-decoder [7] and the demodulator. So-called Extrinsic Information 

generated by the BCJR-decoder is interleaved and then fed back to the demodulator 

as additional a priori information for the received channel symbols. 

It has already been shown in [3] that the BICM-ID concept is optimally 

supported if a so-called Semi Set Partitioning (SSP) symbol labeling is applied. 

For a given signal constel lation the SSP mapping ensures that the so-called 

Harmonic Mean d H [3] is maximized. Table I shows the Harmonic Means for all 

signal constellations under consideration in this paper. 

 

 

(6)


Table I. Harmonic Means for different Signal Constellation Sets 

Restricted SSP 

Free-running SSP 

Signal Constellation Set 

Harmonic Mean d H 

8-PSK (Gray) 0.809 

Star-shaped 32-QAM 0.470 

Diamond-shaped 32 QAM 0.403 

8-PSK 2.877 

Star-shaped 32-QAM 2.501 

Diamond-shaped 32 QAM 2.520 

If we are totally free in designing the symbol labeling the maximum achievable 

values are given in the lower part of Table I (marked as free-running SSP). It can 

be seen that (besides the better α ◊ > α* mentioned above) the diamond-shaped 

32-QAM offers also a slightly higher Harmonic Mean d H if compared to the starshaped 

32-QAM. Note, a comparison to the optimal d H for the 8-PSK constellation 

with a free-running SSP is not fair because due to the higher number of signal 

constellation points in the 32-QAM schemes, the Euclidean distances (which are 

taken into account by the Harmonic Mean) are typically smaller. 

However, since our key objective is to ensure interopera bility to legacy radios 

we are not totally free in the design of the symbol labeling. We have to take into 

account the labeling of the original 8-PSK scheme. For instance, in case of the 

4.8 kbps mode of MIL-STD188-110B Appendix C [4] the original labeling is given 

by a Gray labeling. The left part of Figure 1 shows the corresponding example. In this 

paper, when designing the labelings for the 32-QAM schemes we restrict ourselves 

to the case in which the labeling of the base-layer is part of the overall labels. For 

the given example considered here that means that the Gray labeling of the 8-PSK 

scheme determines the three leftmost bits in the overall five bit long labels for the 

32-QAM schemes. Taking this restriction into account yields the Harmonic Mean 

values mentioned in the upper part of Table I. Obviously, the restriction results 

in significantly smaller Harmonic Mean d H values if compared to the free-running 

optimization approach. Consequently, the attainable performance gains will be 

significantly smaller. The corresponding SSP symbol labelings are shown in the 

center and in the right part of Figure 1. 

III. Simulation environment 

In the following, the pros and cons of considering hierarchical modulation 

in the porting process of legacy waveforms to SDRs shall be analyzed using a simulation 

example. For this purpose we use a simulation environment where the baselayer 

(i.e., the legacy system) resembles the 4.8 kbps mode of MIL-STD188-110B 

Appendix C [4].


179 

A block diagram for the simulation environment under consideration is shown 

in Figure 4. 

Figure 4. Block diagram of simulation environment with base-layer and enhancement-layer 

signal processing 

A. Base-layer 

In the 4.8 kbps mode of MIL-STD188-110B Appendix C the input data is at first 

encoded by a rate R = 1/2, constraint length L = 7 convolutional mother code with 

generator polynomial G = {171,133} 8 . Afterwards the codewords are punctured 

to R p = 3/4 using the puncturing pattern (1 1 1 0 0 1). 

The punctured codewords are block interleaved using one out of several possible 

interleaver sizes. The interleaver size is mainly determined by the maximum 

acceptable delay on the communication link and it must be a multiple i of 768 bits 

(with i = 1,3,9,18,36,72). 

In contrast to the original 4.8 kbps mode of MIL-STD188-110B Appendix C [4] 

in this paper we neglect aspects like scrambling or synchronization. We focus on 

an interleaver size of 6912 only, i.e. i = 9. Instead of a full tail-biting convolutional 

code we use a terminated convolutional code. Anyhow, we assume that none of these 

simplifications has a major impact on the relevance of our findings for an established 

real-world legacy system. 

B. Enhancement-layer 

Similar simulation settings are applied to the enhancement-layer. The only 

difference is that we have to take into account the lower number of bits which are 

transmitted on the enhancement-layer (only 2 bits instead of 3 bits are considered 

in the determination of modulation symbols). From that follows that the interleaver 

size in the enhancement-layer is 4608. It is designed as an S-random interleaver. 

C. Combination of base-layer and enhancement-layer 

After having determined the interleaved blocks of punctured codewords 

for the base-layer and the enhancement-layer, both data streams are multiplexed 

to a single data stream. As mentioned earlier, in this paper we restrict ourselves


to the case in which the 3 bits of the base-layer determine the three leftmost bits 

and in which the 2 bits of the enhancement-layer determine the two rightmost bits 

of the overall 5 bit long labels at each signal constellation point (see labeling 

in Figure 1). 

As transmission channel serves an Additive White Gaussian Noise (AWGN) 

channel with known E S /N O where E S determines the energy per modulation 

symbol and N O the power spectral density of the AWGN. At the receiving end 

of our simulation chain we apply the concept of BICM-ID independently to both, 

the base-layer and the enhancement-layer. Note, the behavior of a legacy radio 

is inherently included by decoding the base-layer without any BICM-ID iteration 

(i.e., no feedback loop in Figure 4). 


Table II summarizes all relevant cases which need to be considered during 

the evaluation. 

Table II. Relevant cases to be considered (TX: transmit; RX: Receive) 

Receiver 

Legacy Radio 

Software 

Defined Radio 

Legacy Radio 

TX: Base-Layer 

RX: Base-Layer 

TX: Base-Layer 

RX: Base- & Enhanc. Layer 

Transmitter 

Software Defined Radio 

TX: Base- & Enhancement-Layer 

RX: Base-Layer 

TX: Base- & Enhancement-Layer 

RX: Base- & Enhancement-Layer 

A. Legacy radio as receiver & arbitrary transmitter 

In a first experiment we want to analyze the performance for the cases in which 

a legacy radio acts as receiver. We can expect a reduced performance if we use an 

SDR as transmitter because of the adverse effects of the enhancement-layer. 

Figure 1 illustrates the reason for our expectation. The demodulator of a legacy 

system will make a decision for the most probable sector in the I/Q-plane. As an 

example the sector for the bit pattern 000 is highlighted in light-grey. This sector 

is exactly the same for all three cases, i.e. for a legacy radio transmitting the baselayer 

only (see leftmost plot in Figure 1) as well as for the star-shaped resp. diamondshaped 

SDR cases (see plots in the center and on the right of Figure 1). However, 

the Euclidean distance of the signal constellation points to the sector resp. decision 

boundaries becomes usually smaller for the SDR cases. Thus, a performance loss 

in terms of Bit Error Rate (BER) behavior has to be tolerated. 

Figure 5 shows the corresponding simulation results.


181 

Figure 5. Bit error rate performance of Legacy Receiver for Legacy resp. Software Defined 

Radio Transmitters 

As expected, it can be seen that the operator at the legacy radio will experience 

a performance loss of up to ~ 4 dB in E S /N O if a SDR with one of the 32-QAM 

schemes serves as transmitter. The star-shaped 32-QAM scheme behaves slightly 

better than the diamond-shaped 32-QAM scheme. One reason for this observation 

can be found in the fact that in the star-shaped 32-QAM scheme only the two 

signal constellation points on radii r * 1 and r * 2 exhibit a smaller distance to the sector 

resp. decision boundary than for the original 8-PSK scheme while in the diamondshaped 

32-QAM scheme these are three, namely the points on radii r ◊ 

1 and r 2 

r 3 

(cmp. to Figure 1). 

Notice, as mentioned earlier the behavior of a legacy receiver is inherently 

included in Figure 4 by decoding the base-layer without any BICM-ID iteration 

(i.e., no feedback loop). In addition, Figure 5 also shows for all three transmitters 

a second curve labeled Error-Free Feedback (EFF). These curves illustrate the best 

possible BER performance attainable in a BICM-ID system. For this purpose, 

during the simulation the values on the feedback loop are directly taken from the 

transmitter side. With this, they can be considered to be error-free. Obviously, due 

to the Gray labeling the perfor mance differences between the non-iterative legacy 

scheme as well as the error-free feedback bound are negligibly small. 

B. SDR as receiver & legacy radio as transmitter 

In this case the base-layer of the SDR receiver reveals a similar performance 

as the legacy receiver because the sector resp. decision boundaries are identical 

(cmp. to Figure 1). The only difference is that the additional interference/noise


being introduced by the enhancement-layer is not added at the transmitter of the 

communication link, but at the receiving end. The corresponding simulation results 

have already been presen ted in Figure 5. Since the transmitter was a legacy 

radio, there is no information on the enhancement-layer. Thus, decoding the 

enhancement-layer does not provide any benefit and can be skipped. The decision, 

if the enhancement-layer needs to be decoded, can be derived from the side information 

whether transmitter is a legacy radio or an SDR. We assume that this side 

information can be provided to all radio operators once in advance, e.g. during the 

network planning/management phase. 

C. SDR as receiver & SDR as transmitter 

Last but not least, we consider the case using an SDR on both ends of the 

communication link. Figure 6 shows the corresponding simulation results. 

Figure 6. BER performance of SDR receiver for SDR transmitter 

As a reference we also plotted a copy of the BER curve for the case where we 

have legacy systems on both ends. In order to allow a fair comparison between all 

curves they are plotted as a function of the mean energy E B per user data bit. As 

a consequence the BER curve for the legacy system shown in Figure 5 is shifted by 

10 • log 10 (4 / 3 • 1 / 3) = –3.52 dB to the left (w.r.t. the punctured code rate R p 

and number of bits per symbol). Since we transmit in total 5 bits per symbol 

using the SDR modes, all the other curves include a shift to the left by 

10 • log 10 (4 / 3 • 1 / 5) = –5.74 dB. 

Figure 6 shows two sets of simulation results, one for the star-shaped 32-QAM 

scheme and one for the diamond-shaped 32-QAM scheme. Each set contains BER


183 

curves for different numbers of iteration I = 0,1,5,10 and for error-free feedback. 

It can easily be seen that performance improvements are possible by iterations for 

both sets. In addition, the diamond-shaped 32-QAM scheme outperforms the starshaped 

32-QAM scheme thanks to the better properties mentioned in Section II.B. 

However, it can also be seen that due to the restrictions in the design of both 

schemes, there is still a considerable gap to the reference curve. The SDR modes 

perform even worse than the original 8.0 kbps mode of MIL-STD188-110B 

Appendix C [4] which also allows transmitting 5 bit per symbol. 

Note, the simulation results for both 32-QAM schemes reveal a plateau at a BER 

of 0.2. The reason for this plateau can be found in the fact that the plotted BER 

curves show the mean BER for the base-layer and the enhancement-layer. Since the 

base-layer is more robust than the enhancement-layer, the three bits transmitted 

on the base-layer exhibit a much better BER performance than the two bits on the 

enhancement-layer. Thus, in this range of E B /N 0, where the plateau appears, the 

mean BER can be approximated by 2 / 5 • 0.5 = 0.2. 

V. Next steps in our ongoing research activity 

In the preceding sections the basic idea, the design, and some first results of our 

ongoing research activity have been described. Unfortunately, these first results 

show that with the given restrictions in the design of the 32-QAM schemes no 

added value, in terms of data throughput, can be offered to the operator at the SDR. 

A. Improved robustness 

Anyways, in parallel we already started some investigations if an improved 

robustness (and with this maybe communication range) can be offered as added 

value, e.g., by transmitting extra error protection information on the enhancementlayer. 

For instance, we can transmit the parity check information which had been 

eliminated from the coded data stream by puncturing. 

Figure 7 shows a first exemplary simulation result where we use a rate 9/20 

FEC code in combination with the 32-QAM scheme instead of the rate 3/4 code 

and the 8-PSK scheme. The effective ratio of data bits carried in each modulation 

symbol remains the same, while at the same time interoperability to the legacy 

system remains preserved. 

The simulation results in Figure 7 show that thanks to the stronger error 

protection scheme an added value in terms of higher robustness can be provided 

to the SDR operator. A single iteration is sufficient to provide a gain if compared 

to the legacy 8-PSK scheme. After 5 Iterations gains of more than 3 dB in E S /N O 

can be realized. 

A more detailed analysis of this case is beyond the scope of this paper and 

will be presented in [8].


Figure 7. BER performance of robust SDR schemes transmitting extra error protection 

information on the enhancement-layer 

B. Further potentials for improvement 

In addition, we see some aspects which offer the potential for improvement 

of both usages of the enhancement layer, either for the higher throughput or the 

higher robustness. 

Currently, the design of the Multiplexer resp. De-Multiplexer shown in Figure 4 

reveals some limiting restrictions. The Multiplexer combines the data streams of the 

base-layer and the enhancement-layer to a single data stream by simply combining 

the 3 bit resp. 2 bit patterns to a single 5 bit long label. As a consequence, the 

BICM-ID decoders for the base-layer as well as the enhancement-layer shown 

in Figure 4 run strictly in parallel and cannot benefit from each other. In addition, 

from the current Multiplexer design it follows that the Harmonic Mean d H of the 

restricted SSP optimization is substantially smaller than the d H of a free-running 

SSP optimization (see Table I). A more sophisticated Multiplexer/Combiner might 

provide higher Harmonic Mean d H values and it might support a joint decoding 

approach of both layers. 

Figure 8 shows a simulation using the free-running SSP labeling. This simulation 

can be considered as an upper practical performance bound for an alternative 

Multiplexer/ Combiner design. Obviously, significant performance gains can be 

realized if compared the preceding simulations using the restricted SSP labeling. 

As a next step of our ongoing research activity, we will analyze how to design 

an alternative Multiplexer/Combiner which comes closer to the practical bound. 

In addition, we will investigate how close this practical performance bound can 

be approached.


185 

Figure 8. BER performance of SDR Receiver for SDR transmitter with an optimal SSP 

symbol labeling 

VI. Conclusions 

In this paper, we analyzed if an added value can be provided to an SDR 

operator using an enhanced legacy waveform. For this purposes the concept 

of hierarchical modulations has been applied in the porting process. The original 

legacy waveform is represented as the base-layer. The focus of the present paper 

has been to present the basic idea, the major design considerations as well as some 

first simulation results. Unfortunately, these results have shown that due to some 

restrictions in the design process no gains in terms of data throughput can be 

realized. However, an outlook on an ongoing research activity has been given 

which allows providing a significant gain in terms of robustness. More details 

will follow in a complementing paper [8]. 

Finally, with respect to the original question being raised in the title of this 

paper, we conclude: partially yes, applying hierarchical modulations in the 

porting process of legacy waveforms allows realizing an improved robustness. 

We also see a potential for providing a higher throughput, but, however, so far 

we have not been able to exploit the additional capacity of the enhancementlayer 

to realize this.


References 

[1] J. Leduc, M. Adrat, M. Antweiler, H. Elders-Boll, „Legacy Waveforms on Software 

Defined Radios: Benefits of Advanced Digital Signal Processing“, in Proc. of NATO 

RTO Information Systems Technology Panel Symposium (IST – 092 / RSY – 022), 

Breslau (Poland), Sept. 2010. 

[2] X. Li and J.A. Ritcey, “Bit Interleaved Coded Modulation with Iterative Decoding”, 

IEEE Communications Letters, pages 169-171, May 1998. 

[3] X. Li, A. Chindapol and J.A. Ritcey, ”Bit-Interleaved Coded Modulation with 

Iterative Decoding and 8-PSK Signalling”, IEEE Transactions on Communications, 

pp. 1250-1257, August 2002. 

[4] U.S. DoD Interface Standard MIL-STD-188-110B “Interoperability and Performance 

Standards for Data Modems”, App. B, April 2000. 

[5] NATO Military Agency for Standardization (MAS), “STANAG 4285: Characteristics 

of 1200/2400/3600 Bits Per Second Single Tone Modulators/Demodulators for HF 

Radio Links”. 

[6] A. Seeger, "A new Signal Constellation for the Hierarchical Transmission of Two 

equally sized data streams" in Proc. IEEE ISIT, p. 169, Ulm, Germany, June 1997. 

[7] L. Bahl, J. Cocke, F. Jelinek, and J. Raviv, ”Optimal decoding of linear codes for 

minimizing symbol error rate”, IEEE Transactions on Information Theory, vol. 20, 

no. 2, pp. 284-287, 1974. 

[8] M. Adrat, T. Osten, J. Leduc, M. Antweiler, H. Elders-Boll, “Can an Added Value 

be offered to SDR Operators in Scenarios where Interoperability to Legacy Radios 

is a Requirement” submitted to Technical Conference of the Wireless Innovation 

Forum SDR’12, Washington, Jan. 2013.

Data Fusion Schemes for Cooperative Spectrum 

Sensing in Cognitive Radio Networks 

Djamel Teguig 1, 2 , Bart Scheers 1 , Vincent Le Nir 1 

1 Royal Military Academy – Department CISS, 

2 

Polytechnic Military School-Algiers-Algeria, 

Renaissance Avenue 30-B1000 Brussels, Belgium, 

{djamel.teguig, bart.scheers}@rma.ac.be, vincent.lenir@elec.rma.ac.be 

Abstract: Cooperative spectrum sensing has proven its efficiency to detect spectrum holes in cognitive 

radio network (CRN) by combining sensing information of multiple cognitive radio users. In this 

paper, we study different fusion schemes that can be implemented in fusion center. Simulation comparison 

between these schemes based on hard, quantized and soft fusion rules are conducted. It is shown 

through computer simulation that the soft combination scheme outperforms the hard one at the cost 

of more complexity; the quantized combination scheme provides a good tradeoff between detection 

performance and complexity. In the paper, we also analyze a quantized combination scheme based on 

a tree-bit quantization and compare its performance with some hard and soft combination schemes. 

Keywords: Cooperative spectrum sensing, cognitive radio (CR), data fusion, soft, quantized and 

hard fusion rules 


In recent years, the demand of spectrum is rapidly increasing with the growth 

of wireless services. The scarcity of the spectrum resource becomes more serious. 

Cognitive radio provides a new way to better use the spectrum resource [1]. Therefore, 

a reliable spectrum sensing technique is needed. Energy detection exhibits 

simplicity and serves as a practical spectrum sensing scheme. As a key technique 

to improve the spectrum sensing for Cognitive Radio Network (CRN), cooperative 

sensing is proposed to combat some sensing problems such as fading, shadowing, 

and receiver uncertainty problems [2]. 

The main idea of cooperation is to improve the detection performance by 

taking advantage of the spatial diversity, in order to better protect a primary user, 

and reduce false alarm to utilize the idle spectrum more efficiently. 

The three steps in the cooperative sensing process are [3]: 

1. The fusion center FC selects a channel or a frequency band of interest for 

sensing and requests all cooperating CR users to individually perform local 

sensing.


2. All cooperating CR users report their sensing results via the control 

channel. 

3. Then the FC fuses the received local sensing information to decide about 

the presence or absence of signal and reports back to the CR users. 

As shown in Fig. 1, CR3 suffers from the receiver uncertainty problem because 

it is located outside the transmission range of primary transmitter and it is unaware 

of the existence of primary receivers. So, transmission from CR3 can interfere with 

the reception at a primary receiver. CR2 suffers from multipath and shadowing 

caused by building and trees. Cooperative spectrum sensing can help to solve these 

problems if secondary users cooperate by sharing their information. 

Figure 1. Sensing problems (receiver uncertainty, multipath and shadowing) 

To implement these three steps, seven elements of cooperative sensing are 

presented [4] as illustrated in Fig. 2. 

Figure 2. Elements of cooperative spectrum sensing [4]


189 

• Cooperation models: is concerned with how CR users cooperate to perform 

sensing. 

• Sensing techniques: this element is crucial in cooperative spectrum sensing 

to sense primary signals by using signal processing techniques. 

• Hypothesis testing: in order to decide on the presence or absence of a primary 

user (PU), a statistical test is performed to get a decision on the presence of PU. 

• Control channel and reporting: is used by CR users to report sensing result 

to the FC. 

• Data fusion: is a process of combining local sensing data to make cooperation 

decision. 

• User selection: in order to maximize the cooperative gain, this element provides 

us the way to optimally select the cooperating CR users. 

• Knowledge base: means a prior knowledge included PU and CR user location, 

PU activity, and models or other information in the aim to facilitate 

PU detection. 

In this paper, we will focus on the data fusion rules. The decision on the presence 

of PU is achieved by combining all individual sensing information of local CR 

users at a central (FC) using various fusion schemes. These schemes can be classified 

as hard decision fusion, soft decision fusion, or quantized (softened hard) decision. 

The hard decision is the one in which the CR users make a one-bit decision 

regarding the existence of the PU, this 1-bit decision will be forwarded to the FC 

for fusion. In [5], a logic OR fusion rule for hard-decision combining is presented 

for cooperative spectrum sensing. In [6], two simple schemes of hard decision combining 

are studied: the OR rule and the AND rule. In [7]-[8], another sub-optimal 

hard decision scheme is used called Counting Rule. In [9] that half-voting rule 

is shown as the optimal hard decision fusion rule in cooperative sensing based on 

energy detection. In the case of soft decision, CR users forward the entire sensing 

result to the center fusion without performing any local decision. In [10] a soft 

decision scheme is described by taking linear combination of the measurements 

of the various cognitive users to decide between the two hypotheses. However, in [11] 

collaborative detection of TV transmissions is studied while using soft decision 

using the likelihood ratio test. It is shown that soft decision combining for spectrum 

sensing achieves more precise detection than hard decision combining. This 

was confirmed in [12] when performing Soft decision combination for cooperative 

sensing based on energy detection. Some soft combining techniques are discussed 

in [13-14-15] as square-law combining (SLC), equal gain combining (EGC) and 

square-law selection (SLS) over AWGN, Rayleigh and Nakagami-m channel. 

The paper is organized as follows. We present in Section II the system model 

related to cooperative spectrum sensing. In Section III, we describe different fusion 

rules for cooperative spectrum sensing; several hard, soft and quantized schemes 

are proposed and discussed. Simulation results in section IV are given to compare 

these fusion rules. We conclude this paper in Section V.


II. System model 

Consider a cognitive radio network, with K cognitive users (indexed by 

k = {1, 2 ... K}) to sense the spectrum in order to detect the existence of the PU. 

Suppose that each CR performs local spectrum sensing independently by using N 

samples of the received signal. The spectrum sensing problem can be formulated 

as a binary hypothesis testing problem with two possible hypothesis H 0 and H 1 . 

H : xk( n) wk( n) 

(1) 

H : x( n) hsn ( ) w ( n) 

k k k 

Where s(n) are samples of the transmitted signal (PU signal), wk 

( n ) is the receiver 

noise for the k th CR user, which is assumed to be an i.i.d. random process with zero 

mean and variance σ 2 n and h k is the complex gain of the channel between the PU and 

the k th CR user. H 0 and H 1 represent whether the signal is absent or present respectively. 

Using energy detector, the k th CR user will calculate the received energy as [16]: 

N 

E x ( n) 

(2) 

k 

In the case of soft decision, each CR user forwards the entire energy result 

E k to the FC. However, for hard decision, the CR users make the one-bit decision 

given by Δ k, by comparing the received energy E k with a predefined threshold λ k . 

1, 

Ek 

k 

k 

 

(3) 

0, otherwise 

Detection probability P d,k and false alarm probability P f,k of the CR user k 

are defined as: 

P d,k = Pr {Δ k = 1|H 1 } = Pr {E k > λ k |H 1 } 

P f,k = Pr {Δ k = 1|H 0 } = Pr {E k > λ k |H 0 } 

Assuming that λ k = λ for all CR users, the detection probability, false alarm 

probability and miss detection P m,k over AWGN channels can be expressed as follows 

respectively [17] 

Pdk 

, 

Qm( 2 , ) 

(4) 

( m, / 2) 

Pf , k ( m) 

(5) 

P 

k 

1 P 

(6) 

mk , dk , 

where γ is the signal to noise ratio (SNR), m = TW is the time bandwidth product, 

Q N (.,.) is the generalized Marcum Q-function, Г(.) and Г(.,.) are complete and 

incomplete gamma function respectively.


191 

III. Fusion rules 

This section describes the fusion rules that are used for the comparison. 

III.1. Hard decision fusion 

In this scheme, each user decides on the presence or absence of the primary 

user and sends a one bit decision to the data fusion center. The main advantage 

of this method is the easiness the fact that it needs limited bandwidth [12]. When 

binary decisions are reported to the common node, three rules of decision can be 

used, the “and”, “or”, and majority rule. Assume that the individual statistics Δ k 

are quantized to one bit with Δ k = 0, 1; is the hard decision from the k th CR user. 

1 means that the signal is present, and 0 means that the signal is absent. 

The AND rule decides that a signal is present if all users have detected a signal. 

The cooperative test using the AND rule can be formulated as follows: 

H 

H 

K 

 

otherwise 

(7) 

The OR rule decides that a signal is present if any of the users detect a signal. 

Hence, the cooperative test using the OR rule can be formulated as follows: 

K 

H : 1 

H 

k 

k 

: otherwise 

The third rule is the voting rule that decides on the signal presence if at 

least M of the K users have detected a signal with 1≤ M ≤ K. The test is formulated 

as: 

(8) 

H 

H 

M 

 

otherwise 

(9) 

A majority decision is a special case of the voting rule for M = K/2, the same 

as the AND and the OR rule which are also special cases of the voting rule for 

M = K and M = 1 respectively.


Cooperative detection probability Q d and cooperative false alarm probability 

Q f are defined as: 

K 

 

 

Qd 

Pr1 H1Prk 

MH 

1 

 

i1 

 

(10) 

K 

 

Qf 

Pr 

1 H0 

Pr 

k MH0 

 

 

i1 

 

Where Δ is the final decision. Note that the OR rule corresponds to the case M = 1, 

hence 

K 

Qd , or 

1 (1 Pd , k 

) 

(11) 

Q 

k1 

1 (1 ) 

(12) 

K 

f , or 

P 

k1 

f , k 

The AND rule can be evaluated by setting M = K. 

K 

Q P 

(13) 

Q 

d , and 

k1 

d , k 

(14) 

K 

f , and 

P 

k1 

f , k 

III.2. Soft data fusion 

In soft data fusion, CR users forward the entire sensing result E k to the center 

fusion without performing any local decision and the decision is made by combining 

these results at the fusion center by using appropriate combining rules such 

as square law combining (SLC), maximal ratio combining (MRC) and selection 

combining (SC). Soft combination provides better performance than hard combination, 

but it requires a larger bandwidth for the control channel [18]. It also 

generates more overhead than the hard combination scheme [12]. 

Square Law Combining (SLC): SLC is one of the simplest linear soft combining 

schemes. In this method the estimated energy in each node is sent to the center 

fusion where they will be added together. Then this summation is compared to 

a threshold to decide on the existence or absence of the PU and a decision statistic 

is given by [19]: 

E E 

(15) 

slc 

where E k denotes the statistic from the k th CR user. The detection probability and 

false alarm probability are formulated as follow [19]: 

k 

k


193 

Qd , SLC 

QmK ( 2 slc 

, ) 

(16) 

Q 

f , SLC 

( mK, / 2) 

(17) 

( mK) 

Where slc k 

and γ k is the received SNR at k th CR user. 

k 

Maximum Ratio Combining (MRC): the difference between this method 

and the SLC is that in this method the energy received in the center fusion from 

each user is ponderated with a normalized weight and then added. The weight 

depends on the received SNR of the different CR user. The statistical test for this 

scheme is given by: 

E 

wE 

(18) 

mrc k k 

k 

Over AWGN channels, the probabilities of false alarm and detection under 

the MRC diversity scheme can be given by [21]: 

Qd , MRC 

Qm( 2 mrc 

, ) 

(19) 

 

Where: mrc k 

k 

Q 

f , MRC 

( m, / 2) 

(20) 

( m) 

Selection Combining (SC): in the SC scheme, the fusion center selects 

the branch with highest SNR [20]. 

max( , ,....., ) 

sc 

Over AWGN channels, the probabilities of false alarm and detection under 

the SC diversity scheme can be written as [21]: 

Qd , SC 

Qm( 2 sc 

, ) 

(21) 

Q 

f , SC 

1 2 

K 

( m, / 2) 

(22) 

( m) 

III.3. Quantized data fusion 

In this scheme, we try to realize a tradeoff between the overhead and the detection 

performance. Instead of one bit hard combining, where there is only one


threshold dividing the whole range of the detected energy into two regions, a better 

detection performance can be obtained if we increase the number of threshold to 

get more regions of observed energy. 

In [12], a two-bit hard combining scheme is proposed in order to divide 

the whole range of the detected energy into four regions. The presence of the signal 

of interest is decided at the FC by using the following equation: 

wn 

i i 

L 

(23) 

i 

where L is the threshold and it is equal to the weight of the upper region, 

n i is the number of observed energies falling in region i and w i is the weight value 

of region i with w 0 = 0 w 1 = 1, w 2 = 2 and w 3 = 4. 

In this paper, we extend the scheme of [12] to a three-bit combining scheme. 

In the three-bit scheme, seven threshold λ 1 , λ 2 … and λ 7 , divide the whole range 

of the statistic into 8 regions, as depicted in Fig. 3. Each CR user forwards 3 bit of information 

to point out the region of the observed energy. Nodes that observe higher 

energies in upper regions will forward a higher value than nodes observing lower 

energies in lower regions. 

The three-bits combining scheme is performed in four steps: 

1: Define a quantization threshold λ i (i = 1 ... 7) for each region according to 

the maximal received energy of the signal. 

Figure 3. Principle of three-bit hard combination scheme 

2: Each user makes a local decision by comparing the received energy with 

the thresholds predefined in 1, and sends 3-bits information to the FC. 

3: The FC sums the local decisions ponderated with weights w i (i = 0.., 7). 

In our case, we have taken: w 0 = 0, w 1 = 1, w 2 = 2, w 3 = 3, w 4 = 4, w 5 = 5, 

w 6 = 6, w 7 = 7. 

4: The final decision is made by comparing this sum with a threshold L. 

wik 

L 

(24) 

k


195 

IV. Simulations and results 

In this section we study the detection performance of our scheme through 

simulations, and compare its performances with soft and hard fusion schemes. 

First, we present the performance of the hard combining schemes as depicted 

in Fig. 4. Secondly, we will compare the performance of the different fusion rules 

in case of soft combining. Next, the two-bit and the three-bit quantized schemes 

are compared in term of detection performance. 

For the hard decision, we present in Fig. 4 the ROC curves of the ‘AND’ and 

the ‘OR’ rule, and compare it to the detection performance of a single CR user. For 

the simulations, we consider 3 CR users. Each user has a SNR of –2 db. As shown 

in Fig. 4, the OR rule has better detection performance than the AND rule, which 

provides slightly better performance at low Pfa than the OR, because the data fusion 

center decide in favor of H1 when at least one CR user detects the PU signal. 

However in the AND rule, to decide of the presence of a primary user, all CR users 

must detect the PU signal. 

Figure 4. ROC for the hard fusion rules under AWGN channel, SNR = –2 dB, K = 3 users, 

and energy detection with N = 1000 

Fig. 5 shows the ROC curves of different soft combination schemes discussed 

in section III.2 under AWGN channel. For the simulations, each CR user sees a different 

SNR. We observe from this figure that the MRC scheme exhibits the best 

detection performance, but it requires channel state information. The SLC scheme 

does not require any channel state information and still present better performance 

than SC. When no channel information is available, the best scheme is SLC.


Figure 5. ROC for soft fusion rules under AWGN channel with K = 3 users, 

and energy detection with m = 5 

Fig. 6 shows the ROC curves for quantized data fusion with 2-bit and 

3-bit quantized combination, the figure indicates that the proposed 3-bit combination 

scheme shows much better performance than the 2-bit combination scheme 

at the cost of one more bit of overhead for each CR user, this scheme can achieve 

a good tradeoff between detection performance and overhead. 

Figure 6. ROC curves for quantized data fusion under AWGN channel with SNR = –2 db, 

K = 4 users and N = 1000 samples


197 

For comparison, we show in Fig. 7 the ROC curves for the different fusion 

rules under AWGN channel. As the figure indicates, all fusions method outperform 

the single node sensing, the soft combining scheme based on SLC rule 

outperforms the hard and quantized combination at the cost of control channel 

overhead, the 3-bit quantized combination scheme shows a comparable detection 

performance with the SLC, with less overhead. 

Figure 7. ROC for combining fusion rules under AWGN channel with K = 3 users, SNR = –2 db 

and energy detection with N = 1000 samples 

V. Conclusion 

In this paper, the effect of fusion rules for cooperative spectrum sensing is investigated. 

It is shown that the soft fusion rules outperform the hard fusion rules. 

However, these benefits are obtained at the cost of a larger bandwidth for the control 

channel. The hard fusion rules occur with less complexity, but also with a lower 

detection performance than soft combination schemes. The proposed quantized 

three-bit combination scheme wins advantage of the soft and the hard decisions 

schemes with a tradeoff between overhead and detection performance. In practical 

application, we can select an appropriate method of data fusion and decision algorithms 

according to the requirement of detection performance and the requirement 

of the available bandwidth for the reporting channel.


References 

[1] S. Haykin, “Cognitive radio: Brain-empowered wireless communications”, IEEE 

Journal Selected Areas in Communications, vol. 23, no. 2, Feb.2005, pp. 201-220. 

[2] T. Yucek & H. Arslan, A. Survey of Spectrum Sensing Algorithms for Cognitive 

Radio Application, IEEE Communications Surveys & Tutorials, 11(1), First Quarter 

2009. 

[3] D. Cabric, S. Mishra, R. Brodersen, Implementation issues in spectrum sensing 

for cognitive radios, in: Proc. of Asilomar Conf. on Signals, Systems, and Computers, 

vol. 1, 2004, pp. 772-776. 

[4] I.F. Akyildiz, B.F. Lo, and R. Balakrishnan, “Cooperative Spectrum Sensing 

in Cognitive Radio Networks: A Survey,” Physical Communication (Elsevier) Journal, 

vol. 4, no. 1, pp. 40-62, March 2011. 

[5] A. Ghasemi, E. Sousa, Collaborative Spectrum Sensing for Opportunistic Access 

in Fading Environments. DySPANn 2005, pp. 131-136, Nov. 2005. 

[6] E. Peh, Y.-Ch. Liang, Optimization for Cooperative Sensing in Cognitive Radio 

Networks, WCNC 11-15, pp. 27-32, March 2007. 

[7] Jayakrishnan Unnikrishnan and Venugopal V. Veeravalli, Cooperative Spectrum 

Sensing and Detection for Cognitive Radio, IEEE GLOBCOM 26-30 Nov. 2007, 

pp. 2972-2976. 

[8] T. Jiang, D. Qu, “On minimum sensing error with spectrum sensing using counting 

rule in cognitive radio networks,” in Proc. 4th Annual Int. Conf.Wireless Internet 

(WICON’08), Brussels, Belgium, 2008, pp. 1-9. 

[9] W. Zhang, R. Mallik, and K. Letaief, “Cooperative spectrum sensing optimization 

in cognitive radio networks,” in Proc. IEEE Int. Conf.Commun., 2008, pp. 3411-3415. 

[10] Zhi Quan, Shuguang Cui, and Ali H. Sayed, Optimal Linear Cooperation for Spectrum 

Sensing in Cognitive Radio Networks, IEEE Journal of Selected Topics in Signal. 

[11] E. Visotsky, S. Kuffner, and R. Peterson, “On collaborative detection of TV 

transmissions in support of dynamic spectrum sharing,” in Proc. 1st IEEE Int. Symp. 

New Frontiers in Dynamic Spectrum Access Netw. (DySPAN), pp. 338-345, 2005. 

[12] J. Ma and Y. Li, “Soft combination and detection for cooperative spectrum 

sensing in cognitive radio networks,” in Proc. IEEE Global Telecomm. Conf., 2007, 

pp. 3139-3143. 

[13] S.P. Herath, N. Rajatheva, C. Tellambura, “On the Energy Detection of Unknown 

Deterministic Signal over Nakagami Channel with Selection Combining,” Proc. IEEE 

Symp. CCECC ’09 Canadian conference on. pp. 745-749, 2009. 

[14] S.P. Herath, N. Rajatheva, “Analysis of equal gain combining in energy detection 

for cognitive radio over Nakagami channels,” IEEE. GLOBECOM 2008, pp. 1-5, 2008. 

[15] Yunxue Liu, Dongfeng Yuan, et al., “Analysis of square-law combining for 

cognitive radios over Nakagami channels,” WiCom’09 5th International Conference 

on, pp. 1-4, 2009. 

[16] V.I. Kostylev, Energy detection of a signal with random amplitude, In Proc. IEEE 

ICC, pp. 1606-1610, New York, Apr. 2002. 

[17] Fadel F. Digham, Mohamed-Slim Alouni, Marvin K. Simon, “On the energy 

detection of unknown signals over fading channels,” IEEE. Trans. Commun., vol. 55, 

pp. 21-24, Jan. 2007.


199 

[18] Zhi Quan, Shuguang Cui, H. Vincent Poor, and Ali H. Sayed, “Collaborative 

wideband sensing for cognitive radios,” IEEE Signal Processing Magazine, vol. 25, 

no. 6, pp. 63-70, 2008. 

[19] Zhengquan Li, Peng Shi, Wanpei Chen, Yan Yan, “Square Law Combining Double 

threshold Energy Detection in Nakagami Channel”, Internation Journal of Digital 

Content Technology and its Application, vol. 5, Number 12, December 2011. 

[20] M.K. Simon and M.-S. Alouini, Digital communication over fading channels. John 

Wiley & Sons, Inc., 2 ed., Dec. 2004. 

[21] Hongjian Sun, Collaborative Spectrum Sensing in Cognitive Radio Networks. 

A doctoral thesis of Philosophy. The University of Edinburgh. January 2011.

Implementation of an Adaptive OFDMA PHY/MAC 

on USRP Platforms for a Cognitive Tactical 

Radio Network 

Vincent Le Nir, Bart Scheers 

Department Communication, Information, Systems and Sensors (CISS), 

Royal Military Academy (RMA), 30, Avenue de la Renaissance, B-1000 Brussels, Belgium, 

{vincent.lenir, bart.scheers}@rma.ac.be 

Abstract: Cognitive radio is envisioned to solve the problem of spectrum scarcity in military networks 

and to autonomously adapt to rapidly changing radio environment conditions and user needs. 

Dynamic spectrum management techniques are needed for the coexistence of multiple cognitive 

tactical radio networks. Previous work has investigated the iterative water-filling algorithm (IWFA) 

as a possible candidate to improve the coexistence of such networks. It has been shown that adding 

a constraint on the number of transmitter’s sub-channels improves the convergence of IWFA. In this 

paper, we propose an adaptive orthogonal frequency division multiple access (OFDMA) physical 

(PHY) and medium access control (MAC) for the coexistence of multiple cognitive tactical radio networks. 

The proposed scheme is implemented on universal software radio peripheral (USRP) platforms 

using Qt4/IT++ and the USRP hardware driver (UHD) application programming interface (API). 

Keywords: Cognitive radio; adaptive OFDMA; distributed bit and power allocation; USRP 


Cognitive radio (CR) has been introduced by Mitola as an extension of software 

radio. In this technology, radio nodes are intelligent agents that search out 

ways to deliver services according to the user needs and the radio environment [1]. 

CR has been an active topic of research since most regulatory bodies found that 

the spectrum is underutilized although most available spectrum is licensed, leaving 

small room for future wireless applications [2]. 

CR can also solve the problem of spectrum scarcity for military networks. 

As the electro-magnetic environment in an operational theater can be very hostile, 

cognitive tactical radio networks can autonomously adapt to rapidly changing 

conditions and user needs [3]. 

The coexistence of cognitive tactical radio networks requires dynamic spectrum 

management techniques in order to reduce the interference and to improve the performance 

in terms of throughput, power (longer battery life), and delay. Dynamic spectrum


management techniques can be centralized/distributed (decisions are made centrally/ 

locally), cooperative/non-cooperative (some information is shared/not shared between 

networks), and can use a horizontal/vertical sharing model (all networks have equal/ 

limited rights to access the spectrum) [4]. Our target application for the coexistence 

of cognitive tactical radio networks corresponds to a mission involving multiple coalition 

nations in a foreign country with no a priori frequency planning, meaning that 

there is no frequency management cell (FMC) to coordinate the different networks. 

Moreover, the networks can’t exchange information between each other’s and they 

have equal priority rights. Therefore, we are interested in a distributed non-cooperative 

technique in a horizontal spectrum sharing model. 

The iterative water-filling algorithm (IWFA) is an adequate dynamic spectrum 

management technique to meet these requirements [5]. Indeed, IWFA is an autonomous 

algorithm solving the distributed power control problem in a frequency selective 

interference channel. Robust versions of the IWFA have been designed to cope with 

dynamic wireless channels [6, 7, 8, 9, 10]. However, IWFA does not converge to a unique 

solution (multiple Nash equilibriums). This aspect is inherent to IWFA because at each 

iteration some power is poured in the best sub-channel regardless the interference 

caused to the other networks, while they have a better benefit avoiding each other by 

taking different sub-channels. The convergence of the IWFA can be improved by adding 

a constraint on the number of transmitter’s sub-channels [11, 12]. 

Random access multi-channel medium access control (MAC) protocols have 

been proposed for CR networks [13]. Multi-channel MAC protocols have also been 

designed for IWFA [6, 14]. These protocols use a dedicated control channel to coordinate 

the radios. It is unlikely that a dedicated control channel can be used in a military 

context since it is a single point of failure. A possible workaround is to use a rendezvous 

multi-channel MAC protocol; however the radios may beacon for a long time before 

establishing a rendezvous. Another alternative is to employ time division multiplexing 

access (TDMA) to obtain a collision-free transmission schedule or frequency division 

multiplexing access (FDMA) as described in [15]. In this paper, we propose an adaptive 

orthogonal frequency division multiple access (OFDMA) PHYsical/MAC to 

allow simultaneous collision-free transmissions in a cognitive tactical radio network. 

The adaptive OFDMA PHY/MAC has the following characteristics: 

• It is based on the IWFA with selection of a single sub-channel [11, 12]. 

It consists of grouping several OFDM sub-carriers to form a sub-channel. 

In a first mode, it uses a fixed bit-loading per sub-carrier. In a second 

mode, it uses an adaptive bit-loading related to the spectrum sensing and 

the channel estimation on each OFDM sub-carrier. 

• It is robust against multi-path due to the insertion of a cyclic prefix and 

allows a single-tap equalizer due to the orthogonality of the sub-carriers. 

• It uses a blind demodulation chain, meaning that it uses the cyclic prefix to 

estimate blindly the timing offset and to detect the presence of an OFDM 

signal. The timing offset estimate is used to synchronize control packets


203 

and data packets, and to estimate blindly the frequency offset, transmission 

channel, phase offset and transmitted bits. The residual ambiguity 

given by the BPSK or the adaptive QAM modulation schemes is solved by 

transmitting a single sub-carrier pilot or by using differential encoding. 

The adaptive OFDMA PHY/MAC has been implemented on universal software 

radio peripheral (USRP) platforms using Qt4/IT++ and the USRP hardware 

driver (UHD) application programming interface (API). Qt is a cross-platform 

application framework that is widely used for developing application software with 

a graphical user interface (GUI) [16]. IT++ is a C++ library of mathematical, signal 

processing and communication classes and functions. Its main use is in simulation 

of communication systems and for performing research in the area of communications 

[17]. The goal of the UHD is to provide a host driver and API for current and 

future USRP products. The UHD driver can be used standalone or with 3rd party 

applications such as Gnuradio, Labview, or Simulink [18]. 

This paper is organized as follows. First, the adaptive OFDMA MAC protocol 

is described in Section II. Second, the adaptive OFDMA PHysical layer functions 

are described in Section III. The implementation of the adaptive OFDMA PHY/ 

MAC on USRP platforms using Qt4/IT++ and the UHD API is described in Section 

IV. Finally, Section V concludes the paper. 

II. Adaptive OFDMA MAC protocol 

In this Section, the adaptive OFDMA MAC protocol is described. It is assumed 

that the CRs have agreed on the same front-end parameters to transmit and 

receive, i.e. carrier frequencies, sampling rates and bandwidths. It is also assumed 

that the CRs have the same OFDM parameters, i.e. number of sub-carriers, cyclic 

prefix (CP) size, and number of sub-channels. The adaptive OFDMA MAC protocol 

uses control packets for the handshaking between two CRs. These control 

packets are request-to-send (RTS) and clear-to-send (CTS) packets. In a first mode, 

the two control packets include source address, destination address and best subchannel 

sensed by the CR. In a second mode, control packets also include target 

rate constraint, bit and power allocation. Therefore, unlike carrier sense multiple 

access (CSMA) scheme and other random access multi-channel MAC protocols, 

the two control packets convey some information and need to be aligned with 

the timing of OFDM symbols transmitted in the same bandwidth of interest to 

keep the orthogonality between sub-carriers. 

A. First mode 

The first mode uses a fixed bit-loading, e.g. a fixed BPSK or QAM modulation 

over the different sub-carriers. In the following, we suppose a transmission from CR 1 

to CR 2. CR 1 and CR 2 perform spectrum sensing in the bandwidth of interest and


determine their best sub-channel based on an energy estimate. The RTS includes 

the source address CR 1, the destination address CR 2, and the best sub-channel 

of CR 1 given by spectrum sensing. CR 1 transmits the RTS on its best sub-channel. 

CR 2 demodulates all the sub-channels in parallel (see Section III.C) and discovers 

that a RTS is sent based on the destination address of CR 1. CR 2 gets the RTS source 

address and the best sub-channel of CR 1. The CTS includes the CR 2 source address, 

the CR 1 destination address, and the best sub-channel of CR 2 by spectrum sensing. 

CR 2 transmits the CTS on CR 1 best sub-channel. CR 1 knows to receive on its best 

sub-channel. CR 1 demodulates the CTS and gets CR 2 best sub-channel. Finally, 

CR 1 transmits the data on CR 2 best sub-channel. 

B. Second mode 

The second mode uses an adaptive QAM over the different sub-carriers. CR 1 and 

CR 2 determine their best sub-channel based on an energy estimate. The RTS includes 

the CR 1 source address, the CR 2 destination address, the CR 1 best sub-channel, 

and the target rate constraint. CR 1 transmits the RTS on its best sub-channel. CR 2 

demodulates all the sub-channels in parallel and discovers that a RTS is sent based 

on the destination address of CR 1. CR 2 gets the RTS source address, the target rate 

constraint, and the best sub-channel of CR 1. If the CR 1 best sub-channel is the same 

as the CR 2 best sub-channel, the CTS includes the CR 2 source address, the CR 1 

destination address, the CR 2 best sub-channel, and the bit and power allocation for 

CR 1 target rate constraint. CR 2 transmits the CTS on CR 1 best sub-channels. CR 1 

knows to receive on its best sub-channel. CR 1 demodulates the CTS and gets CR 2 

best sub-channel, as well as the bit and power allocation for its target rate constraint. 

Finally, CR 1 transmits the data on their common best sub-channel with adaptive QAM. 

If the CR 1 best sub-channel is different from the CR 2 best sub-channels, CR 1 gets 

the best sub-channel of CR 2 by the CTS, and send a second RTS on CR 2 best subchannel 

to allow CR 2 to compute the bit and power allocation. CR 2 sends a second 

CTS with this information on CR 1 best sub-channel. Finally, CR 1 transmits the data 

on CR 2 best sub-channel with adaptive QAM. 

III. Adaptive OFDMA physical layer 

In this Section, the key functions of the adaptive OFDMA PHY are described, 

i.e. the spectrum sensing, the OFDM signal detection, the blind OFDM demodulation 

and the distributed bit and power allocation. 

A. Spectrum sensing 

A CR needs to determine its best sub-channel and communicate this information 

to another CR. The bandwidth of the signal of interest B is divided into


205 

a number of sub-channels S of width B/S. As there is no assumption about the type 

of the observed signals, a nonparametric method should be used. We propose 

to use the classical Barlett’s method for power spectrum estimation known also 

as the method of averaged periodograms [19]. We assume that the complex sampling 

rate equals the bandwidth of the signal of interest. The averaged periodograms for N 

sub-carriers of a complex baseband signal with K blocks of N samples y = [y(kN),… 

,y((k+1)N-1)] with k = [0,…,K-1] is given by 

K1 N1 j2in 

1 

 

2 

 

N 

E i | y( kN n) e | 

(1) 

KN 

k0 n0 

with i = [0,…,N-1] bins. The best sub-channel selection consists of integrating 

the estimated power spectrum of the frequency bins corresponding to the width B/S 

for all sub-channels and determines the sub-channel which have the lowest energy 

( m1) 

N 1 

S 

opt 

S = min Ei ( ) 

(2) 

m 

mN 

i 

S 

with m = [0,…,S-1]. This sensing procedure is used first by CR 1 for the selection 

of its best sub-channel for RTS transmission, and is further embedded in the RTS 

control packet. Secondly, this sensing procedure is used by CR 2 to communicate 

its best sub-channel to CR 1 in the CTS control packet via CR 1 best sub-channel. 

B. OFDM signal detection 

The CR also needs to detect within the bandwidth of interest B the presence 

of an OFDM signal even if the signal is sent over a single sub-channel. A survey 

of OFDM signal detection techniques can be found in [20, 21, 22, 23]. Some 

techniques require the knowledge of a pilot sequence, OFDM parameters, and 

noise variance to determine the detection threshold. Detection techniques requiring 

the knowledge of a pilot sequence have better detection performance, but 

worse bandwidth/power/complexity efficiency. Detection techniques requiring 

the knowledge of the noise variance assume a white noise assumption and degrade 

severely in the presence of colored noise. Moreover, when the theoretical values 

of the threshold are not known, they have to be empirically computed by feeding 

the detector with pure noise signals and calculating the test statistic. 

In the proposed adaptive OFDMA MAC protocol, control and data packets 

are sent only over a subset of the available sub-carriers. Moreover, other unknown 

signals or colored noise can be present in the bandwidth of interest. We propose 

to use a modified version of the cyclic prefix based sliding window correlation detector 

to determine the presence of an OFDM signal in the bandwidth of interest. 

This detector does not require the knowledge of a pilot sequence but only requires


the knowledge of the OFDM parameters, i.e. number of subcarriers N, cyclic prefix 

size P. The detection threshold is based on the non-correlated part of the cyclic 

prefix sliding window correlation estimate instead of the noise variance. This allows 

detecting the presence of an OFDM signal even in the presence of an unknown 

signal in the bandwidth of interest. Moreover, this detector also gives an estimate 

of the OFDM timing offset when it is detected. Assuming a complex baseband 

signal with K blocks of N+P samples samples y = [y(k(N+P)),…,y((k+1)(N+P)-1)] 

with k = [0,…,K-1], the cyclic prefix sliding window correlation estimate is given by 

( ) 

K2 P1 

 

* 

| y( k( NP) j) y ( k( NP) jN)| 

k0 

j 

( K1) 

P 

2 

y 

(3) 

with σ 2 y the variance of the received complex baseband signal. The absolute value 

of the correlation estimate is able to cope with frequency and phase offsets introduced 

by Doppler shifts and clock mismatches. The estimate of the timing offset 

is given by 

opt 

max ( ) 

(4) 

{0,..., NP1} 

In order to determine the presence of an OFDM signal, the estimate of the timing 

offset is compared with the non-correlated part of the cyclic prefix sliding 

window correlation estimate. Assuming a channel delay spread spanning the entire 

cyclic prefix, the non-correlated part of the cyclic prefix sliding window correlation 

estimate is given by ρ(mod(θ opt + 2 P, N + P)). Assuming that the non-correlated 

part is a Gaussian distribution with mean m nc and variance σ 2 nc , the detection 

threshold η is given by 

m nc 

nc 

(5) 

with α an integer corresponding to the number of standard deviations necessary 

to discriminate between an OFDM signal and a non-OFDM signal using only 

the cyclic prefix sliding window correlation estimate ρ(θ). The detector scheme is 

opt 

( ) Presence of an OFDM signal 

(6) 

opt 

( ) Absence of an OFDM signal 

Using K = 50 blocks in the application, α has been set to 40, allowing very few 

false alarms and mis-detections. 

C. Blind OFDM demodulation 

The cyclic prefix sliding window correlation estimate can be used to estimate 

the frequency offset [24]


207 

opt 1 

opt 

( ) 

2N 

After time and frequency offset corrections, the OFDM symbols are transformed 

to the frequency domain by the discrete Fourier transform (DFT) operation. 

As there is no interference between two consecutive OFDM symbols, we obtain independent 

sub-carriers with the following channel model 

Y ( kN i) H ( kN i) X ( kN i) N ( kN i) 

(8) 

with k = [0,…,K-1] and i = [0,…,N-1], in which Y(kN+i), H(kN+i), X(kN+i), and 

N(kN+i) are respectively the demodulated data, the channel frequency response, 

the transmitted symbol and the noise for the block k and sub-carrier i. Assuming 

the channel invariant over the K blocks, a blind estimate of the channel amplitude 

is given by 

K1 

(7) 

opt 2 1 

2 

| H i 

| | Y ( kN i)| 

(9) 

K 

k0 

Assuming the channel invariant over the K blocks, a blind estimate of the phase 

offset can be obtained for M-PSK signals [25] by the following expression 

K1 

est 1 

M 

() i Y ( kN i) 

(10) 

K 

k0 

Modulation stripping has also been investigated for M-QAM signals in [26]. 

Knowing the phase offset estimate for each sub-carrier, it is possible to correct 

linear shift of the phase in the frequency domain due to an incorrect timing offset 

estimate belonging to the ISI free region, as well as abrupt changes of the phase 

in the frequency domain due to the phase ambiguity introduced by the blind phase 

offset algorithm (phase unwrapping). The phase correction for M-PSK signals 

is performed using the Algorithm 1.


D. Distributed adaptive bit and power allocation 

The distributed bit and power allocation (adaptive QAM) can only be determined 

when a CR sends its control packet on the other CR best sub-channel. If CR 1 

best sub-channel is the same as CR 2 best 

sub-channel, then only one handshake 

is necessary (RTS-CTS) to inform CR 1 

about CR 2 best sub-channel, bit and 

power allocation. However, if CR 1 best 

sub-channel is different from CR 2 best 

sub-channel, two handshakes are necessary 

(RTS-CTS-RTS-CTS) because CR 2 

informs CR 1 about its best sub-channel 

in the first CTS and the bit and power 

allocation in the second CTS. The distributed 

bit and power allocation is based on 

the waterfilling algorithm, in which an inner 

loop maximizes the bit allocation for 

a transmit power constraint, and an outer 

loop minimizes the transmit power for 

a target rate constraint. The algorithm 

for distributed bit and power allocation 

is described in Algorithm 2, in which 

λ is the Lagrangian parameter [27], 

Г is the SNR gap which measures the loss with respect to theoretically optimum 

performance [29], p = [p(0),…,p(N-1)] and p opt = [p(0) opt ,…, p(N-1) opt ] are the power 

allocation vectors, b opt = [b(0) opt ,…,b(N-1) opt ] is the bit allocation vector, P tot is the total 

power constraint, Δf is the sub-carrier bandwidth, R is the data rate and R target is the 

target rate constraint. As shown in [11, 12], when the number of sub-channels is lower 

than the number of concurrent links, CR users have to share the same sub-channels 

and iterative updates of bit and power allocation can lead to convergence problems 

due to multiple Nash equilibriums. To avoid this problem, the number of sub-channels 

is larger than the number of concurrent links to ensure the convergence to a single 

Nash equilibrium. This leads to a distributed adaptive OFDMA solution. 

IV. Implementation using QT4/IT++ and the UHD API 

In this Section, the implementation of the adaptive OFDMA PHY/MAC on 

USRP platforms using Qt4/IT++ and the UHD API is described. Several classes 

and procedures have been implemented for this application using QThread to enable 

multi-threading, Qwt to enable plotting, IT++ to enable mathematical operations, 

and Gstreamer to enable video/audio transmission and reception.


209 

A. Names and description of the implemented classes 

• Class BitWaterfilling. This class does the bit and power allocation according to 

spectrum sensed and estimated channel for a target rate constraint or a total 

power constraint. 

• Class BlindOFDM. This class gets a vector of bits from a named pipe (FIFO) 

coming from text, video, or audio in the application. It modulates this vector 

of bits into a fixed BPSK, QAM or an adaptive QAM OFDM vector according 

to its best group of subcarriers. It also performs a blind detection of a received 

OFDM vector based on the cyclic prefix, and blindly demodulates a received 

fixed BPSK, QAM or an adaptive QAM OFDM vector into a vector of bits 

according to its best group of subcarriers (blind time offset correction, blind 

frequency offset correction, blind phase offset estimation). Finally, it puts 

a vector of bits into a named pipe (FIFO) to be read by a text reader, video 

reader, audio reader in the application. 

• Class File. This class converts from/to characters to/from bits. It also reads/ 

writes a vector of bits from/to a file. 

• Class MainWindow. This class is dedicated to the GUI thread. This class 

has a first push Button to Start/Stop Tx, a second push Button to Start/Stop 

Rx, and a third push Button to Start/Stop Video. It uses a lineEdit to take command 

or to input text and textEdit to display some text. It also uses multiple 

lineEdits to control Tx rate, Tx frequency, Tx gain, Tx amplitude, Rx rate, Rx 

frequency, Rx gain, FFT size, CP size, and number of sub-channels. 

• Class Packets. This class does a conversion between vector of a double, float, 

integer and a vector of bits to include in the RTS/CTS packets. It encodes/ 

decodes the RTS/CTS packets with necessary information (source address, 

destination address, best sub-channel, bit and power allocation). 

• Class Plot. This class plots some information (spectrum sensing, best group 

of sub-channels, target rate constraint, bit and power allocation). 

• Class Protocols. This class is dedicated to the worker thread (QThread). 

It can reinitialize the parameters as requested by the GUI (FFT size, CP size, 

number of sub-channels). This class implements a state machine for a Tx/Rx 

handshaking adaptive OFDMA MAC protocol based on RTS sending/listening, 

CTS listening/sending and data sending/listening. It also performs a time offset 

estimation before RTS sending, CTS sending to avoid inter-carrier interference. 

• Class Sensing. This class estimates the spectrum based on averaged FFT. It takes 

a decision based on the mean spectrum estimated shifted by a fixed amount 

of dB, and selects the best group of sub-carriers according to the number 

of sub-channels. 

• Class Text. This class is implemented as a separate thread (QThread). It reads 

some input text in the GUI and put it in a named pipe (FIFO). It also displays 

some text in the GUI from a named pipe (FIFO).


• Class UHDDevice. This class reinitializes USRP parameters as requested 

in the GUI (Tx rate, Tx frequency, Tx gain, amplitude, Rx rate, Rx frequency, 

Rx gain). It reads a certain amount of samples at a particular time (timestamp). 

It writes a certain amount of samples at a particular time (timestamp). 

It also checks for errors after sending/receiving data. The procedures are 

implemented such that the received and transmitted samples are in a steady 

state for the USRP (first samples should be discarded because of the time to 

power up). 

• Class Video. This class is implemented as a separate thread (QThread). It uses 

a Gstreamer transmit pipeline showing the video transmitted on screen and 

writing the same data in a named pipe (FIFO) which will be read by the OFDM 

modulator for transmission. It also uses a Gstreamer receive pipeline showing 

the video received on screen if a video container is detected. 

B. Description of the application 

Figure 1 shows the application window for spectrum sensing. The GUI parameters 

can be chosen during run-time. The transmission parameters are the Tx 

Rate in Msps, the Tx Frequency in MHz, the Tx Gain in dB, the Tx Amplitude 

corresponding to a linear multiplication factor. The reception parameters are the Rx 

rate in Msps, the Rx frequency in MHz, the Rx Gain in dB. The OFDM parameters 

are the FFT size, the CP size, and the number of sub-channels. There are two 

buttons to Start Tx and to Start Rx, which correspond to different state machines. 

The Tx state machine does a spectrum sensing, send RTS, receive CTS, and send 

data procedure, while the Rx state machine does a spectrum sensing, receive RTS, 

send CTS, and receive data procedure. The spectrum sensing is performed using 

the method of averaged periodograms (Barlett’s method) described in Section II. 

The red plot corresponds to the sub-carriers whose powers are larger than a decision 

threshold (in this case a mean decision threshold shifted by a fixed amount 

of dB). One can see a DC offset due to the USRP and some occupied sub-carriers 

between sub-carriers 320 and 340. 

Figure 2 shows one of the plots provided by the application. The different 

plots are the spectrum sensing and decision as shown in Figure 1 (Tab 1), the best 

sub-channel plot (Tab 2), the estimated channel amplitude plot (Tab 3), the power 

allocation plot (Tab 4), the bit allocation plot (Tab 5). For the system to work well, 

a good compromise should be taken between the FFT size, CP size, the expected 

frequency offset, and the number of OFDM symbols used for estimation of the channel. 

As shown on Figure 2, a good compromise has been found with an FFT 

size 512, CP size 128, and number of OFDM symbols 50 such that we get a very 

good estimate of the channel. This serves the bit and power allocation function to 

have accurate values since the spectrum sensing is based on average periodograms 

and the estimate of the channel is based on the average of multiple OFDM symbols.


211 

Figure 1. Application window for spectrum sensing 

Figure 2. Application window for channel estimation


Figure 3 shows the application window for text, video and audio (Tab 6). This 

tab is used to input some text, video or audio in a named pipe called ‘inputpipe’ 

whenever the Tx state machine is launched. Once the enter command or the video 

button is pressed, the Tx state machine switches from a spectrum sensing state to 

a RTS sending state to initiate a communication. It is also used to output some text, 

video or audio from a named pipe called ‘outputpipe’ whenever the Rx state machine 

is launched. Once the enter command or the video button is pressed, the Rx state 

machine switches from a spectrum sensing state to a RTS listening state to receive 

a communication. After handshaking between the two CRs, data can be received 

as shown on Figure 3. 

Figure 3. Application window for text, video and audio 

Future improvements of this application would be to implement other algorithms 

for spectrum estimates (Multitaper method, Welch, Hamming, Hanning...) 

and to implement decision algorithms based on noise estimation. However, 

a whitening approach might take a long time because of eigenvalue decomposition. 

Secondly, a comparison of the new OFDM signal detection algorithm with existing 

algorithms in the literature could also lead to an improvement of the application. 

Thirdly, it would be interesting to implement other algorithms for OFDM demodulation 

to compare different non-data-aided and data-aided approaches.


213 

V. Conclusion 

In this paper, we have proposed an adaptive OFDMA PHY/MAC on USRP platforms 

for a cognitive tactical radio network. In the first part of the paper, the adaptive 

OFDMA MAC protocol has been described, as well as the key function of the adaptive 

OFDMA PHY, i.e. the spectrum sensing, the OFDM signal detection, the blind 

OFDM demodulation and the distributed bit and power allocation. In the second part 

of the paper, the adaptive OFDMA PHY/MAC implementation on USRP platforms 

using Qt4/IT++ and the UHD API has been described. Several classes have been 

implemented, as well as support for text, video and audio transmission. 

References 

[1] J. Mitola III and G.Q. Maguire Jr., Cognitive Radio: Making Software Radios More 

Personal. IEEE Personal Communications, 6(4):13-18, 2009. 

[2] FCC. Spectrum Policy Task Force Report. ET Docket, (02-135), 2002. 

[3] O. Younis, L. Kant, A. McAuley, K. Manousakis, D. Shallcross, K. Sinkar, 

K. Chang, K. Young, C. Graff and M. Patel, Cognitive Tactical Network Models. 

IEEE Communications Magazine, 48(10):70-77, 2010. 

[4] E. Hossain and V. Bhargava, Eds., Cognitive Wireless Communication Networks. 

Springer, 2007. 

[5] W. Yu, G. Ginis and J. M. Cioffi, Distributed Multiuser Power Control for Digital 

Subscriber Lines. IEEE Journal on Selected Areas in Communications, 20(5): 

1105-1115, 2002. 

[6] F. Wang, M. Krunz and S.G. Cui, Price-based spectrum management in cognitive 

radio networks. IEEE Journal on Selected Areas in Communications, 2(1):74-87, 2008. 

[7] G. Scutari, D.P. Palomar and S. Barbarossa, Optimal linear precoding strategies 

for wideband noncooperative systems based on game theory, part II: Algorithms. 

IEEE Transactions on Signal Processing, 56(3):1250-1267, 2008. 

[8] P. Setoodeh and S. Haykin, Robust Transmit Power Control for Cognitive Radio. 

Proceedings of the IEEE, 97(5):915-939, 2009. 

[9] R.H. Gohary and T.J. Willink, Robust IWFA for open-spectrum communications. 

IEEE Transactions on Signal Processing, 57(12):4964-4970, 2009. 

[10] M.i Hong and A. Garcia, Averaged Iterative Water-Filling Algorithm: Robustness 

and Convergence. IEEE Transactions on Signal Processing, 2011. 

[11] V. Le Nir and B. Scheers, Improved Coexistence between Multiple Cognitive 

Tactical Radio Networks by an Expert Rule based on Sub-channel Selection. Wireless 

Innovation Forum European Conference on Communications Technologies and 

Software Defined Radio (SDR’11-WInnComm-Europe), Brussels, Belgium, 2011. 

[12] V. Le Nir and B. Scheers, Iterative Waterfilling Algorithm with Sub-channel 

Selection for the Coexistence of Multiple Cognitive Tactical Radio Networks. Military 

Communications and Information Systems Conference (MCC’2011), Amsterdam, 

Nederlands, 2011.


[13] J. Mo, H.W. So and J. Walrand, Comparison of Multichannel MAC Protocols. IEEE 

Transactions on Mobile Computing, 7(1):50-65, 2008. 

[14] Y. Lin, K. Liu and H. Hsieh, Design of Power Control Protocols for Spectrum Sharing 

in Cognitive Radio Networks: A Game-Theoretic Perspective. IEEE Conference on 

Communications (ICC’2010), Cape Town, South Africa, 2010. 

[15] L. Yang, W. Hou, L. Cao, B.Y. Zhao and H. Zheng, Supporting Demanding Wireless 

Applications with Frequency-agile Radios. Proceedings of the 7th USENIX Symposium 

on Networked Systems Design and Implementation, 2010. 

[16] J. Blanchette and M. Summerfiled, Eds., C++ GUI Programming with Qt 4, 

Second Edition. Prentice Hall, 2008. 

[17] IT++ project. http://itpp.sourceforge.net/current/index.html 

[18] “Universal Software Radio Peripheral” hardware driver. http://code.ettus.com/redmine/ 

ettus/projects/uhd/wiki. 

[19] J.G. Proakis and D.G. Manolakis, Eds., Digital Signal Processing. Principles, 

Algorithms and Applications. 3rd Edition, Prentice Hall, 1996. 

[20] V. Le Nir, T. van Waterschoot, M. Moonen and J. Duplicy, Blind CP-OFDM and 

ZP-OFDM parameter estimation in frequency selective channels. EURASIP Journal 

on Wireless Communications and Networking vol. 2009, Article ID 315765, 10 pages 

doi:10.1155/2009/315765, 2009. 

[21] S. Chaudhari, V. Koivunen and H.V. Poor Autocorrelation-Based Decentralized 

Sequential Detection of OFDM Signals in Cognitive Radios. IEEE Transactions on 

Signal Processing, 57(7):2690-2700, 2009. 

[22] D. Danev, On Signal Detection Techniques for the DVB-T Standard. Proceedings 

of the 4th International Symposium on Communications, Control and Signal Processing 

(ISCCSP’2010), Limassol, Cyprus, 2010. 

[23] D. Danev, E. Axell and E.G. Larsson, Spectrum sensing methods for detection 

of DVB-T signals in AWGN and fading channels. Proceedings of the IEEE 21st 

International Symposium on Personal Indoor and Mobile Radio Communications 

(PIMRC’2010, Istanbul, Turkey, 2010. 

[24] J. van de Beek, M. Sandell and P.O. Borjesson, ML Estimation of Time and 

Frequency Offset in OFDM Systems. IEEE Transactions on Signal Processing, 

45(7):1800-1805, 1997. 

[25] H. Meyr, M. Moeneclaey, and S.A. Fechtek, Eds., Digital Communications 

Receivers: Synchronization, Channel Estimation, and Signal Processing. John Wiley 

and Sons, 1998. 

[26] M. Rezki, L. Sabel and I. Kale, High Order Modulation Stripping for Use 

by Synchronisation Algorithms in Communication Systems. Proceedings 

of the Instrumentation and Measurement Technology Conference (IMTC’2003, 

Vail, USA, 2003. 

[27] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge University Press, 

2004. 

[28] John M. Cioffi, A Multicarrier Primer. ANSI Contribution T1E1.4/91-157, 1991.

Validation of the ITU 1546 Land-Sea 

Propagation Model for the 900 MHz Band 

Krzysztof Bronk 1 , Rafał Niski 1 , Jerzy Żurek 1 , Maciej J. Grzybkowski 2 

1 National Institute of Telecommunication, 

Wireless Systems and Networks Department, Gdansk, Poland, 

{K.Bronk, R.Niski, J.Zurek}@itl.waw.pl 

2 National Institute of Telecommunication, 

Electromagnetic Compatibility Department, Wroclaw, Poland, mag@il.wroc.pl 

Abstract: The article discusses the accuracy of the ITU-1546 land-sea propagation model for 

the 900 MHz band. This analysis is mainly based on the measurement results which were compared 

with the data resulting from the theoretical model. This paper is comprised of four sections. Firstly, 

the measurement methodology and scenarios are described. In the next section, the method of field 

strength calculations for the mixed paths, introduced by the ITU, is briefly explained. The next, and 

most important, part is filled with the comparison (and statistical analysis) of the theoretical and 

measured field strength values, which allows to evaluate the model. Finally, in the last section, a short 

conclusion is presented. 

Keywords: component; land-sea propagation model, validation, measurements, GSM 900 


The following article presents the selected results of the extensive measurement 

campaign at the Baltic Sea waters conducted as a part of the EfficienSea project [5, 6]. 

All the activities described in this paper were carried out by the National Institute 

of Telecommunications (NIT) in cooperation with other EfficienSea partners, i.e.: 

Maritime Office in Gdynia – MOG and Gdynia Maritime University – GMU. 

The campaign took place in the second half of 2011 and comprised four 

parts lasting from one to five days, during which a total of over 40 000 measurement 

points have been gathered. For the purpose of this campaign two ships have 

been utilised: MS “Tucana” (operated by the MOG) – during the 1st and 2nd part 

of the campaign and MS “Horyzont II” (operated by the GMU) – during the 3rd 

and 4th part respectively. 

The whole measurement campaign covered selected radiocommunications 

systems available at sea, including 2G/3G, TETRA, Telenor VHF Data, Iridium. 

In this paper the authors present some results for the GSM 900 system only.


In the subsequent part of this paper the measurement methodology and measurement 

devices set will be presented. After that, the method of propagation loss 

evaluation for the mixed propagation paths, based on the ITU 1546 recommendation, 

will be discussed. Finally, a comparative analysis of the measurement results and 

the results of the predicted communication ranges, obtained from the computations 

(performed with the software tool created by the NIT) will be presented. 

II. Field strength measurements 

As the part of the EfficienSea project activities, the measurement campaigns 

at the Baltic Sea waters were conducted. The main goal of them was to analyse 

the coverage, availability, type and quality of the 2G/3G data transmission services 

offered by one of the Polish cellular providers at the Polish coastal waters (especially 

at the Bay of Gdansk and along the Polish coast line, see Fig. 7). The most 

important output of this measurements was an indication which data transmission 

services (EDGE, UMTS or HSPA) were available in the particular area of interest. 

Besides that, a big number of additional information have been obtained, such as: 

the actual achievable throughputs (for uplink and downlink), ping values, signal 

levels, Cell-IDs, etc. 

The measurement set for this part of the campaign can be schematically 

depicted as in Fig. 1. 

Figure 1. The measurement set for 2G/3G systems tests


217 

The central and most important element of the measurement set was an application 

module. This software tool (developed by the NIT for the purposes of this 

campaign) handled all the functions connected with control over the equipment and 

FTP client, triggered the subsequent steps of the measurement algorithm, performed 

the necessary calculations and stored the results. It also had a very strong capability 

of detecting and handling errors and unexpected situations, so consequently 

the whole measurement process was as automatic as possible. 

The user interface of this application is presented in Fig. 2. In this picture 

the main functions of each of the modules are indicated. 

Figure 2. User interface of measurement software tool


The second part of the measurement set was the GSM/UMTS module, based 

on the Dev Kit platform by Sierra Wireless, which was originally manufactured for 

telecommunication devices testing. This platform was equipped with a card slot 

supporting both regular modems as well as professional ones for measurement 

purposes. On the Dev Kit platform, a GSM/UMTS MC-8795V card was installed, 

which supported HSPA data transmission with rates up to 7.2 Mb/s (downlink) 

and 5.76 Mb/s (uplink). The communication between the application module and 

the GSM/UMTS module was based on the AT commands. 

The Dev Kit platform and the MC-8795V card are depicted in Fig. 3. 

Figure 3. The Dev Kit platform and the MC-8795V card 

Another part of the measurement set was the GPS module Holux M-215 

– a USB wired GPS receiver equipped with a MTK chipset, which supported the data 

transmission protocol NMEA0183 v.3.01. The module’s sensitivity was 159 dBm 

and the cold-start time was 36 seconds. 

The additional parts of the measurement set were the Anritsu MS2721B 

Spectrum Analyzer (high-performance handheld spectrum analyzer, operating 

in the frequency range of 9 kHz ÷ 7.1 GHz, suitable for a great variety of RF, microwave 

or cellular signal measurements, even in the harsh physical environment) 

and a highly-precise, calibrated antenna SAS-521F-7 by A.H. Systems. 

The methodology of the measurement was as follows. In the NIT premises 

in Gdansk, an external FTP server (connected to the Internet via a very fast fibre 

connection) has been activated, and 13 test files have been uploaded into it. 

Those files contained some random data and they varied in size (the possible sizes 

of the test files were: 10 kB, 25 kB, 50 kB, 100 kB, 250 kB, 500 kB, 1 MB, 2,5 MB, 

5 MB, 10 MB, 25 MB, 50 MB and 100 MB). The rest of the necessary measurement 

equipment was placed aboard the ship. 

Using the GSM/UMTS/HSPA Module, a radio link between the ship and 

the External FTP server was established. The module was able to work in a few 

modes: preferred 2G or 3G or the best available service (EDGE/UMTS/HSPA), 

which gave an instant information about the services available in a given area.


219 

After establishing the connection, it was possible to start a bi-directional data 

transmission between FTP ship and to measure its duration. In order to test 

the downlink transmission, the application module started to download a file 

from the FTP server; to test the uplink transmission, the same file was uploaded 

into the FTP. The size of the file that was being transmitted in a given moment 

was not random, but selected by the application module according to the special 

algorithm. If the available throughput was low, the file size was gradually decreased, 

so that the duration of a single measurement was not excessively long; 

on the other hand, if the throughput increased, it was also possible to increase 

the size of a test file, so the file size was being changed adaptively. When the transmission 

was complete, the software calculated the real throughput (separately for 

downlink and uplink), which was done through dividing the (known) file size by 

the (measured) transmission duration. The GSM/UMTS/HSPA modem allowed 

to obtain some additional parameters as well, i.e.: signal level (in dBm), number 

of the utilized GSM/UMTS channel and Cell ID. Through the ICMP protocol, 

the value of the ping was also extracted. 

Additionally, the application module was connected with the Anritsu MS2721B 

programmable spectrum analyzer to enable the optional channel power measurements. 

An integral part of the measurement set was a GPS module which provided 

a precise location- and time-stamp for every single measurement record (additionally, 

the current speed in km/h, course in degrees and altitude in m AMSL were obtained). 

After finishing a complete set of a measurement, the application module created 

a record with the results. Every record comprised the following information: 

• Date and time of the measurement; 

• Geographic coordinates (longitude and latitude); 

• Direction of radio link (downlink/uplink); 

• Measured transmission throughput [in kb/s]; 

• Measurement duration [in ms]; 

• Amount of transmitted data [in bytes]; 

• Ping [in ms]; 

• Available service [EDGE/HSPA/UMTS]; 

• Signal level [in dBm]; 

• Speed [in km/h]; 

• Course [in deg]; 

• Altitude [in m AMSL]; 

• Cell ID; 

• Nr of GSM/UMTS channel; 

• Channel power [in dBm]. 

Since the spectrum analyzer made its own measurements as well, it also created 

a log file with the results. The analyzer was also equipped with an in-built GPS 

receiver, so every record in this log file was marked with a time-stamp and precise 

geographic coordinates of the point where the current measurement took place.


In some measurement campaigns it was not possible to use the spectrum 

analyser, therefore to obtain the same conditions of received signal lever measurements, 

the modem measurement method was chosen. Moreover, the modem 

provided additional parameters which were important part of the measurements. 

It should also be noted that all measurements were performed on the move, 

and the ship’s speed was not greater than 20 km/h. 

III. Method of mixed path field strength calculations 

The results of the radio wave field strength measurements in the 900 MHz 

band, carried out for the mixture of land and see paths, have been compared to 

the field strength calculations performed according to the method provided by 

the ITU-R P.1546-4 Recommendation [1]. In this Recommendation, the mixedpath 

method of calculations utilises the values of E l (d) and E s (d) to represent 

the field strength at a specific distance d from the transmitting antenna. The additional 

factor, i.e. the representative clutter height, R, for all-land and all-sea paths 

respectively, interpolated for transmitting/base antenna height h 1 , as well as for 

frequency (900 MHz) and percentage time (50%), has been used as required. 

The receiving/mobile antenna height used for the purpose of the calculations 

was equal to the real height of the antenna mounted at the shipboard on the level 

of 10 meters over sea surface. 

The mixed path field strength, E [dB(μV/m)], has been evaluated in line 

of the ITU method [1] as: 

 

E 1A El dT A Es dT 

(1) 

where: 

d T – total distance between transmitter and receiver [km], 

E l – field strength calculated for land path zone at the distance d T [dB(μV/m)], 

E s – field strength calculated for see path zone at the distance d T [dB(μV/m)], 

A – mixed path interpolation factor, given by formula: 

2/3 

where the fraction of path over sea, F sea , is given by: 

and the factor V is expressed by: 

A[1 1 F ] V 

sea 

(2) 

F 

d 

sT 

sea 

(3) 

dT 

 

V max1.0,1.0 

40.0 

 

(4)


221 

with: 

 

N 

d 

d 

M 

s 

l 

sn 

lm 

Esn dT Elm dT 

 

(5) 

n1 sT m1 

where: 

E sn (d T ) – field-strength value [dB(µV/m)] for distance d T , assumed to be all of sea 

or coastal-land zone type n, 

E lm (d T ) – field-strength value [dB(µV/m)] for distance d T , assumed to be all of land 

zone type m. 

The length of total propagation path d T is equal to: 

d 

d 

dT dsT dlT 

(6) 

where: 

d d – total length of sea and coastal land paths traversed by radio 

d 

sT 

lT 

n 

m 

sn 

lm 

wave [km] 

lT 

(7a) 

d – total length of land paths traversed by radio wave [km] (7b) 

When the radio path consists of two parts of land or sea zones, then N s =2 

and M s =2 respectively. When a mixed path includes only one land and only one 

sea zone, then N s =M s =1. Only those types of paths (see Fig. 7) were taken into 

consideration during the mixed radio path field strength calculations. A detailed 

method to determine the particular field strength values E (E l or E s ), using the field 

strength versus distance curves, valid for the land and/or see paths, is described 

in the ITU-R P.1546-4 Recommendation. 

The mixed path field strength calculations were conducted by means of the Digital 

Terrain Model (DTM) of Polish coast and Baltic Sea. DTM model SRTM-3 with 

the 3 seconds resolution, which means, for areas in Poland, a rectangle of the size 

60x90 meters, has been used during these calculations. 

The comparison of the mixed path field strength calculations’ results obtained 

for the 900 MHz frequency band at the distance up to 37 km from the transmitting 

antenna are illustrated in the next section. These results were necessary 

to validate the quality of the mixed path field strength prediction provided by 

the ITU 1546 method. 

IV. Comparison of results of measurements and field strength 

calculations 

All the presented measurements were carried out on board the MS Horizon II 

ship (see Fig. 4). The antennas of the measurement set were placed 10 m AMSL


and the measurement route was within the Bay of Gdansk and along the east part 

of the Polish coastline – measurement locations are depicted in Fig. 7. For the further 

analysis, eight GSM 900 base stations (for which the distance from the sea was at 

least 5 km) have been selected. The BTSs’ parameters obtained from the network 

operator are given in the Table 1. In the considered case, the measurement points 

for seven stations correspond to the path profile type 1 (land – sea) and for one 

station the type 2 of the profile is more suitable (land – sea – land – sea). For all 

of the stations the statistical analysis of the obtained field strength measurements 

results has been conducted. The input parameter, i.e. the difference between the calculated 

and measured field strength values, ΔE, is given as follows: 

EE1546 E meas 

[dB] 

(8) 

where E 1546 denotes the median of the field strength calculated on the basis of the 

ITU 1546 model and E meas indicates the field strength value obtained in the measurements. 

The levels (values) of the electric field strength E were calculated on the basis 

of the parameters shown in table 1 (including BTS antenna’s characteristics), 

in the software tool developed by the NIT [2], in which the newest version of the 

ITU 1546-4 Recommendation was implemented. 

In Figs. 6 and 7 the values of ΔE as a function of distance for different 

types of the path profile were shown. It is noteworthy that the red line denotes 

the mean value and the green ones represent the standard deviation. 

Figure 4. Horyzont II – the ship utilized during the measurement campaign


223 

Figure 5. Dependence of ΔE on distance for propagation path type 1 

Figure 6. Dependence of ΔE on distance for propagation path type 2 

Figure 7. Measurement area


Table I. BTS information obtained form the network operator 

BTS Name BTS1 BTS2 BTS3 BTS4 BTS5 BTS6 BTS7 BTS8 

Longitude [deg] 16,55 17,90 17,90 18,46 18,47 18,56 18,64 19,61 

Latitude [deg] 54,48 54,74 54,74 54,50 54,47 54,34 54,36 54,29 

Terrain height [m] AMSL 59 67 67 142 140 127 43 102 

GSM channel 66 59 49 65 50 65 56 65 

Antenna height [m] AGL 46 48 49 39 35 16 23 49 

Antenna azimuth [deg] 330 330 60 70 145 60 60 300 

Antenna type K 739650 K 739650 K 730691 K 739 685 K 742 272 K 739622 K 742264 K 730691 

EIRP [dBW] 27,5 26,9 27,2 27,1 27,9 26,3 25,2 26,8 

Table II. Detailed analysis results 

BTS Name BTS1 BTS2 BTS3 BTS4 BTS5 BTS6 BTS7 BTS8 

Measurements number 390 75 118 141 24 26 19 53 

Path type 1 1 1 1 1 1 1 2 

Min path length 29,8 15,7 16,4 26,9 26,6 19,7 16,9 28,1 

Max path length 36,7 16,4 30,2 35,6 33,4 23,5 34,1 33,2 

ΔE min [dB] -6,5 -11,5 -12,5 -10,5 -10,5 -11,5 -13,5 -6,5 

ΔE max [dB] 8,5 -0,5 -0,5 -5,5 0,5 -4,5 -0,5 8,5 

average of ΔE [dB] 2,3 -5,2 -7,1 -8,1 -4,4 -7,9 -4,6 -0,2 

RMSE [dB] 3,0 2,6 2,9 1,1 4,1 1,7 3,7 2,9


225 

Table 2 presents the results of the analysis, i.e.: number of measurements, 

minimum and maximum distance between measurement points and BTSs (length 

of the radio path), minimum and maximum ΔE values as well as the average value 

of the ΔE and RMSE. 

In the Table 3 the summary results of the measurement errors, for both types 

of the path profile as well as for all measurement points, are presented. 

Table III. Overall analysis results 

BTS Name BTS1-7 BTS8 All 

Measurements number 793 53 846 

Path type 1 2 1 & 2 

Min path length 15,7 28,1 15,7 

Max path length 36,7 33,2 36,7 

ΔE min [dB] –13,5 –6,5 –13,5 

ΔE max [dB] 8,5 8,5 8,5 

average of ΔE [dB] –2,4 –0,2 –2,2 

RMSE [dB] 5,4 2,9 5,3 

The obtained results show a very good agreement of the measurement data 

with the results obtained from the mathematical model introduced by the ITU 

in the recommendation 1546-4. It is proved by the small values of the average error 

ΔE, which amounts to –2,4 dB for the propagation paths of type 1 (land-sea) and 

–0,2 dB for the propagation paths of type 2 (land-sea-land-sea), as well as –2,2 dB 

on average. Negative values of the average ΔE mean that in most cases the measured 

level of the signal strength was higher than expected. 

Additionally, the relatively small values of the RMS error, which are respectively: 

5,4 dB for propagation path type 1, 2,9 dB for type 2 and 5,3 dB on average 

verify positively the correctness of the measurement method and reproducibility 

of the measurements. Both values (average and RMS errors) are smaller for 

the propagation paths type 2, which might suggest that for these type the correctness 

of the model is better. On the other hand, it must be mentioned that the number 

of the measurements for the propagation path type 1 was 793, whereas for type 2 

it was almost 15 times less, merely 53. 


In the paper the comparative analysis of the ITU 1546 land-sea propagation 

model for the 900 MHz band and the actual field strength measurements results 

has been presented. An important part of this research was a statistical analysis, 

which allowed to validate the accuracy of the ITU 1546 model.


The comparative research and measurement [3,4], which has been carried 

out earlier by the authors in the maritime VHF band, showed that the measured 

signal level was higher than the level calculated according to the method described 

in the recommendation ITU 1546. In this case the differences were from several to 

dozen dBs. It proves, yet again, that for the land-sea paths in the 900 MHz band, 

the discussed model is slightly underestimated. 

To make the comparison more reliable it would be advisable to assess and 

consider the type of the environment in which BTSs are located. This issue is addressed 

in the authors’ current research activities. 


The measurement results utilised in this paper have been obtained as part 

of the EfficienSea project, partially financed by the EU Baltic Sea Region 

2007-2013 Programme. 

References 

[1] ITU-R, Method for Point-to-Area Predictions for Terrestrial Services in the Frequency 

Range 30 MHz to 3000 MHz, Recommendation ITU-R P.1546-4, Geneva, 10/2009. 

[2] R. Niski, K. Bronk, J. Stefański, Opracowanie narzędzia programowego do 

prognozowania zasięgów stacji brzegowych dla potrzeb radiokomunikacji morskiej 

(nr 08300047) Instytut Łączności – Państwowy Instytut Badawczy, 2007. 

[3] R. Niski, J. Żurek, New empirical model of propagation path loss for the Baltic Sea 

in the marine VHF frequency band (Polish Journal of Environmental Studies) HARD, 

Olsztyn 2007, no. 4B, vol. 16, pp. 143-145. 

[4] R. Niski, J. Żurek, Empiryczna weryfikacja modeli propagacyjnych w strefie 

przybrzeżnej Morza Bałtyckiego (Zeszyty Naukowe Wydziału Elektroniki, 

Telekomunikacji i Informatyki Politechniki Gdańskiej. Radiokomunikacja, Radiofonia, 

Telewizja) Politechnika Gdańska, Gdańsk 2007, nr 1, s. 117-120. 

[5] Communication for e-Navigation – results of the tests and measurements, E-Navigation 

Underway 2012 Conference, 18-20 January 2012, Copenhagern – Oslo – Copenhagen. 

[6] A. Lipka, K. Bronk, R. Niski, Measurement campaign at the Baltic Sea, 2012, 

EfficienSea report.

Chapter 7 

Mobile Ad-hoc 

and Wireless Sensor Networks

Algorithms for Channel and Power Allocation 

in Clustered Ad hoc Networks 

Luca Rose 1 , Christophe J. Le Martret 1 , Mérouane Debbah 2 

1 Thales Communications and Security, France, 

{luca.rose, christophe.le_martret}@thalesgroup.com 

2 Alcatel – Lucent Chair in Flexible Radio, Supelec, France, 

merouane.debbah@supelec.fr 

Abstract: In the context of mobile clustered ad hoc networks, this paper proposes and studies a self- 

-configuring algorithm which is able to jointly set the channel frequency and power level of the transmitting 

nodes, by exploiting one bit of feedback per receiver. This algorithm is based upon a learning 

algorithm, namely trial and error, that is cast into a game theoretical framework in order to study its 

theoretical performance. We consider two different feedback solutions, one based on the SINR level 

estimation, and one based on the outcome of a CRC check. We analytically prove that this algorithm 

selects a suitable configuration for the network, and analyse its performance through numerical 

simulations under various scenarios. 


In recent times, the interest for technological solutions which allow communications 

to happen in difficult conditions, e.g. without the aid of a central controller, 

has gained much momentum. The development of cognitive radios (CR), devices 

able to sense their environment and to modify their configuration in accordance, 

has made this a reality. 

On operational theatres, the presence of a fixed central controller infrastructure, 

for instance a base station, configuring the whole network is difficult to implement 

and is not desirable for the weakness it presents against potential enemies. Moreover, 

one can expect future equipments on the battlefield to be able to exploit the free 

spectrum to communicate and to keep their transmit power as low as possible. 

The goal is both minimizing their spatial frequency footprint, avoiding to pollute 

transceivers from other networks, and reducing the battery drain while achieving 

a certain Quality of Service (QoS). The concept of cognitive, self-configuring ad hoc 

network, thus, is a candidate solution to all of the above challenges. 

In our work, we consider clustered ad hoc networks where the nodes are grouped 

into subsets (clusters), each of which is led by a cluster head (CH). We assume


that all the clusters share the same frequency band, each CH being in charge of allocating 

sub-channels of the common resource to the multiple transmitter-receiver 

links that need to be operated within its cluster. 

The CH, basically, fulfils two purposes: () i it selects a frequency-channel and 

a power level to be employed by the devices within its control zone, ( ii ) it manages 

the intra-cluster communication by allocating logical sub-channels to each link. 

Thus we can consider our system as locally centralized, and globally distributed. 

In order to do so, we assume that the CH only relies on local information, without 

any form of cooperation or explicit coordination with the other CHs. This reduces 

the amount of signalling demanded and makes the network more resistant to jamming 

attacks. For the same reasons, we need to minimize the amount of feedback 

between the CH and the nodes under its control. 

The closest works to ours are [1-4]. In [1] an algorithm for interference avoidance 

is presented assuming an underlying clustered ad hoc network. The algorithm 

sets the frequency channel, leaving to the CH the duty to choose the power based 

on the needs of the cluster’s devices. The authors assume the clusters to be far 

apart from each other in such a way that the interference created form one cluster 

to another does not depend on the actual transmitters location. In [2], authors 

consider and present a trial and error (TE) algorithm, and analytically study its 

convergence properties. There, the scenario under analysis is composed of a group 

of communicating links, without considering the structure of a clustered network. 

In [3], authors suggest the use of iterative water filling (IWF) to allocate sub-channels 

and power in order to achieve a certain QoS, measured in terms of achievable rate. 

The authors assume a system with low interference, i.e., interferers very distant 

from each other, such that the convergence of the IWF could be insured. In [4], 

authors consider a clustered network where, in each cluster, a single transmitter 

broadcasts to the other nodes. In this work, each transmitter allocates its power 

using an IWF strategy aiming at maximizing the weighted mean of the throughputs. 

In a clustered network with many transmitters and only one decision maker, it is 

not practical to implement such a water-filling strategy. Indeed, this would require 

all the receivers to feedback to the decision maker their channel state information. 

Thus, this strategy requires a large amount of signalling to allow the CH to evaluate 

the correct power allocation. Moreover, there exists a sufficient literature, e.g. [4-6] 

showing that, in decentralized networks, the operating point achieved through 

IWF is often less efficient than the one achieved through spectrum segregation, 

i.e. forcing each link to operate only on a small fraction of the available bandwidth. 

In our paper, we present and detail an algorithm which, when employed by all 

the CHs, is able to set the network channel and power configuration by exploiting 

the information of only one bit feedback per receiver. This algorithm, namely trial 

and error learning algorithm, has been studied in [2] under the assumption of a static 

scenario (i.e., time invariant channel, fixed power gains and network topology), 

with the transceivers aiming at achieving a certain SINR to fulfil a given QoS.

Chapter 7: Mobile Ad-hoc and Wireless Sensor Networks 

231 

In this paper, we study several scenarios taking into consideration cluster 

mobility as well as more realistic communication performance metrics. We show 

the capability of the proposed algorithm to statistically steer the network into a state 

where clusters next to each other employ different channels. 

Therefore, the main contributions of this paper are the following: () i we detail 

a self configuring algorithm by defining all its parameters; ( ii ) we study its behaviour 

under several scenarios; ( iii ) we compare two ways of measuring transmission 

success, either by comparing the estimated SINR to a target or by considering 

packet integrity through cyclic redundancy check (CRC) of the transmitted packet; 

( iv ) through numerical simulations, we estimate the optimal number of spectral 

resources, (i.e., channels) the network should be providing for the algorithm to well 

perform. 

The paper is organized as follows. In Sec. II we present the general model of an 

ad hoc network and provide its associated game-theoretical model in Sec. III. In Sec. 

IV we briefly describe the resource allocation algorithm and we show the test bench 

scenarios in Sec. V providing the results of the experiment in Sec. VI. Finally, we 

conclude our work in Sec. VII. 

II. System model 

In this work, we consider a network populated with 

of which composed by 

links (transmitter-receiver pairs), with 

clusters, each 

N 

N 

K 

N . 

k k k 

Let 1, 2, , K 

 

k 

1, 2, , 

N 

the set 

k 

of links within an arbitrary cluster . The nodes communicate by sharing a common 

spectrum, thus creating mutual interference. The overall spectrum is divided 

into channels, and we denote by 1, 2, ,C 

the set of available channels. 

Each cluster, say , is managed by its CH, which selects its transmission setting, 

i.e., a channel and a power level , to be used by all the devices belonging 

to the cluster. The power level is chosen among a finite set of possible power 

levels 0, , P MAX 

, where P 

MAX 

is the maximum amount of power that can be 

used by a transmitter device. The CH divides the selected channel, , into N 

sc 

orthogonal logical sub-channels and assigns them to the links to avoid intra-cluster 

interference. Assuming a time division multiple access scheme (slotted frame), 

each CH allocates to each link a set of sub-channels per slot, as depicted in Fig. 1. 

We define by the set of sub-channels allocated to link and by an arbitrary 

element of . In every cluster we also assume that the transmit power on each 

sub-channel is constant for all the links. 

indicates the set of clusters and 

k 

k


Figure 1. Sub-channel assignment instance. Each different colour corresponds to a different link 

We consider flat and block fading channels, i.e., channels power gain is both 

time and frequency invariant for the whole duration of one transmission. As such, 

the level of multiple access interference (MAI) in each sub-channel suffered by 

a receiving node, for instance the receiving node of link , on the sub-channel 

is given by the sum of the interference created by all the transmitters which employ 

the same sub-channels at the same time, that is: 

N 

x 

k 

MAI k 1 

( , ) 

pxg( l, m)1 t 

 

. 

m s ckc 

 

(1) 

 

x 

s s 

N 

 

sc 

xk lx 

k 

In (1), gl (, 

m) 

indicates the channel power gain between the transmitting 

node of link and the receiving node of link , and 1 

is the indicator function. 

Therefore, the level of the SINR experienced by the receiver of link on 

sub-channel is given by: 

N pg( , ) 

SINR 

, 

( 

k , s) 

m N 

k k 

k k m m 

2 

sc 

MAI 

k 

( m , s) 

k k 

where g( m, 

m) 

indicates link power gain, which is modelled by the two-ray 

model [7], i.e. 

2 2 

k j k j 

m l m l 

 

4 

d k j 

( m , l ) 

(2) 

G G h h 

k j 

g( , ) . 

(3) 

m 

l 

In (3), and represent the antenna gains, , the height of the antennas 

of nodes and respectively, and d 

( 

k j 

m , 

is the distance between the two 

l ) 

nodes. In order to study the performance of the network, we assume the queue of each 

transmitter to be not empty, i.e., we analyse the system in a fully loaded situation. 

For the sake of simplicity, we consider an uncoded binary phase shift keying 

(BPSK) modulation scheme for each sub-channel transmission. Since the transmitters 

may use multiple sub-channels per link to perform their communication, we 

introduce an equivalent SINR that accounts for all the sub-channels in order to 

assess the link performance. We define our equivalent SINR based on a bit error


233 

rate (BER) point of view. Here, we consider the interference as Gaussian noise, thus, 

the equivalent SINR may be expressed, by applying uncoded BPSK BER formula, as: 

 

N 

 

SINR ( ) erfc erfcSINR 

 

, 

 

k 

1 

k 

eq m 

( 

k , s) 

N 

sc m 

s 

where erfc is the complementary error function. 

III. Game formulation 

In this section, we model the scenario presented in Sec. II under a normalform 

formulation [8]. 

(4) 

A. Normal-form 

A game in a normal-form is defined by a triplet: 

 

 

 

 

 

, , uk (5) 

k 

where, represents the set of players, 12 ... K 

is the joint set of actions 

with , i.e., ak ( pk, ck) 

. Since the utility is a measure of the individual 

quality of the chosen action, its formulation strongly depends on the type 

of feedback chosen. Here, we formulate our utility function as 

1 

 

p 

 

k 

uk( a) 1 Feedback 

x( ) , 

1 N 

a 

 

k 

P 

MAX x 

 

where Feedback 

x( a ) is a one bit value, which depends on the nature of the feedback 

chosen in the network, as described in the following section. This utility function 

is chosen to be monotonically decreasing with the power consumption 

and increasing with the number of successful transmission Feedback 

x( a ). The parameter 

tunes the interest we have in satisfying the constraints over the power 

consumption. 

Definition 1. (Interdependent game). The game is said to be interdependent 

if for every not empty subset and every action profile a 

( a , a ) 

 

it holds that: 

' 

' 

i , a a : u ( a , a ) u 

( a , a ). (7) 

k 

i i 

 

(6) 

In the following, we assume that game is interdependent. This is a reasonable 

assumption, since, physically, this means that no cluster is electromagnetically 

isolated. Under a normal-form formulation, the solution concept used is the Nash 

equilibrium (NE), which we define as follows:


Definition 2. (Nash equilibrium in pure strategies). An action profile 

is a NE of game if and a 

k 

k 

* * ' * 

u ( a , a ) u ( a , a ). 

(8) 

k k k k k k 

Generally speaking, a game can have an arbitrary number of NE, thus, to 

measure the efficiency of each one, we introduce the social welfare function, defined 

by the sum of all individual utilities: W( a) u ( a ). 

B. QoS and feedback strategies 

K 

 

In this work, we express the QoS constraints in terms of SINR, which means 

that we fix a given SINR target for each link. For simplicity sake, this value will be 

assumed here constant, i.e. equal to , for all links. As explained in the previous 

section, the utility function design (6) allows the system to take these constraints 

into account. 

We discuss now two different feedback strategies that can be applied in real 

systems. 

1) SINR-based feedback: 

This is the first strategy that naturally arises, given that the QoS is expressed 

in terms of SINR. Most of communication systems estimate the received SNR 

based on pilot sequences, and thus the SINR when MAI is present. Relying on this 

capability, we define the feedback as: 

Feedback a 1 . 

(9) 

x 

k 

k 

SINR 

a 

This formulation was proposed and studied in [9]. There, authors proved that 

with a utility function such as (6), the action profile which maximizes the social 

welfare is the one which () i maximizes the number of links which simultaneously 

satisfy the SINR condition, ( ii ) minimizes the network power consumption. Tuning 

the parameter allows to favour either the QoS constraints satisfaction for 

large values, or the consumed power for small values. 

2) CRC-based feedback: 

Usually, communication systems implement a CRC to check the integrity 

of the received packets. From this information, it is thus possible to infer the quality 

of the communication link, and this allows us to consider another kind of feedback 

defined as: 

Feedback ( a ) 1 . 

(10) 

x 

 

x 

 

CRC x ( a) 0 

Here, the receivers feedback a 1 if the packet is received without errors and 

a 0 otherwise. Note that, in this case the result in [9] does not apply, especially since


235 

the CRC is a stochastic function of the action profile . In this case, the theoretical 

framework is not able to predict the exact point of convergence of the algorithm. 

However, simulation results, illustrated in Sec. VI, indicate that this way of evaluating 

the feedback results in better performance. 

IV. Trial and error 

In this section, we briefly summarize the TE algorithm, introduced in [10], [11], 

and applied to wireless networks in [2]. TE is a state machine which selects, in a fully 

decentralized way, a strategy for a player such that, when every player is using 

the same scheme, the system is at an optimal NE a large proportion of the time 

with high probability. A state of a player is defined as a triplet zk ( mk, ak, uk) 

, 

where mk, ak, 

u 

k 

represent, respectively, the mood, the benchmark action and 

the benchmark utility of player . There are four possible moods, each implying 

a different behaviour and depending on different responses by the network. 

• Content 

If player is content, then it plays action a 

k 

with probability (1 ) , and 

another action (chosen randomly according to some probability distribution) 

with probability Here, 0 1, namely the experimentation probability, 

0.02 

is a parameter of the system. Numerical simulations suggest as a value 

K 

with a good trade off between stability and experimentation. At each iteration, 

each player compares the actual utility with the benchmark utility u 

k. 

There 

are four possible outcomes: () i if uk 

uk 

, and it did not experiment, i.e., ak 

ak 

, 

player mood becomes hopeful, ( ii ) if uk 

uk 

, and it experimented, i.e., ak 

ak 

, 

( ( k ( ) k )) 

then, with probability 

F u a 

u , becomes the new benchmark action, and 

the new benchmark utility; ( iii ) if uk 

uk 

and ak bar ak 

then the player mood 

turns to watchful; ( iv ) if uk 

uk 

and ak 

ak 

, then nothing changes. Here F() 

, 

is a non increasing function as explained in [11]. 

• Hopeful 

If player is hopeful it evaluates its utility and compares it with the benchmark 

utility u 

k 

. If uk 

uk 

, then the player mood becomes content and the benchmark 

becomes the new benchmark utility. If uk 

ruk, 

then the player becomes 

watchful. 

• Watchful 

If player is watchful it evaluates its utility and compares it with the benchmark 

utility u 

k. 

If uk 

uk, then the player mood becomes discontent. If uk 

uk, 

then the player becomes hopeful.


• Discontent 

If player is discontent, it experiments a random action and evaluates its 

G( u ) 

corresponding utility . Then, with probability k 

the player mood becomes 

content, with and as new benchmark action and utility. Here, G( ), is a non 

increasing function as explained in [11]. 

A. Trial and error properties 

The theoretical properties of TE have been thoroughly analysed in precedent 

works. In this section, we report two among the most relevant results with our 

notations. 

Theorem 1. Let be an interdependent game, and let it have at least one NE 

and let each player employ TE, then a NE that maximizes the social welfare among 

all equilibrium states is played a large proportion of the time. 

This theorem, shown in [11], states that the algorithm does not only look for 

individual optimality (the NE) but, among the states individually optimal, it searches 

the one which maximizes the global outcome. 

Theorem 2. Let and let game be interdependent with at least one NE. 

Then, TE converges to the NE where the number of links satisfied is maximized and 

the power employed to obtain this result is minimized. 

This result, proven in [9], shows that TE is able to select among all the possibilities 

an optimal working point for the network under analysis, at least for a large 

proportion of the time. 

V. Scenario description 

The scope of this section is to present and describe the scenarios used to 

run the simulations and study the performance of TE. First, we consider a static 

dense scenario. Second, we consider a mobile scenario with one cluster moving 

around four static clusters. We aim at illustrating that TE is suitable for configuring 

networks even in mobility, where channels are, thus, no more time-invariant. 

In the following, we set K 1 , to comply with the conditions in Theorem 2. 

A. Static scenario 

In this scenario, we consider a square field of 5 km per side populated with 

K 16 equally dimensioned square clusters, each of which has a side of 5 km. 

4 

In each cluster, 8 nodes are randomly positioned as in Fig. 2. The clusters are 

not overlapping, the nodes belonging to each cluster are coloured with different 

colours, and the role (transmitter or receiver) is decided once and for all. In this


237 

scenario, each cluster has N 

sc 

sub-channels, which are randomly associated 

with the links. This means that, between two TE loops there will be three time slots, 

N 

sc 

and three feedbacks. For each of these packets 2 sub-channels are randomly 

Nk 

assigned for each link. 

Figure 2. Square scenario setting with K = 16 clusters and N k = 4 pairs. Clusters and nodes 

are static with SINR-based feedback. CH, AVG PW, and AVG SAT indicate respectively the most 

frequently selected channel, the APC and the AS. 

B. Mobility scenario 

In this scenario, we evaluate the performance of TE in the presence of a moving 

cluster. We assume clusters to be aligned and sharing the spectrum 

while a fifth cluster is far enough to be creating little interference. An instance 

of this starting situation is depicted in Fig. 3. In this case the topology is such that, 

between the four static clusters, there exists an empty space for the fifth cluster to 

pass. Therefore, when all the five clusters are aligned, no cluster is overlapping with 

another. This happens after around 2250 iterations. Later, the cluster in mobility 

reaches the end of the field after 3000 iterations. Here, the number of available channels 

is restricted to .


Figure 3. Cluster positions at the beginning of the mobility scenario with K=5 clusters in a field 

of 1 km side. Four clusters are static and aligned, the cluster at the bottom is the one in mobility. 

VI. Simulation results 

In this section, we evaluate the performance of the TE for the scenarios 

introduced in Sec. II according to some metrics defined in the following section. 

A. Performance metrics 

In order to evaluate the performance and the behaviour of the proposed 

algorithm, we have selected the following metrics: 

• Average satisfaction (AS): defined as the average number of positive feedbacks 

the receivers send to their CH, for each iteration of the TE. It evaluates 

how much the algorithm enables to satisfy the criterion selected by 

the feedback (either SINR or CRC). 

• Average power consumption (APC): defined as the average amount of power 

used by the transmitters in a cluster to achieve the corresponding satisfaction 

level. It captures how much power is consumed per cluster. 

• Packet error rate (PER): defined as the average dropped packets, it helps 

evaluating the link quality and thus if the algorithm is correctly configuring 

the network.


239 

• Channel switch per iteration (CSpI): defined as the average number of channels 

that have changed for each TE iteration and thus captures the channel 

allocation stability. 

B. Static scenario, SINR-based feedback 

In this section, we analyse the performance of TE, in terms of satisfaction and 

power consumption, applied to the square scenario described in Sec. V-A. Here, 

receivers feedback their satisfaction based on the comparison between the received 

SINR and the threshold , fixed in the simulation equal to 10 dB. 

In Fig. 4, we plot AS in the network and the APC by the nodes as a function 

of the iteration number. As we can see, full satisfaction is not reached. This is due 

to the scarcity of resources in the network that does not permit full satisfaction. 

This can be understood intuitively since, in a network with K 16 clusters sharing 

channels, each cluster has on the average two neighbour clusters which 

employ the same channel. 

Figure 4. Achieved AS and APC as a function of the TE iterations for a square static scenario, 

with SINR-based feedback 

In Fig. 2, we show the node localizations on the field and the corresponding 

links with the AS and APC along with the most often chosen channel for each


cluster. Note that having the same channel as the most used ones does not imply 

a collision, since the channels might be used in different time slots. On the contrary, 

having two different channels as the most used one implies no interference 

for a large part of the simulation. 

C. Mobility scenario 

In this simulation, we refer to the scenario presented in Sec. V-B. First we 

consider the case where receivers feedback their satisfaction based on the comparison 

between the received SINR and the threshold 10 dB. Then, we consider 

the case where receivers send a CRC-based feedback. 

In Fig. 5, we plot the global performance of the system in terms of AS and APC. 

It is possible to see the drop down of the system performance after 2000 iterations. 

The algorithm reacts by increasing the power level and by modifying the channel 

configuration. The satisfaction level, then, increases when the algorithm rearranges 

the channel and power allocation scheme in order to suit the new topology. Note 

that, when the mutual interference is too high, TE turns off one cluster by selecting 

zero power. The rationale behind this is that, if the desired level of SINR is not reachable 

by the current topological configuration, then the algorithm prefers to stop one 

of the clusters to improve the individual utility. When the algorithm reaches a different 

channel assignation pattern it is, again, possible to achieve a higher level of satisfaction. 

Figure 5. Achieved AS and APC as a function of the TE iterations for a square static scenario, 

with SINR-based feedback


241 

In Fig. 6, we plot the AS and APC in a similar scenario, where the feedback 

is based on the evaluation of a CRC over a packet of 256 bytes. Note that, 

here, the reaction to the approach of the moving cluster appears to be a sudden 

increment in the power level. The power level increment is larger then when 

using an SINR-based feedback. Intuitively, this is due to the fact that the CRC 

test is more tolerant on the SINR decrement than the SINR test. Therefore, 

the transmission power increment is more effective to insure the compliance 

with the constraints when considering a CRC-based feedback than when considering 

a SINR-based one. 

Figure 6. Achieved AS and APC as a function of the TE iterations for a the mobility scenario, 

with SINR-based feedback 

In Fig. 7 we plot a summary of the simulation run. Here each colour represents 

one of the possible two channels, while the height of the bins represents the used 

power. The static clusters are indexed with numbers 1, 2, 4, and 5 and the moving 

cluster is indexed with the number 3. When the system reaches time instant () i 

the 3rd cluster is close enough to create interference to the other clusters. This forces 

the system to reorganize the power-channel pattern. When the moving cluster 

is completely aligned with the others ( ii ) the system starts working in an orthogonal 

way and the power starts decreasing. At ( iii ) the cluster is far enough to stop 

creating interference.


Figure 7. Channel-power allocation as a function of the TE iterations for the mobility scenario with 

two channels. Each colour represents a different channel, and the heights of the graph the transmit 

power level. Clusters 1, 2, 4, 5 are static, cluster 3 is in mobility. (i) beginning of the interference from 

the 3rd cluster, (ii) Five clusters are aligned, (iii) end of interference from the 3rd cluster. The blue 

solid lines represent P MAX = 50 W. 

D. Static scenario, CRC-based feedback 

In this section, we analyse the performance, in terms of satisfaction and power 

consumption when the TE is applied to the square scenario described in Sec. V-A. 

We recall that, in the following graphs, the upper curve measures the AS, where 

the feedbacks are calculated with a CRC on the received packets. We recall that in this 

simulation each packet is considered to be 256 bytes long. In Fig. 8, the performance 

of such a system is summarized. The upper curve represents the AS reached in the network, 

while the lower curve represents the APC. Note that, it is not possible to directly 

deduce the PER from the satisfaction. Especially, low levels of AS do not automatically 

translate into high levels of PER. This is because, when the transmitter is employing 

zero power, which may happen especially if the satisfaction level is low, the feedback 

is zero, but it cannot be considered as an unsuccessful transmission. Therefore, to 

evaluate the PER, we need to reduce the level of no-satisfaction by the amount 

of time the transmitters were using zero power. On the other hand, a high level 

of AS, can guarantee a high number of packets received correctly, which translates 

in a low PER. In this system, simulation results indicate an average PER = 2.8 10 -3 . 

Note that, when we employ an SINR-based feedback, we obtain PER = 0.23, which 

is much higher for equivalent average employed power.


243 

Figure 8. AS and APC as a function of the TE iterations for a square static scenario, 

with CRC-based feedback 

In Fig. 9, the performance of the algorithm on a single node is reported. It is 

possible to see that, generally, most of the transmitted packets are correctly received. 

Moreover, it appears that packets errors increase during some particular time windows, 

i.e., errors appear in burst. This is probably due to a change in the network 

(for instance another cluster starts employing the same channel) which makes 

the power-channel pair chosen by the CH inappropriate for the transmission. 

Figure 9. Fraction of packet correctly received. Single node CRC outcome for scenario 

V-A CRC-based feedback


E. Channel switch per second 

The stability of a network configuration is an important parameter to evaluate 

the performance of a self configuring algorithm. TE attempts to steer the network 

to a NE, which is inherently stable point. Nonetheless, the stochastic nature of TE, 

the incompleteness of the information and the lack of CH cooperation leave 

space for interference and collisions. To evaluate this instability we have defined 

the CSpI metric in Sec. VI-A. To compute it, we run 20 simulations on the scenario 

described in Sec. V-A and we count the number of time a CH switches its channel. 

We performed this evaluation both in the case of a SINR-based feedback and 

of a CRC-based feedback and found CSpI SINR = 4.5 10 -3 and CSpI CRC = 4.3 10 -3 . 

As we can see the results are very close one to each other. This is due to the fact that 

avoiding other clusters interference is important independently from the nature 

of the feedback. As a consequence, in both cases clusters try to employ good (low 

interference) channels. 

F. Average satisfaction versus available channels 

Here, we aim at evaluating the variation of TE’s performance as a function 

of the available channels. In this simulation we use the scenario depicted in Sec. V-A, 

where we set a CRC-based feedback. In this scenario, we have K 16 clusters and 

we vary the number of available channels for the network from 4 to 18. For each 

Figure 10. Expected satisfaction versus available channels. This plot has been realized assuming 

a square field as the one described in V-A, assuming a SINR-based feedback


245 

of this values, we run 20 tests, each of which lasts 6000 TE iterations. We recall 

that, three packets are sent for each iteration, and each packet has 256 bytes length. 

The result is depicted in Fig. 10. It is possible to see that the curve does not reach 

the full satisfaction. This is due to the stochastic nature of the algorithm. Since 

clusters are experimenting, and the CH have no way of cooperating one with each 

other, a certain, even if low, level of unsatisfaction is unavoidable. From these 

results, it appears that the optimum number of channels should be 10. Here, we 

mean optimum as the minimum number of channels needed to keep the network 

satisfied at least 90% of the time. 

VII. Conclusion 

In this paper, we have presented and studied the performance of a resource 

allocation algorithm, namely the trial and error (TE) learning algorithm. We have 

shown that it is effectively capable of setting the transmission parameters (channel 

and power) of clustered ad hoc network, using only one bit feedback per receiver. 

This feedback must be an evaluation of the quality of the transmission link. In our 

settings, we have proposed two different types of feedback strategies: one based 

upon the measurement of the SINR at the receiver, the other reporting the CRC 

check status of the transmitted packet over the link. 

In a crowded network, when several clusters try to share a few spectral resources, 

TE is able to find a setting such that the largest part of the cluster fulfils its 

QoS constraints, employing a low level of power. The clusters which are not able to 

fulfil their QoS constraints are automatically turned off, saving battery power and 

avoiding useless interference. 

When clusters are moving, the changes in the topology force the algorithm to 

react quickly and to find a different channel and power allocation scheme, such as to 

satisfy the new conditions. This may be done by a temporary increase in the power 

level, or by a reorganization of the channel assignment. 

Several paths could be followed to extend this contribution. Experimentation 

parameters which adapt on the satisfaction levels, for instance, could be used to let 

the algorithm discriminate between almost static or high mobility situations. Moreover, 

a study on a more effective probability distribution for the experimentation, 

and on the effect of the values of the parameters and could bring insight on 

ways to improve the performance. Finally, the case when an action profile depends 

upon stochastic parameters would need to be investigated to study convergence 

properties of the game when CRC-based feedback is used. 

VIII. Acknowledgment 

This research work was carried out in the framework of the CORASMA EDA 

Project B-0781-IAP4-GC.


References 

[1] B. Babadi and V. Tarokh, “GADIA: A greedy asynchronous distributed interference 

avoidance algorithm,” IEEE Transaction on Information Theory, vol. 56, pp. 6228-6252, 

Dec. 2010. 

[2] L. Rose, S.M. Perlaza, M. Debbah, and C.L. Martret, “Distributed power 

allocation with SINR constraints using trial and error learning,” in IEEE Wireless 

Communications and Networking Conference, WCNC, Paris, France, Apr. 2012. 

[3] J.S. Pang, G. Scutari, D.P. Palomar, and F. Facchinei, “Design of cognitive radio 

systems under temperature-interference constraints: A variational inequality approach,” 

IEEE Transaction on Signal Processing, vol. 58, no. 6, pp. 3251-3271, Jun. 2010. 

[4] V. Le Nir and B. Scheers, “Autonomous dynamic spectrum management for 

coexistence of multiple cognitive tactical radio networks,” in Proceedings of the Fifth 

International Conference on Cognitive Radio Oriented Wireless Networks 

Communications (CROWNCOM), Jun. 2010. 

[5] O. Popescu and C. Rose, “Water filling may not good neighbors make,” in IEEE 

Global Telecommunications Conference – GLOBECOM, San Francisco, CA, USA, 

Dec. 2003. 

[6] L. Rose, S.M. Perlaza, and M. Debbah, “On the Nash equilibria in decentralized 

parallel interference channels,” in IEEE Workshop on Game Theory and Resource 

Allocation for 4G, Kyoto, Japan, Jun. 2011. 

[7] T. Rappaport, Wireless Communications: Principles and Practice, 2nd ed. Upper 

Saddle River, NJ, USA: Prentice Hall PTR, 2001. 

[8] D. Fudenberg and J. Tirole, “Game theory,” MIT Press, 1991. 

[9] L. Rose, S.M. Perlaza, M. Debbah, and C.L. Martret, “Achieving Pareto optimal 

equilibria in energy efficient clustered ad hoc networks,” in Military Communication 

Conference, Milcom, Orlando, FL, USA, 2012. 

[10] H.P. Young, “Learning by trial and error,” University of Oxford, Department 

of Economics, Economics Series Working Papers 384, 2008. 

[11] B.S. Pradelski and H.P. Young, “Efficiency and equilibrium in trial and error 

learning,” University of Oxford, Department of Economics, Economics Series Working 

Papers 480, 2010.

High Spatial-Reuse Distributed Slot Assignment 

Protocol for Wireless Ad-hoc Networks 

Muhammad Hafeez Chaudhary 1 , Bart Scheers 2 

1 Royal Military Academy, Belgium, mh.chaudhary@rma.ac.be 

2 Royal Military Academy, Belgium, bart.scheers@rma.ac.be 

Abstract: Application of ad hoc networks in mission-critical environments requires wireless connectivity 

that meets certain quality-of-service (QoS). In such networks mechanism to control access to 

the shared wireless channel is crucial to ensure efficient channel utilization and to provide the QoS. 

TDMA based MAC protocols are considered to be appropriate for this kind of applications; however, 

finding an efficient and distributed slot assignment protocol is crucial. In this paper, a distributed 

slot assignment protocol is developed which gives high spatial reuse of the channel. To assign slots, 

the protocol does not need global topology information: Each node assigns slots based on the local 

topology information. The protocol can find the conflict-free slot assignment with limited message 

overhead. We evaluate the performance of the proposed protocol and show that the protocol gives 

considerably better channel utilization efficiency than exiting distributed slot assignment protocols. 


The last decade has seen an explosive growth in applications of wireless communications 

and networking technology bringing ubiquitous mobile service into 

the everyday realm. A recent forecast by CISCO suggests that during the current 

year there will be more wirelessly connected devices than the total human population. 

Moreover, the demand for high-speed wireless data transfer is increasing at 

astronomical rates. The phenomenal increase in the wirelessly connected devices 

and the applications running on them have put an extremely high premium on 

the communications spectrum, and thus placing great demand on designing spectrum 

efficient communication and networking protocols to meet the requirements 

of the current and emerging applications in wireless networking. Currently, a key 

area of research is mobile ad hoc networking, which is driven by the requirement 

of having a technology that enables a disparate set of mobile devices/nodes create 

a network on demand, as the need arises, to accomplish an assigned mission. 

In a wireless network, simultaneous transmissions of two or more nodes 

in the same channel may not be successful if their intended receivers are in the radio 

interference range of more than one transmitter. A mechanism to control access to


the shared wireless channel is crucial to ensure efficient channel utilization and to 

provide quality-of-service (QoS). Ad hoc networks for mission-critical applications 

and emergency response services require that the data be delivered to the destination 

node reliably and within certain time limits. To support such QoS, schedule based 

medium access control protocols using TDMA scheme are deemed suitable [1]. 

There are numerous algorithms dealing with TDMA slot scheduling in ad hoc 

networks, whereby nodes can find conflict-free slot assignment. In [1], a TDMA 

slot scheduling protocol named GinMAC is presented; the protocol requires 

the network be arranged in a tree-like structure. Building on the GinMAC, in [2] 

the BurstProbe slot assignment protocol is proposed. These slot scheduling protocols 

require global topology information which may not be available in ad hoc 

networks that are inherently bereft of any central coordinating and controlling 

infrastructure. In [3, 4] a slot assignment protocol called USAP is proposed which 

allows nodes to get conflict-free slot assignment in a distributed way using local 

topology information. Kanzaki and his colleagues proposed a protocol named 

ASAP in [5, 6] which can be viewed as an extension of the USAP, adding details 

on dynamic frame-size selection and more detailed procedures about the nodes 

joining/leaving the network. Another related protocol named DRAND is proposed 

in [7], and later on extended to Z-MAC [8] which combines artifacts of both TDMA 

and CSMA medium access schemes. The main focus of these protocols (and others 

like in [9, 10]) is to find conflict-free slot assignment to nodes in a distributed 

way. The maximization of the channel utilization efficiency (i.e., the spatial reuse 

of the slots) is not explicitly considered. Thus from channel utilization viewpoint, 

these protocols may give suboptimal performance. 

In this paper, we propose a high spatial-reuse distributed slot assignment protocol 

(HUDSAP). In the protocol, the nodes find slot assignment using their local 

topology information. The protocol introduces a priority mechanism by which 

nodes having higher number of one-hop neighbors (NoNs) assign slots first. Each 

node computes its priority index independently using only the information form 

the nodes within its contention zone. We show that the protocol achieves substantially 

higher channel utilization efficiency than the DRAND and related protocols. 

The remainder of the paper is organized as follows: Section II gives preliminaries 

on slot assignment and the problem formulation; Section III presents details 

of the proposed slot assignment protocol; Section IV outlines an adaptive frame 

length selection scheme; Section V illustrates the performance of the protocol by 

simulation examples; and finally Section VI gives some concluding remarks. 

II. Preliminaries and problem formulation 

For the slot assignment we represent the network by a graph G =(ϑ, ξ), where 

ϑ is the set of vertices that correspond to the nodes in the network and ξ is the set 

of edges representing the wireless links between the nodes. We assume that for any


249 

two distinct vertices i, j Î ϑ, an edge (i, j) exits in ξ if and only if i and j can hear 

each other – that is, all edges are bidirectional. 

Given a graphical representation of the network, the TDMA slot assignment 

problem with the associated assignment constraints can be defined as an equivalent 

graph coloring optimization problem. The equivalence between the two problems 

is one-to-one; that is, two nodes receive different slots if and only if the corresponding 

vertices have received different colors. Moreover, the total number of slots is equal 

to the total number of colors used. 

The objective of the slot assignment optimization problem is to minimize 

the number of slots used to find the transmission schedule for all nodes in the network 

under the constraint that a node can only assign a slot that has not been used 

within the contention zone of the node. The contention zone of a node is assumed to 

be limited to its two-hop neighbors. As is often assumed in designing MAC protocols, 

such a definition of the contention zone is imposed to remove the hidden-terminal 

problem. The hidden-terminal problem arises when two nodes cannot hear transmissions 

of each other but a third node can hear transmissions of both of them. 

An optimal solution for slot scheduling problem is known to be NP-complete 

[11, 12]: That is, the computational complexity of finding the solution increases 

exponentially as the number of nodes increases; which means it becomes prohibitively 

time consuming to find the optimal slot schedule. That is why to solve the slot 

assignment problem, in literature heuristic-based suboptimal solutions are proposed 

that vary in their schedule length, the convergence time, and the message overhead. 

To this end, in [13], three greedy heuristic-based slot assignment procedures are 

proposed: namely, the RAND (random), the MNF (minimum neighbor first), 

and the PMNF (progressive minimum neighbor first), listed in increasing order 

of complexity. The basic principle underlying each of these schemes is essentially 

the following: first, give a unique label to each node, and then assign slots to nodes 

in decreasing order of their labels. In RAND, the nodes are labeled in a random 

way; in MNF, the node with minimum number of neighbors is labeled first; and 

in PMNF, the nodes are labeled as in MNF with a difference that after labeling 

a node, the node and its edges are removed. Effectively that means, at each step 

among the nodes that have not been assigned slot yet, the RAND takes a node at 

random and allots time slot to it; the MNF takes the node with maximum number 

of neighbors and allots slot to it; and the PMNF first removes the nodes and 

the associated edges that have already been assigned slots, then within the updated 

network assign slot to the node with maximum number of neighbors. 

It has been shown in [13] that the schedule length (SL) achieved with the three 

schemes is in the order SL PMNF ≤ SL MNF ≤ SL RAND . That is, from the spatial reuse 

point of view, the PMNF is the most efficient and the RAND the least efficient 

among the three labeling schemes. However, the problem with these schemes 

is that they require knowledge of the global network topology; that is, they are 

centralized schemes. For ad hoc networks, distributed slot assignment schemes


are sought because such networks are devoid of any central coordinating and 

controlling infrastructure and the global topology knowledge is hard to come by 

at individual nodes. 

Recently a distributed implementation of RAND, known as DRAND, is proposed 

in [7] which can achieve the same channel utilization efficiency as the RAND. 

The RAND ordering can be viewed as archetype of the channel scheduling in wireless 

networks. The roots of many heuristics for slot scheduling algorithms in the literature 

can be found to be equivalent to the RAND [3, 5, 7, 8, 14]. The appeal 

of RAND lies in its simplicity and the ease with which its distributed version 

can be implemented. However, as shown in [13], the performance (in terms of SL) 

of the RAND can be significantly inferior to the MNF and PMNF. In this work, we 

present a distributed implementation of MNF, which we call as the HUDSAP that 

can achieve same channel utilization efficiency as the MNF by using only the local 

topology knowledge at individual nodes. Towards this end, the ensuing section 

gives details of the proposed protocol. 

III. Slot assignment protocol 

We assume that the time is divided into slots and the nodes are synchronized 

on the slot boundaries. The protocol operates in two main phases: the neighborhood 

discovery phase and the slot assignment phase. Fig. 1 shows the state diagram 

of the protocol, where the slot assignments phase is divided into four states: node classification, 

waiting slot assignment, active slot assignment, and completed slot assignment. 

In the neighborhood discovery phase, a node collects information about the nodes 

within its two-hop neighborhood, that is, one-hop neighbors (ONs) and two-hop 

neighbors (TNs), the NoNs of these nodes, and their assigned slots. The two-hop 

neighbors are strict two-hop neighbors, that is, it excludes the one-hop neighbors 

that can also be reached by another one-hop node. Based on that, the node constructs 

neighbor information table (NiT), an example of which is shown in Fig. 2. In the slot 

assignment phase, the nodes assign slots in a distributed way as explained next. 

Each node compares its NoNs with the NoNs of the nodes, in its NiT, which 

have not yet assigned slots, and based on that, the node classifies itself in one 

of the following three groups: 

1. In node group I (NG-I) if the NoNs of the node is greater than the NoNs 

of all nodes in its NiT that have not yet assigned slots; for instance, in Fig. 2 

initially nodes e and p will place themselves in this group. 

2. In NG-II if the NoNs of the node is equal to the NoNs of some (one or 

more) nodes in its NiT that have not yet assigned slots; for example, in Fig. 2, 

node b and l will initially place themselves in this node group. 

3. In NG-III if the NoNs of the node is less than the NoNs of some (one or more) 

nodes in its NiT that have not yet assigned slots; for instance, in Fig. 2, all nodes 

will initially place themselves in this node group except nodes b, e, l, and p.


251 

Figure 1. State transition diagram of the HUDSAP 

The classification at each node is done independently solely based on the local 

topology information available at the node in the form of NiT. After the classification, 

the node assigns slot to itself and the procedure of which depends on 

its node group. The slot assignment is managed differently in the three groups, 

as shall be explained in the ensuing sections. During the slot assignment phase, 

following control messages are exchanged between the nodes: slot assignment 

request (SAR), slot assignment grant (SAG), slot assignment confirmation (SAC), 

slot assignment denial (SAD), and slot assignment failure (SAF). The SAR, SAF, 

and SAC messages are transmitted by the node that attempts to assign a slot, and 

the SAG and SAD are response messages to the SAR message. These response 

messages are transmitted by the nodes within the contention area of the node that 

has sent out the SAR message. 

A. Slot assignment in NG-I 

All nodes in this group assign the first free slot that has not yet been assigned 

to any node within their two-hop neighborhood. For the network of Fig. 2, initially 

the nodes e and p are in this group and there is no slot assigned to any node 

in the network; so both of these nodes assign slot 0 to themselves. After assigning 

the slot, the nodes announce their slot assignment to their neighbors by sending 

the SAC message as shown in Fig. 3. In this case the nodes do not have to wait for 

any confirmation from their one-hop and two-hop neighbors about their slot 

assignment, because the nodes in this group are sure that there is no other node 

within their contention area currently assigning slot until their slot assignment


is complete; the other nodes in the contention area are prohibited from initializing 

slot assignment by the slot assignment procedures for NG-II and NG-III, as we 

shall see in the ensuing sections. Each one-hop neighbor after receiving the SAC 

announcement does the following things: updates its slot assignment information, 

forwards the SAC to its ONs, removes the node which transmitted the announcement 

from its NiT for further consideration during the classification step, and 

changes its state accordingly as shown in Fig. 1. It is interesting to note that all nodes 

in this group can complete slot assignment in one time-slot, as no slot assignment 

permission is required from neighbors. 

Figure 2. Network topology with NoNs of each node within {.}. An example of NiT 

of node p is also shown


253 

Figure 3. Successful slot assignment to nodes e and p in NG-I 

Figure 4. Successful slot assignment to node b in NG-II 

Figure 5. Failed slot assignment to node b in NG-II 

B. Slot assignment in NG-II 

For slot assignment in NG-II, we propose two procedures: one based on 

an elaborate exchange of control messages, and the other based on a prioritization 

mechanism using node IDs.


1) Message exchange based slot assignment: 

Each node in this group has at least one or more nodes (that have not 

assigned slots yet) in its NiT with the same number of NoNs as that node itself. 

For node i, in this group, let N i be the number of nodes with the same 

NoNs as the node i that have not yet assigned slots. To assign slot, the node 

runs a lottery for which the probability of success is chosen as P (i) =1 / (N i +1). 

In case a node does not win the lottery, it will wait for certain time slots chosen 

at random before trying the lottery again. If the node wins the lottery, it assigns 

the minimum possible slot that has not been assigned to any node within its 

contention-zone and sends the SAR message to its neighbors. For the given network 

topology in Fig. 2, initially nodes b and l are in this group and the probability 

of winning the lottery for each of them is 1/3. The probability that only one of them 

wins the lottery in a given slot and thus avoid collision of their announcement 

messages is 2/9. 

For the slot assignment to be complete, the node has to wait for the SAG messages 

from nodes within its contention area. After the slot assignment is granted by 

the nodes in the contention area, the node i sends out a SAC message; an example 

of which is shown in Fig. 4. Similar to the NG-I, after receiving the confirmation 

message, the one- and two-hop nodes update their slot assignment information, and 

remove node i from their NiTs for further consideration in the node classification 

step. Interestingly, the node i does not have to wait for response messages from 

nodes that have NoNs different than the node itself as well as from nodes having 

same NoNs that have completed slot assignment. For example, if node i has only 

one node with the same NoNs, then a confirmation from only that node is needed: 

if that node is a one-hop neighbor then the slot assignment can be completed 

in three slots – one to send SAR, second to receive SAG, and third to send SAC; 

if that node is a two-hop neighbor, then can be done in five slots – two additional 

slots are used by the one-hop node to relay SAR/SAG messages; and in either 

case, a unicast addressing can be employed. This can reduce the traffic overhead 

substantially, and thereby reduces the chances of collision of the slot assignment 

control messages in the network. 

When a node receives a SAR message of a one-hop node, the receiving node 

performs one of the following tasks: 

I. If the receiving node and all of its ONs have NoNs which are different 

than the NoNs of the SAR transmitter, then the receiving node neither 

forwards the SAR to its ONs nor sends a reply message (i.e., SAG/SAD). 

The receiving node simply waits for the SAC/SAF message form the transmitter 

of SAR. 

II. In case the receiving node has same NoNs as the transmitter of SAR but 

the NoNs of all of its ONs are different, the receiver node sends a SAG 

message to the transmitter. In this case also, the receiver does not forward 

the SAR to its ONs.


255 

III. If the NoNs of the receiving node are different than the NoNs of the transmitter 

of the SAR message, but one (or more) of its ON(s) has (have) 

same NoNs, the receiver forwards the SAR to those ONs. In this case, 

the receiver has to relay back any reply message (i.e., SAG/SAD) from its 

ONs: The receiver node sends a SAG message if all of its ONs with same 

NoNs reply with SAG, and otherwise sends a SAD message. 

IV. If the receiving node and one (or more) of its on-hop nodes have same 

NoNs as the transmitter of the SAR message, the receiver node sends a SAD 

message to the transmitter if it has already sent a SAG for this particular 

slot to one of its ONs other than the transmitter of recently received SAR. 

Otherwise, the receiver first forwards the SAR to its ONs with the same 

NoNs as the transmitter of SAR and waits for their reply. Once the receiver 

receives responses of those nodes, it fuses them and replies with a SAG 

message if all those nodes send SAG, else it replies with a SAD message. 

When a node receives a forwarded SAR message of a two-hop node, the receiving 

node performs one of the following tasks: 

I. If the NoNs of the receiver are different than the NoNs of the originator 

of the SAR message, the receiver discards the SAR message and does 

nothing else. 

II. In case the NoNs of the receiver are same as the NoNs of the originator 

of the SAR message, the receiver replies with a SAD message if it has already 

sent its own SAR message to its neighbors for the same slot, otherwise 

it replies with a SAG message. 

When a node receives a SAF message originated from a one-hop node it executes 

one of the following tasks: 

I. If the NoNs of the receiver and all of its ONs are different than the transmitter 

of the SAF message, the receiver discards the message. 

II. In case the NoNs of the receiver are different than the NoNs of the transmitter 

but some (one or more) of its ONs have same NoNs as the transmitter, 

then the receiver forwards the SAF to those ONs. 

III. If the NoNs of the receiver are same as the transmitter, but the NoNs 

of all of its ONs are different than the transmitter, then the receiver frees 

the slot for which it has earlier sent the SAG message to the transmitter 

of SAF. After which, the receiver discards the SAF. 

IV. In case the receiver and some (one or more) of its ONs have same NoNs 

as the transmitter of SAF, the receiver releases the slot and forwards 

the SAF to those ONs having same NoNs. 

When a node receives SAF originated from a two-hop node, the receiver 

discards the SAF if its NoNs are different than the transmitter of the SAF, else 

it releases the slot related to the SAF. 

When a node receives a SAC message it updates its NiT; if the SAC is originated 

from a one-hop node, then the receiver also forwards the SAC to its ONs.


Remark 1: The slot assignment in NG-II by the preceding dialog based mechanism 

though requires message exchange between a subset of nodes in the contention 

area of a node, however, despite this the control message overhead could be high. 

For instance, if multiple nodes in a contention area try to assign slots at the same 

time, then none of them will succeed and they have to try again after a random 

back-off time, which would slow down the convergence of the algorithm. In this 

regard, next we present an alternative slot assignment procedure for NG-II which 

avoids the preceding detailed message exchange routine when assigning slots to 

the nodes. 

Figure 6. For linear network topology all nodes except the two nodes at the extremities are in NG-II. 

When the number of nodes is large and the nodes in the network are in increasing (or decreasing) 

order of their IDs, for slot assignment, the nodes on the right (left) edge have to wait until all nodes 

on their left (right) side have completed slot assignment 

2) Alternative priority based slot assignment procedure: 

When all nodes have unique IDs, the slot assignment in NG-II can be handled 

in a much simpler way, very much like in the NG-I. Let there be a one-to-one 

function Ψ i which could map a node ID i to a unique numeric number µ i , that is, 

Ψ i : i → µ i , µ i ≠ µ j , i, j Î J. (1) 

Now any node i in NG-II, instead of randomly deciding about when to initiate 

slot assignment, compares its ID µ i with the IDs of the nodes (that have not yet 

assigned slots) having same NoNs in its NiT. If its ID is greater than these nodes, 

it assigns the minimum possible slot(s) to itself which is(are) not yet taken by 

the nodes in its contention area; otherwise, the node does not try to assign slot(s) 

unless the preceding condition is true. After assigning the slot(s), the node sends 

out the SAC message to neighboring nodes. Note that, like the nodes in NG-I, for 

slot assignment by this procedure the nodes in NG-II are not required to exchange 

messages SAR/SAG/SAD/SAF. 

Remark 2: The control message overhead of the alternative priority based slot 

assignment procedure is quite low compared to the message exchange based procedure. 

However, under the alternative slot assignment procedure, in certain cases, 

some nodes in NG-II may have to wait inordinate amount of time for slot assignment, 

for example, as shown in Fig. 6, the nodes in NG-II on the right (left) hand 

side have to wait until all other nodes in the group have completed their slot assignment. 

In such scenarios where a node in NG-II has to wait excessively long time to 

initiate slot assignment, the message exchange based slot assignment mechanism 

could be employed. To be more specific, if a node in NG-II does not receive new 

slot assignment information, during a predefined time duration, about any node


257 

within its contention area having the same NoNs as the given node then the node 

initiates slot assignment according to the message exchange based procedure. 

In this regard, to avoid more than two nodes to initiate slot assignment at the same 

time, after expiration of the predefined time duration, the nodes employ random 

back-off mechanism whereby a node waits for randomly chosen time-slots before 

starting slot allocation. If during this time, the node receives new slot assignment 

information, it will abort the message exchange based slot assignment procedure 

and will revert to the priority based slot assignment procedure. 

C. Slot assignment in NG-III 

The nodes in this group do not try to assign slots to themselves. They listen to 

the messages from their neighbors, assist in slot assignment of their neighbors by 

forwarding slot assignment control messages as discussed in the preceding sections, 

and update their slot assignment information. When a node, in this group, receives 

a SAC message about any node in its contention area, it removes that node from 

its NiT for further consideration in the node classification step. 

Each node, in NG-II and NG-III, that has not yet assigned a slot to itself, 

whenever updates its slot assignment information and removes any node from 

consideration in its NiT, it reruns the node classification test shown in Fig. 1. Note 

that after the classification test, a node in NG-II that has not yet assigned a slot to 

itself may find itself in NG-I, and a node in NG-III may find itself in either NG-II 

or NG-I (cannot be in both because the groups are mutually exclusive). For example, 

when nodes e and p (which were in NG-I) finish slot assignment, their neighbors 

d, j, i, and o (which were in NG-III) reclassify themselves in NG-II. Each node 

handles the slot assignment according to the procedure specific to its current node 

group. It is interesting to note that the nodes can only upgrade their groups and 

thus the slot assignment protocol is bound to converge within limited time; that is, 

the slot assignment to all nodes will be completed within a bounded time. We will 

analyze the convergence time and the message complexity of the protocol in more 

details in our future work. 

Proposition: The execution of the HUDSAP produces conflict-free slot assignment 

schedule. 

Proof: To show that the slot assignment schedule produced by the HUDSAP 

is conflict-free, it suffices to note the following: 1) At any given time only nodes 

in NG-I and NG-II are assigning slots and the two groups are mutually exclusive; 2) 

By definition, all neighboring nodes of each node in NG-I do not assign slots until 

slot assignment is completed for the nodes in NG-I; and 3) Each node in NG-II 

assign slot which is not assigned to any other node in its neighborhood by exchanging 

the control messages, or by the prioritization mechanism.


IV. Frame length selection 

With uniform frame size across the network, although the design and implementation 

of the MAC protocol will be simplified, however, the channel resources 

will remain underutilized. In the conventional TDMA slot assignment protocols, 

frame length is fixed based on the maximum expected number of nodes in the network, 

for instance, to ensure that each node gets at least one slot [8]. These protocols 

show poor channel utilization as they must leave enough unused slots for new 

coming nodes. Another possibility is to dynamically set the frame length for each 

node which is equal to the maximum slot number (MSN) assigned within the network. 

However, this would require each node to know the MSN. The propagation 

of the MSN within the entire network would not be adaptive to the local slot assignment 

changes – any change in the slot assignment may change the MSN and 

the new value has to be propagated throughout the network. Although by setting 

the network-wide same frame length based on the MSN effectively removes the requirement 

of a priori fixing the number of slots in a frame, however, it would still 

give lower channel utilization efficiency. 

The channel utilization can be improved by variable frame length for each 

node depending on the slot assignment in its neighborhood, that is, to change 

the frame length dynamically according to the slot assignment to the nodes within its 

contention area. If the contention area of a node is limited to its two-hop neighborhood 

and reuse of slots is allowed outside this area, then each node can set its 

frame length which is a function of the MSN assigned within the contention area 

(instead of the entire network). Note that, the MSN allocated within the contention 

area cannot exceed the two-hop neighborhood size of the node. To have conflictfree 

transmission among nodes with different frame lengths, usually the lengths 

of frames are chosen as multiple of two [4-8]. 

To set the frame length L i of node i according to the local MSN, the Z-MAC 

protocol in [8] proposed the following rule: 

L i = 2 k , (2) 

where k is a non-negative integer. The value of k is selected such that the following 

holds: 

2 k–1 ≤ A i ≤ 2 k –1, (3) 

where A i is the MSN within the two-hop contention area of the node i. This scheme 

although could achieve better channel reuse than the uniform frame length rule 

across the entire network, however this is not optimal from the point of view of channel 

utilization efficiency. In this regard, we propose an alternative scheme by which 

local framing rule varies depending on the node connectivity. Specifically, we classify 

nodes in two groups: leaf nodes and non-leaf nodes. For leaf nodes – a leaf node 

is a node which has only single one-hop neighbor; otherwise the node is a non-leaf 

node – the frame length is selected as in the Z-MAC. For non-leaf nodes, the framing


259 

rule is modified as follows. Let Ã i be the MSN within the one-hop neighborhood 

of the non-leaf node i. The frame length is set as in (2) and (3) with A i replaced by Ã i . 

An example of the heterogeneous frame length (Z-MAC and proposed) and 

the uniform frame length across the network is given in Fig. 7. From the figure, 

we can observe increase in the channel reuse due to the variable length frame size. 

The increase in channel reuse directly translates into higher channel utilization 

efficiency as well decrease in the data transfer delays. 

Figure 7. An example of variable and fixed length TDMA frames for a given conflict-free slot assignment: 

In the case of uniform length frame, all nodes have 8-slot frame length; in the Z-MAC 

variable length frame strategy, node a has 4-slot frame length whereas all other nodes have 8-slot 

frame length; and in the HUDSAP variable length frame strategy, nodes a and b have 4-slot frame 

length whereas all other nodes have 8-slot frame length 

Once the nodes have decided their frame lengths, the information are exchanged 

with nodes within their respective contention areas. After which the nodes 

can start data transmission within their assigned time-slots. The Z-MAC framing rule


produces slot assignment which is always conflict-free. However, in the proposed 

framing scheme, there could be occasional conflicts in assigned slots. Such conflicts 

can be detected by the nodes once frame-length information is available. When 

a node detects a slot conflict due to frame size selection, it compares its assigned 

slot with the slot assigned to the node causing the conflict. If the slot of the given 

node is less than the conflicting node, then the given node selects its frame length 

according to the MSN within its two-hop area (instead of one-hop area); otherwise, 

it leaves the conflict resolution to the conflicting node which would use the same 

procedure to resolve the conflict. 

V. Simulation examples and discussion 

In this section, with some numerical experiments, we evaluate channel utilization 

efficiency (i.e., the spatial reuse of slots) of the proposed HUDSAP and compare 

it with the centralized schemes RAND and MNF of [13]. Note that the DRAND 

is proven to achieve the same channel utilization efficiency as RAND [7]; so the comparison 

of HUDSAP with DRAND is not performed here. For a given network, 

assuming uniform frame length, the channel utilization efficiency is measured 

in terms of the minimum number of slots used by the protocol to find the conflictfree 

slot assignment schedule. We deploy the nodes in a 400-by-400 planar region. 

We conduct two numerical experiments: In the first, we fix the transmission range 

of nodes to 40 and vary the number of nodes from 50 to 400; in the second, we fix 

the number of nodes to 300 and vary the transmission range from 10 to 50. Each 

point in the numerical results is obtained by averaging over 10 4 random deployments 

(uniform distribution) of the nodes in the region. 

The results are plotted in Fig. 8 and Fig. 9. The figures show that the HUDSAP 

gives conflict-free slot assignment schedule that requires substantially less number 

of slots than the RAND (and consequently of the DRAND). That means, the spatial 

reuse of the slots, and consequently the channel utilization efficiency, is higher 

in HUDSAP. It should be noted that we deployed the node in a planar region of fixed 

area. Therefore, when either the number of nodes is small or the transmission range 

is small, or both, the performance gap is negligibly small. This is because, in such 

scenarios, the network is divided into small sub-networks (each comprising a few 

nodes) that are disconnected from each other – here we do not impose the condition 

that the network is connected. The performance of slot allocation protocols 

from spatial reuse point of view in such small networks does not differ much. 

However, if we impose the condition that the network is always connected, that is, 

any node can be reached from any other node (via multi-hops), then there would 

be a noticeable performance gap between the HUDSAP and the RAND even for 

a network comprising 50 sensors or less, which we observed in simulations that are 

not included here due to space constraints. Regarding the topological information, 

in DRAND each node requires knowledge about the identities of the nodes within its


261 

two-hop contention area and their slot assignment. Compared to that, in HUDSAP, 

each node also requires knowledge about the NoNs of the two-hop neighbors. Note 

that, although the NoNs of the one-hop neighbors is available in DRAND but that 

knowledge is not employed in slot scheduling. 

Figure 8. Channel utilization efficiency comparison for fixed number of nodes 

Figure 9. Channel utilization efficiency comparison for fixed transmission range 

For MNF two labeling schemes are considered, one based on the the NoNs 

and the other based on the number of two-hop neighbors (NtN) which are, respectively, 

denoted as MNF-NoNs and MNF-NtNs. The figures show that there is no 

appreciable gap between the schedule length of the HUDSAP and the MNF-NoNs.


However, there is a marginal gap between the performance of the HUDSAP and 

the MNF-NtNs. The reason for this performance gap is that the HUDSAP is designed 

for ordering based on NoNs whereas the MNF-NtNs is based on the NtNs. 

Even though the performance gap is not substantial, it is straightforward to extend 

the HUDSAP where prioritization of slot assignment is based on the NtNs, 

in which case it is expected that the performance of the HUDSAP and the MNF- 

NtNs will converge. However, that will require each node to know the NtNs of all 

nodes within its contention area, which would entail additional protocol overhead. 

Given that there is a marginal gain in channel utilization efficiency, the additional 

overhead may not justify the gain. Besides, it should be noted that the HUDSAP 

is a distributed slot assignment protocol which relies on local topology knowledge 

at each node, whereas the MNF is a centralized protocol which requires global 

topology information. 

Next we evaluate the impact of variable frame-length selection on the channel 

utilization efficiency. In this regard, for the preceding two network deployment 

scenarios, we compare the number of additional packets that can be 

transmitted within the frame length of Z-MAC when the frame-length is selected 

according to the proposed framing scheme. In Fig. 10 and Fig. 11 we 

plot the additional packets: average over 10 4 random deployments of the nodes 

and the maximum over the deployments. The figures show that with the proposed 

framing scheme, we could transmit more packets, which when seen for 

the network use over considerably longer time window could translate into 

substantially higher throughput. 

Figure 10. Number of additional packets (maximum and average over deployments of nodes) 

that can be transmitted within the frame length of Z-MAC under the proposed adaptive frame size 

selection scheme. Comparison is for fixed transmission range


263 

Figure 11. Number of additional packets (maximum and average over deployments of nodes) that 

can be transmitted within the frame length of Z-MAC under the proposed adaptive frame size 

selection scheme. Comparison is for fixed number of nodes 

VI. Concluding remarks 

In this work, we proposed a distributed TDMA slot assignment protocol to 

schedule access of the wireless nodes to the shared wireless channel. The nodes, 

by this protocol, can find conflict-free slot assignment using only the local topology 

information. The proposed medium access scheduling protocol is suitable for 

wireless ad hoc networks for mission-critical applications and emergency response 

services which require wireless connectivity be provided that meet certain QoS 

requirements. We showed that the protocol can achieve higher channel utilization 

efficiency compared to the existing protocols like RAND and DRAND. We also 

proposed an adaptive frame size selection scheme which gives higher channel 

utilization than the framing scheme proposed in Z-MAC protocol. 

In this work we give a qualitative description of our protocol and substantiate 

it with performance evaluation with a few numerical examples. In our planned ongoing 

work, to better understand the behavior of the protocol, we would do further 

analysis based on analytical modeling and event driven simulations. We plan to 

compare the convergence time and the control message overhead of the protocol. 

We also plan to study the behavior of the algorithm for dynamic topology changes 

due to the nodes joining and leaving the network, and the movement of the nodes 

within the network.


References 

[1] P. Suriyachai, J. Brown, and U. Roedig, “Time-critical data delivery in wireless 

sensor networks,” Proc. of 6th IEEE Int. Conf. on Distributed Computing in Sensor 

Systems (DOCSS’10), pp. 216-229, June 2010. 

[2] J. Brown et al., “BurstProbe: debugging time-critical data delivery in WSNs,” Proc. 

of EU. Conf. on WSNs, pp. 195-210, Feb. 2011. 

[3] C.D. Young, “USAP: A unifying dynamic distributed multichannel TDMA slot 

assignment protocol,” Proc. of IEEE MILCOM, pp. 235-239, Oct. 1996. 

[4] C.D. Young, “USAP multiple access: dynamic resource allocation for mobile multihop 

multichannel wireless networking,” Proc. of IEEE MILCOM, pp. 271-275, Nov. 1996. 

[5] A. Kanzaki, T. Uemukai, T. Hara, and S. Nishio, “Dynamic TDMA slot assignment 

in ad hoc networks,” Proc. of 17th Int. Conf. on Advanced Inform. Net. and Applications 

(AINA ’03), pp. 330-335, Mar. 2003. 

[6] A. Kanzaki, T. Hara, and S. Nishio, “An efficient TDMA slot assignment protocol 

in mobile ad hoc networks,” Proc. of ACM Symp. on Applied Computing (SAC ’07), 

pp. 891-895, Mar. 2007. 

[7] I. Rhee, A. Warrier, J. Min, and L. Xu, “DRAND: Distributed randomized TDMA 

scheduling for wireless ad hoc networks,” IEEE Trans. on Mobile Computing, vol. 8, 

no. 10, pp. 1384-1396, Oct. 2009. 

[8] I. Rhee et al., “Z-MAC: A hybrid MAC for wireless sensor networks,” IEEE/ACM 

Trans. on Net., vol. 16, no. 3, pp. 511-524, June 2008. 

[9] L. Bao, and J.J. Garcia-Luna-Aceves, “A new approach to channel access ccheduling 

for ad hoc networks,” Proc. of 7th ACM Int. Conf. on Mobile Comp. and Net. 

(SIGMOBILE’01), pp. 210-221, Jul. 2001. 

[10] A. Rao, and I. Stoica, “An overlay MAC layer for 802.11 networks,” Proc. of ACM 

Int. Conf. on Mobile Sys., Applications, and Services, pp. 135-148, 2005. 

[11] E.L. Lloyd, and S. Ramanathan, “On the complexity of link scheduling in multihop 

radio networks,” Proc. of 26th Conf. on Inform. Science and System, 1992. 

[12] S. Even et al., “On the NP-completeness of certain network testing problems,” 

Jr. on Networks, vol. 14, no. 1, pp. 1-24, 1984. 

[13] S. Ramanathan, “A unified framework and algorithm for (T/F/C)DMA channel 

assignment in wireless networks,” Proc. of 16th IEEE INFOCOM, pp. 900-907, 

Apr. 1997. 

[14] A. Ephremedis, and T. Truong, “Scheduling broadcasts in multihop radio networks,” 

IEEE Trans. on Commun., vol. COM-38, pp. 456-460, Apr. 1990.

Hybrid Network Synchronization for MANETs 

Harri Saarnisaari, Teemu Vanninen 

Centre for Wireless Communications, University of Oulu, Oulu, Finland 

harri.saarnisaari@ee.oulu.fi, teemu.vanninen@ee.oulu.fi 

Abstract: Military radio networks such as mobile ad hoc networks (MANETs) usually rely on global 

satellite navigation systems (GNSSs) for network time synchronization. However, in reality all nodes 

may not have a GNSS timing device or they do not have a direct link to a GNSS timed node. In addition, 

in some situations GNSSs may fail. Therefore, additional supporting mechanisms are needed 

for synchronization. This paper provides a hybrid algorithm that uses GNSS timed nodes as master 

time servers if they are available but automatically turns to a distributed mode if GNSS time is not 

available providing robustness and survivability. Furthermore, the distributed mode can be used also 

as a stand-alone synchronization mechanism, i.e., a GNSS time reference is not necessarily needed 

at all. Simulations show the behavior of the algorithm in different scenarios but as main conclusions 

it can be said that the availability of GNSS timed nodes speeds up the initial convergence, the convergence 

rate depends on the number of nodes and the number of hops and that if GNSS time is lost, 

the network still maintains synchronism but time is not necessarily GNSS time. 

Keywords: component; network; synchronization; hybrid 


All the pictures are created by the authors 

Military mobile ad hoc networks (MANETs) usually use time division multiple 

access (TDMA) as a channel access protocol and often frequency hopping to 

provide robustness. These operations require time synchronized nodes. There are 

several possible strategies for network time synchronization [1]. One option is to 

have full autonomy, where the clocks function independently without affecting 

to each other. This option requires frequent calibrations since clocks tend to drift 

from each other. Precise clocks that provide autonomy for a period [2] or external 

precise time source such as global satellite navigation system (GNSS) are also possibilities 

herein. A second option is the (centralized) master-slave structure. This 

is a hierarchical system where the lower level nodes synchronize with the higher 

level nodes but not vice versa. A drawback of this method is that a fault in a master 

(or sub-master) node affects the whole (rest of the) network. This makes the network 

vulnerable. The advantages of this method are its simplicity and that clock quality 

requirements are reduced to the higher hierarchy levels, which lowers the costs. 

The third alternative is the mutual (distributed, decentralized) synchronization


in which the nodes synchronize themselves based on mutual cooperation without 

a master. Naturally, there are also hybrid strategies where a master (or a group 

of masters) leads the game but the rest of the nodes cooperate in a mutual fashion. 

Possibly, there is a hierarchical structure and cooperation exists within a hierarchy 

level. Furthermore, the master could be one that is not permanent but it can be 

replaced by another node in the case of failure or if it is not available for some 

other reason. 

Inside the strategies are protocols, which describe how timing messages are 

distributed in a network and what and how many messages are needed. Several 

protocols have been proposed for network time synchronization (NTS) of wireless 

sensor and ad-hoc networks. References [3-9] provide a good snapshot of these 

protocols. They follow the above mentioned general strategies in a way or another. 

Actions one needs are not usually concentrated in a single spot (system). Instead, 

one may use information coming from different systems all around the globe. 

In order to keep different systems on the same time base it is natural to assume, 

or require, that if GNSS time is available it has to be used since it is a provider 

of the Universal Coordinated Time (UTC). This paper considers a novel hybrid 

synchronization approach proposing an algorithm which uses a GNSS timed 

node or nodes as a master or masters if they are available. If those are not available 

it uses a distributed approach. This increases robustness since lost of masters 

does not destroy synchronization. Both use of GNSS time whenever available and 

robustness are features sought by military. It is also very possible that all nodes 

do not have a GNSS time source device or they are not direct neighbors of such 

a node. The proposed algorithm is able to synchronize nodes in a multi hop fashion. 

It uses nodes that have heard GNSS timed nodes or their nearest or further distant 

neighbors as masters if those are available. Furthermore, the algorithm distributes 

knowledge of existence of masters all over the network, and neglects this information 

if masters are lost. This information may be utilized in the merging case. The way 

how network time synchronization messages are delivered (which channel, how 

often, etc) is not proposed, neither there are special requirements on that meaning 

that the algorithm is very generic. Simulations demonstrate convergence speed for 

different network size, (normalized) network synchronization accuracy and effect 

of losing all the masters. 

II. Algorithm 

The principles of the proposed algorithm are simple. First, use GNSS time 

as a master time reference if it is available. Second, if GNSS time is not available 

in the network, use a distributed algorithm. A consequence of the first principle 

is that even if only one GNSS timed node is available, its existence must be distributed 

all over the network and its time must be used as a reference. In practice this 

is implemented by maintaining a heard master (HM) flag in the synchronization


267 

message. It is on if GNSS time is available and off otherwise. It is set on if a node 

receives message from a GNSS timed node (master) or from a node whose HM 

flag is on. A consequence of the second principle is that if a node is not a GNSS 

timed node or does not receive any time synchronization messages where the HM 

flag is on, it uses a distributed time synchronization algorithm. 

The state machine of the algorithm is shown in Fig. 1. The shown algorithm 

includes the integrity check not implemented in the simulation chain. The integrity 

check is essential to self-monitor the network for malfunctioning of GNSS timed 

nodes (or detect misleading GNSS timed nodes) and inform the operator if needed. 

Herein, this aspect was not considered and all GNSS timed nodes were assumed 

to perform properly, i.e., they all show the same time. Neither was inform-mycurrent-network 

feature implemented. That can ease and speed up merging cases 

that is discussed briefly later on. 

Figure 1. The state machine of the proposed network synchronization algorithm


A node initially listens a period for network synchronization messages 

(NETSYNC MSG). If it does not receive any it initiates its own message. Then it waits 

again. The waiting time may include some randomness around an initial value or 

in an interval. It may also be shorter in the initialization phase than in the communications 

phase. It may be shorter for GNSS timed nodes than for other nodes 

since that would favor GNSS timed nodes that are anyway used as master nodes. 

If it receives a message its behavior depends on the contents of the message and its 

own state. If the node is a GNSS timed node (master) it does not adjust itself. If it is 

non GNSS timed node its behavior depends on contents of the message as clearly 

described on the state machine. However, some principles may be unclear from 

the machine. Therefore, some aspects are considered more detailed in what follows 

in this section. 

Time tuning at the receiver is based on time difference of the local clock 

reading and received time readings [10]. Time tuning is done at the every adjustment 

instant, i.e., discretely at certain points, not continuously after receiving any 

single message. This is because of robustness. If all messages would be used for 

time tuning after they arrive it may result a ping pong effect in time or between 

master-slave and distributed mode. In addition, if a receiver could receive several 

messages it can select the best ones for its own purpose. For example, it receives 

a few messages from HM-flag-off nodes and then finally at the end of the listening 

period a message from a HM-flag-on node. Then, it can be use the best time 

source on that period, i.e., the HM-flag-on node and ignore the HM-flag-off nodes. 

Observed time difference is used as such if the message is from a master node 

(node with a GNSS time device) or from a node with the HM lag on. If several 

masters are observed during a period, only one is used. This could be the one 

with the smallest time since that is closest to the receiver and, as a consequence, 

the bias from uncompensated propagation delay would be the smallest. In the simulations 

the first master was selected for simplicity. If several HM-flag-on nodes are 

observed, average of their time difference to the local time is used to set the time. 

In both the cases the observed (and selected) time differences are used in the masterslave 

fashion, i.e., directly to set the time. In the distributed mode the average time 

difference is weighted by a coefficient (0.5 herein) and then added to the previous 

time [10]. 

An adjustment interval should be selected such that a new adjustment is done 

before clocks drift too far away from each other. Or, indeed, well before to leave 

some margin. What affects to this interval are the initial synchronization error after 

previous adjustment, discussed in the next subsection, and clock skew caused by 

frequency errors. For example, 1 ppm (part per million) clocks cause 1 µs error 

in every second. Since this may be in both directions, the total maximum error is 2 µs. 

One has to count how many seconds clocks could be without adjustment and select 

adjustment period based on this. At the initial phase the adjustment period, as well 

as listening period, could be shorter. In the communications phase, at least, these


269 

two coincide. It would also be advantageous if time information (NETSYNC MSG) 

could be transmitted more than once in the adjustment period since that increases 

robustness, e.g., since some messages may be lost. However, this may not be sensible 

since that increases overhead in the system. Therefore, final selection is a tradeoff 

between overhead and robustness. 

Time information is sent in a network synchronization message. The message 

includes a time stamp, a master flag and the HM flag information. The master flag 

informs that a node is a master (GNSS timed node). The synchronization messages 

are send in a control channel (either logical or physical), but in this paper we do 

not care how the control channel is actually implemented. It is just assumed that 

the network synchronization messages are send at certain intervals and that these 

messages can be heard by other nodes. This is a sensible assumption since otherwise 

synchronization is impossible. 

Possible difficult points in MANETs are late entry and merging. These should 

be made as automatic and fast as possible, i.e., the operator should not need to 

manage this and it should not take tens of minutes. As the network synchronization 

process is part of this, it should be fast too. It is obvious from the state machine that 

the GNSS timed node case is trivial. If two GNSS timed networks merge they are 

readily synchronized. If a non-GNSS timed node or network merges with a GNSS 

timed node or network it adapts GNSS time. However, if two non-GNSS timed 

networks merge the situation is more problematic since use of pure distributed 

algorithm would yield to very long convergence period. Therefore, the process 

should be speeded and that means, e.g., merging case detection and decision which 

of the networks has to change its time and which preserves its time. 

A. Performance of NTS 

It can be concluded from [10-13] that the accuracy of the algorithm depends 

on the accuracy of time delivery and biases caused by uncompensated delays and 

clock frequency offsets (skew). Since the beating frequency of clocks is not usually 

controlled, skew occurs in every adjustment period. This term is mitigated by 

properly chosen the adjustment interval. Contrary, propagation delay could be 

compensated. In master-slave networks bias by uncompensated delays cumulates 

on each hop the master’s message has to jump whereas in distributed networks 

this effect is somewhat averaged away. Therefore, delay compensation might be 

a critical issue especially if the network synchronization requirements are high 

and also if distances (measured in seconds) between nodes are large, quite close to 

the synchronization requirement. There are several means for the compensation. 

In the master-slave scheme master and slave may interact pairwise to measure 

the propagation delay. This means that interaction has to be repeated for each pair. 

An example of this is the IEEE 1588 precise timing protocol. In two way schemes 

nodes send measured time differences back to the origin which use them to mitigate


the delay. If feedback fails this kind of system reduces to uncompensated systems, 

i.e., it still works but has bias. The system is robust in the sense that it does not rely 

on delay compensation (assuming that provided synchronization accuracy is sufficient). 

Node positions, if they are known, can be used to calculate propagation delays 

and used for delay compensation. In this sense use of situation awareness picture 

(concerning node positions) could be a valuable tool in network synchronization. 

Note that the positioning accuracy does not necessarily need to be extremely accurate 

since 1 µs corresponds to 300 m. If one or few µs is tolerable compensation 

accuracy, then 300 m to 1 km positioning accuracy would be sufficient. 

III. Simulations 

Simulations were implemented using MATLAB. The simulation area is a X 

by X square with N randomly deployed nodes. The communication range is αX. 

There are M masters that could be switched off at the moment T. Results are shown 

as a function of adjustment intervals. All time errors and distances are normalized 

to time-of-arrival (TOA) measurement accuracy which is typically about one tenth 

of the inverse of the signal bandwidth or, equally, one tenth of the symbol duration. 

Therefore, 1 MHz signal provides (about) 0.1 µs TOA accuracy (or 30 m) and 5 MHz 

signal 0.02 µs (6 m) accuracy. With these numbers, if X is 500, it corresponds to 

15 000 m or 3000 m, respectively. Correspondingly one founds effects of skew and 

distances. One counts the effect of skew in TOA accuracy units during the adjustment 

period. So, if β is the skew and T adj the adjustment period, then the effect of skew 

is βT adj /σ, where σ is the accuracy. For example, if the skew is 1 ppm and T adj is 1 s 

or 100 s, then for 1 MHz signal this corresponds to 10 and 1000, respectively, and 

for 5 MHz signal these are five times larger. 

If required synchronization accuracy (worst case) in frequency hopping 

systems would be half the hop duration and hop rate would be 1000 hops/s, then 

the required worst case synchronization accuracy would be 5000 units or 25 000 

units, respectively using the above numerical values. This would be nice to remember 

when one interprets the simulation results. Even without reference to hopping systems 

one should calculate worst case accuracy related to delay estimation accuracy. 

Furthermore, we calculate clock time errors with respect a reference node since 

this is the meaningful way to do it [10]. If there are masters, a master is selected 

as a reference that is quite natural. In the other case node number 1 is selected. 

The former selection causes some harm when lost of GNSS nodes is investigated 

since error is still measured with respect to a master. However, most significant 

conclusions can be drawn even with this small inconvenience. Results are averaged 

over 100 runs with same node deployment. The initial time offsets, skews and TOA 

errors are altered. 

The first result shows ultimate accuracy limits since delays are perfectly compensated 

and skews are zero. There are 32 nodes including 5 masters and X = 500.


271 

The initial offset (of non-masters) is from range ±3×10 8 . The communication range 

is 250. A random deployment is shown in Fig. 2 and resulting time errors at Fig. 3. 

Since the communication range is long and masters quite uniformly distributed 

over the area, convergence is fast, at the first round, i.e., all nodes are direct neighbors 

of masters. The not shown root means square error (RMSE) values are equal 

to one, i.e., equal to the time delivery accuracy as it should be. Therefore, the time 

delivery accuracy is the ultimate limit. 

500 

Connectivity plot. Masters are red diamonds. 

450 

400 

350 

300 

250 

200 

150 

100 

50 

0 

0 100 200 300 400 500 

Figure 2. Example random node deployment into area and existing connections. 

X = 500, N = 32, M = 5, α = 0.5 

10 

8 

6 

4 

time error term 

2 

0 

−2 

−4 

−6 

−8 

−10 

0 5 10 15 20 

adjustment instants 

Figure 3. Ultimate timing errors and convergence speed. X = 500, N = 32, M = 5, α = 0.5 

The results in Fig. 4 show the same setup but with uncompensated delays. 

The maximum bias is close to the maximum communication range as well as is 

RMSE, which value, indeed, is the limit in the single hop case. In the multiple hops 

case uncompensated delays would cumulate.


If the skew is from interval ±500 (for non masters) the total bias still increases 

as shown in Fig. 5 as well as RMSE, that goes up to level 400. 

300 

250 

200 


150 

100 

50 

0 

0 5 10 15 20 


Figure 4. Timing errors when delays are uncompensated. X = 500, N = 32, M = 5, α = 0.5 

500 

450 

400 

350 


300 

250 

200 

150 

100 

50 

0 

0 5 10 15 20 


Figure 5. Timing errors when delays are uncompensated and skews are present. 

X = 500, N = 32, M = 5, α = 0.5 

Shorter communications ranges often yield to separated subnets and are thus 

meaningless to study in this paper where the focus is in the behavior of the whole 

network. To study this shorter communication range effect we increase the number 

of nodes to 128 and kept the rest parameters. The results are in Fig. 6. It can be seen 

that some nodes are one hop and some two hop neighbors of masters. Then the range 

was decreased to 100, i.e., one fifth of the area size. Although not easily seen from 

the Fig. 7, the convergence rate decreases to four rounds (but depends naturally 

on the largest number of hops between a master and a node). After convergence 

the error is increased. It is believed that this is due to averaging in the algorithm


273 

since a node computes an average based on all received HM-flag-on messages. This 

averaging increases robustness. It would be possible to modify the algorithm to accept 

only one HM-flag-on message. It is expected that if this selection could be done 

wisely, the error would be smaller. However, this feature was not tested. The total 

bias is also increased due to this averaging and accumulated skew and delay terms. 


X = 500, N = 128, M = 5, α = 0.5 


X = 500, N = 128, M = 5, α = 0.2 

The effect of losing the masters at the adjustment point 100 is investigated 

when X = 500, N = 32, M = 5, α = 0.5. The results in Fig. 8 show that after the lost 

the nodes' times diverge from zero (that is the masters' time). However, the nodes' 

times stay very close together and their times go to the same direction. Therefore, 

they remain synchronized, but that is what is expected. The novelty was to include


an automated step from the master-slave mode into the distributed mode (and 

back if needed). 

Finally, convergence speed and error without master nodes is investigated. 

The RMSE results in Fig. 9 show that convergence takes about 80 adjustment periods, 

i.e., much much more than in the master-slave mode. However, the timing 

error is at the same level. 

6000 

5000 

4000 


3000 

2000 

1000 

0 

0 50 100 150 200 


Figure 8. Timing errors when delays are uncompensated and skews are present and masters 

are lost at point 100. X = 500, N = 32, M = 5, α = 0.5 

1000 

900 

800 

rmse of time error wrt ref 

700 

600 

500 

400 

300 

200 

100 

0 

0 50 100 150 200 


Figure 9. RMSE timing errors when delays are uncompensated and skews are present without 

master nodes. X = 500, N = 32, M = 5, α = 0.5 

IV. Conclusions 

This paper proposed a novel hybrid master-slave distributed network time 

synchronization algorithm that is robust to lost of masters. The algorithm uses GNSS


275 

timed nodes as masters if they are available and is in a robust distributed mode otherwise. 

The algorithm has flexibility for further developments. The basic properties 

of the algorithm were demonstrated by simulations. The ultimate synchronization 

accuracy depends on the time delivery accuracy, but in real life uncompensated 

delays or errors in delay compensation and unequal clock frequencies cause larger 

time errors. The convergence speed in terms of required adjustments depends on 

the number of hops between masters and slaves. The convergence speed in distributed 

mode may be much slower than in the master-slave mode. In addition, it is 

slower for larger networks. Therefore, some kind of master-slave structure would 

be beneficial in distributed networks at least in the initial phase. In addition, the adjustment 

period could be shorter in the initialization phase as usual. 


This work was done for European Defence Agency (EDA) projects WOLF 

(Wireless Robust Link for Urban Force Operations) and ETARE (Enabling Technology 

for Advanced Radio in Europe). 

References 

[1] W.C. Lindsey, F. Ghazvinian, W.C. Hagman, and K. Dessouky, “Network 

synchronization,” Proceedings of the IEEE, vol. 73, no. 10, pp. 1445-1467, October 

1985. 

[2] H.A. Stover, “Network Timing/Synchronization for defence communications,” 

IEEE Transactions on Communications, vol. 28, no. 8, pp. 1234-1244, August 1980, 

in a special issue of network synchronization. 

[3] K. Römer, P. Blum, and L. Meier, “Time synchronization and calibration in wireless 

sensor networks,” in Handbook of Sensor Networks: Algorithms and Architectures, 

I. Stojmenovic, Ed. John Wiley & Sons, 2005. 

[4] B. Sundararaman, U. Buy, and A. Kshemkalyani, “Clock synchronization for 

wireless sensor networks: a survey,” Ad Hoc Networks, vol. 3, pp. 281-323, 2005. 

[5] C. Rentel, T. Kunz, “Network synchronization in wireless ad hoc networks,” Carleton 

University, Systems and Computer Engineering Department, Ottawa, Canada, 

Technical report SCE-04-08, 2004. 

[6] C. Rentel, T. Kunz, “A clock-sampling mutual network time-synchronization 

algorithm for wireless ad hoc networks,” in IEEE Wireless Communications and 

Networking Conference (WCNC), vol. 1, 2004, pp. 638-644. 

[7] Q. Li, D. Rus, “Global clock synchronization in sensor networks,” IEEE Transactions 

on Computers, vol. 55, no. 2, pp. 214-226, February 2006. 

[8] D. Mills, “Improved algorithms for synchronizing computer network clocks,” IEEE 

Transactions on Networking, vol. 3, no. 3, pp. 245-254, June 1995.


[9] E. Serpedin, Q.M. Chaudhari, Synchronization in Wireless Sensor Networks: 

Parameter Estimation, performance Benchmarks and Protocols. Cambridge University 

Press, 2009. 

[10] H. Saarnisaari, “Analysis of a discrete network synchronization algorithm,” 

in Proceedings of the IEEE Military Communications Conference, Atlantic City, NJ, 

USA, 2005. 

[11] A. Fasano, G. Scutari, “The effect of additive noise on consensus achievement 

in wireless sensor networks,” in Proceedings of the IEEE International Conference 

on Acoustics, Speech, and Signal Processing, April 2008, pp. 2277-2280. 

[12] A. Davies, “Discrete-time synchronization of communications networks having 

variable delays,” IEEE Transactions on Communications, vol. 23, no. 7, pp. 782-785, 

July 1975. 

[13] G. Scurati, S. Barbarossa, and L. Pescosolido, “Distributed decision through selfsynchronizing 

sensor networks in the precence of propagation delays and asymmetric 

channels,” IEEE Transactions on Signal Processing, vol. 56, no. 4, pp. 1667-1684, 

April 2008.

Application of Dezert-Smarandache Theory 

for Tactical MANET Security Enhancement 

Joanna Głowacka, Marek Amanowicz 

Faculty of Electronics, Military University of Technology, Warsaw, Poland, 

{jglowacka, mamanowicz}@wat.edu.pl 

Abstract: The article presents a concept of Dezert-Smarandache theory application for enhancing 

security in tactical mobile ad-hoc network. Tactical MANET, due to its specification, requires collection 

and processing of information from different sources of diverse security and trust metrics. 

The authors specify the needs for building a node’s situational awareness and identify data sources 

used for calculations of trust metrics. They provide some examples of related works and present their 

own conception of Dezert-Smarandache theory applicability for trust evaluation in mobile hostile 

environment. 

Keywords: situational awareness, trust, inference methods, tactical MANET, security, Dezert-Smarandache 

theory 


The mobile ad-hoc networks are collections of independent nodes that 

can communicate via radio channels. These networks are often developed in conditions 

of limited or total lack of access to fixed infrastructure. 

MANETs are characterized by high dynamic changes in the location of each 

node and the vulnerabilities of various types of attacks. Due to the open medium, 

ad-hoc networks are more susceptible to eavesdropping and data injections. A dynamic 

change of network topology contributes to the frequent connecting and 

disconnecting nodes, and no central network monitoring makes it difficult to 

detect malicious behaviour of nodes. In addition, the network resource limitations 

contribute to the selfish attacks. They are aimed at consuming a large amount 

of bandwidth. One of the selfish behaviours is the failure to transfer the packages 

by a node to conserve its own energy. 

Security ensuring is particularly difficult for a tactical ad-hoc network, due to 

the necessity of dealing with a hostile environment, strict capacity constraints, the re- 

This article was written as a part of a scientific research project financed by polish government budget for 

science in 2010-2013 “An advanced methods and techniques of traffic control in tactical ad-hoc networks” 

No. O N517 274 839.


quirements for services, very rapid changes of network topology and dynamically 

forming groups of common interests, which cannot be pre-defined by trust relationships 

[1]. These networks are characterized by simple capability of adding new nodes, 

which may be of diverse nature, such as the allies, neutral or hostile nodes. 

Figure 1. A sample mobile ad-hoc network structure 

One method of ensuring the security is user authentication. Only the authorized 

nodes and those verified as allies can have access to the network. However, 

during the mission, a node can be taken over by the enemy, or change the nature 

of its behaviour – behaving to the detriment of the mission. 

Due to the lack of a central management system it is needed for nodes to 

cooperate. Each of them is in fact a router ensuring cooperation between subnets 

and nodes located at a distance greater than the radio range. 

Restrictions on ad-hoc networks contribute to the need of using other means 

than in wired networks to satisfy the safety requirements. In addition to authorization 

and authentication mechanisms, it is necessary for a node to have the knowledge 

on the behaviour of other nodes in the network, determining safety routes for data 

transfer and knowledge concerning the reaction manners in certain situations. 

The situational awareness building method will be complement of standard security 

mechanisms in mobile ad-hoc networks. 

II. Node’s situational awareness 

A. Definitions 

To identify opportunities of secure cooperation between nodes in ad-hoc 

networks, it is necessary to collect information about other nodes in the network.


279 

The ability to have accurate information about the surrounding reality and interpretation 

of the current situation in terms of the performed tasks is defined as a node’s 

situational awareness. 

The main product of the node’s situational awareness mechanism is information 

on the node trust levels. 

Trust is an interdisciplinary concept, characterized by a variety of definitions. 

It is understood as relying on the integrity, strength and ability of a person or thing. 

In the case of ad-hoc network it is translated as a set of relationships between people 

who use similar communication protocols [1]. These relations are defined based on 

previous interactions of individuals. In [2], trust is treated as the degree of belief 

about the behaviour of other entities. Trust can also be understood as reputation, 

opinion, or the probability of correct behaviour [3]. 

In MANET, trust is the level of faith, which can be assigned by the node to 

its surroundings on the basis of observations and opinions coming from the other 

nodes in the network [4]. 

B. Benefits 

Building node’s awareness is essential to achieve the mission. In heterogeneous 

networks, the completion of the mission is dependent on the integrity of individuals. 

The knowledge gained from building node’s awareness can ensure cooperation 

only between trusted entities that do not behave suspiciously. 

Secure exchange of information between nodes requires proper selection 

of the route of data transfer. Sending data via routes that are not safe may contribute 

to the leak or acquisition of data by unauthorized persons. Lack of metrics 

allowing for choosing the path depending on the level of confidence in nodes and 

the risk, that exists in choosing the path of data transfer and cooperation between 

the nodes, may contribute to the failure of the operation. 

Figure 2. The possibility of application of the knowledge from building node’s awareness 

in different OSI model layers


The dynamic process of creating a current situational view of node can be 

the basis for decisions on how to control traffic. 

The knowledge about the surrounding environment gained through the mechanism 

of building node’s awareness can be applied in different layers in order to protect 

the communication between nodes. In the data link layer it can be used to define 

a parameter indicating the possibility of cooperation with the nodes or the need 

for failure to communicate with nodes characterizing a low level of confidence. 

In the third layer level of trust, it can be used as metric routing protocol that will 

allow you to safely share data. Specified nodes confidence level can be used also 

in the application layer, where the nodes of questionable confidence level will be 

forced to certain behaviour for performing its final assessment assignment. 

C. Data sources 

Node’s situational awareness in most cases is built based on direct interactions, 

indirect observations and recommendations. 

Trust determined by the node based on direct interaction and observation 

of behaviour of other nodes is called direct trust. 

Trust determined on the basis of indirect observations and recommendations 

is called indirect trust. Recommendations shall be understood as opinions of other 

nodes on the node for which the level of confidence is being specified. 

Figure 3. Direct and indirect trust 

In many cases information from various sources may be incomplete, inconsistent 

or conflicting. This requires the selection of appropriate methods of inference, 

which would allow clear and accurate assessment of the current environment 

in which network node operates. 

III. Related works 

The problem of gathering information about the surrounding node reality and 

determining the nodes trust in ad-hoc networks has recently been very popular and 

widely developed in the literature, which demonstrates the importance of this topic.


281 

Probabilistic inference is the most frequently used method in the literature to 

determine the node trust level. The information about node behaviour is evaluated 

as 0 or 1, match or success. 

In [5], the theoretical concept was presented to assess the level of trust and 

its propagation. Trust in this approach is considered as a measure of uncertainty 

expressed in a measure of entropy. In the case of entropy based trust model, promoted 

trust is calculated on the basis of individual trust values. In the trust model 

based on the probability, the value of propagated trust is calculated by the probability 

of trust relationships. In this model, the probability of correct relations 

assessment is also included. This concept also includes an assessment of trust on 

the basis of observation. 

In [6], new concept of the TMF (Trust Management Framework) was presented. 

TMF is used for nodes to obey protocol and cooperate with each other. 

There are two types of TMF: 

• reputation based – trust is assessed based on direct observations and information 

of the second hand, Bayesian approach based on the distribution 

of β is being used, 

• trust establishment – trust is assessed based on direct observation and 

the relationship established between nodes without regard to previous 

opinions of intermediate nodes. 

Both types of TMF are immune to numerous attacks, therefore, OTMF (Objective 

TMF) has been proposed to prevent them. This solution is based on modified 

Bayesian approach, in which different weights are assigned to different information 

given at the time of their occurrence and that concerning their supplier. Influence 

of the previous observations decreases exponentially, and the trust is used as a weight 

for second-hand information. The two parameters – “trust value” and “confidence 

value” – are combined in OTMF into one metric called “trustworthiness”. 

The article [7] presents Hermes framework determination of node trust, which 

helps in ensuring the reliability of packet transmission. Framework ensures that 

the source sends packets only by the trusted intermediate nodes. In this solution, 

each node determines the reliability metric of neighbouring nodes based on direct 

observation of transmitted packets. Reliability is further extended with opinions 

from other nodes. The proposed solution uses a Bayesian approach to determine 

the value of trust. Trust is calculated on the basis of the beta probability distribution. 

Beta distribution parameters are determined from observations gathered 

during the packet forwarding behaviour. A new metric called trustworthiness, 

being a combination of trust and confidence metrics, is introduced. 

In [8], the trust model was created for the DSR protocol in order to take a decision 

on acceptance or rejection of the route. Decisions are made based on the estimated 

trust respectively. Trust is determined on the basis of the direct trust and recommendations 

from other nodes. Direct trust is determined by the sum of experiences 

on a given node and the recommendation trust is the sum of the recommendations


received from the nodes. In determining the total trust value both values are taken 

into account, but the direct trust has the higher weight. Direct trust is determined 

by observation of packets forwarded by the evaluated node. Collected observations 

are evaluated positively if they are not modified. The assessment is proportional to 

the type of transmitted packet (e.g. route request, data packet). If sent packages are 

modified, the observation is evaluated negatively. Negative evaluation is proportional 

to the type of packet transmitted and the type of modification (e.g. modification 

of information about the source, recipient, sequence number, route). 

The use of probabilistic inference provides an easy way to determine the node 

trust level but also has some disadvantages. Classical logic is based only on two 

values represented by 0 and 1 or true and false. The border between them is clearly 

defined and unchanging. In addition, classical probability theory does not allow 

for distinguishing uncertainty (expressed in terms of probability) from incomplete 

knowledge (lack of knowledge on the topic). 

The other inference method used to evaluate and combine knowledge about 

node behaviours is fuzzy logic. 

Trust model based on the recommendation similarities (RFSTrust) calculated 

with the use of fuzzy mathematics for MANET environment was presented in [9]. 

The fuzzy trust model is proposed to quantify and evaluate the trustworthiness 

of nodes, which includes five types of fuzzy trust recommendation relationships. 

Theoretical analysis and the simulation results show that RFSTrust model can effectively 

prevent selfish nodes and improves the performance of the entire MANET. 

As in the case of inference based on classical logic, fuzzy inference does not 

allow separation of uncertain knowledge from lack of knowledge. 

Another method of enabling the representation of uncertainty is the mathematical 

theory of evidence. 

The use one of the mathematical evidence methods – the Dempster-Shafer 

theory (DST)[10] – for the determination of selfish behaviour of each node is presented 

in [11]. Node cooperation rating is based on the observation of correct packet 

delivery. If a source node receives the information about arrival of the package, it will 

mean that all nodes in the path behave correctly. In this case, the source node defines 

the m() function for each path node, which is named basic belief assignment, as: 

0 A{ SELFISH} 

mA ( ) 

. (1) 

1 A{ UNSELFISH} 

In the absence of proof of information delivery and the lack of error messages, 

the source node finds that a path includes not cooperating nodes. Unfortunately, 

a number of these nodes and information about which nodes are selfish is not 

known. The basic belief assignment is defined as: 

P A{ SELFISH} 

mA ( ) 

. 

(2) 

1 P A { UNSELFISH, SELFISH}


283 

Each of the nodes in the network is equipped with a dedicated component 

implementing an algorithm based on the Dempster-Shafer theory, which uses 

the received recommendations and results of nodes observation. The solution 

has defined two types of trust. The first determines the extent to which the source 

node trusts another node that it will send the package correctly. It was used to 

determine the belief function defined by the Dempster-Shafer theory. The second 

value indicates the degree of trust in which a node trusts that recommendations 

generated by another node are correct. 

IV. Concept description 

Dynamic evaluation of the environment surrounding the node is possible 

by continuously monitoring the node behaviour, their analysis and information 

inference. 

Figure 4. Node evaluation process 

In many cases, the knowledge acquired by a single node is insufficient to fully 

assess the current situation, therefore it must be able to exchange information about 

situational awareness built between nodes. Nodes can have different access to data 

about other nodes, so their passing information may be incomplete or uncertain. 

In the solutions described in section III, in most cases it is impossible to distinguish 

ignorance from uncertain knowledge, taking into account incomplete and 

conflicted knowledge derived from various sources. 

The Dezert-Smarandache theory [12-14] allows combining information from 

multiple sources. It focuses on the problems of combining uncertain, conflicted 

and inaccurate information [15]. DSmT overcomes the limitations of applying 

the inference methods used so far in the assessment of trust: probabilistic inference, 

fuzzy logic or DST. These methods enable the binary evaluation of nodes or 

creating hypotheses that cannot penetrate. By using DSmT it is possible to create 

any number of hypotheses that do not have to be exclusive, and thus more accurate 

assessment of the nodes. In addition, this method enables to distinguish the uncertain 

knowledge from ignorance. 

This theory rejects the main limitations of the Dempster-Shafer theory: 

• frame of discernment is a finite, exhausted and exclusive set of hypotheses, 

• the application of the excluded middle rule, 

• acceptance of the Dempster’s rule as a rule a combination of views, 

• acceptance of the Dempster’s conditioning rule.


The DSmT distinguished two types of models: 

• free model – where frame of discernment (Θ) consists of extensive but not 

exclusive items, so components can be mutually overlapping. This model 

is called free because of the lack of assumptions imposed on the hypothesis. 

• hybrid model – it allows the modelling of imprecise-views and exclusivity 

constraints Θ elements. In this case, the elements may overlap, but they 

do not have to. 

The DSmT introduces the concept of hyper-power set, which is denoted by D Θ . 

This collection is understood as the set of all proposals that were created from elements 

of Θ with the use of operators and . For example: 

 

for D 

 

, , ,..., ,| D | 5 

1 2 0 1 4 

, , , , . 

0 1 1 2 1 3 1 2 4 1 2 

(3) 

A. Events monitoring 

Node assessment is made based on direct node observation and information 

from neighbouring nodes. Examples of observed events by which nodes can be 

evaluated are: 

• provision of information – some of the nodes in ad-hoc networks are 

characterized by self-interested behaviour in order to deprive other nodes 

of the shares, for example by failing to forward packets for selfish node 

to the other nodes. Validation of packet transmission is possible through 

the analysis of incoming acknowledgments, when transmission of acknowledgments 

is enabled in the network or by tracking the packages sent by 

the monitoring node. 

• compliance of safety rules – in tactical networks information may have different 

levels of sensitivity, for example: secret, confidential, non-confidential. 

Data on a certain level of sensitivity can be sent only to nodes that have 

access to information about a specific level or a higher level. Based on information 

collected on nodes access levels and data contained in the labels, 

it can be verified if a node observes the principles of safety, i.e. whether 

it has access only to data which is authorized and makes it available only 

to the authorized users. 

• recommendation correctness – in the case when trust level is determined 

by recommendations from other nodes in the network, it is necessary 

to provide protection against “liar” nodes. A “liar” shall construe nodes, 

which transmit incorrect recommendations on other nodes, the objective 

of re-routing packet forwarding, intercepting or preventing delivery to 

the destination node.


285 

The observed events can be evaluated as 0, 1 – using the classical theory of probability. 

However, in many cases, the observed behaviour provides some indication 

of both hypotheses, which would require omitting the evaluation of the event or 

a need to assign two assessments – which would misrepresent the two behaviours. 

Each behaviour is treated equally and the designated level of trust makes it impossible 

to identify the appropriate response to behaviour. 

B. Nodes classification 

Application of the Dezert-Smarandache theory provides for more hypotheses, 

which enable more accurate assessment of behaviour. Additionally, through the creation 

of secondary hypotheses using sum and product operators, it can constitute 

representation of imprecise and uncertain hypotheses. 

During the observation of nodes behaviour they can be evaluated as: 

• cooperating node (C) – the node transmitting information, 

• egoistic node (E) – the node is not transmitting information, 

• honest node (H) – the node transmitting the proper recommendations, 

• liar node (L) – the node transmitting incorrect recommendations, 

• secure node (S) – the node adhering to safety rules, 

• unsecure node (U) – the node is not adhering to safety rules. 

The set of basic assumptions in some cases may be insufficient for correct classification 

of nodes. Apart from the hypotheses, it is possible to determine the basal 

intermediate hypotheses developed from the basal hypothesis with the sum and 

logical product operators. The secondary hypotheses can distinguish: 

• uncertain cooperating node (UC) – the node to which correctness of packet 

forwarding was tested, but it is not possible to take clear decision whether 

it is a cooperating or selfish node. This situation can occur if a node did 

not receive confirmation of the package transfer – for each of the nodes 

in the path the uncertain cooperating node hypothesis is taken. 

UC C E 

• suspect liar node (SL) – the node whose recommendations may be biased, 

the value reported earlier, recommendation differs from the accumulated 

knowledge and the other recommendations, however, this difference does 

not yet allow for finding that they are wrong and biased. 

SL H L 

• uncertain honest node (UH) – the node to which you cannot determine 

whether the recommendations forwarded by it are correct, because 

of the lack of previously accumulated knowledge. 

UH H L


• suspect unsecure node (SU) – the node whose behaviour indicates partial 

compliance with security rules, for example, a node has access to the resources 

which are not eligible, but only make them available to the authorized 

individuals. 

SU S U 

• uncertain secure node (US) – the node in the case of which you cannot 

determine if it complies with the security rules, due to lack of knowledge 

regarding the node's resource access level. 

US S U 

With such specific hypotheses, it is possible to refine the assessment of nodes 

indicating the possibility of exchanging data with the node, but it does not include an incoming 

recommendation and needs more detailed observation of node’s behaviour, for 

example by using an additional mechanism for including a node to certain behaviours. 

C. Sample evaluations 

Information fusion is done separately for each type of event – co-operation 

between the nodes- following the security rules and recommendation correctness. 

Each hypothesis is assigned with a value of m(), depending on the number of observed 

events, which were assigned to a particular hypothesis. The m() is described 

by conditions defined by the following formula (4): 

m( ) 0 

mA ( ) 1. 

(4) 

 

AD 

The set of hypotheses for each type of event allows using the DSm rule of combination 

for free-DSm models: 

m ( A) m ( X ). 

M( ) 

i i 

 

X1, X2,..., 

Xk 

D 

i1 

( X1X2... Xk 

) A 

k 

(5) 

Tables 1 and 2 show some example values of the received recommendations 

on node’s cooperation observation and compliance with security policies. 

TABLE I. Recommendations about node cooperation 

cooperating egoistic uncertain cooperating suspect egoistic 

m 1 0,650 0,030 0,320 – 

m 2 0,720 0,050 0,230 – 

m 3 0,690 0,040 0,270 – 

m M(Θ) 0,865 0,011 0,020 0,105


287 

TABLE II. Recommendations about security policy compliance 

secure unsecure uncertain secure suspect unsecure 

m 1 0,230 0,265 0,150 0,350 

m 2 0,270 0,290 0,070 0,370 

m 3 0,320 0,280 0,100 0,300 

m M(Θ) 0,136 0,130 0,007 0,727 

It can be specified based on the collected recommendations, if the node is cooperating 

and suspect unsecure. This information allows a node to take a decision on further 

node observation. The node can forward low sensitivity information, whose transmission 

to unauthorized units will not contribute to the realization of carried out actions. 

V. Conclusion 

Ensuring security in tactical MANET requires gathering and processing information 

about the node surrounding reality. Information from various sources, 

however, is often uncertain, incomplete and even conflicting. The method ensuring 

coverage of all of this information is Dezert-Smarandache theory, which allows 

representing of imprecise hypotheses. By applying the Dezert-Smarandache theory 

it is possible to identify specific and general hypotheses, which can combine data 

from different sources with access to information on the behaviour of nodes. As part 

of further work a function that enables combining data including their update time 

and weight of data sources will be determined. 

References 

[1] K. Seshadri Ramana, A.A. Chari, N. Kasiviswanth, “A Survey on Trust Management 

for Mobile Ad Hoc Networks”, International Journal of Network Security & Its 

Applications (IJNSA), vol. 2, no. 2, April 2010. 

[2] L. Capra, “Towards a Human Trust Model for Mobile Ad-hoc Networks”, Dept. 

of Computer Science, University College London. 

[3] Z. Han, K.J.R. Liu, Y.L. Sun, W. Yu, “A Trust Evaluation Framework in Distributed 

Networks: Vulnerability Analysis and Defence Against Attacks”, INFOCOM 2006. 

25th IEEE International Conference on Computer Communications, April 2006. 

[4] J. Głowacka, “Procedures of building nodes’ awareness for security in tactical 

ad-hoc networks, KKRRiT 2011, Poznań 2011, Telecommunication Review 

– Telecommunication News 2011 [CD], no. 6, pp. 405-408 (in Polish). 

[5] Zhu Han, Yan Lindsay, K.J. Ray Liu, Wei Yu, “Information Theoretic Framework 

of Trust Modelling and Evaluation for Ad Hoc Networks”, IEEE Journal on Selected 

Areas in Communications, vol. 24, no. 2, February 2006.


[6] Jien Kato, Jie Li, Ruidong Li, “Future Trust Management Framework for Mobile 

Ad Hoc Networks”, IEEE Communications Magazine, April 2008, pp. 108-114. 

[7] C. Zouridaki, B.L. Mark, M. Hejmo, R.K. Thomas, “A Quantitative Trust 

Establishment Framework for Reliable Data Packet Delivery in MANETs”, In SASN ’05: 

Proceedings of the 3rd ACM workshop on Security of ad hoc and sensor networks, 

pp. 1-10, New York, NY, USA, 2005. 

[8] V. Balakrishnan, V. Varadharajan, U.K. Tupakula, P. Lucs, “Trust and 

Recommendations in Mobile Ad hoc Networks”, Third International Conference on 

Networking and Services, IEEE 2007. 

[9] Junhai Luo, Xue Liu, Mingyu Fan, “A trust model based on fuzzy recommendation 

for mobile ad-hoc networks”, Computer Networks 53 (2009), pp. 2396-2407. 

[10] G. Shafer, “A mathematical theory of evidence”, Princeton U.P., Princeton, NJ, 1976. 

[11] J. Konorski, R. Orlikowski, “DST-Based Detection of Noncooperative Forwarding 

Behavior of MANET and WSN Nodes”, Proc. 2nd Joint IFIP WMNC., Gdansk, 

Poland, 2009. 

[12] F. Smarandache, J. Dezert, “Advances and Applications of DSmT for Information 

Fusion”, vol. 1, American Research Press Rehoboth, 2004. 





[15] J. Głowacka, M. Amanowicz, „Situational awareness of a military MANET node 

– the basis” („Podstawy tworzenia świadomości sytuacyjnej węzła wojskowej sieci 

MANET”), Telecommunication Review – Telecommunication News 2012, no. 2-3, 

pp. 59-62 (in Polish).

Mechanisms of Ad-hoc Networks Supporting 

Network Centric Warfare 

Rafał Bryś, Jacek Pszczółkowski, Mirosław Ruszkowski 

Systems Elements Designing Section, Military Communication Institute, 

Zegrze Poludniowe, Poland, 

{r.brys, j.pszczolkowski, m.ruszkowski}@wil.waw.pl 

Abstract: In this article are presented the results of Ad-hoc networks mechanisms analyzes (identification, 

authorization, autoconfiguration and data exchange). These mechanisms were analyzed taking into 

account specifications of Ad-hoc networks using in netcentric operations. After that, have been selected 

the most appropriate mechanisms specified to be used in the special conditions. The Ad-hoc networks 

mechanisms specifications will be helpful in designing and building of networks for netcentric operations. 

Keywords: ad-hoc networks; NCW; autoconfiguration; authorization 


The way of modern warfare is strongly focused on the informations of enemy 

status and activities. It prodeuces the need for new and more sophisticated communications 

systems, which will guarantee fast and secure way to exchange data 

during Network-Centric Warfare. The overall military systems architecture provides 

MANET (Mobile Ad-hoc NETwork) for a military communications networks, 

at the lowest levels of command. 

The essential features of Ad-hoc networks intended for use in netcentric 

environments are: 

• Network decentralization. Each node in Ad-hoc network can perform 

the services as well as participate in the data transfer to the recipient. 

• Ability for dynamic topology changes. Network nodes are independent 

from each other and can arbitrarily changes its location, and thereby in their 

mutual relations. 

• Radio links usage. It eliminates the need to develop telecommunications 

infrastructure. 

• High network reliability – in case of failure of any network components, 

others can automatically take over their functions. 

• Good scalability (ease of expantion). Nodes joining the network that fulfill 

certain safety requirements are able to realize services almost immediately.


According to given above Ad-hoc networks advantages and communications 

requirements for netcentric systems, it can be concluded that these networks can be 

used in military systems but must fulfill some additional requirements for special 

communication systems. Therefore, the Ad-hoc networks functions should be 

enhanced with additional mechanisms: adaptation, identification, authentication, 

auto-configuration, data exchange and routing. 

In next paragraph of this article is presented the specification of above mechanisms 

in terms of special use (netcentric operations) of Ad-hoc networks. 

II. Autoconfiguration mechanisms 

Addressing in Mobile Ad-Hoc networks can be classified in terms of nodes 

addresses managing, as follows: 

• state accumulation addressing (so-called: stateful), 

• stateless addressing. 

In statefull approach, addresses are accumulated in a network entity that 

has knowledge of assigned and not assigned IP addresses, so it can avoid duplication 

of addresses. In stateless approach, each node randomly selects its own IP address 

and then executes duplicate address detection test to ensure that selected address 

is not yet in use on the network. 

The accumulation state approach seems to be the most optimal to prevent 

addresses duplication, but taking into account network vulnerability to destructive 

factors, this solution is not acceptable. Therefore, for the netcentric Ad-hoc networks 

it is recommended to use of stateless mechanisms with emphasis on the Ad-Hoc 

IP Address Autoconfiguration mechanism. 

A. Ad-Hoc IP Address Autoconfiguration 

The Ad-Hoc IP Address Autoconfiguration mechanism combines mechanisms 

SDAD (Strong Duplicate Address Detection) and WDAD (Weak Duplicate Address 

Detection). In this way, the duplicate address detection mechanism checks 

duplication occurence during the initialization of the node, and detect and solve 

duplicated addresses by analysis of routing messages in intermediary nodes. 

The combination of two mechanisms allows continuously operate for nodes during 

network splitting and merging. As in WDAD, each node selects the key of 128 bits 

length and appends it to the routing protocol control packets. Intermediary nodes 

must retain the key value for each address in the routing table or cache. Automatic 

configuration procedure is exactly the same as in the SDAD mechanism. When 

a node receives a routing packet, examines all the IP addresses and keys values 

(contained in this packet) and compares it with the addresses and keys contained 

in the address table or cache. If more than one key value will be found for the IP 

address, the conflict of addresses is stated. In this case the node sends the unicast


291 

error message to the address with lower value key. During normal operation, 

if the node receives information about the error of his address, it releases this address 

and starts reconfiguring procedure to obtain a new IP address. In the next 

subsections is presented idea of the WDAD and SDAD mechanisms. 

B. Strong Duplicate Address Detection (SDAD) 

SDAD mechanism is the basis for all stateless solutions. It consists of a simple 

mechanism that allows an Ad-Hoc node to select IP address and then check to 

if it is already used by another node. Although, SDAD produces some problems 

and limitations. Performing of duplicate address detection procedure is limited 

to the nodes initialization phase. So, for some reasons (such as temporary 

loss of connection to network), autoconfiguration process leads to duplication 

of addresses, and the network is not able to function properly. This protocol 

does not guarantee the uniqueness of the network addresses and the probability 

of address duplication increases with the size of the network. In addition, 

the protocol generates a lot of traffic – every joining node sends a few broadcast 

packets in flood mode. 

C. Weak Duplicate Address Detection (WDAD) 

The WDAD mechanism is intended to extend the mechanism for detecting 

duplicate addresses on whole life cycle of the network. The idea is that duplicate 

addresses can be tolerated as long as packets reach the destination node indicated 

by the sender, even if the destination address is used by another node. Therefore, 

each node selects an identification key using for the routing protocol to distinguish 

the potential duplicate IP addresses. 

The main disadvantage of WDAD mechanism is its dependence on routing 

protocol. The WDAD requires some changes in the routing layer to support the key 

identifier of the node. In the routing layer, each node is identified by a virtual address 

involving the combination of IP address and key value. In addition, WDAD 

duplicate addresses detection relies on local routing information, which makes 

it completely adaptable only to proactive routing (each node uses a full routing 

table). In case of reactive routing, the possibility of duplicate address detection 

in a realitvely short time is reduced – delays increase. WDAD does not generate 

additional traffic caused by the automatic configuration mechanism, but produces 

the overhead of routing protocol packets. 

III. Identification and authorization mechanisms 

The security issues are carried out in two phases: authentication of participants 

of data exchange session, data authentication, and integrity guarantee. This


article presents the mechanisms used in the first stage. There can be distinguished 

mechanisms for different functional purposes, as follow: 

• identification, 

• authentication, 

• authorization. 

The user authenticity checking process and giving the authority for a specific set 

of services is carried out by subsequent performing of the mechanisms mentioned 

above. First of all must be carried out user identification, what provides recognition 

of the unique features assigned to a person or object. This process allows to identify 

user or object of automated data processing system. The result of the identification 

process is so-called electronic identity. The next step is to carry out the authentication 

process, which aim is to provide a correlation of the electronic identity with 

real world. The aim of the authentication process is to ensure at least one of the two 

parties of data exchange about the identity of the other. After successfully authenticating 

the parties of data exchange session, the authorization process is executed. 

The aim is to give the user or program or process appropriate permissions (access 

and resource use rights) associated with the authenticated part. 

The following chapters are presented mechanisms used in the WLAN, WMAN 

and WPAN network techniques. 

A. IEEE 802.11 

The IEEE802.11 technique as the most often used, has a number of mechanisms 

proposals: 

• MAC filtration, 

• WEP (Wired Equivalent Privacy), 

• EAP (Extensible Authentication Protocol) and 802.1X, 

• WPA/WPA2 (WIFI Protected Access). 

The most powerful security mechanism used in the WLAN is WPA and WPA2 

(an upgraded version). It is also recommended to use in Ad-hoc networks dedicated to 

support netcentric operations. With regard to identification and authentication, both 

protocol versions use the same mechanisms. The difference is only in used to the encryption 

data protocol. The first version uses the TKIP protocol, while the second – AES. 

WPA (WiFi Protected Access) is the successive WEP mechanism and provides 

implementation of safety procedures at a higher level. It inherits from its predecessors 

some parts of properties and also introduces new, such as data encryption or 

integrity checking mechanisms. It is assumed that the WPA2 mechanism (IEEE 

802.11i) is a combination of: WPA2 = 802.1x + EAP + AES + CCMP. 

The EAP (Extensible Authentication Protocol), described in recommendation 

RFC2284, was intended to authenticate endpoints of PPP connection for remote 

access to network resources. The objective of this standard is to control network 

access, based on ports.


293 

In terms of authentication, 802.11i mechanism implemented two policies, 

namely: 

• WPA Personal – often denoted as WPA-PSK (Pre-Shared Key), where 

both parties identify together and encrypt data using previously set key of 

256-bit length, 

• WPA-Enterprise – is dedicated to the larger networks, where the station 

authentication is carried out using the RADIUS server and mechanisms 

described in 802.1X recommendation. 

This standard also set a new security architecture – RSN (Robust Security 

Network). In this architecture the process of secure connection establishing is divided 

into four phases: agreeing of security policy, authentication, key generation 

and distribution, and safe data exchange ensuring data integrity and confidentiality. 

From the netcentric operations point of view, the WPA2-PSK mechanism seems 

to be optimal, because there is no need to use network authentication server. In case 

of WPA2-Enterprise mode, each new station that joining to network or disconnected 

and connected (due to changes in propagation) is obligated to authenticate with 

entity (authentication server), which can be inaccessible due to a server crash. For 

this reason, in 802.11 networks should be applied WPA2 Pre Shared Key mechanism, 

taking into account all the principles of key material distribution and protection. 

In this case, the PMK (Pairwise Master Key) used to determine the PTK (Pairwise 

Transient Key), during the 4-way handshake, is determined on the basis of three 

values: SSID, SSID length and shared secrets – WPA password. 

B. IEEE 802.15 series 

1) IEEE 802.15.4 

The ZigBee system architecture developed by the ZigBee Alliance is based on 

the physical layer PHY and data link layer – MAC layer specification, as is described 

in Recommendation 802.15.4:2004 and the later version 802.15.4:2006. In the structure 

is the network coordinator. This entity has full functionality in the star structure 

taking part in relay of data between devices. In others, network coordinator functions 

are limited to determining the basic parameters of the network. 

In terms of secure communication, the ZigBee can operate in several modes: 

open mode without security mechanisms included, only authentication and integrity 

check – AES-CBC-MAC, only encryption – AES-CTR, and authentication, integrity 

check and encryption AES-CCM. The authorization is thus realized in two above 

modes, and is made in the data link layer MAC. This is performing by calculating 

a common secret (MIC – Media Integrity Code) using CBC-MAC mechanism 

basing on a common cryptographic key held by cooperating stations. This secret 

is calculated based on a symmetric key (known by two parties of data exchange 

session) and the bits in the entire data frame (including the header and data field).


Therefore, apart from the sender and recipient authentication, there is the frame 

integrity check performed. The MIC code is transmitted in each 802.15.4 frame 

except confirmations frames – ACK. 

2) IEEE 802.15.3 

The 802.15.3 networks are organized in a picocells (called piconet), are managed 

by the picocell supervisor PNC (PicoNet Controller). PNC is selected from 

the terminals located in the area of mutual radio range. PNC is responsible for 

synchronizing the piconet elements (using “beacon”), management of QoS parameters, 

power saving modes, and network access control. PNC also performs security 

functions relating to all piconet members. Secure relationships DEV-DEV can be 

created independently, without the use of PNC. 

The user’s identification, authentication and authorization are done in a similar 

way as for 802.15.4 networks, using CBC-MAC algorithm. It is realized based 

on knowledge of the symmetric key by both parties of exchanging data session 

and the “nonce” value, which carrying the identifiers of the sender and recipient. 

The last block of CBC-MAC algorithm result is a message integrity code calculated 

on the basis of identification data and common cryptographic key. In the 802.15.3 

networks, all the messages (including the beacon message) are authorized. 

3) IEEE 802.15.1 

IEEE 802.15.1 networks are organized in piconets, managed by the terminal 

selected as the Master, which oversees the complex piconet consist of up to 255 slave 

devices, where only up to 7 may have an active state. Communication between slave 

devices always takes place through the master device. Security (encryption and authentication) 

is done through a pairing process for secure communication between 

terminals, using a cryptographic key for the SAFER+ algorithm (algorithm E). 

In the first phase of the pairing, initial key K INIT (Initialization Key) is generated 

using the shared 128-bits pseudorandom string (IN_RAND) and PIN number (up 

to 128 bits of length). If the PIN is shorter, is completed with the hardware address 

of the master device (BD_ADDR). Figure 1 shows the key generation process. 

Figure 1. K INIT key negotiation from PIN number 

Than, the devices encrypts the data carrying in time slots and carry out negotiations 

of new 128-bits cryptographic key K LINK – Link Key. This key is stored


295 

by the devices and is used to encrypt the proper data exchange. The possession 

of this key by a pair of devices is equivalent to their logical association (pairing). 

The last stage is authentication of pairing devices – Figure 2. 

Figure 2. Authentication process. 

It involves the generation of SRES’ word, which is compared with the SRES 

recalculated locally (the original). The authentication process is as follows: 

1. Station A (authenticator), sends 128-bits pseudo-random sequence as a clear 

text to station B AU_RAND A (authenticated). 

2. Station B using the E1 algorithm calculates the 32-bits SRES’ word based 

on its hardware address (the remaining 96 bits (ACO) are used to calculate 

the encryption key), K LINK cryptographic key and the AU_RAND A word. 

3. Station B sends the SRES’ to station A. 

4. Station A calculates locally the SRES word and compares it with 

SRES’ – SRES=SRES’. 

5. The positive result of comparison indicates a positive result of authentication 

station B by station A. 

6. The process is repeated but station B is the authenticator. 

C. IEEE 802.16 

WiMAX (Worldwide Interoperability for Microwave Access – 802.16) is intended 

for the building of urban networks and MANs, were designed as part 

of the last mile users access. WiMAX security architecture includes elements and 

mechanisms such as stations digital signature (X.509), the security associations (SA), 

encryption protocols, key management protocols (also responsible for confirming 

the identity of co-operating stations). 

Security Association is a logical link between two terminals (stations) and 

consist of safety parameters, such as cryptographic keys, certificates, etc. Due to 

the three-phased secure data exchange process between stations, there are two 

basic types of association: authentication security association (Authorization 

Security Association) and the data exchange security association (Data Security


Association). The phases of secure data exchange process are: authentication, negotiation 

of cryptographic keys and encryption. 

Authentication security associations are used during station authentication 

stage and includes security parameters used for this stage: 

• X.509 certificate used to identify the station, 

• authorization key (AK – Authorization Key) used to generate the KEK 

(Key Encryption Key) – KEK is a cryptographic key to encrypt the TEK 

negotiation process (Traffic Encryption Key), 

• hash – computed by HMAC-Digest function, used to data integrity check 

during the negotiation of cryptographic material, 

• sequence number of AK, 

• the period validity of AK, 

• association descriptor. 

The station authorization phase is performed in four steps: 

1. The connecting station sends authentication informations in AuthenticationInfMess, 

containing a manufacturer certificate, to verify its credibility. 

2. Simultaneously, the station sends an authentication request in a AuthorizationReqMess 

message, which contains a request of authentication key 

and security association descriptor. Also, there are send informations 

of supported cryptographic algorithms, data authentication mechanisms 

and connection identifier BCID (Basic Connection ID). BCID is assigned 

by the station B during the initial phase of connection establishing (Initial 

Ranging). Station A is authorized after its certificate confirmation. 

3. Station B sets security association descriptors (the initial association 

properties / Primary SA / and existing static association / Static SA /). 

Then activates the authentication key and sends the response to station 

A in the AuthorizationRepMess, containing : 

a. AK key (encrypted by public key of station A), 

b. 4-bits AK key number used to distinguish the following keys, 

c. validity period of AK key, 

d. SA descriptors – previously established 

4. In the last phase is calculated the cryptographic key KEK and message authentication 

keys (HMAC_Key_D and HMAC_Key_U) based on key AK. 

They will be used during the negotiation phase of TEK keys. Because the AK 

key has a limited period of validity than the station should periodically 

renew the authorization process by sending an AuthenticationReqMess, 

before the AK validity period expires. 

The figure 3 shows that the station identification and authentication process 

is carried out in two stages of the symmetric and asymmetric encryption, using 

PKM protocol version 2. This protocol is responsible for normal and cyclic station 

authentication processes and exchange of cryptographic material. This protocol 

operates in a client/server mode. The station identification is performed basing


297 

on X.509 certificates. Each WiMAX station has the X.509 certificate issued and 

implemented by the equipment manufacturer. 

Figure 3. IEEE 802.16 authorization phase 

IV. Adaptative ad-hoc technologies 

Network centric operations requires such Ad-hoc networks organization that 

provide the conditions for data exchange as far as possible to guarantee reliable delivery 

of informations and ensure accuracy of the transmissions. 

In order to fulfill these conditions, technologies used for wireless networks 

must meet two fundamental conditions: 

– Support as far as possible creating and maintaining a Ad-hoc network 

topology; 

– Offer the mechanisms of adaptation to terrain conditions, propagation, 

the nature of the end-user services and quality requirements for 

these services and types of users of such networks (mobile, nomadic, 

stationary). 

With regard to the abovementioned factors was carried out analysis of network 

techniques enabling the building of Ad-hoc networks and adaptation to 

the changing conditions in a radio channel. The analysis concerned the network 

technologies standardized within the IEEE and covered standards for the Personal 

Area (WPAN), Local (LAN) and Metropolitan Network (WMAN). 

A. IEEE 802.11 

WLAN wireless technologies standardized within the IEEE 802.11 group is currently 

the most prevalent. WLANs based on IEEE 802.11 can be found in a number 

of private and commercial installations.


Originally, it was developed version labeled as 802.11 operating in 2,4 GHz 

ISM band. However, due to the growing needs of the telecommunications market 

and technological development have been designed subsequent versions of this 

technique. Currently, as WLAN solutions are used four base protocols prepared by 

IEEE 802.11 group (a, b, g, n) specifying the encoding of radio signals, the operating 

bands and offered rate. Original version is currently not used. 

Name 

Max. 

bitrate 

[Mbit/s] 

Bandwidth 

[GHz] 

Table I. IEEE 802.11 standards [3] 

Modulation 

Supported 

MIMO 

streams 

Range 

[m] 

Publication/ 

comments 

802.11 2 2,4 FHSS, DSSS, IR 1 100 June 1997 

802.11a 54 5 OFDM 1 120 September 1999 

802.11b 11 2,4 HR-DSSS, CCK 1 140 September 1999 

802.11g 54 2,4 

HR-DSSS, CCK, 

OFDM 

1 140 

June 2003 

compatible downwards 

with 802.11b, 

802.11n 540 2,4 or 5 OFDM 4 250 November 2009 

Significant role for high transmission rate ensuring have OFDM modulation, 

which allows more efficient use of radio spectrum. The IEEE 802.11n extension 

allows obtaining higher speeds and transmission ranges by using MIMO technology 

(Multiple-Input Multiple-Output). For MIMO technology are used more 

than one antenna transmission or reception and a special encoding techniques. 

WLANs based on 802.11 can run in infrastructure mode, where users 

communicate with each other through the AP (Access Point) or independent, 

in Ad-hoc mode. The main directions of development of the 802.11 series 

of standards focused on solutions for network infrastructure, while working 

in Ad-hoc mode was not considered or undertaken works in this direction are 

on a limited scale. 

Prevalence of use cards compliant with the IEEE 802.11 standard for office 

and home has made the necessary research on the use of these technologies for 

building and organization of wireless networks in Ad-hoc mode. 

Comprehensive solution for this type of network is proposed in extension IEEE 

802.11s – “ESS – Extended Service Set Mesh Networking”. The solution assumes 

the possibility of creating a self-configuring 802.11 wireless network operating 

in Ad-hoc mode, where each device can have several interfaces. One of the basic 

assumptions of the developed solution was to provide extensibility and flexibility, 

consisting of applying a variety of mechanisms to fulfill the same functions in different 

nodes of the network. 

For IEEE 802.11s standard assumed use of the underlying mechanisms for 

the physical layer, the security of information and access to the transmission


299 

medium used in previous versions of IEEE 802.11. Additionally, assumes the use 

of new, additional solutions. The required functionality described by the IEEE 

802.11s includes: 

a) PHY – Physical Layer, based on the IEEE 802.11a /b/g/n. 

b) Medium Access Coordination – mechanisms for medium access control, 

based on standard solutions adopted for the network compatible with IEEE 

802.11 with QoS extensions described in the IEEE 802.11e. To further 

use are expected new features such as the ability to dynamically change 

the operating channel. 

c) Mesh Security – authentication of users, nodes and trunks cryptographic 

protection, in large part based on the extension of EEE 802.11i. 

d) Mesh Configuration & Management – mechanisms for automatic configuration 

of radio parameters (operating channel frequency selection, 

transmit power, etc.), QoS management policies. 

e) Discovery & Asscociation – mechanisms for detecting the presence of neighboring 

nodes and mesh networks and to identify parameters of mesh network, 

the procedures for connecting to the network nodes and the statement 

of logical links with its neighbors. 

f) Mesh Topology, Learning, Routing & Forwarding – a group of mechanisms 

and protocols for the control and the creation of the current topology, and 

data forwarding, it is assumed the use the Hybrid Wireless Mesh Protocol 

(HWMPA) for routing, which is a combination of mandatory reactive 

protocol mechanisms Ad-hoc Radio Metric on-demand Distance Vector 

(RM-AODV), and optional proactive mechanisms – Tree Based Routing 

(TBR). It is also assumed to use a proactive routing protocol Metric Radio 

Optimized Link State Routing (RM-OLSR). 

g) Internetworking – mechanisms for ensuring mesh network cooperation 

with external 802 series networks, including the wired networks. 

h) Mesh Measurement – mechanisms and procedures for monitoring the mesh 

network and radio environment, e.g. for setting routes in mesh network. 

In this standard was assumed using of several types of mesh network nodes, 

which can change operation mode during its functionality. They are: 

a) Mesh Point (MP) – this is the basic type of the node in the network 802.11s. 

It is used to carry out the procedures for establishing and maintaining 

networks. Arranges and maintains the logical links of detected neighbors 

and forward transit traffic. 

b) Mesh Access Point (MAP) – an extended version of the node MP, also 

supports the function of an access point (AP) to support standard WLAN 

client station (not supporting 802.11s standard). 

c) Mesh Portal Point (MPP) – used to work as node in mesh networks to 

the external networks compatible with the IEEE 802 (wired, wireless, 

and also compatible with the 802.11s). In addition to features for the MP


can perform the functions of the bridge (network bridge) and support 

a proactive routing within their own mesh network. 

d) Lightweight Mesh Point (LWMP) – nodes implementing the functionality 

of a Mesh Point node only in an environment of mutual audibility of nodes 

(as the classic WLAN nodes in Ad-hoc mode). 

e) Client Station (STA) – a classic client station designed to work with the AP 

and mechanisms not supporting 802.11s standard. 

For each of these nodes is assumed that it will work as terminal and offer users’ 

access to services at the OSI application layer. 

B. IEEE 802.16 

Standard for wireless metropolitan structures WMAN defined by a group 

of IEEE 802.16. Mobile WiMAX provides support for stationary equipment, 

nomadic and full mobility for the devices carried or transported at speeds up to 

120 km/h. This functionality of WiMAX technology is extremely important for 

dynamic operation conducted with the use of vehicles. 

Assuming the possibility of using battery-powered portable devices IEEE 

802.16e authors have developed the two modes operation of subscriber stations 

for energy conservation, that are: idle and sleep mode. 

IEEE 802.16e introduces the possibility of scalable bandwidth utilization 

of the radio channel from 1.25 MHz to 20 MHz, with modulation SOFDMA 

(Scalable Orthogonal Frequency Division Multiple Access), where the number 

of subcarriers varies with the width of the channel. This property, greatly increases 

the feasibility of efficient transmission with high speeds. 

Mobile WiMAX is ready to use multipath reception. For this purpose are 

used OFDM modulation and MIMO (Multiple Input – Multiple Output) solutions, 

involving the use of multiple antennas on the receiving and transmitting sites. 

It has a great importance in the case of communication systems in urban areas, 

where the multipath phenomenon is the most noticeable. 

In response to the possibility of variable propagation conditions experienced 

by mobile subscribers, the IEEE 802.16e introduces the hybrid automatic repeat 

request H-ARQ (Hybrid Automatic Repeat called Request), where the first correction 

coding is used, the detection coding (used for ARQ) can be used, if necessary, 

in the next step. 

IEEE 802.16e also implies the possibility of using smart antennas, where 

the adaptive gain control of selected direction of the transmission is used. It can increase 

the useful range and reduce interference. 

Current implementation of WiMAX technology in the great majority basing 

on the solutions with the use of base stations. The creators of the IEEE 802.16 standard 

established the possibility of network functioning without using base stations. 

Devices working in cooperation mode on the network (without the use of WiMAX


301 

base stations) has been identified as a “mesh” mode and presented in the extension of 

IEEE 802.16, and then included in the version IEEE 802.16-2004 (extension “a” 

has been incorporated into the version of 2004). Mesh mode can also be used in cooperation 

of base stations. In this mode, client stations are referred as MSS (mesh 

subscriber station), and base stations as MBS (mesh base station). In the mesh 

mode, unlike the communication mode managed by the base station, where 

in the transmission frame structure is distinguished direction from the base station 

(downlink) and up to her (uplink), transmission between mesh networks 

elements (MSS or BSS) is carried out as bi-directional which is established during 

initialization of the SS. 

These links are described by an 8-bits connection identifier (Link ID). If the data 

is transmitted in the direction to the MSS located closer to the MBS, traffic is treated 

as uplink traffic, otherwise it is seen as the downlink channel. The mesh mode uses 

three methods of data transmission known as a method of scheduling: 

• coordinated distributed scheduling; 

• uncoordinated distributed scheduling; 

• centralized scheduling. 

In the case of distributed scheduling, each node transmits the current and 

proposed schedule of data transmission to neighboring nodes in one hop distance. 

If the destination node provides support for data transfer according to this schedule, 

it responds to the source node in the control subframe slot. Ultimately the source 

node sends to destination node acknowledgment of receipt of approval for transmission 

according to the desired schedule. For distributed coordinated scheduling, 

the scheduling messages are sent with using a scheme providing collision avoidance. 

Centralized scheduling method is similar to mode where the transmission 

is directly managed by the base station. In this mode, the MSS points are used only 

as relay stations to the nearest MBS. 

Described in IEEE 802.16-2004 mesh mode is not included in the extension 

known as the IEEE 802.16e. Due to the fact that the use of mobile WiMAX technology 

requires other devices than based on the IEEE 802.16d, mesh mode is not 

supported by network equipment manufacturers for the IEEE 802.16e. However, 

are created theoretical attempts to modify the IEEE 802.16e devices for mesh mode. 

One of the proposals to use mesh mode in a mobile version of WiMAX has been 

presented for the tactical communications networks in [4]. 

The most recent proposal to improve the functionality of mobile WiMAX 

subscriber station is in extension IEEE 802.16j, which introduces the concept 

of a relay station – MRS (multihop relay station). MRS perform a similar function 

as the MSS in a centralized mode, passing data from other SS subscriber station to 

base station BS, in the case of absence of direct contact SS to BS. MRS is used also 

in the IEEE 802.16m version (Advanced Air Interface with data rates of 100 Mbit/s 

mobile & 1 Gbit/s fixed), known as WiMAX2 designed to provide transmission 

rates up to 1 Gbit/s.


C. IEEE 802.15 

Personal Area Networks (WPAN) are created to provide a wireless user access 

to facilities and services provided by them and co-operation devices in the network. 

Due to their use, a network device must be simple to use, characterized by small 

size and weight and low power consumption. The most popular standard for WPAN 

is defined in the IEEE 802.15 group. These are the techniques: Bluetooth (IEEE 

802.15.1). UWB (IEEE 802.15.3), ZigBee (802.15.4). In Table II, developed in [1], 

are presented the most important properties of these techniques. 

Table II. Properties of WPAN technologies 

Bluetooth 

802.15.3 802.15.4 

WiMedia UWB Forum standard ZigBee 

Bandwidth 2,4 GHz 3,1-10,6 GHz 

3,1-4,85 GHz, 

6,2-9,7 GHz 

2,4 GHz and 868/915 MHz 

Modulation 

Channel 

Access 

Bitrate 

GFSK, 

π/4-DQPSK, 

8DPSK 

Polling, master- 

-slave, TDD (Time 

Division Duplex) 

21 kbit/s do 

40 Mbit/s 

QPSK, DQPSK, 16-QAM, 

32-QAM, 64-QAM 

CSMA-CA, 

OFDM, 

optionally TDD 

53,3, 80, 110, 

160, 200, 320, 

400, 480 Mbit/s 

No data 

DSSS with BPSK or MSK 

CSMA/CA and guaranteed 

time slots GTS in superframes 

< 2 Gbit/s 2-250 kbit/s 

Range [m] 0,1-100 < 10 1-100 

Transmiter 

Power 

Network 

topologies 

QoS 

Security 

Usability 

< 100 mW 

Piconet, 

scatternet, mesh 

SDP 

(Service Discovery 

Protocol) 

SAFER+, 

authentication, 

encryption 

Dedicated 

applications 

Bandwidth depended 

1 mW 

Star, 

cluster, mesh 

Guaranteed time slots 

AES-128 

AES-128, 

authentication 

Sensor networks, 

automation systems 

1) UWB (IEEE 802.15.3) 

The IEEE 802.15.3 specification defines the physical layer and data link of high 

speed (high rate) WPANs (Wireless Personal Area Network), that offers above 

20 Mbit/s data rate.


303 

The basic element of the IEEE 802.15.3 wireless network is piconet, defined 

as a wireless communication system, operating in Ad-hoc mode, which allows 

mutual communication of many independent devices. It is assumed that the size 

of a typical piconet covers an area with a radius of 10 m. The IEEE 802.15.3 standard 

defines several functional elements in piconet structure, shown in Figure 4. 

Figure 4. IEEE 802.15.3 piconet elements [11] 

The basic component of this structure is the data transfer object – DEV, which 

in general should be identified as a terminal. DEV can also be a group of devices 

in one location. In piconet, the one DEV element must act as PNC (Piconet Coordinator). 

PNC is responsible for synchronizing the piconet elements (using 

“beacon” messages), the management of QoS parameters, power saving modes, 

and network access control. PNC also performs security functions relating to all 

piconet members. 

Piconet is not a spatial structure, but rather logical. Crucial in the piconet 

creation have PNC. The precondition, which is adopted for the piconet existence 

is DEV act as PNC sending a messages “beacon” containing the information required 

to piconet maintain. The process of piconet creation is done by sending 

joining notification to the PNC, by the DEV in the surrounding area, interested 

in piconet membership. Standard assumes dynamic piconet membership for DEV. 

DEV can attach and detach during the transmission. In situations where the device 

acts as a PNC leaves piconet, must be designated the new PNC. 

Standard provides parallel operation of different piconets. DEV can belong 

to the parent piconet, where have guaranteed access to bandwidth, offspring 

piconet (where piconet is managed by the unit included in the home network), 

and the neighboring piconet when it is managed by the device, that there have no 

contact with home piconet. 

Messages between piconets are sent in superframe, in frames transmitted 

during the transmission time labeled CFP (Contention Free Period). As addresses 

in the so-constructed network are proposed to use the PNID (Piconet ID) identifiers,


and device IDs DevID (Device ID). To avoid ambiguity in the allocation of these 

addresses is assumed to use the parent element defined as MC (mesh PNC). 

2) Bluetooth (IEEE 802.15.1) 

Bluetooth is a standard for wireless networks described by the recommendation 

IEEE 802.15.1. To work using unlicensed 2.4 GHz ISM band. Provides transmission 

range in open space defined by three transmit power classes respectively: 100, 10 

and 1 meter. Depending on the version of the standard offer bit rate of 21 kbit/s 

for 1.0 to 40 Mbit/s for version 3.1. 

The technique enables “point-to-point” and “point-to-multipoint” connection 

in Ad-hoc mode for devices on a small area. The 802.15.1 standard define Bluetooth 

profiles i.e. scenarios for the using Bluetooth technology. Profile supports handling 

application-layer data and voice, print data, access to local and global networks, 

in uniform way. 

The standard assumes that the communication between Bluetooth devices 

is carried out in the logical structure as for the IEEE 802.15.3 i.e. piconet. It is 

assumed that at the same time can be active seven slaves and one master. Status 

of the master unit performs machine that initiates the process of creating a network. 

Data exchange can take place only between the parent node (master) and slave 

(slave). Direct data exchange between slave devices is not possible. 

In addition, devices can participate in piconet in power save mode-Park, except 

seven active slaves. By enabling and disabling the device in this mode, the master 

can provide communication for multiple devices. To ensure cooperation between 

piconets the ability of working in bridge mode was adopted. Thus, piconets can be 

combined into larger structures called scatternet. The IEEE 802.15.1 standard does 

not define the principles of scatternet creation. There were designed only custom 

proposals including Bluetrees, Bluestar, Bluemesh. 

3) ZigBee (IEEE 802.15.4) 

IEEE 802.15.4 was developed for sensor networks. In relation to the Bluetooth 

and UWB offers a relatively low data rate (up to 250 kbit/s). The standards assumed 

the ability to create complex network topologies. IEEE 802.15.4 standard defines 

the format of the physical and data link layer. Based on these recommendations, 

the organization ZigBee Alliance (assembling more than 150 companies) extended 

the functionality of this solution by introducing a standard named ZigBee, which 

also includes recommendations for network and application layers. 

ZigBee defines three types of devices that can be used to create network connections. 

These include the following types: 

• The coordinator – the parent node of ZigBee network; 

• Router – a node that can be used to create network and realizing routing 

and transit data functions; 

• Terminal – devices that do not offer the data relay functionality.


305 

The standard defines the following elements: 

• Application Object (APO) – endpoints (processes) that work together and 

carrying out certain tasks utility. It was assumed ability to run up to 240 

objects in each of the nodes. APOs can work together within a single node, 

and through the network infrastructure; 

• ZigBee Default Object (ZDO) – the objects responsible for network maintaining. 

Determines the type of device (coordinator, router, terminal 

equipment), discovers the environment, network management, creates 

associations between the APO at different nodes; 

• Application Support Sublayer (APS) – responsible for APOs communication 

in the network and maintain established (using ZDO) network connections 

between the APOs; 

• Network Layer (NWK) – uses the functions provided by the data link 

layer. NWK is used in the processes for the network addresses allocation, 

routing, attaching and removal of network nodes. 

• Security Service Provider (SSP) – provides mechanisms for the cryptographic 

protection for APS and NWK layer; 

In the network, ZigBee addresses are determined hierarchically, from the coordinator. 

ZigBee network routes are designated in reactive manner. Source node 

sends request about the destination node to the nearest neighboring node. The request 

is advertised until it reaches the destination. Routing informations are stored 

in intermediate nodes, and can be used for future transmissions. 

It may be noted that the greatest support for the operation of wireless devices 

in a network organized in Ad-hoc mode are WPANs networks. Due to their short 

range and requirements to reduce the cost of specialized devices, do not use superior 

devices specialized in managing the wireless infrastructure. 


In article are presented the mechanisms and communication technologies 

of Ad-hoc networks using for net centric operations. The analysis 

indicates that the most appropriate is IEEE802.11s technology with MIMO 

antennas and modern coding techniques. Taking into account the applications 

of this technique – NCW, it is necessary to use WPA2-PSK and 

Ad-hoc IP Address Authoconfiguration mechanisms. The next step of our work is to 

perform simulations experiments and than build a technology demonstrator.


References 

[1] K. Gierłowski, T. Klajbor, J. Woźniak, “Analiza sieci bezprzewodowych serii IEEE 

802.15.x – Bluetooth (BT), UWB i ZigBee – z transmisją wieloetapową. Część I”, 

Przegląd Telekomunikacyjny. Wiadomości Telekomunikacyjne. 2008, nr 10. 

[2] K. Gierłowski, T. Klajbor, J. Woźniak, “Analiza sieci bezprzewodowych serii IEEE 

802.15.x – Bluetooth (BT), UWB i ZigBee – z transmisją wieloetapową. Część II”, 

Przegląd Telekomunikacyjny. Wiadomości Telekomunikacyjne. 2008, nr 11. 

[3] K. Gierłowski, J. Woźniak, “Analiza szerokopasmowych sieci bezprzewodowych 

serii IEEE 802.11 i 16 (WiFi i WiMAX) z transmisją wieloetapową”, Przegląd 

Telekomunikacyjny. Wiadomości Telekomunikacyjne. 2008, nr 8-9. 

[4] BRYON KEITH HARTZOG, “WiMAX potential commercial off – the – shelf solution 

for tactical mobile mesh communications”, Military Communications Conference, 

2006. MILCOM 2006. 

[5] Evren Eren, Kai-Oliver Detken, “WiMAX-Security – Assessment of the Security 

Mechanisms in IEEE 802.16d/e”, WMSCI 2008. 

[6] IEEE Std 802.15.3b-2005, “Wireless Medium Access Control (MAC) and Physical 

Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPANs)”. 

[7] Naveen Sastry, David Wagner, “Security Considerations for IEEE 802.15.4 

Networks”, University of California. 

[8] J. Jeong, J. Park, H. Kim and D. Kim, “Ad Hoc IP Address Autoconfiguration”, 

I-D draft-jeong-adhoc-ip-addr-autoconf-02.txt, February 2004. 

[9] P. Paakkonen, M. Rantonen and J. Latvakoski, “IPv6 addressing in a heterogeneous 

MANET-network”, I-D draft-paakkonen-addressing-htr-manet-00.txt, December 

2003. 

[10] C. Jelger, T. Noel, and A. Frey, “Gateway and address autoconfiguration for IPv6 

adhoc networks”, I-D draft-jelger-manet-gateway-autoconf-v6-02.txt, April 2004. 

[11] IEEE Standards, “IEEE Standard for Information technology – Telecommunications 

and information exchange between systems – Local and metropolitan area networks 

– Specific requirements. Part 15.3: Wireless Medium Access Control (MAC) and 

Physical Layer (PHY). Specifications for High Rate Wireless Personal Area Networks 

(WPANs).”, IEEE Std 802.15.3-2003, 29 September 2003.

Using Network Coding in 6LoWPAN WSNs 

Jarosław Krygier 

Military University of Technology, Faculty of Electronics, 

Telecommunications Institute, Warsaw, Poland, jkrygier@wat.edu.pl 

Abstract: The low power wireless sensor devices which usually uses the low power wireless private 

area network (IEEE 802.15.4) standard are being widely deployed for various purposes and in different 

scenarios both in civil and military world. IPv6 low power wireless private area network 

(6LoWPAN) was adopted as part of the IETF standard for the wireless sensor devices in order to 

decrease the IP overhead and to adapt big IPv6 packets to small maximum transmission unit (MTU) 

offered by the IEEE 802.15.4 standard. This paper is focused on the adoption of the network coding 

to 6LoWPAN/IEEE 802.15.4 network. The solution based on COPE mechanism is proposed and 

described here. Also the implementation problems are discussed. 

Keywords: component; network codding, WSN, 6LoWPAN 


Tactical communication networks are evolving toward complex heterogeneous 

ad-hoc networks where mobile nodes can simultaneously embed very different kinds 

of communication technologies such as HF or VHF tactical radios (including legacy 

systems), UHF (microwave) ad-hoc radios, lightweight satellite ground stations and 

also wireless sensor networks (WSNs) [1]. Considering the mobile and temporary 

nature of military missions, developing sensor networks that operate in wireless 

mode are a necessity. These conditions pose a number of challenges comprising 

sensor hardware platforms, communications and networks, and information management 

systems. In order to cope with a wide heterogeneity of hardware devices 

as well as to meet the requirements on the end-to-end communication (including 

sensor networks), the Internet Protocol (IP) should be applied in the network elements 

– also in highly mobile and sensor part of the network [9]. 

WSNs are currently the most common type of so called Low-power Wireless 

Personal Area Network LoWPAN (LoWPAN), which are self-configuring ad-hoc 

networks composed of wirelessly connected, autonomous devices, usually characterized 

by constrained computational and memory resources and low consumption. 

On the other hand, the LoWPAN tiny nodes have to be equipped with the IP protocol. 

Unfortunately, the IP traffic causes significant overhead, which is especially


relevant in WSNs and military mobile ad-hoc networks (in low data rates modes). 

In order to adapt the IP to the LoWPANs some solutions are proposed, from which 

the most commonly used are: uIP (micro IP), nano IP and IPv6 low power wireless 

private area network (6LoWPAN). 6LoWPAN was adopted as the part of the IETF 

standard for the sensor devices [2]. Although the 6LoWPAN is not restricted to 

the IEEE 802.15.4-based devices, but it is commonly used with such standard. It is 

also proposed for tactical networks in [3]. 

The 6LoWPAN significantly reduces the IPv6 overhead in the network using 

header compression mechanisms, but additional reduction is required in order 

to transfer high amount of data across the sensor network effectively. Meanwhile, 

network coding has recently emerged as an effective strategy to provide significant 

performance improvements in wireless networks. Many researchers have shown 

that by applying network coding in wireless networks we can improve the network 

throughput and reliability, minimize energy and decrease the network congestion 

[4] [5] [6]. Most of the network coding solutions have been verified using 

mathematical analysis and by the simulation experiments, while modest amount 

have been implemented and tested in limited network environment. Moreover, 

the network coding theory can be applied practically in each layer of the communications 

protocols stack. Also most of the solutions concern the physical (PHY), 

medium access control (MAC) and the application (APPL) layers. This paper 

is focused on application of the network coding theory in the 6LoWPAN-based 

system, taking into account the network (NET), 6LoWPAN adaptation (AL) and 

MAC layers, while the wireless links are built using the IEEE 802.15.4 standard. 

Proposed solution is tailored to the Contiki 6LoWPAN and MAC implementation 

run on the AVR RAVEN sensor motes [10]. 

The remainder of the paper is organized as follows. Section II presents the concept 

of the network coding mechanism for 6LoWPAN. Section III explains the implementation 

details. Description of the tests and the results discussion is performed 

in section IV. General conclusions and future work plans are given in section V. 

II. Concept of network coding for 6LoWPAN/IEEE 802.15.4 

network 

Network coding is known to improve network throughput by mixing information 

from different flows and conveying more information in each transmission. 

Though the idea of network coding is not new, in the past, it has been 

applied mainly in the context of multicasting in traditional wired networks. It is 

because the data packets are sent using multicast and broadcast addresses and they 

can visit many intermediate nodes in the same time, and than some streams often 

meet together and can be mixed (coded) and decoded in other nodes. Wireless 

networks are equipped with similar features, where sent frames can be received by 

many nodes simultaneously, because of the broadcast nature of the transmission


309 

medium. In order to shortly explain the idea of the network coding in a wireless 

multihop network let us assume simple scenario composed of three wireless 

nodes illustrated in Fig. 1 [8]. Nodes 1 and 3 have to send their packets (‘a’ and ‘b’ 

respectively). Packet ‘a’ has to be delivered to node 3 and packet ‘b’ has to reach 

node 1. The left picture shows transfer of the packets without network coding. Each 

node performs standard packet transmission, know as store and forward. It means 

that to transfer those two packets from the source to the destination, four wireless 

transmissions are required. If the network coding is employed (right picture), after 

receiving both packets (‘a’ and ‘b’) by node 2, it can transmit a single coded packet 

(a Å b) in such a manner that both destination nodes can receive it. It is possible 

because of the broadcast nature of the wireless network. Two packets are coded 

using simple xor function. It means that in order to extract one native (original) 

packet from coded packet the receiver needs the second packet. For example, 

the node 3 is able to extract packet ‘a’ from packet ‘a Å b’ if it stores the packet ‘b’ 

(a = aÅbÅb). This exchange reduces the total number of wireless transmission 

from 4 to 3. Generally, network coding can be based on linear composition of two 

or more native packets [7]. 

Figure 1. An example of information exchange with network coding 

A practical scheme, referred to as COPE (Coding Opportunistically), based 

on the network coding for wireless networks is proposed in [6]. The solution tailored 

to the 6LoWPAN network presented in this paper is based on COPE. Each 

node in the network has to store transmitted and received packets during predefined 

time as shown at the network in Fig. 2. For example, node 2 should store transmitted 

packet P4 and received packets P3 and P5 (marked in rectangles). 

Figure 2. Collecting of sent and received packets


For simplicity, it was also assumed that packets will be coded using simple 

xor function and they can be extracted just in the next hope nodes. The native 

packets hidden in the coded packet can be extracted in each destination 

node if the node has stored n-1 packets (where n – number of native packets included 

in the coded packet). It means that each node has also to store the table with 

information about the packets stored in the neighboring nodes. Fig. 3. shows that 

node 1 keeps the neighbor table. The entries in this table are collected based on 

the packets identifiers constructed from the native packets source addresses and 

sent in coded packets in additional header. If the data packets are not in the network 

during some time, signaling packets should be used to inform neighboring nodes 

about stored packets. If the sender sent a coded packet which is a composition of too 

many native packets, some receivers could not extract their missing native packet 

and they should be resent by the sender. Such operation increases the network load 

and decreases the network coding gain. 

Figure 3. Storing of network coding neighbor table in each node 

The network coding gain in opportunistic coding scheme is achievable due 

to, as it has been already mentioned, the broadcasting nature of the wireless network. 

Majority of MAC and PHY radio-based protocols make possible to transfer 

and receive the broadcast frames. The same, considered IEEE 802.15.4 standard 

gives an opportunity to broadcast traffic handling. Assuming that node 1 from 

Fig. 4 wants to transfer the coded packet P1ÅP2ÅP4ÅP5, it ought to broadcast 

the packet. 

Unfortunately, broadcast transmission is not confirmed by the receivers in MAC 

layer, than coded packet transmission is not reliable. Even if some mechanism 

was employed in the network layer that can confirm the coded packets, it would 

be significant, additional signaling traffic. If the transmission reliability is not rel-


311 

evant or ensured by the upper layer protocols the coded packets can be sent using 

broadcasting technique (as in Fig. 4). 

Figure 4. Sending coded packet in broadcasting mode 

As an alternative, the coded packet can be sent in unicast transmission to one 

of the next hope node which confirms its reception using MAC acknowledgment 

(MAC ACK) and the other nodes can receive the packet, interpret it but they cannot 

acknowledge its correct reception (as in Fig. 5). 

Figure 5. Sending coded packet in unicast mode with MAC ACK 

The third possibility is to confirm the coded packet on the fly. Fig. 6. shows 

that node 2 confirms the coded packet using MAC ACK, but nodes 3 and 4 confirm 

reception of given coded packet while transferring its native or other coded 

packets (NC_ACK).


Figure 6. Sending coded packet in unicast mode with mixed acknowledgment 

Tailoring the network coding scheme explained above to the 6LoWPAN/IEEE 

802.15.4 network some assumptions are taken. Since the solution is dedicated to 

sensor motes, it should not consume too much memory during implementation and 

operation. It also should use both IEEE 802.15.4 MAC and network layer features 

(6LoWPAN, uIP6 – microIPv6, RPL – Routing for Low Power and Lossy Networks, 

IPv6 ND – Neighbor Discovery). Fig. 7 presents typical multihop transmission 

of 6LoWPAN datagrams (compressed or fragmented packets) between node 1 

and 3. Depending on the implementation, the 6LoWPAN packets are equipped with 

the addresses (‘Src’, ‘Dest’) which comes from the IPv6 layer and can be carried inline 

or can be extracted from the MAC addresses. Thus, from node 1 the 6LoWPAN 

packet is sent to the IP ‘Dest’ address of the node 3, but the IEEE 802.15.4 frames 

are equipped with the MAC ‘Src’ and ‘Dest’ addresses of node 1 and 2 respectively. 

Both the data and ACK frames are also overheard by the neighboring nodes. 

Figure 7. Typical multihop transmission in 6LoWPAN 

Also depending on the 6LoWPAN implementation, the packets can be routed 

in the IPv6 layer or forwarded in the 6LoWPAN layer. The first case is shown in Fig. 8,


313 

where three nodes are represented by the layered architecture similar to those used 

by the 6LoWPAN implementation in the Contiki operating system [10]. The application 

(Appl), routing (RPL) and neighbor discovery segments are sent using TCP/ 

IPv6 protocol (uIP6 implementation) and routed in the intermediate node using 

tiny IPv6 routing table, collected by the RPL protocol. The 6LoWPAN layer is supported 

also by the IPv6 neighbor discovery which delivers the neighboring nodes 

MAC addresses and IPv6 addresses (string them in the ND cache). 

Figure 8. Routing over 6LoWPAN 

The 6LoWPAN defines also the mesh routing (Fig. 9.). It means that routing 

is performed using the 6LoWPAN mesh header. Moreover, if fragmentation is performed, 

each fragment is routed from source to the destination independently, what 

is different than during routing over the 6LoWPAN, where all fragments have to 

be collected in the intermediate node before subsequent forwarding. 

Figure 9. 6LoWPAN mesh routing 

The network coding operations in 6LoWPAN depends on the routing scheme. 

If the forwarding mechanism is located in the IPv6 layer the coding has to be


performed after header decompression and all fragments collection. This is because 

the 6LoWPAN fragments are not equipped with IP destination address (except the first 

one, if the address is carried inline). If the mesh forwarding is used, the coding operation 

can be located in 6LoWPAN layer before decompression and fragments collection. 

During network coding operation two or more native packets are mixed using 

linear combination. But the native packets can differ in lengths. It is confirmed 

in [6], that coding of native packets with significantly different lengths gives small 

gain. It means that it is required to collect the native packets with similar lengths. 

A format of the coded packet is shown in Fig. 10. Example two 6LoWPAN packets 

with header compression (IPHC – IP header compression) and IP addresses 

carried inline (s – source, d – destination) are xored to receive the coded packet 

with additional network coding header (NC Header). The coded packet is sent to 

an appropriate MAC address, depending on the transmission rule. The example 

in Fig. 10 shows that coded packet is sent to broadcast address. During network 

coding operation it is required to ensure an appropriate length of the native packets 

in order to do not exceed the MAX IEEE 802.15.4 MAC SDU. 

Figure 10. Format of coded packet 

Figure 11. Network coding header format (NCi – NC identifier ‘0001’, A – address type, 

N – number of MAC addresses, Pad L – Pad length) 

The network coding header format is presented in Fig. 11. It is composed 

of the Dispatch field and destination nodes identifiers (MAC addresses) which in-


315 

dicate the destination nodes responsible for decoding the native packets. Dispatch 

field begins from the NC identifier which is set to binary 000 and is allocated by 

the 6LoWPAN recommendation [2] to Non-IPv6 packets. 

III. Implementation 

The network coding proposal presented in section II was implemented in Contiki 

operating system (OS) [10]. The Contiki OS is an open source operating system 

for networked embedded systems in general, and wireless sensor nodes in particular. 

It is developed by a team of developers from the industry and academia. 

The Contiki system incorporates uIPv6 – the IPv6 protocols stack. Both the Contiki 

system and applications for the system are implemented in the C programming 

language. Contiki has been ported to a number of microcontroller architectures, 

including the Texas Instruments MSP430 and the Atmel AVR [11]. The network 

coding implementation was tested on the 8-bit Amtel AVR motes. 

Currently, the Contiki OS allows running the 6LoWPAN implementation 

working on the IEEE 802.15-based MAC, but with some limitations. The most 

important is that the 6LoWPAN implementation is limited to IP routing. It means 

that all fragments have to be collected in each node before transferring the packet to 

the next node. It is shown in Fig. 12. This limitation is very important from network 

coding point of view. Since the sensor motes have memory limitations, they cannot 

allocate to big memory for stored packets. It is because the remaining memory must 

be used to store the fragments before the node rebuild a full IPv6 packet. 

Figure 12. Collecting the fragmented packets in Contiki 6LoWPAN implementation 

The network coding functions (net_coding(), net_decoding()) are located 

in functions defined in sicslowpan.c file, and they allows the network coding before


the packets reach the uIP6 layer. A location of these functions is shown in Fig. 13 

against the other Contiki protocols stack functions. 

Figure 13. Location of the net_coding() and net_decoding() functions in contiki protocols stack 

(routing over 6LoWPAN) 

IV. Verification results discussion 

Implemented network coding mechanisms were verified in a simple testbed 

based on the Atmel evaluation and starter kit that includes two AVR Raven boards 

with a 2.4 GHz transceiver, on-board picoPower AVR application processors with 

LCD display, and one USB stick with a 2.4 GHz transceiver to allow USB connections 

to a PC. The maximum transmission power of the nodes were decreased 

and the nodes were deployed in such a way that using the RPL routing protocol 

the network were configured to the structure presented in Fig. 7, where node 1 

is the USB stick connected to a PC. Such configuration corresponds to the real sensor 

networks in which the nodes transmit the sensor data to the gateway (node 1). 

Instead of the real sensor data generated by the node 3, the IPv6 Ping streams 

(30 packets in each stream) with packet intervals from 0.4 to 0.45 s were transmitted 

between nodes 1 and 3. The length of the Ping packets were set up to 10 Bytes. 

The number of streams were changed from 1 to 7. Also a number of native pack-


317 

ets that can be buffered in each node were changed during verification experiments 

from 0 (without network coding) to 5. Each native packet were buffered 

in the node by 0.5 second. Fig. 14 shows the overall traffic captured in the air 

depending on the number of streams and number of buffered native packets 

in the node (0 – means without network coding). 

Figure 14. Overall traffic captured in the air 

For clarity, detailed values of overall traffic are presented for the network 

without network coding procedures and while 5 native packets can be buffered 

in the node for coding purposes. The results confirm that using network coding 

the overall traffic in the air is decreased. Additionally, the level of traffic decreasing 

is directly dependent on the traffic generated by the nodes and the network coding 

buffers capacity. The cost of the network coding in the network was the end-to-end 

packet delay increasing. In the case of presented network the end-to-end packet 

delay was increased by about 0.5 s (native packet buffering time). By increasing 

the number of buffered native packets in the node the we can attain additional 

gain but the limitation is a memory of the sensor motes. 

V. Conclusions and future work 

The paper described the possibility of using the network coding mechanisms 

in 6LoWPAN wireless sensor networks. The proposal is based on relatively simple 

mechanism relying on the linear combination of two or more 6LoWPAN packets 

and on transmission coded packet to one hop neighbours using broadcasted nature 

of WSN. The proposed mechanisms are tailored to the Contiki 6LoWPAN and 

MAC implementation run on the AVR RAVEN sensor motes. 

The first verification results confirm that using network coding in presented 

network the overall traffic can be decreased. This endeavor would require more


work in the future, especially in terms of implementation of all described here routing 

cases and performing detailed tests in the wider network with more realistic 

traffic generated by the nodes. 

References 

[1] EDA IAP4 Raoadmap, “Future IAP4 European Defense stratEgic Research Agenda”, 

EDA Final Report, September 2011. 

[2] G. Montenegro, N. Kushalnagar, Intel Corp, J. Hui, D. Culler, “Transmission 

of IPv6 Packets over IEEE 802.15.4 Networks”, Request for Comments: 4944, September 

2007. 

[3] H. Song, S.H. Lee, S. Lee, H.S. Lee, “6LoWpan-based tactical wireless sensor network 

architecture for remote large-scale random deployment scenarios”, IEEE Military 

Communications Conference, MILCOM, October 2009. 

[4] R. Ahlswede, N. Cai, S.-Y.R. Li, and R.W. Yeung, “Network information flow,” 

IEEE Transactions on Information Theory, vol. 46, no. 4, pp. 1204-1216, 2000. 

[5] Y. Chen, G. Feng, L. Zhou, “Using network coding to improve robustness and 

persistence for data transmission in sensor networks”, 6th International ICST 

Conference on Communications and Networking in China, CINACOM, August 2011. 

[6] S. Katti, H. Rahul, H. Wenjun, D. Katabi, M. Medard, J. Crowcroft, “XORs 

in the Air: Practical Wireless Network Coding “, IEEE/ACM Transactions on 

Networking, vol. 16, Issue 3, pp. 497-510, June 2008. 

[7] S.-Y.R. Li, R.W. Yeung, and N. Cai, “Linear network coding”. IEEE Transactions on 

Information Theory, February, 2003. 

[8] B. Ni, N. Santhapuri, Z. Zhong, S. Nelakuditi, “Routing with opportunistically 

coded exchanges in wireless mesh networks”, Proceedings of the 2nd IEEE Workshop 

on Wireless Mesh Networks (WiMesh 2006) (2006), pp. 157-159. 

[9] J. Krygier, M. Bednarczyk, K. Maślanka, Integration of MAC Relaying and 

IP Routing Protocols in Ad-Hoc Networks: Multimetric Approach, Military CIS 

Conference MCC’2010, Wroclaw, Poland, September 2010. 

[10] A. Dunkels, B. Grönvall, T. Voigt, “Contiki – a lightweight and flexible operating 

system for tiny networked sensors”, Proceedings of the First IEEE Workshop on 

Embedded Networked Sensors (Emnets-I), Tampa, Florida, USA, November 2004. 

[11] Thang Vu Chien, Hung Nguyen Chan, and Thanh Nguyen Huu, “A Comparative 

Study on Operating System for Wireless Sensor Networks”, International Conference 

on Advance Computer Science and Information System, ICACSIS 2011, Jakarta 2011.

Testbed Implementation of Energy Aware Wireless 

Sensor Network 

Ewa Niewiadomska-Szynkiewicz 1, 2 , Michał Marks 1, 2 , Filip Nabrdalik 2 

1 Research and Academic Computer Network (NASK), Warsaw, Poland, 

ewan@nask.pl, Michal.Marks@nask.pl 

2 Institute of Control and Computation Engineering, 

Warsaw University of Technology, Warsaw, Poland, 

ens@ia.pw.edu.pl, mmarks@elka.pw.edu.pl, F.Nabrdalik@stud.elka.pw.edu.pl 

Abstract: Wireless sensor networks (WSNs) are autonomous ad hoc networks designed and developed 

for potential applications in monitoring, surveillance, security, etc. The sensor devices that are 

battery powered should have lifetime of months or years. Therefore, energy efficiency is a crucial 

design challenge in WSN. In this paper the energy efficient communication techniques, i.e., activity 

control and power control protocols are presented and discussed. We focus on the implementation 

of energy aware algorithm for WSN – Geographical Adaptive Fidelity (GAF) – in our testbed network 

formed by the Maxfor devices. The results of experiments confirm significant energy savings that 

lead to network lifetime increase. 

Keywords: wireless sensor networks; WSN; energy aware communication; activity protocols; power 

control protocols 

I. Introduction to WSN 

The last decade has seen tremendous interest in all aspects of wireless sensor 

networks (WSNs) – distributed systems composed of numerous smart, embedded and 

inexpensive sensor devices deployed densely in a sensing area [1], [2], [15]. The nodes 

of a network equipped with CPU, battery, sensing units and radio transceiver, networked 

through wireless links can be used in applications, in which traditional networks are 

inadequate. The important property of WSN is the ability to operate in harsh and hostile 

environments, in which human monitoring is risky, and often impossible. The lack 

of fixed network infrastructure components allows creating unique topologies and 

enables the dynamic adjustment of individual nodes to the current network structure 

in order to execute assigned tasks. Wireless sensor networks are deployed in various 

environments and are used in large number of practical applications concerned 

with monitoring, rescue missions and military actions. Ad hoc architecture of WSN 

has many benefits, however its flexibility come at a price. A number of complexities and 

design constraints are concerned with the characteristics of wireless communication,


i.e., limited transmission range and throughput, limited QoS, limited resources, and 

multihop nature of a network. Moreover, the quality of wireless transmission depends 

on numerous external factors, like weather conditions or landform features. Most 

of these factors can change in time. 

For these reasons, WSNs need some special treatment. The most important 

constraint is concerned with limited power source. Nodes forming the network are 

often small battery-fed devices. Each device, participating in WSN needs to manage 

its power in order to perform duties as long and as effective as possible. Moreover, 

devices are often deployed in remote locations where replacing batteries is difficult 

or even impossible, and usually a high disparity between expected and real power 

drawn can be observed. Thus, power management is very important. A numerous 

energy aware communication methods and strategies have been developed, and 

are presented in literature [2], [12], [14]. In general, power management is a topic 

that has been a subject of intensive research in ad hoc networking in recent years. 

In this paper we discuss the approaches to design energy aware WSN topologies. 

The main contribution of our work is to show the benefits of application 

of the energy aware communication protocol in real network system. We have 

implemented and verified the modified version of commonly known protocol GAF 

in our testbed network formed by the Maxfor devices. The results of tests in our 

laboratory were compared with simulation results presented by Xu et al. in [16]. 

In section II, we describe the energy consumption characteristics in WSN. In section 

III, we investigate some energy aware methods and algorithms, in section IV, 

the GAF protocol is described. The results of the performance evaluation of GAF 

through testbed implementation are presented and discussed in section V. 

II. Energy consumption in WSN 

A lifetime of WSN is measured by the time interval before all devices have 

been drained out of their battery power or WSN is unserviceable – no longer 

provides an acceptable event detection ratio. To maximize the lifetime, all aspects 

such as architecture, circuits and communication protocols must be made energy 

efficient. Let us focus on the energy consumption characteristics of a sensor device 

(WSN node). Three main components that consume the energy are: microprocessor, 

sensing circuit and radio transceiver. 

The energy consumed by microprocessor is determined by the sum of dynamic P d 

and static P s power, i.e., P = P d + P s . Many techniques have been developed to 

minimize the energy consumption, such as: dynamic voltage scaling, modulation 

scaling, and energy aware embedded, event driven operating systems. 

The objective of a sensing unit is to translate physical measurements to electrical 

signals. Several sources of energy consumption can be listed: signal sampling, 

conversion of physical signals to electrical ones, analog to digital conversion, signal 

conditioning. The energy usage in this unit is relatively constant. In some applica-


321 

tions the detectors can work in the event driven mode, and can be switched off for 

some time to save energy. 

The communication system is the major energy consumer during wireless 

network operation. The radio transceiver unit can operate in one of four modes, 

which differ in the consumption of energy: transmission – signal is transmitted to 

other nodes (greatest power consumption), receiving – message from other nodes 

is received (medium power consumption), stand-by (idle) – inactive transceiver, 

turned on and ready to change to transmission or receiving mode (low power 

consumption) and sleep – transceiver is switched off. In order to extend the lifetime 

of a network, it is frequent practice that radio transceivers of some nodes are 

deactivated. The nodes remain inactive for most time and are activated only to 

transmit or receive messages from the network. 

The energy consumption strongly depends on the transmission range and 

modulation parameters. Generally, short transmissions in a network are desired. 

They involve smaller power usage and cause less interference in a network, simultaneously 

effected transmissions, thus increasing the network throughput. 

III. Energy aware communication 

A. Methods and algorithms 

In this paper we focus on energy aware protocols for IEEE 802.15.4 (ZigBee) 

based networks. The protocol used by the node of WSN consists of the application, 

transport, network, data link and physical layers [1]. The energy management 

in WSN is emphasized in data link and network layers. The MAC protocol guarantees 

efficient access to the transmission media while carefully managing the energy 

allotted to the nodes. Typically, this objective is achieved by switching the radio to 

a low-power mode based on the current transmission schedule. 

Energy efficiency is considered mostly by protocols provided by network 

layer. Energy aware routing ensures the survivability of low energy networks. 

In commonly used wireless senor networks it is assumed that the receiver is not 

located within the transmitter’s range. The transmitter must transmit data to the receiver 

by means of intermediate nodes. Thus, the natural communication method 

in WSNs is a multi-hop routing. This is a certain limitation, but on the other hand 

it enables the construction of network of greater capacity – a multi-hop network 

enables simultaneous transmission via many independent routes, and managing 

available energy. Moreover, each node of a network can attribute the level of power 

used to send a message to the other node in order to minimize the amount of energy 

received from the battery, while at the same time maintaining the coherence 

of the network. Due to a significant node redundancy (nodes are densely deployed), 

and the assumption that each node of WSN has impact on the power used to transmit 

a message numerous low energy consumption communication methods and


energy conservation techniques have been developed, and are described in literature. 

In general, two main approaches for power saving in WSN can be distinguished: 

• power control protocols, 

• activity control protocols. 

B. Power control protocols 

The popular approach to energy efficient communication is controlling a transmit 

power of each node in a network. In general, short transmissions involve smaller 

power consumption and cause less interference and latency. Power control protocols 

(PCs) are responsible for providing the routing protocols with the list of the nodes’ 

neighbors, and making decisions about the ranges of transmission power utilized 

in each transmission. Therefore, the PC protocols are placed partially in the OSI 

network layer and the OSI data link layer [14]. A numerous PC protocols have 

been developed and are described in literature. They utilize various information 

about a network and network nodes, i.e., location of nodes and its neighbors or 

direction from which the signal was received. Based on these information power 

control techniques may be divided into following groups: location-based (nodes 

are able to determine their exact positions), direction-based (relay on the ability 

of all nodes to estimate relative directions of their neighbors) and neighbor-based 

(determine all neighbors within the maximal transmission range). The detailed 

survey of PC protocols can be found in [14]. The results of performance evaluation 

of selected techniques through simulation are presented in [12]. 

C. Activity control protocols 

The other approach to energy efficient communication is controlling a number 

of active nodes in a network. Due to a significant node redundancy and multiple 

paths between nodes we can turn off selected intermediate nodes while still guarantee 

full connectivity and maximum link utilization constraints. Dynamic power 

management is an efficient approach to reduce system power consumption and 

extend the lifetime of individual node without significantly degrading the network 

performance. The basic idea is to deactivate the radio transceiver of selected 

nodes when not needed and wake them up when necessary. Hence, radio devices 

remain inactive for most working time and are activated only to transmit or receive 

messages from other nodes. In general, dynamic power management is a complex 

problem. It can involve the limitation of accessible band, and can also interrupt 

the data transfer in the network. Therefore, implementing the correct policy for 

radio switching and estimating the optimal value of radio transceiver’s switch-off 

time are critical for a network performance. 

The activity control protocols (AC) employ dynamic management of radio 

devices. The objective is to limit the power consumption while simultaneously


323 

minimizing the negative impact on the network throughput and on the efficiency 

of data transmission routing. Different types of AC protocols are presented 

in literature, and are applied to WSN systems. Similarly to PC protocols, they 

utilize various information about a network and network nodes, i.e., location 

of nodes in a network, nodes density, connectivity, etc. Many of them implement 

clustering techniques. It was observed that grouping nodes into clusters 

can reduce the overall energy usage in a network. Based on utilized information 

activity control techniques may be divided into several groups: locationbased 

(GAF [16], GeRaF [6]), connectivity-based (Span [7]), clustering-based 

([LEACH [9], HEED [17], PANEL [5], PRWST [3]) and hierarchical (EEHC [4], 

CGPS [12]). 

It should be pointed that activity control protocols should be capable of buffering 

traffic destined to the sleeping nodes and forwarding data in the partial network 

defined by the covering set. The covering set membership needs to be rotated between 

all nodes in the network in order to maximize the lifetime of the network. 

IV. GAF algorithm 

The Geographic Adaptive Fidelity (GAF) algorithm developed by Y. Xu et al., 

and described in [16] selects nodes responsible for relaying traffic in the network 

based on their geographical position estimated using the GPS system or calculated 

using any other location system [2], [11]. GAF assumes covering the network deployment 

area with a virtual grid. The location information is employed to form clusters. 

The GAF protocol relies on the concept of ‘’node equivalence’’. Two nodes are 

equivalent when they are equally useful as relays in communication between other 

nodes. Problem of selecting equivalent nodes is nontrivial. It can be easily observed 

that network nodes equivalent in communication between a given pair of nodes do 

not have to be equivalent in communication between other pairs of nodes. 

To select equivalent nodes GAF divides a spatial domain, where nodes are 

distributed into cells that form a grid, see Fig. 1. The size d of each cell is calculated 

due to transmission ranges of nodes, i.e., d 

r/ 5 , where r denotes the maximal 

transmission range assigned to network nodes. It is assumed that each node in a cell 

is in transmission range of all other nodes within adjacent cell. The construction 

of such a grid allows to preserve the network connectivity. All nodes in a network 

may switch between one of three states: active, discovery and sleep. In the active 

state a node is responsible for relaying traffic on behalf of its cell. In the discovery 

state nodes exchange discovery messages, trying to detect other nodes with higher 

energy in the same cell. However, the overhead due to discovery messages is not 

very high. The following load balancing energy usage is proposed. After spending 

a fixed amount of time T A in the active state, a node switches to the discovery state, 

and another node from the same cell switches to the active state. After spending 

a fixed amount of time T D in the discovery state, the node backs to the active state.


Whenever a node changes the state, it sends a message containing its identifier (ID) 

and its ranking value (RV). The ranking value is used to select the relay node to 

transmission a message. Several rules are proposed to determine a node’s RV. In general, 

a node that is in the active state has a higher rank than a node in the discovery 

state, and nodes with longer expected lifetimes (calculated with respect to energy 

consumption characteristics) have higher rank than the others. 

Figure 1. Communication using GAF. The data transmission from the source to the target 

is realized using different relay nodes for different time slots 

Nodes in the active and discovery states may switch to the sleep state whenever 

they find a node in the same cell with higher RV. When a node enters the sleep state, 

it cancels all pending timers, and powers down the radio. After spending the fixed 

amount of time T S in the sleep state the node turns on its radio and switches to 

the discovery state. The sole concept of GAF is to maintain only one node with its 

radio transceiver turned on per cell, see Fig. 1. The mentioned parameters, i.e., T A , 

T D , T S , are used to tune the algorithm. In our improved version of GAF (GAF-M) 

the value of the interval T D is estimated independently for each cell, and depends 

on the number of nodes that form a given cell. The bigger the number of nodes 

the shorter T D time. Moreover, we prohibit switching between different states 

the nodes with very low battery level. 

The GAF algorithm was designed for IEEE 802.11 networks. It can run over 

any routing protocol for ad hoc networks. Y. Xu et al., discuss the performance 

of GAF combined with two reactive routing protocols AODV (Ad-hoc On-Demand 

Distance Vector) [13] and DSR (Dynamic Source Routing) [10]. We adopted GAF 

to work in 802.15.4 networks. In our implementation we used DYMO (Dynamic 

Manet On-demand) [8] routing protocol that is a successor of AODV. DYMO 

shares many benefits of AODV but is slightly easier to implement. 

Y. Xu et al. claim that GAF provides longer lifetime of WSN with minimal loss 

in data delivery rates compared to pure AODV and DSR protocols. The simulation 

results presented in [16] confirm the good performance of the algorithm. It is 

worth to note that simulation results show that GAF extends the network lifetime 

proportionally to the increase of nodes density in the deployment area.


325 

V. GAF performance in testbed network 

The main objective of our research was to implement and validate the GAF 

algorithm using testbed implementation in our laboratory formed by physical 

sensor devices. 

A. Testbed WSN – hardware and software 

The experiments in our WSN laboratory were performed using testbed implementation 

involving MTM-CM5000 motes (http://www.maxfor.co.kr/eng/ 

en_sub5_1.html) manufactured by Maxfor (see Fig. 2). 

Figure 2. The MTM-CM5000 mote 

The MTM-CM5000 mote is IEEE 802.15.4 compliant wireless sensor node 

based on the original open-source “TelosB” platform design, developed and published 

by the University of California, Berkeley. The mote’s architecture is presented 

in Fig. 3 and the general specification is given in Table I. 

Figure 3. Architecture of the MTM-CM5000 mote 

The testbed networks were formed by three to eleven MTM-CM5000 motes and 

one base station, all operating under TinyOS system. TinyOS (http://www.tinyos.net/)


is an open source, highly portable operating system that can be used to on-line 

operation of WSN formed of real low-power wireless devices. Our application 

was written in nesC (Network Embedded Systems C) language provided by TinyOS 

system. TYMO (http://tymo.sourceforge.net/) – the implementation of DYMO protocol 

on the TinyOS system was used to work with GAF. The preliminary version 

of our implementation was done in TOSSIM simulator, and next transformed into 

physical network. TOSSIM (http://docs.tinyos.net/tinywiki/ index.php/TOSSIM) 

is a discrete-events simulator for TinyOS wireless sensor networks. By exploiting 

the sensor network domain and TinyOS’s design, TOSSIM can capture network 

behavior at a high fidelity while scaling to thousands of nodes. The same code 

can be used for simulation and real WSN operating in TinyOS. 

Table I. Specification of MTM-CM5000 Mote 

Processor 

RF chip 

RF power 

Power supply 

Antenna 

RF current draw 

Range 

Sensors 

TI MSP430F1611 

TI CC2420 

–25 dBm~0 dBm 

2.1 V~3.6 V – (AA or AAA battery) 

Dipole antenna / PCB antenna 

Receive mode: 18.8 mA 

Transmit mode: 17.4 mA 

Sleep mode: 1.0 μA 

~150 m (outdoor), 20~30 m (indoor) 

Light, humidity, temperature 

B. Results of experiments 

Multiple experiments were performed in our laboratory. The goal of the first 

series of experiments was to evaluate the performance of the GAF algorithm 

in a physical wireless network. The wireless sensor networks implementing GAF 

and DYMO protocols were compared with networks with no power capabilities at 

all (implementing pure DYMO). The key metric for evaluating examined networks 

was the lifetime of the network. To compare the performance of a given network 

we used the following characteristic: 

LTI = Lifetime GAF+DYMO / Lifetime DYMO (1) 

where Lifetime GAF+DYMO denotes the lifetime of a network implementing GAF and 

DYMO protocols, and Lifetime DYMO is the lifetime of a network using only DYMO. 

The goal of the second series of experiments was to test the influence 

of the nodes density into a lifetime of a given network. 

In this paper the results of experiments performed for nine network configurations, 

i.e., examples E1-E9 describing different model size and topology are


327 

presented and discussed. The detailed description of examined networks is given 

in Table II. The variable N_cells denotes the number of cells that form a grid covering 

a deployment area, and S_cell the number of motes in each cell. Figures 4 and 5 

show the exampled network configurations. 

Table II. Specificatiojn of Eight Testbed Networks 

Testbed Networks (examples) 

E1 E2 E3 E4 E5 E6 E7 E8 E9 

N_cells 1 1 1 1 1 2 2 2 3 

S_cell 2 3 4 5 6 2 3 4 3 

Figure 4. Network E4 formed by 7 motes (source, sink and one cell of 5 nodes) 

Figure 5. Network E9 formed by 11 motes (source, sink and three cells of 3 nodes each) 

During the tests, we calculated the lifetime of a network measured by the time 

interval before WSN was unserviceable (all motes in one cell were drained out 

of their battery power). The performance of examined algorithms GAF+DYMO 

and pure DYMO for networks E4 and E9 are presented in Fig. 6 and 7. The summary 

of results – extensions of network lifetimes using GAF and DYMO over pure 

DYMO for eight experiments is presented in Table III.


Figure 6. Extension of WSN lifetime using GAF; network E4 

Figure 7. Extension of WSN lifetime using GAF; network E9 

Table III. Summary of Results (Values of LTI For All Testbed Networks) 

Method 


E1 E2 E3 E4 E5 E6 E7 E8 E9 

DYMO 1 1 1 1 1 1 1 1 1 

GAF+DYMO 1.78 2.18 3.28 5.18 5.90 1.71 2.57 3.40 2.56 

The goal of the third series of experiments was to compare the performance 

of the original version of GAF with the modified version GAF-M implementing our 

scheme for T D time interval calculation. The performance of examined algorithms 

GAF-M and pure DYMO for network E4 is presented in Fig. 8.


329 

Figure 8. Extension of WSN lifetime using GAF-M; network E4 

The summary of results for this case study are presented in Table IV and Fig. 9. 

The modified GAF-M allows to extend the network lifetime up to 30%. 

Table IV. Summary of Results (Values of LTI For 5 Testbed Networks) 

Method 


E1 E2 E3 E4 E5 

DYMO 1 1 1 1 1 

GAF+DYMO 1.78 2.18 3.28 5.18 5.90 

GAF-M + DYMO 1.93 2.78 4.40 5.48 6.50 

Figure 9. Sensitivity to network density


The results of experiments confirm the strength of the GAF algorithm. GAF 

provides longer lifetime compared to pure DYMO routing protocol. The results 

of tests performed in our testbed networks formed by the physical devices 

were not worse than simulation results described in [16]. We can even say that 

the combination of GAF and DYMO provides better results than the application 

that combines GAF and AODV. Similarly to the conclusions presented in [16], 

we could observe that GAF extends the network lifetime proportionally to 

the increase of nodes density in the deployment area. Tables III and IV, and Fig. 9 

show that time of accurate network operation strongly depends on the number 

of nodes that form each cell of a grid covering the deployment region. Moreover, 

our modifications of GAF improve the performance of the algorithm that leads 

to network lifetime increase. 

VI. Summary 

Many challenges arise from ad hoc networking and development of real life 

wireless sensor systems. In this paper we focused on energy aware networks. We 

described a functionality of the popular activity control protocol GAF for power 

saving in WSNs, and our testbed implementation that combines GAF and routing 

protocol DYMO. We evaluated the performance of our application in the laboratory, 

and improved the algorithm performance w.r.t. original version. The results 

of experiments confirm good performance of GAF in real life networks. In our 

future work we plan to make experiments with higher dimension networks and 

compare GAF with other activity control protocols described in literature. 

As a final observation we can say that the design of wireless sensor networks 

should account for trade-offs between several attributes such energy consumption 

(due to mobility, sensing, and communication), reliability, fault-tolerance, data 

collection latency, and quality of information, and their impact on mission objectives. 

Therefore, strategies and techniques for energy efficient, reliable and secure 

communication in wireless sensor network has become a hot debate nowadays. 


This work was partially supported by National Science Centre grant 

NN514 672940.


331 

References 

[1] I.F. Akyildiz, W. Su, Y. Sankasubramaniam, and E. Cayirci, “Wireless sensor 

networks. A survey”, Computer Networks, vol. 38(4), pp. 393-422, 2002. 

[2] I. Akyildiz, and M. Vuran, “Wireless sensor networks”, John Wiley & Sons, Ltd, 

West, Sussex, UK, 2010. 

[3] N. Asiam, W. Philips, W. Robertson, and S. Sivakumar, “A multi-criterion 

optimization technique for energy efficient cluster formation in wireless sensor 

networks”, Information Fusion, vol. 12(3), pp. 202-212, 2011. 

[4] S. Bandyopadhyay, and E. Coyle, “Minimizing communication costs in hierarchicallyclustered 

networks of wireless sensors”, Computer Networks, vol. 44(1), pp. 1-16, 2004. 

[5] L. Buttyan, and P. Schaer, “Position-based aggregator node election in wireless sensor 

networks”, International Journal of Distributed Sensor Networks, pp. 1-15, 2010. 

[6] M. Casari, A. Marcucci, M. Nati, C. Petrioli, and M. Zorzi, “A detailed simulation 

study of geographic random forwarding”, Proc. of Military Communications Conference 

MILCOM. vol. 1, pp. 59-68, 2005. 

[7] B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris, “Span: An energy-efficient 

coordination algorithm for topology maintenance in ad hoc wireless networks”, ACM 

Wireless Networks, vol. 8(5), pp. 481-494, 2002. 

[8] I. Chakeres and C. Perkins, “Dynamic manet on-demand (dymo) routing”, 2006, 

http://www.ietf.org/internet-drafts/draft-ietfmanet-dymo-05.txt. 

[9] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energy-efficiet 

communication protocol for wireless sensor networks”, Proc. of the 33rd Hawaii 

International Conference on System Sciences. pp. 1-10, 2000. 

[10] D.B. Johnson, “Routing in Ad Hoc Networks of Mobile Hosts”, Proc. of the IEEE 

Workshop on Mobile Computing Systems and Applications, pp. 158-163, 1994. 

[11] E. Niewiadomska-Szynkiewicz, and M. Marks, “Optimization schemes for wireless 

sensor network localization”, Journal of Applied Mathematics and Computer Science, 

vol. 19(2), pp. 291-302, 2009. 

[12] E. Niewiadomska-Szynkiewicz, P. Kwaśniewski, and I. Windyga, “Comparative 

study of wireless sensor networks energy-efficient topologies and power save protocols”, 

Journal of Telecommunications and Information Technology, no. 3/2009. pp. 68-75, 

2009. 

[13] C.E. Perkins and E.M. Royer, “The ad hoc on-demand distance vector protocol” 

in C. E. Perkins, editor, Ad hoc Networking, Addison-Wesley, pp. 173-219, 2000. 

[14] P. Santi, “Topology control in wireless ad hoc and sensor networks”, John Wiley 

& Sons, Ltd, West Sussex, UK, 2006. 

[15] R. Verdone, D. Dardari, G. Mazzini, and A. Conti, “Wireless sensor networks 

and actuator networks. Technologies, analysis and design”, Elsevier, USA, 2008. 

[16] Y. Xu, J. Heidemann, and D. Estrin, “Geography-informed energy conservation 

for ad hoc routing”, Proc. of the 7th Annual International Conference on Mobile 

Computing and Networking (MobiCom ’01). pp. 70-84, 2001. 

[17] O. Younis, and S. Fahmy, “Distributed clustering in ad-hoc sensor networks: A hybrid, 

energy-efficient approach”, Proc. of the IEEE INFOCOM, vol. 1, pp. 629-640, 2004.

An Energy Aware Self-Configured Wireless 

Sensor Network 

Marcin Wawryszczuk, Marek Amanowicz 

Electronics Faculty, Military University of Technology, Warsaw, Poland, 

marcin.wawryszczuk@gmail.com, marek.amanowicz@wat.edu.pl 

Abstract: Recent years have shown that wireless communication are becoming more popular 

in professional, academic and every day applications. A wireless sensor node is usually autonomous, 

advanced device with limited power source. This limited power source makes the topology control 

a crucial technique, which allow the network to obtain energy efficiency without affecting the network’s 

connectivity and sensing coverage. The article demonstrates the energy efficient topology 

control mechanism, which allow a majority of network nodes stay in sleeping mode and do not affect 

the connectivity and sensing coverage. The result in energy consumption of the nodes and network 

lifetime are exposed and compared to different approaches. 

Keywords: wireless sensor network, topology control, energy efficiency 


Wireless sensor networks have become an emerging technology that represents 

the next evolutionary step in environment monitoring, traffic control, objects detection 

and tracking, etc. Typically, a network consists of a large number of nodes, 

distributed over some specific area and organized into multi hop system. A node 

of wireless sensor network is a simple device equipped with control, communication, 

sensing units as well as a power supply. The power supply is usually a small, 

low capacity battery which is very hard to replace. It makes the energy factor 

of the sensor node a critical one. As a result, topology control mechanisms of sensor 

network have to be equipped with efficient power management procedures. 

The network designer is responsible for these power management procedures to 

protect the energy resources what imply that the network lifetime is elongated. On 

the other hand, the network has to realize the primary objectives what is often out 

of whack with the energy conservation. The suitable compromise between both 

is called “topology control” (TC). It adjusts desired network’s parameters (like connectivity, 

coverage, etc.) while reducing energy consumption. 

There are two approaches to the topology control in wireless sensor networks. 

One of them is the nodes activity control, where only the small subgroup of nodes


is operative (active) and the rest of them run in sleep mode. The other method 

is called sleep/wake-up algorithms. The nodes with implemented sleep/wake-up 

algorithm stay active for a small period of time and then turn into sleep mode, 

where the communication, sensing and control units are deactivated. The active 

and inactive cycles are repeated during the whole node lifecycle. This approach 

minimize the energy consumption of every single node, but it does not guarantee 

the connectivity and/or coverage at any time of the network life. Obviously, these 

methods reduce the global energy consumption, as a result of all nodes energy conservation. 

Both methods are capable of elongating the network lifetime significantly. 


Much of the related research are addressed to WSN that are dense, battery 

powered and self configured. Because of these requirements, most of the authors 

are concentrated on finding solution at various levels of the communication protocol, 

including extremely energy efficient aspects [1]. Some of the most popular 

solutions in energy efficient topology control are stated below. 

• GAF [2] (Geographical Adaptive Fidelity) is the nodes activity control 

method. This approach groups nodes in virtual grids. Each grid is defined 

such that, for any two adjacent grids A and B, all nodes in grid A are capable 

of communicating with any node in grid B, and vice-versa. The nodes 

in the same grid are equivalent in terms of routing, consequently just 

one node at a time needs to be active. The rest of the nodes can switch to 

the sleep mode and save the energy in this way. 

• GeRaF [3] (Geographic Random Forwarding) is the nodes activity control 

method. In this method, each node posses the information about the neighborhood. 

Each node follows to the sleep/active pattern, starting from 

channel listening. The data transmission starts when the source node w s 

has any packet to send. The source node broadcasts the data with the location 

of itself and the location of the destination node w d . The node, which 

is inside the communication range of w s and is closest to the destination 

node w d simultaneously, continues the process of the data forwarding. 

The process is accomplished when the end point w d is reached. GeRaF 

protocol can cause a lot of collisions in a network [3][4]. 

• SPAN [5] is the nodes activity control method. In the method, only a small 

group of coordinators stay active, while the rest of nodes switch into sleep 

mode and only periodically check if they are requested to wake up and become 

the coordinators. The coordinators are responsible for passing the data 

from a source node to a sink node. To guarantee a sufficient number of coordinators, 

SPAN uses so called coordinator eligibility rule: if two neighbors 

of a non-coordinator node cannot communicate with each other directly 

or via coordinators, it means that the node should become a coordinator.


335 

However, it may happen that several nodes discover the lack of a coordinator 

at the same time and, thus, they all decide to become a coordinators. 

• ASCENT [6] (Adaptive Self Configuring Sensor Network Protocol) 

is the node activity control method. In ASCENT a node decides when to 

join the network or stay dormant. The decision is taken based on the connectivity 

and packet loss ratio measured locally. 

• STEM [7] (Sparse Topology and Energy Measurement) is the sleep/wakeup 

method. This approach exploits two different communication units: one for 

wakeup signaling and other for data transmission. The wakeup radio is not 

a low power consumption unit. Both radios work with the same communication 

range. Periodically each node turns its wakeup radio on for T active 

time, every T duration. When a source node wants to communicate with 

a neighboring node, it sends a stream of signals via the wakeup channel. 

As soon as the destination node receives the signal message, it acknowledges 

that fact and turns the data transmission unit on. 

• FSP [8] (Fully Synchronized Pattern) is the sleep/wakeup method. In this 

approach, all nodes in network wake up at the same time, following 

the pattern. It means that each node goes periodically into active mode 

for T active time, every T wakeup . After that, nodes switch to the sleep mode 

until the next T wakeup . 

• SWP [9] (Staggered Wakeup Pattern) is the sleep/wakeup method. In this 

approach nodes create the specific data acquisition tree structure. The nodes 

located at different levels of data gathering tree, wake up in a different time. 

Obviously, the active time T active of the adjacent tree levels must partially 

overlap to guarantee that children and parent nodes in the mentioned tree 

structure are able to communicate. 

Figure 1 shows the mechanism of SWP, where T active is the time the node stays 

active, T sleep is the time the node stays dormant, T is the sum of T active and T sleep . 

Figure 1. Staggered sleep/wakeup pattern 

• 802.11 PSM [10] is the sleep/wakeup method. In the approach, when two 

nodes want to communicate, their activity times T active , have to overlap. 

The method uses backoff mechanism for signaling and sampling the channel. 

When a node has a packet to send, it waits random period of time, 

assigned by backoff mechanism, before the action is performed.


Comparing the methods described above, GAF and SPAN can reduce the energy 

consumption 3 times [11], GeRaF 2-3 times [4], ASCENT even more than 5 times [11]. 

In group of sleep/wakeup methods, the energy conservation depends on 

the T active to T ratio. Therefore, it is difficult to estimate the exact energy conservation. 

Definitely STEM method is worse than the rest of the methods, because 

the method requires two communication units. For small T active /T ratio, the energy 

consumption is decreased to 1-2% of the regular energy consumption (without any 

method implemented). 

III. Assumptions 

In our work, we assume that the sensor network is dense and composed with 

MicaZ nodes, equipped with one communication module Chipcon CC2420. Nodes 

positions are known and every node is aware of its neighborhood in 2-hops distance. 

These information can be acquired through procedures introduced in [12]. 

Communication and sensing ranges are modeled as a uniform disks. The radiuses 

of the disks are labeled as r c and r s respectively. The communication radius is not 

greater than the sensing radius. The nodes are placed on a flat area, so the network 

is considered in two dimensional perspective. 

IV. Problem statement 

We would like to consider the network purposed to detect ground moving 

objects and track them afterwards. In this case, it is necessary to monitor only 

the border layer of the network. Accordingly, the devices placed on the network 

edge have to maintain their sensing units active, while the rest of the devices 

can turn them off. They adhere to the set S E (external layer). The nodes localized 

on the network edge, can drowse communication units temporally to 

conserve the energy. The nodes placed inside the network are mainly inactive 

(all modules are switched off) and their control and communication units are 

active only for a short period of time. These nodes belong to the set S I (internal 

layer). The time when some node units are switched off is called T S . The time 

when the units are active and the device is capable of communicating with 

others is called T A . The sum of T A and T S is called cycle duration T. The T A to 

T ratio is called duty cycling: 

T 

DC = A 

100% 

(1) 

T 

It is obvious that the lower value of DC implies the lower energy consumption 

in particular node [12].


337 

The set of all nodes participating in the network structures S is defined as: 

S S S 

(2) 

E 

Obviously the nodes shaping the internal layer consume significantly less 

energy in comparison with the external layer nodes. What is more, it can exist k max 

disjoint sub-layers of environment at the same time. Each external layer node w E 

belongs to only one sub-layer: 

w S S : w S , i 1,..., k 

(3) 

E E Ei E Ei 

where: i is the sub-layer number, i=1,…,k max . Each layer has to follow the consistency 

rule: 

w S d( w , w ) r d( w , w ) r 

(4) 

z 

I 

max 

z Ei z j c z l c 

where: w z is a node, which belongs to the external layer, w j , w l are the closest 

nodes to the node w z , S Ei is the i-th sub-layer, i is the number of the sub-layer, 

i=1,…,k max . The consistency rule guarantees that the distance between any two 

nodes in a sub-layer is lower that the communication range (and the sensing 

range as well). 

The external sub-layers are active interchangeably. This approach allows 

the network to work longer in initial size of deployment. 

In the following chapters we answer the question how to assign the nodes to 

the proper layer and how to assure the consistency in external sub-layers during 

the whole network lifecycle in fully distributed manner. 

Figure 2. One of the sub-layers of external layers


V. Energy consumption 

The overall energy consumption is defined as temporal energy consumption 

of control unit E MC , communication unit E R and sensing unit E S in specified period 

of time Δt: 

te te te 

(5) 

E E dt 

E dt E dt 

t MC R S 

ts ts ts 

where: E Δt is the energy consumption in the analyzed period of time Δt, 

E MC is energy consumption of the control unit, E R is energy consumption of the communication 

unit, E S is the energy consumption of the sensing unit, t s is the start 

time, t k is the end time, Δt = t k -t s . 

The energy consumption of the control unit is defined as: 

EMC TMC IMC 

V 

(6) 

where: T MC is the time of measurement, I MC is a current supply, V is a voltage. 

Obviously, E MC consists of: 

E E E 

(7) 

A S 

MC MC MC 

where: E A MC is energy consumption in active mode and E S MC is energy consumption 

in sleep mode. 

The energy consumption of communication unit is defined as: 

ERTRIR V 

(8) 

where: T R is time of measurement, I R is current supply, V is voltage. On the other 

hand, E R consists of energy consumption in four different states of communication 

module: sleep E S R , listening EL R , transmission ETX R and reception ERX R : 

E E E E E 

(9) 

S L TX RX 

R R R R R 

The energy consumption of the sensing unit is defined as: 

EDTDID V 

(10) 

where: T D is the time, when the communication unit stays active, I D is a current 

supply, V is a voltage. 

VI. Procedures description 

To configure the network structures, self-configured algorithms were developed 

to classify the nodes to the internal layer or one of the external layers. 

Each node w in a startup phase of the network lifetime, checks if the conditions 

described by the equation (11) are met to determine whether the node belongs 

initially to the external layer:


 

 

 

 

w S : x x y y 

k s k k 

w S : x x y y 

k s k k 

w S : x x y y 

k s k k 

w S : x x y y 

k s k k 

 

 

 

 

339 

(11) 

where: S S is a set of neighborhood nodes in 2-hops distance, w k is a node in S S , 

x is x coordinate of the node w, y is y coordinate of the node w, x k is x coordinate 

of the node w k , y k is y coordinate of the node w k . The procedure below demonstrates 

how the sub-layer number one is shaped. The procedure is performed by each node 

in the network. 

Shaping of sub-layer one of external layer 

1: If condition (11) is fulfilled then 

2: label as a sub-layer one node 

3: broadcast the information about sub-layer affiliation 

4: wait Δt wi 

5: If k max > 1 then 

6: broadcast the request for formation the next sub-layer 

7: End-if 

8: Else 

9: label as the internal layer node 

10:End-if 

In this way the first sub-layer of external layer is developed. It was mentioned 

that it can exist k max sub-layers because of the density of the network. The process 

of shaping the next sub-layer is always initiated by the previous sub-layer. The procedure 

of forming next sub-layers is described by the procedure below. 

Shaping of next sub-layer of the external layer 

1: If request for formation the sub-layer i received then 

2: wait Δt wi 

3: If the equation (11) is fulfilled then 

4: label as a sub-layer i node 

5: broadcast the information about sub-layer i affiliation 

6: wait Δt wj 

7: If i+1


In the procedures two timeouts: Δt wi and Δt wj Are used. The first one guarantees 

that the node is waiting long enough to receive the information from all nodes shaping 

previous sub-layer. The second timeout guarantees that the node will wait long 

enough to collect the information about the nodes at the same sub-layer. 

After the sub-layer is established, the correction procedure is performed. The correction 

procedure is responsible for securing the consistency in a sub-layer and switching 

redundant nodes off. The sub-layer consistency is assured by the consistency rule 

(eq. 4). Therefore, the local consistency of the sub-layer between any two nodes w j 

and w l is guaranteed. It means that the node evaluates the consistency rule and selects 

the set S d i of nodes among all neighboring nodes. It can exist several sets Sd i , satisfying 

the consistency rule requirement, but the desired one is with the lowest cardinality. 

To switch the redundant nodes off, the procedure exploits the redundancy rule: 

d( w , w ) r d( w , w ) r d( w , w ) r 

(12) 

z j c z l c j l c 

where: w z is the node checking whether it is redundant, w j and w l are the closest 

nodes to the node w z . The procedure is presented below. 

Sub-layer correction 

1: If the information about creation the sub-layer sent then 

2: wait Δt wj 

3: If the consistency rule is unfulfilled (eq. 4) then 

4: engage additional nodes 

5: End-if 

6: If the redundancy rule fulfilled (eq. 12) then 

7: label as the internal layer node 

8: broadcast the information about the internal layer affiliation 

9: End-if 

10: End-if 

The external layer is composed of one or several sub-layers. Among the sublayers, 

only the one is active (in case of sensing the environment). After certain time, 

the active sub-layer i activates the sub-layer i+1 (or sub-layer 1) and swaps to inactive 

state. After that, only the one sub-layer (number i+1) is still active. 

When the network structures are established, the sub-layer consistency procedures 

run. The procedures are responsible for sub-layers reconfiguration in case 

of the sub-layer inconsistency detection and when the inconsistency is at risk. 

In the first case each node exchange their neighborhood table S ng with the neighbors 

within the communication range r c . This neighborhood tables synchronization take 

place every Δt a . In case one of the closest nodes do not answer in synchronization 

process, and both of them belong to the same sub-layer, the node which has not 

received the information starts the sub-layer reconfiguration. The algorithm below 

describes this situation (from external sub-layer node perspective).


341 

Sub-layer reconfiguration in case of inconsistency detection 

1: If the consistency rule is not fulfilled (eq. 4) then 

2: engage the additional neighbor nodes to assure the consistency of the sub-layer 

3: End-if 

The second group of the reconfiguration procedures run when the inconsistency 

can appear. It can happen when the power source capacity of the sub-layer 

node is at risk. When the node discovers that: 

C t C 

(13) 

Z 

where: C Z (t) is the power source capacity, C G is the border power source capacity, 

it starts reconfiguration process. The algorithm above describes this situation. 

Sub-layer reconfiguration in case of inconsistency detection 

1: If the power source capacity is lower than the border power source 

capacity (eq. 13) then 

2: engage the additional neighbor nodes capable to replaceitself 

3: End-if 

VII. Simulation results 

We evaluated the performance of proposed solution. The model of node used 

in the simulation is previously mentioned MicaZ node, equipped with ZigBee 

protocols stack. The energy consumption parameters of micaZ node are shown 

in table 1and described in more details in [13] and [14]. 

Table 1. The energy consumption of MicaZ node 

G 

Control unit 

Communication unit 

Sensing unit 

Active 

8 mA 

Sleep 

15 μA 

Memory write 

15 mA 

Memory read 

4 mA 

Receive (RX) 

19.7 mA 

Send (TX) (at 0 dBm) 17.4 

Sleep 

1 μA 

Active 

5,5 mA 

Sleep 

0 mA


The simulation was run for three types of nodes deployment: 

A) Nodes with antennas put 7 cm above the ground covered by grass vegetation, 

r c = 10.5 m, the network size is equal 190 m × 190 m, 

B) Nodes with antennas put 30 cm above the ground covered by grass vegetation 

l, r c = 50 m, the network size is equal 730 m × 730 m, 

C) Nodes with antennas put 30 cm above the ground covered by forest vegetation 

l, r c = 30 m, the network size is equal 300 m × 300 m. 

The duty cycling (DC) is set differently for the external and internal layers 

nodes. External layer nodes work with DC = 25% and internal layer nodes work 

with DC = 10%. What is more, during the activity time T A , the external layer nodes, 

actively sensing the environment have control, communication and sensing units 

switched on. During the sleep time T U , the communication module of these nodes 

is switched off, but the sensing function is maintained. The external layer nodes, 

which are not responsible for sensing and the internal layer nodes, retain the control 

and communication units active at time T A . At the sleep time T S , both groups 

of nodes have all units switched off. 

Table 2 shows the number of nodes in the network with external layer consisting 

of three sub-layers in comparison with the number of all nodes in the network. 

The result is split by deployment type and the sub-layer number. 

Table 2. The reduction of active nodes 

Deployment type Sub-layer number # of sub-layer nodes # of all nodes 

A) 1 80 

B) 2 83 

1474 

C) 3 82 

A) 1 62 

B) 2 60 

920 

C) 3 63 

A) 1 41 

B) 2 43 

412 

C) 3 42 

The average energy consumption of the nodes is shown in table 3. 

As a result, the average lifetime of the external layer nodes is about 21 days 

(having three sub-layers switching every one hour) and internal layer nodes lifetime 

is about 100 days. 

Figure 3 depicts the lifetime of the networks: without any topology control 

mechanism implemented, one of the node activity control mechanism presented 

in section II and the topology control procedure introduced in the article. The approaches 

are evaluated according to two criterions: global lifetime and the time 

while the network consistency is guaranteed. In our approach the internal layer


343 

nodes consume significantly less energy in comparison with the external layer 

nodes. When the external layer nodes run out of the energy, they are replaced by 

the internal layer nodes. In the earlier mentioned methods, the active nodes cover 

the whole network. Therefore, after certain period of time some active nodes run 

out of energy. It means, it is possible that some uncovered areas inside the network 

will appear. 

Table 3. Average energy consumption in mWh 

Active sub-layer node Inactive sub-layer node Internal layer node 

25.2 5.1 2.15 

Figure 3. The comparison of networks lifetimes 

We also checked the reconfiguration procedures in the simulations (if do 

they work and how effective, from the reconfiguration time perspective, they are). 

The average times needed to reinstate the sub-layers t rec are evaluated for either: 

reconfiguration in case of inconsistency detection and when the inconsistency 

is jeopardized. The result is shown on picture 4. 

Figure 4. The time of a sublayer reconfiguration


Maximal time the sub-layer can be inconsistent is equal to: 

t 

t sec 

(14) 

a 

rek 

where: Δt a is the time when the nodes exchange the information about their 

neighborhood (mentioned above), t rek is the time necessary to accomplish the reconfiguration. 

Assuming the Δt a time equal to 180 sec, the time of the sub-layer 

inconsistency is round 181.5 sec. 

Demonstrated layer configuration elongate the network lifetime in the initial 

size of deployment. The simulation experiment showed that the network shrinks 

to inside at the borders (due to higher energy consumption of the external layer 

nodes). As a result, after 100 days of work, the network size is reduced by r c meters 

at each border. 

VIII. Conclusion and future work 

The article demonstrates the self-configured, energy aware method of topology 

control. The procedures of network structures establishment and maintenance 

are provided and explained. The methods are evaluated in terms of energy consumption, 

the network lifetime and ability to guarantee the sub-layers consistency 

and repair. The network shaped in this way is capable of detecting object 

moving on the ground. It gives excellent basis for the creation of such a system, 

because the coverage consistency and maintenance are assured. What is more, 

the network lifetime is significantly improved, even 10 times, in comparison with 

a network without any topology control mechanism. The value of DC parameter 

is adjusted to the network purposes. The greater values of DC parameter used 

in the simulation has no serious impact on the communication effectiveness. 

Evidently, there is a possibility to use lower values for DC parameter, but it would 

degrade the communication performance and could have an impact on the detection 

operation. 

There exist still many areas to explore within this research engagement. 

It would be worth to check the network performance in different types of the nodes 

deployment and environments of deployment. Verification the ability of the network 

to objects tracking is another very desired research which we plan to practice 

in near future.


345 

References 

[1] C. Patel, S.M. Chai, S. Yalamanchili, and D.E. Schimmel, Power/Performance Tradeoffs 

for Direct Networks. In Parallel Computer Routing & CommunicationWorkshop, 

pages 193-206, July 1997. 

[2] Y. Xu, J. Heidemann, D. Estrin, Geography informed Energy Conservation for Ad Hoc. 

Proc. ACM MobiCom 2001, pp. 70-84. Rome, 2001. 

[3] P. Vasari, A. Marcucci, M. Nati, C. Petrioli, M. Zorzi, A Detailed Simulation 

Study of Geographic Random Forwarding (GeRaF) in Wireless Sensor Networks. Proc. 

of Military Communications Conference 2005 (MILCOM 2005), vol. 1, pp. 59-68, 

17-20 Oct. 2005. 

[4] M. Zorzi, R.R. Rao, Geographic Random Forwarding (GeRaF) for Ad Hoc and Sensor 

Networks: Multihop Performance. IEEE Transactions Mobile Computing, vol. 2, no. 4, 

pp. 337-348, 2003. 

[5] B. Chen, K. Jamieson, H. Balakrishnan, R. Morris, Span: An Energy-Efficient 

Coordination Algorithm for Topology Maintenance in Ad Hoc Wireless Networks. ACM 

Wireless Networks, vol. 8, no. 5, September 2002. 

[6] A. Cerpa, D. Estrin, Ascent: Adaptive Self-Configuring Sensor Network Topologies. 

Proc. IEEE INFOCOM 2002. 

[7] C. Schurgers, V. Tsiatsis, M.B. Srivastava, STEM: Topology Management for 

Energy Efficient Sensor Networks. IEEE Aerospace Conference ’02, Big Sky, MT, 

March 10-15, 2002. 

[8] J. Ansari, D. Pankin, P. Mähönen, Radio-Triggered Wake-ups with Addressing 

Capabilities for Extremely Low Power Sensor Network Applications. Proc. of the 5th 

European conference on Wireless Sensor Networks (EWSN 2008), Bologna (Italy), 

Jan. 30 – Feb. 1, 2008. 

[9] A. Keshavarzian, H. Lee, L. Venkatraman, Wakeup Scheduling in Wireless Sensor 

Networks. Proc. ACM MobiHoc 2006, pp. 322-333, Florence (Italy), May 2006. 

[10] IEEE 802.11, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer 

(PHY) Specifications. [online], document dostępny na: http://standards.ieee.org/about/ 

get/802/802.11.html 

[11] A. Arora, Topology Control. [online], dokument dostępny na: www.cse.ohio-state. 

edu/~anish/.../Lecture5.ppt 

[12] M. Wawryszczuk, The Method of Self-organizing Wireless Sensor Network 

Implementation with Extended Durability (in Polish). PhD thesis, Military University 

of Technology, September 2011. 

[13] MicaZ Datasheet, [online]. The article accessible on: https://www.eol.ucar.edu/rtf/ 

facilities/isa/internal/CrossBow/Doc/MPR-MIB_Series_User_Manual_7430-0021- 

05_A.pdf 

[14] Xmesh User’s Manual, [online]. The article accessible on: http://www.memsic. 

com/support/documentation/wireless-sensor-networks/category/6-user-manuals. 

htmldownload=95%3Axmesh-user-s-manual

Chapter 8 

Localization Techniques

Enhanced Location Tracking 

for Tactical MANETs Based on Particle Filters 

and Additional Information Sources 

Peter Ebinger *1 , Arjan Kuijper 2 , Stephen D. Wolthusen 3 

1 AGT Group (R&D) GmbH, Darmstadt, Germany, pebinger@agtinternational.com 

2 Fraunhofer IGD, Darmstadt, Germany, arjan.kuijper@igd.fhg.de 

3 Information Security Group, Department of Mathematics, Royal Holloway, U. of London, 

Norwegian Information Security Lab., Gjøvik University College, Norway, 

stephen.wolthusen@rhul.ac.uk 

Abstract: The networking capabilities of tactical mobile ad-hoc networks (MANETs) provide the basis 

to enhance robustness and accuracy of Blue Force Tracking (BFT) where existing BFT mechanisms are 

unavailable, unreliable, or simply not sufficiently accurate (owing to factors such as update frequencies 

and the need for back-link availability). BFT is an essential element to any tactical environment given 

its ability to contribute to situational awareness at all levels. 

Tactical environments are characterized by spectrum contention, jamming and other factors limiting 

the ability of naïve approaches, e.g. in urban environments and broken terrain. Unlike previous work 

this paper aims to provide MANET-based BFT without the requirement of line-of-sight (LOS) links 

or back-end infrastructure which is robust against temporal disruption of network connectivity. These 

results are achieved by distributedly fusing sensor data and additional information sources across 

the tactical MANET using techniques also employed in robotics and object tracking. 

Our contribution is the provision of enhanced BFT mechanisms exploiting networking capabilities 

of tactical MANETs and data fusion mechanisms based on Sequential Monte Carlo methods, specifically 

particle filters, incorporating additional information such as mission information (e.g. mobility 

models) and topographic data. We demonstrate that the use of these techniques enhances both accuracy 

and robustness as compared to standard BFT by using a simulation environment with various 

mobility and radio propagation characteristics. 

Keywords: Mobile Ad hoc Networks, Location Tracking, Blue Force Tracking 


Blue Force Tracking (BFT) is a tactical term for a localization system based 

on Global Positioning System (GPS) sensors with the objective to provide location 

estimates of friendly military forces (blue forces). BFT is an essential element to 

*Work performed while working at Competence Center for Identification and Biometrics, Fraunhofer IGD, 

Darmstadt, Germany.


any tactical environment given its ability to contribute to situational awareness at 

all levels [1]. Decreasing the friction in complex geopolitical contexts, particularly 

in urban environments, may help to support decision-makers and lower the risk 

of errors [2]. This has not only been incorporated into the doctrine of modern 

forces’ in support of Command and Control (C2), but has also resulted in an overall 

reduction of causalities [3], [1]. 

There are several challenges to BFT that render it a complex problem: Nodes 

(military units) are typically scattered on the operation area, the environment 

changes dynamically due to node mobility and operations need to be coordinated 

not only within the armed forced of a specific country, but also with allies. For 

rapidly evolving tactical situations in confined areas (such as those encountered 

in urban areas) existing mechanisms (e.g. satellite-based systems) are limited by 

several factors including update frequency, accuracy, and robustness to communication 

channel disruption (due to jamming or capacity limitations). 

Satellite-based BFT systems highly depend on the availability of a complex 

infrastructure and backend. They are therefore vulnerable to attacks on the global 

infrastructure, e.g. jamming or denial of service (DoS) attacks. BFT systems based 

on mobile ad hoc networks (MANETs) that are locally deployed reduce the dependencies 

and can improve this situation. 

MANETs in conjunction with sensors and information sources found 

in mobile computing components provide redundant, mutually supporting 

information on individual node locations. Data sources comprise geolocation 

(e.g. GPS) receivers, electronic compasses and gyroscopes, but also the ability 

to perform trilateration based on radio-frequency signals used for communication 

[4]. The types of terrain and radio frequency environments to be expected 

in tactical MANETs implies that one cannot assume continuous connectivity 

across the MANET, but rather that the MANET will be partitioned frequently 

in an arbitrary manner [5]. 

Although MANET-based BFT systems have been proposed [6], [7], these are 

typically limited to basic communication mechanisms that do not provide additional 

means to increase the robustness and the accuracy of BFT and are therefore 

inappropriate for challenging environments, e.g. urban operations or similarly 

constrained terrain. 

In this paper we propose a mechanism which can be used as a supplement 

or even substitute for existing infrastructure-based BFT systems but which is robust 

against the aforementioned problems affecting naïve designs for MANETs or 

similar mesh-type networks. Robustness and accuracy of location estimation are 

increased within the proposed BFT system utilizing additional information sources 

such as mission information (group structure and tactical mobility patterns) and 

topographic data. Employing a combination of mobility models adapted to the tactical 

domain, the interpolation of likely node positions in the absence of immediate 

updates becomes possible.

Chapter 8: Localization Techniques 

351 

It is important to note that a key requirement for robust BFT in the tactical 

domain is that each node should retain information on current and likely positions 

of other network nodes. Our proposal outlined in this paper employs Sequential 

Monte Carlo (SMC) methods in a distributed manner to achieve this purpose. 

In this environment Particle Filters (PF) are particularly suitable as they provide 

a natural, scalable mechanism for mutual updates of location and motion estimations 

which decay gracefully in cases of missing updates [8]. Equally important 

is their high tolerance level regarding faulty or deliberately incorrect data whilst 

permitting the inclusion of additional information. 

In the remainder of the paper we therefore briefly outline related work on BFT 

and geolocation tracking and review background terminology for particle filters. 

This provides the basis for discussion on several assumptions used in the design 

of the BFT algorithm. An elaboration of the proposed particle filter algorithm 

(SMC-BFT), first conceptually and then at the algorithmic level, precedes a discussion 

of a simulative evaluation of the algorithm based on the network simulator 

developed by [9]. A comparison to a naïve baseline algorithm and a global motion 

estimation algorithm concludes the paper, demonstrating significantly improved 

results of the outlined algorithm with respect to the average estimation error. 


Riblett and Wiseman [6] present a MANET-based system for tactical environments 

named TacNet. They point out that a MANET-based communication 

approach provides the capabilities to the network to dynamically self-heal and 

to overcome some line-of-sight constraints that are typical limitations of radio 

networks. The TacNet system provides secure access to critical data such as realtime 

maps of resource positions and could be used as an underlying communication 

network for Blue Force tracking. However, no enhanced data fusion and 

location estimation mechanisms are proposed and no evaluation of the system 

performance is given [7]. 

Suri et al. [1] describe how BFT can benefit from a distributed approach 

in comparison to centralized architectures due to the dynamic, geographically 

sensitive nature of most tactical situations. They present observations from several 

tactical networking experiments and discuss the requirements for and the advantages 

of peer-to-peer (P2P) approaches for tactical network environments. Even 

in cases where connectivity to a central BFT server is lost, nodes in close proximity 

continue to communicate in order to share critical location information. 

A number of research papers have been published about localization schemes 

for MANETs, in particular, for scenarios where no GPS sensors are available or where 

only a subset of (anchor) nodes are equipped with a GPS sensor. Biaz and Ji [10] 

present a survey and comparison on localization algorithms for wireless ad hoc networks, 

e.g. based on radio signal characteristics such as time of arrival (TOA), time


difference of arrival (TDOA), angle of arrival (AOA) and received signal strength 

indicator (RSSI). However, most of these approaches address a different scenario 

in comparison to BFT as in tactical MANETs it is typically assumed that all nodes are 

equipped with GPS sensors. The challenge of BFT is not to determine the location 

of a local node but to have accurate location estimations for all other network nodes. 

Some MANET localization approaches based on SMC are presented in the following, 

although they also address different scenarios in comparison to BFT. Hu 

and Evans [9] introduce a SMC localization method for a MANET setup where only 

a small number of seed nodes know their locations. The other nodes estimate their 

location based on location messages that they receive from seed nodes. The evaluation 

results of the SMC-based approach are promising, even in scenarios where 

network transmissions are highly irregular and severe computation and memory 

limits are in place. Huang et al. [11] present a MANET localization framework 

based on particle filters based on AoA and RSSI. They study how different types 

of sensor data can be incorporated into a common localization framework and 

claim that they provide the first localization framework based on particle-filters 

that integrates heterogeneous sensor types. 

Ristic et al. [12] outline the concept of Terrain-Aided Tracking. They propose 

to exploit knowledge of the environment or limitations in the dynamic motion 

of the target to produce more accurate location estimation results. They show 

an example how additional information sources can be used in order to improve 

the accuracy of state estimation based on particle filters. However, they do not address 

the communication and distributed processing challenges of a MANET scenario. 

Rosencrantz et al. [13] present a decentralized state estimation architecture 

based on particle filters for a robotic system. They address the problem that 

arises in a distributed estimation systems if observation arrive that refer to a time 

in the past. They introduce the concept of re-simulation in order to reduce the variance 

of importance weights. In cases of re-simulation part of the history of an agent’s 

particle filter is erased and then re-simulated to the current time, incorporating all 

collected observations at appropriate time intervals. 

III. Background: Sequential Monte Carlo models / particle filters 

In this section we outline the mathematical background of state estimation 

using particle filters [8] – also known as sequential Monte Carlo (SMC) methods 

– and how they can be applied to our application scenario: 

A. System state 

Current properties and status of a tactical MANET can be described using 

a set of state variables. For simplicity the number of nodes in the MANET 

is assumed to be fixed and denoted as .


353 

Notation: The current state of a node at time is denoted as 

The state is (at least partially) not directly observable but can be inferred from 

sensor data. The measurements (related to target state) at time are denoted as 

. The sequence of all available measurements up to time is denoted 

as z1:k 

{ z 

t 

, t1, , k} 

. 

The posterior probability distribution function (pdf) regarding all available 

measurements up to time is denoted as p( xk| z 

1: k) 

and is also known 

as filtering distribution. 

State information including BFT tracking location and other information 

is collected, evaluated and transmitted to other network nodes. The current state 

pos 

vel 

information at time includes the location x 

k 

and the velocity x 

k 

of all other 

nodes. Sensors estimate the GPS coordinates of a node and the speed and direction 

of movement. 

B. State evolution and estimation problem 

Assumptions The target state evolves according to a discrete-time stochastic 

model defined by the following equation: 

x f ( x , v ) 

(1) 

k k k1 k1 

nx n nx 

where f : 

k 

is a known, possibly nonlinear function of the state 

and and a process noise sequence. 

Objective: The overall objective of our architecture is to recursively estimate 

based on measurements that are related to the target state via the measurement 

equation: 

z ( , ) 

k 

hk xk 

w 

k 

(2) 

nx n nz 

where hk : is a known, possibly nonlinear function of the state 

and the measurement noise sequence . 

From a Bayesian perspective the objective is to recursively quantify some 

degree of belief in the state at time given data z 

1:k 

up to time . Thus it is 

required to construct the posterior pdf p( xk| z 

1: k) 

which can be obtained recursively 

in two stages: prediction and update. 

C. Particle filters 

In the following a short overview of the basic concept of particle filters including 

importance (re-)sampling is given [14], [12]. The key concept of particle filters 

is to represent probability density functions by a set of samples (also referred to 

as particles) and their associated weights. The pdf p( xk| z 

1: k) 

is therefore approximated 

with an empirical density function:


N 

N 

i i 

i i 

k 1: k qk 

k k 

qk i qk 

i1 i1 

p( x | z ) x x , 1, : 0 

(3) 

where is the Dirac delta function and denotes the weight associated with 

particle . 

Monte Carlo Integration: Monte Carlo (MC) integration is the basis for SMC 

methods. Its objective is the numerical evaluation of a multidimensional integral: 

I g g x 

dx 

(4) 

where . 

Monte Carlo methods factorize gx 

f x x 

in such a way that x 

is interpreted as a probability density satisfying x 0 and x, dx1. 

i 

The assumptions is that it is possible to draw samples { x , i1, , N} 

distributed according to x 

. An approximation of the integral 

is the sample mean: 

I g f xxdx 

(5) 

N 

1 

I g f x . 

(6) 

N 

i 

 

N 

i 1 

Importance Sampling: Ideally we want to generate samples directly from x 

and estimate using equation (6). Suppose we can only generate samples from 

a density qx which is similar to x 

, then a correct weighting of samples 

makes SMC estimation still possible. The pdf qx is referred to as the importance 

or proposal density, where qx and x 

have the same support. Equation (5) 

can be rewritten as: 

x 

I gf xxdx f x 

qxdx 

(7) 

qx 

x 

provided that is upper bounded. 

qx 

A Monte Carlo estimate of can be computed by generating independent 

samples { x , i1, , N} 

distributed according to qx and forming 

i 

the weighted sum: 

are the importance weights. 

N 

i 

1 

{ x 

i i i 

I 

N 

g f { x q { x , q { 

x 

i 

(8) 

N q { x 

i1


355 

A problem for consideration – in particular for distributed and cooperative 

state estimation – is that measurements can only be combined if they refer 

to the same state, and therefore the same instance in time. Only simultaneously 

measured observations can be directly combined. In the following section we 

introduce our novel SMC-based BFT that overcomes these issues using resimulation 

whenever measurements need to be combined that refer to different 

moments in time. 

IV. SMC-Based BFT (SMC-BFT) 

In the following we describe some basic assumptions and present some examples 

of additional information that can be incorporated in order to enhance BFT 

in tactical MANETs. We show how re-simulation can be used to apply particle 

filters in distributed environments and finally present the concept of the proposed 

SMC-based BFT (SMC-BFT) in detail. 

A. Basic assumptions 

The overall objective of our proposed SMC-BFT approach is to derive a probabilistic 

estimation of a node’s location and velocity for enhanced BFT. The posterior 

distribution for is represented by a set of weighted samples { x , i 1, , N} 

which are periodically updated using importance sampling. Time is divided into 

discrete time units (as described in section III) where the current state of a node 

at time is denoted as . Node movement and messages exchange are modeled 

as discrete steps within each time unit. 

SMC-BFT State Evolution Model: Location and velocity state variables evolve 

according to a discrete-time stochastic model defined by equation (1) where the system 

state variable represents both node location x 

loc 

vel 

k 

and velocity x 

k 

. 

GPS Measurement Equation: Sensor measurements of GPS sensors are modeled 

according to equation (2) where the measurement represents both location 

loc 

vel 

measurement z 

k 

and velocity measurement z 

k 

of the GPS sensor. The measurements 

noise sequences are modeled as Gaussian white noise with 

mean zero and standard deviations and respectively. 

loc 

GPS 

vel 

GPS 

B. Incorporation of additional information sources 

An important tenet of the SMC-BFT concept is the inclusion of additional 

knowledge sources such as information about the tactical mission and a topographical 

model of the environment. 

Mission Information: Knowledge of mission objectives, participants and group 

structure of a specific mission are promising information sources for improving BFT. 

Typical formation patterns and typical (minimum, average and maximum) velocities


can be derived from this form of information. Such properties may either belong 

to specific nodes or reflect a specific tactical situation. 

In section V-C2 an example of how group structure and movement formation 

information can be used for enhanced BFT (Reference Point Group Mobility Model). 

Topographical Model: A topographical model of the environment can be used 

to improve estimation of node movement. The direction of movement is probabilistically 

dependent on properties of the area of movement [12]. There are areas with 

restricted movement (e.g. water, mountains), areas of free movement (e.g. open 

field) and areas which are characterized with a high likelihood of following specific 

paths (e.g. roads, pathways, or stairs). 

In section V-C2 an example of how such information can be used to significantly 

enhance BFT in a setup where nodes follow specific paths or streets (Manhattan 

Grid Model) is demonstrated. 

Additional information sources enable the preference of more likely state 

values and the exclusion of implausible estimations. For example, impossible 

locations can be removed from the set of potential locations based on node type 

and topography model, e.g. lakes or buildings for regular cars. Additionally impossible 

velocity values can be excluded based on node properties and physical laws, 

e.g. based on upper bounds for acceleration. 

C. Re-simulation 

An important aspect of distributed data fusion is their temporal and causal 

relationship: how can measurements that refer to different nodes be combined and 

integrated if they were performed at different times 

The sequence of events must be considered in order to arrange measurements 

of different nodes in a distributed system. Observations retrieved from other nodes 

at a specific time may refer to a moment that is several steps in the past, each node 

therefore records current measurement values as well as a history of received and 

local measurements. This allows node’s particle filter to periodically re-simulated 

when data arrives at time that refers to an earlier time t l 

t k 

(cf. [13]), which 

is more recent than previously received measurements. 

When this occurs a section of the filter’s history is erased and re-run to the current 

time , incorporating all collected observations at appropriate intervals. Measurements 

that refer to a moment in the past that are outside the history window 

( tk tl thistory 

) are discarded. This leads to improved re-sampling of particles 

and improved performance of the particle filter. 

D. SMC-BFT concept 

The basic concept of SMC-BFT is that each network node utilizes a set of particle 

filters to probabilistically model the location information of the other network


357 

nodes. Each particle filter is initialized when first measurement arrive and continuously 

updated afterwards. 

The following actions performed in each round, on each node for each 

of the other nodes: 

1) Initialization (when first measurements arrive) 

2) Prediction and Filtering 

An overview of these actions is shown in figure 1 and the individual steps 

of each action are described in more detail in the following sections. 

1) Initialization: The particle filter for BFT of another node is initialized at time 

when the first measurements (referring to some time ) of that specific 

node arrive: 

i 

• Generation of samples x 

l 

, i1, , 

N 

equally distributed around newly 

available measurements , incorporating additional information sources, e.g. 

– topography model for deletion of invalid samples or alignment of samples, 

– mission information for group movement patterns. 

• If required ( t l 

t k 

): Re-simulation from time to current time for 

this node based on system evolution model (cf. equation (1)), incorporating 

additional information sources. 

Figure 1. Basic Concept of SMC-BFT


2) Prediction and Filtering: All newly arriving incoming location messages are 

preliminarily processed. The output of this process is a set of new measurements 

(referring to some time t 

l 

tk 

in the past). For some nodes new 

measurements are available in a specific round and for other nodes not. 

Based on the availability of new measurements for another network node 

in a specific round the BFT particle filter for a specific node is reset to a moment 

in time tl 1in the past or not. Subsequently the system state is updated according 

to the procedure described below. The basic input for the prediction and filtering 

i 

step is the set of samples x , 1, , 

l1 i N 

representing a probabilistic estimation 

of the location and velocity of another node at a specific time tl tk 

(or if no new 

measurements are available is equal to ). 

• System update based on previous system state estimation x 

l1 

and system 

evolution model (cf. equation (1), incorporating additional information source 

(similar to “re-simulation” of one step). 

• If new measurements are available: 

1) Elimination of impossible system states: Remove all samples that are invalid 

due to newly arrived measurement , e.g. that are 

– too far away from the measured location, 

– in an invalid location according to a topography model or 

– too far away from the center of a group formation. 

2) If not enough samples are left: Generation of new samples x 

i 

l 

, i1, , 

N 

equally distributed around the newly available measurement , incorporating 

additional information sources. 

3) If required ( t l 

t k 

): Re-Simulation from time to current time for this 

node based on system evolution model (cf. equation (1)), incorporating 

additional information source. 

The goal of re-simulation is to generate new samples for the current time 

x i 

, i k 

1, , N by probabilistically updating all previously calculated samples 

x j, l jki , 1, , 

N 

stored in the history list up to estimation time , 

i 

e.g. applying the mobility model and incorporating mission information and 

topographical model. 

V. Evaluation 

The key metric of BFT is the accuracy of location estimation. Estimation errors 

should be minimized while limiting communication overhead to an acceptable level. 

In this section we evaluate how our proposed SMC-BFT mechanism performs 

and its robustness with respect to various network parameters in comparison to 

standard BFT approaches.


359 

A. Evaluation setup 

A MANET simulator developed in [9] is used as a basis for evaluation (also 

used and extended in [15]). The basic setup is that each mobile node contains 

a computing device, has GPS sensor capability and topography model and mission 

information are available on each node. 

Nodes regularly exchange data (geo-location, velocity) to direct neighbors 

via multi-hop flooding with all other nodes (e.g. piggyback with routing information). 

Location measurement and message processing are performed in discrete 

time steps (round based simulation) where the duration of each step was set to 1 s. 

B. Standard BFT approaches 

We compare our newly proposed SMC-BFT method with two standard BFT 

mechanisms: A simple straight forward BFT mechanism, which simply takes 

a measurement “as is” and a more elaborate mechanism based on dead reckoning 

(cf. [16]) considering the progress in time between last measurements and current 

time as well as node velocity. 

a) Standard BFT (STD-BFT): The most recently received measurement , (GPS 

coordinates and velocity) is taken as an estimate of current location and velocity 

of a node: 

x z 

k 

b) Velocity Based BFT (VEL-BFT): The location accuracy is improved by predicting 

the current location based on available measurement data (cf. [2],[17]). 

Previous node location, time difference t tk 

t between measurement 

time and current time as well as node velocity are considered: 

pos pos vel 

x z t 

z 

k l l 

x 

vel 

k 

z 

l 

vel 

l 

C. Experimental setup and simulation parameters 

In the following sections we describe the default parameter set used for our 

evaluation. It is explicitly mentioned in the description whenever default values 

are not used. 

1) Basic Parameters: The default setup contains 20 nodes that move within a simulation 

area of 1000 x 1000 m. Nodes locally determine their location and 

velocity using the GPS sensor. The GPS measurement standard deviation for 

loc 

vel 

location is GPS 

3.0 m and for velocity and GPS 

1.0 m/s. Every 10 s each 

node generates a location message and forwards it to all its neighbor nodes.


Each simulation is carried out for 250 s (i. e. 250 rounds) of which the first 50 s 

are left out as initialization and stabilization period and therefore are not considered 

for the evaluation. Each simulation setup is run in 50 iterations to determine 

stable average values. The number of particles for the SMC-BFT approach is 50. 

2) Mobility Model: For the evaluation we implemented a mobility model reflecting 

the topography of the environment and a group based model reflecting 

tactical formations. 

Manhattan Grid Model: The Manhattan Grid Model [18] is based on road 

topology. Roads are located in a grid structure and nodes move in horizontal or 

vertical directions on these roads. The model follows a probabilistic approach: each 

node probabilistically chooses at each intersection to keep moving in the same 

direction or to turn left or right. 

For the evaluation we implemented the Manhattan Grid Model based on 

the BonnMotion [19] algorithm. This model is used as default mobility model for 

the evaluation. There are 10 blocks in each direction (i. e. each block is 100 x 100 m) 

and the probability for mobile nodes to turn at a crossing is 0.5. The probability 

for a node to change its speed (in a simulation round) is 0.2. The minimum speed 

is 0.5 m/s, the mean 2.0 m/s and the standard deviation of a normally distributed 

random speed 2.0 m/s. The pause probability is zero. 

Reference Point Group Mobility (RPGM): The Reference Point Group Mobility 

(RPGM) Model [20], [21] can represent tactical relationships among a group 

of mobile nodes. Applications of the RPGM model include complex scenarios such 

as a military maneuver with joint aircraft, tank and infantry operations or two-level 

scenarios, e.g. infantry and helicopters, with slow and fast nodes. 

For the evaluation the following default parameter set is used. There is a group 

of 20 nodes that move with a common average speed. The minimum group speed 

is 4 m/s and the maximum 10 m/s. The overall group movement follows a random 

waypoint mobility model. Each node chooses a destination randomly distributed around 

a group destination in a distance of up to 25 m from this specified location. In each 

step a node may also randomly move up to 30 percent of this maximum distance. 

3) Radio Propagation Model: During evaluation a deterministic and a probabilistic 

radio propagation model are used. 

Free Space Model: The Free Space Model is a basic deterministic model where 

only line-of-sight radio transmissions through free space (without any obstacles, 

reflection or diffraction disturbances) are taken into account. For the evaluation 

we use the available simulator implementation [9]. 

Shadowing Model: In reality radio propagation is a probabilistic process due 

to multipath propagation effects. The shadowing model consists of two distinct 

parts: a path loss model (which predicts the mean received power) and a second 

that reflects the variation of the received power at a certain distance. 

For the evaluation an implementation based on ns-2 [22] is used. This model 

is used as default radio propagation model with a radio transmission range of 250 m.


361 

The pathloss exponent is set to 4.0 (recommendation for outdoor environment/ 

shadowed urban area: 2.7 to 5 [22]) and a shadowing deviation 

dB 

8.0 is used 

(recommendation for outdoor environments: 4 to 12 [22]). 

D. Accuracy of location estimation 

40 

Field Size (Node Number, Node Density) 

Estimation Error [m] 

30 

20 

10 

STD-BFT 

VEL-BFT 

SMC-BFT 

0 

250 500 750 1000 1250 1500 

Field Size [m] 

Figure 2. Average Estimation Error vs. Node Density for Manhattan Grid Model 

1) Field Size, Node Number and Node Density: In this evaluation setup we increase 

the field size (incrementally from 250 x 250 m to 1500 x 1500 m) proportional to 

the number of nodes (5 to 30). Therefore node density decreases with increasing 

field size and node numbers. Figure 2 shows that the estimation error stays on 

a similar level as long as node density provides some basic connectivity of all 

nodes, but decreases when network connectivity is disturbed. The SMC-BFT 

method outperforms other methods in all setups by 16 to 41 percent. 

2) Mobility Model Parameters: In figure 3 and figure 4 simulation results are 

shown for different speed settings for Manhattan Grid and RPGM models 

respectively. The results for both mobility models show that STD-BFT works 

well for scenarios with low mobility. 

40 

Node Speed 


30 

20 

10 

STD-BFT 

VEL-BFT 

SMC-BFT 

0 

1 2 3 4 

Average Speed [m/s] 

Figure 3. Average Estimation Error vs. Node Speed for Manhattan Grid Model


40 

Group Speed 


30 

20 

10 

STD-BFT 

VEL-BFT 

SMC-BFT 

0 

2 4 6 8 

Maximum Speed [m/s] 

Figure 4. Average Estimation Error vs. Group Speed for RPGM Model 

In figure 3 the minimum speed for the Manhattan Grid Model is set to 0.5 m/s 

for speeds below 3 m/s and to 1 m/s otherwise. Mean speed as well as the standard 

deviation are increased from 1 m/s to 4 m/s. Estimation errors grow relatively 

proportional to speed. Average estimation errors of SMC-BFT are 23 to 48 percent 

lower than for both standard BFT methods. 

In figure 4 maximum group speed of RPGM is increased from 2 m/s to 8 m/s, 

minimum group speed is increased proportionally from 0.5 m/s to 2 m/s. Average 

estimation errors of SMC-BFT are 12 to 55 percent lower than for the other methods. 


150 

125 

100 

75 

50 

Radio Range and Radio Propagation Model 

STD-BFT: Free Space 

VEL-BFT: Free Space 

SMC-BFT: Free Space 

STD-BFT: Shadowing 

VEL-BFT: Shadowing 

SMC-BFT: Shadowing 

25 

0 

150 200 300 400 

Radio Range [m] 

Figure 5. Radio Propagation Model and Radio Range 

3) Radio Propagation Model Parameters: The following setup evaluates the radio 

propagation models influence on BFT performance. For these purposes two 

radio propagation models (Free Space and Shadowing) are used with varying 

radio ranges from 150 m to 400 m where radio range is the maximum transmission 

range for the Free Space model and the average transmission range for 

the Shadowing model. Figure 5 shows that error estimation is significantly lower


363 

for the Shadowing model. When even a few messages are transmitted beyond 

“normal” (Free Space) radio range it significantly improves location estimation 

due to an increase in valuable information that would otherwise not be available to 

that part of the network. The results show that the average estimation error is lower 

for SMC-BFT for all evaluated setups than for both standard BFT methods. 

20 

Accuracy of GPS Messearument 


15 

10 

5 

STD-BFT 

VEL-BFT 

SMC-BFT 

0 

2 3 4 5 

Standard Deviation of GPS Measurements [m] 

Figure 6. GPS Measurement Accuracy 

4) GPS Measurement Accuracy: Accuracy of sensor measurements has a direct 

impact on BFT performance. Figure 6 shows results for varying GPS precisions 

(GPS standard deviation 2 m to 5 m). Estimation error increases are relatively 

proportional to GPS accuracy where the estimation error of the proposed 

SMC-BFT is 25 to 38 percent than for the other methods. 

Table I. Number of Messages and Average Estimation Error vs. Message Generation Interval 

Message Generation Interval 

5 s 10 s 20 s 

Average Number of Generated Messages 481 240 120 

Average Number of Forwarded Mess. 33512 16576 8488 

Average Estimation Error 

STD-BFT 8.4 m 13.7 m 24.5 m 

VEL-BFT 6.7 m 10.9 m 20.9 m 

SMC-BFT 4.6 m 8.1 m 15.0 m 

E. Communication overhead 

In this section we evaluate BFT communication overhead and the resulting 

location estimation errors in dependency of message generation intervals. As message 

generation, forwarding and processing do not depend on the data fusion 

algorithm, the number of messages that are generated and exchanged are equivalent 

for all three methods.


Table I shows that the quantity of messages significantly decreases when location 

message generation time intervals are increased. Similarly location estimation 

error increases proportionally to a decrease in message generation intervals where 

SMC-BFT outperforms the other approaches in all scenarios by 25 to 45 percent. 

VI. Conclusion and outlook 

In this paper we demonstrated how BFT can be significantly enhanced 

by using networking capabilities of tactical MANETs and data fusion based on 

particle filters. Using advanced data analysis and fusion mechanisms for multiple 

sensor and data sources based on techniques that are employed in robotics and 

object tracking, we developed a model for enhanced BFT in tactical MANETs by 

incorporating additional information such as mission information (e.g. mobility 

models) and topographic data. As shown in the simulation environment using 

various mobility and radio propagation characteristics (cf. figure 2 to 6), our 

proposed SMC-BFT model enhanced both accuracy and robustness as compared 

to existing models. 

This paper has outlined mechanisms that can be used to implement more accurate 

and robust BFT systems for tactical MANETs. Our proposed model can be 

used as a replacement if no satellite infrastructure is available, or as a supplement 

to an existing backend system thereby improving local precision while maintaining 

a coarse global overview, e.g. using a satellite system. 

References 

[1] N. Suri, G. Benincasa, M. Tortonesi, C. Stefanelli, J. Kovach, R.Winkler, 

U.S. Kohler, J. Hanna, L. Pochet, and S. Watson, “Peer-to-Peer Communications 

for Tactical Environments: Observations, Requirements, and Experiences,” IEEE 

Communications Magazine, vol. 48, pp. 60-69, 2010. 

[2] P. Labbe, L. Lamont, Y. Ge, and L. Li, “Creating a Dynamic Picture of Network 

Participant Geospatial Information in Complex Terrains,” in Proceedings of the Second 

International Conference on Internet Monitoring and Protection. Washington, DC, 

USA: IEEE Computer Society, 2007, pp. 39. 

[3] K.R. Chevli, P.Y. Kim, A.A. Kagel, D.W. Moy, R.S. Pattay, R.A. Nichols, and 

A.D. Goldfinger, “Blue Force Tracking Network Modeling and Simulation,” in IEEE 

Military Communications Conference (MILCOM), 2006. 

[4] R. Filler, S. Ganop, P. Olson, and S. Sokolowski, “Positioning, Navigation and 

Timing: The Foundation of Command and Control,” in Command and Control 

Research and Technology Symposium (CCRTS), 2004. 

[5] S. Reidt and S.D. Wolthusen, “Connectivity Augmentation in Tactical Mobile 

Ad hoc Networks,” in IEEE Military Communication Conference (MILCOM), 2008.


365 

[6] L. Riblett. and J. Wiseman, “TACNET: Mobile ad hoc Secure Communications 

Network,” in 41st Annual IEEE International Carnahan Conference on Security 

Technology, 2007. 

[7] L. Riblett and J. Wiseman, “TacNet Tracker©: Built-in Capabilities for Situational 

Awareness,” in 42nd Annual IEEE International Carnahan Conference on Security 

Technology, 2008. 

[8] P. Ebinger and S. Wolthusen, “Efficient State Estimation and Byzantine Behavior 

Identification in Tactical MANETs,” in IEEE Military Communications Conference 

(MILCOM), 2009. 

[9] L. Hu and D. Evans, “Localization for Mobile Sensor Networks,” in 10th ACM Annual 

International Conference on Mobile Computing and Networking (MobiCom), 2004. 

[10] S. Biaz and Y. Ji, “A Survey and Comparison on Localisation Algorithms for Wireless 

Ad Hoc Networks,” Int. J. Mob. Commun., vol. 3, pp. 374-410, May 2005. 

[11] R. Huang and G.V. Zaruba, “Incorporating Data from Multiple Sensors for Localizing 

Nodes in Mobile Ad Hoc Networks,” IEEE Transactions on Mobile Computing, vol. 6, 

pp. 1090-1104, September 2007. 

[12] B. Ristic, S. Arulampalam, and N. Gordon, Beyond the Kalman Filter: Particle 

Filters for Tracking Applications. Artech House, 2004. 

[13] M. Rosencrantz, G. Gordon, and S. Thrun, “Decentralized Sensor Fusion with 

Distributed Particle Filters,” in 19th Conference on Uncertainty in Artificial Intelligence 

(UAI), 2003. 

[14] A. Doucet, N.D. Freitas, and N. Gordon, Eds., Sequential Monte Carlo Methods 

in Practice. Springer, 2001. 

[15] A. Baggio and K. Langendoen, “Monte Carlo Localization for Mobile Wireless 

Sensor Networks,” Ad Hoc Networks, vol. 6, no. 5, pp. 718-733, Jul. 2008. 

[16] A. Agarwal and S.R. Das, “Dead Reckoning in Mobile Ad Hoc Networks,” in IEEE 

Wireless Communications and Networking (WCNC), vol. 3, 2003, pp. 1838-1843. 

[17] E. Amar and S. Boumerdassi, “Enhancing Location Services with Prediction,” 

in Proceedings of the 2009 International Conference on Wireless Communications 

and Mobile Computing: Connecting the World Wirelessly, ser. IWCMC ’09. New 

York, NY, USA: ACM, 2009, pp. 1025-1029. 

[18] Special Mobile Group (SMG), “Universal Mobile Telecommunicatios System (UMTS) – 

Selection Procedures for the Choice of Radio Transmission Technologies of the UMTS,” 

European Telecommunications Standards Institute (ETSI), Tech. Rep., 1998. 

[19] N. Aschenbruck, R. Ernst, E. Gerhards-Padilla, and M. Schwamborn, 

“BonnMotion: A Mobility Scenario Generation and Analysis Tool,” in 3rd International 

ICST Conference on Simulation Tools and Techniques, 2010. 

[20] X. Hong, M. Gerla, G. Pei, and C.-C. Chiang, “A Group Mobility Model for Ad Hoc 

Wireless Networks,” in 2nd ACM International Conference on Modeling, Analysis 

and Simulation of Wireless and Mobile Systems (MSWiM ’99), 1999. 

[21] T. Camp, J. Boleng, and V. Davies, “A Survey of Mobility Models for Ad Hoc 

Network Research,” Wireless Communications and Mobile Computing, vol. 2, no. 5, 

pp. 483-502, 2002. 

[22] K. Fall and K. Varadhan, The ns Manual, May 2010, available online at 

http://www.isi.edu/nsnam/ns/doc/ns_doc.pdf; accessed 2012-07-26.

Spatial Localization of Radio Wave Emission Sources 

Using SDF Technology 

Jan M. Kelner, Piotr Gajewski, Cezary Ziółkowski 

Military University of Technology, Warsaw, Poland, 

{jkelner, pgajewski, cziolkowski}@wat.edu.pl 

Abstract: The paper is devoted to the question of locating of radio wave emission sources by using 

of SDF technology. The generalized algorithm to estimate the radio transmitter coordinates has been 

presented. The Doppler characteristics are the basis to determine the location coordinates of a source. 

They are obtained as the result of measuring of instantaneous frequency of a signal received on 

board of an airplane. 

Keywords: location techniques, Doppler shift, Signal Doppler Frequency (SDF), 3D localization 


The localization of mobile radio waves transmitters is more and more interesting 

service both for networks users of and for operators of Electronic Warfare 

(EW) systems. Many location techniques have been already studied and described, 

mainly [6]-[12]: CoO (Cell of Origin), AoA (Angle of Arrival), RSS (Received Signal 

Strength), TOA (Time of Arrival), TDoA (Time Difference of Arrival), GPS/A-GPS 

(Assisted GPS), FDoA (Frequency Difference of Arrival) as well as the so-called 

hybrid methods. 

Above methods have some advantages and disadvantages for using in practical 

systems. Mostly, the location in two dimensions (2D) is these methods main limitation. 

The article presents an idea of the method for location in 3D that use 

the values of signal Doppler shift which is measured on the mobile flying platform 

(helicopters, planes, drones). 

The SDF (Signal Doppler Frequency) technology [2] is based on the analytical 

description of the Doppler effect [1]. It uses the distinctive nature of Doppler curves 

resulting from the reciprocal location of signal sources and receiver in relation to 

the movement of the objects trajectory. 

Because of using the Doppler effect, the SDF method is compared to FDoA 

one. However, the SDF and FDoA methods present considerably different approaches 

to the location techniques. The possibility of object spatial 3D localizations 

is a main advantage of the SDF technology, which is presented below.


Moreover, the main propriety of SDF method is its high precision as well as independency 

for the time-frequency structure of signals emitted by objects being 

localised. It gives the possibility to use the method universally and autonomously 

in numerous practical applications. Several applications of the SDF technology have 

been presented till now, among others in the area of: reconnaissance and electronic 

warfare, spectrum monitoring, sea and land rescue, and navigation. 

II. Characteristic of SDF method 

The SDF technology bases on the analytical description of the Doppler frequency 

[1]. The expression describing the variability of the Doppler frequency 

has the following form [1]: 

k 

 

xvt 

 

f 

D 

x , t 

2 

k 

2 

f 

2 2 2 0 

1k xvt 1k y z 

 

, (1) 

 

 

where: k vc, – the difference of velocity between a signal source and the receiver, 

– the electromagnetic wave propagation speed in the medium, – the 

carrier frequency of the emitted signal, x xyz 

, , – coordinates of the signal 

source location. 

Above formula shows that the value of the Doppler frequency shift is a function 

of signal source location coordinates. This fact has become the essence of elaborated 

method is a measurement of the instantaneous value of the signal frequency 

by a moving measurement receiver. Basing on (1), particular coordinates x, y, z 

of the transmitter location can be determined as the functions of the Doppler 

frequency shift measured in various time moments. In the case, when the receiver 

is moving on a fixed height above the flat terrain, the coordinates are known and 

fixed values. Making elementary transformations of the equation (1) we achieve 

formulas describing the coordinates x and y of the signal source [2]-[5]: 

 

 

x v tAt 

At 

 

 

tAt 

1 1 2 2 

At 

1 2 

, (2) 

y 

1 

1 

v 

t t At At 

 

 

1 2 1 2 2 

2 

 

k At1 At2 

 

2 

 

z 

, (3) 

where: 

2 

2 

1 

F t 

f 

D 

t 1k 

At , Ft k . (4) 

Ft 

f k 

So, in order to determine coordinates of signal source, it is necessary to measure 

the Doppler frequency value in two moments and . Therefore, it is 

0


369 

possible to determine locations of emitting radio wave sources during the receiver 

movement, basing on the measurement of the Doppler frequency shift. 

The idea of SDF 3D method is described below. 

III. SDF spatial localization method 

The algorithm of localization procedure is presented in the Fig. 1. The following 

simplifications were assumed in the analysed scenario [2]: 

• the signal generated by the localised transmitter is received at any point 

of the space where the localization stand (LS) is placed; 

• the localised signal source is immovable and it is placed on the flat Earth 

surface; 

• the LS is moving at a defined height over the Earth, what means that only 

the direct component is taken into account in the signal coming from 

the source to the receiver; 

• the operator of the LS knows the frequency of the signal transmitted 

by the localised source. 

Figure 1. Algorithm of the SDF spatial localization procedure


Here, the several possible situations of localization using an aircraft are taken 

into account. They depend on the knowledge of an approximated direction for 

the transmitter. So, Algorithm contents three blocks. Block (A) corresponds to 

the situation, when the LS operator knows the approximated direction to the signal 

source. Block (B) is using if the LS operator does not know the approximated 

direction to the signal source, and the stand is moving in the direction, for which 

is not possible the coordination determining. And finally block (C) corresponds 

to the situation where the LS operator does not know the approximate direction 

to signal source, but he is able to determine transmitter coordinates x and y on 

the flying route. 

The task of part B is to define the precise coordinates of the transmitter using 

procedure Z, presented in the Fig. 2 [2]. 

Figure 2. Illustration of Z-procedure 

The Z procedure is brought to define 3-phase flight route of LS between 

the points (1)→(2)→(3)→(4). In the first phase between the points (1)→(2), the ap-


371 

proximate transmitter’s coordinations can be determined. Knowing the approximated 

direction to the localised transmitter, it is possible to select a suitable flight 

direction within the first phase of the procedure Z. This direction should be established 

so that the Doppler frequency shift changed [2]. In the point (2), the change 

of the movement direction takes place in the OXY plane by the angle (Fig. 2) 

defined by the formula [2]: 

y 

s 

1 xy 

s 

arctg arccos 

xvt D 

xvt s 

arccos arccos , 

D D 

xy 

where: 2 2 

D xvt y , s – the length of the route segment between point (2) 

xy 

and point (X) (Fig. 2), where the Doppler frequency is equal to zero. 

This phase of the Z procedure allows to achieve the coordinate values 

of the transmitter position with high accuracy. It is resulted from the suitable selection 

of the route where almost symmetrical part of the Doppler curve is achieved 

around the value of . The point (X) in the Fig. 2 is called the Point of the Closest 

Approach – PCA [12]. The selection of the length 2s of the route segment between 

the points (2)→(3) depends on the distance of the localization stand to the signal 

source. The question of selection of the measurement segment 2s is presented 

in the references [2,13]. In the point (3) the next change in the movement direction 

takes place in the plane OXY by the angle of (Fig. 2) described by formula [2]: 

xy 

xy 

(5) 

y ' 

vt ' x' 

 

2 

arctg 

 

y' 2 y' 

, (6) 

where: x', 

y', z ' – coordinates of the signal source position determined in the second 

phase of the procedure Z, – the time measured from the moment of changing 

the movement direction in the point (2), y' y ' – a factor responsible for the sign 

of the angle and providing the proper change of the direction (to the right or 

to the left) in the point (3). 

Within the third phase of the procedure Z the plane flights in the direction 

to the transmitter. The change in the direction for the route (3)→(5) (Fig. 2) is executed 

similarly to the point (2) of the procedure Z. The precise coordinate values 

of the source position are determined in this way. 

When the direction to the transmitter is not known and Doppler shift does not 

changed, the block (B) is worked out. Here, three situations can be distinguished: 

a) the stand moves away from the transmitter that is in a large distant from 

the receiver ( f 

D 

t fDmax 

), 

b) the stand moves closer to the transmitter, but is in a large distant from it 

( f 

D 

t fDmax 

),


c) the stand moves (closer or away) in small distant of the source, but the flight 

direction overlaps (with a margin) the direction to the signal source 

( f 

D 

t fDmax 

). 

The Fig. 2 presents an example of such situation by moving the LS between 

the points (6)→(7). Then, the diametrical change in the movement direction is required 

in the plane OXY by the angle of ±90° related to the present movement 

direction. Next, the procedure Z is executed. The segment (7)→(2) corresponds to 

the (1)→(2) one in the procedure Z as for the situation (A). 

The last block (C) corresponds to the situation where the LS operator does 

not know the approximate direction to signal source, but he is able to determine 

coordinates x and y of the source position on the route of his movement (the segment 

(1)→(2) in the Fig. 2. The coordinate y is then defined ambiguously. It is not 

known if the signal source is on the right or on the left side of the LS movement 

trajectory (the segment (1)→(2) in the Fig. 2). Therefore, the LS direction change 

in the point (2) can be done in the incorrect direction. In such a case, the movement 

of the receiver in the direction (2)→(8) will cause that f 

D 

t fDmax 

. The return 

(8)→(2) and further continuation of the flight according to the procedure Z is required 

in such situation. 


The SDF method was verified by simulations and measurements for 2D localization. 

Some results have been presented in [4],[5]. The procedure Z has been 

positively verified by simulation tests for the rescue action scenario [14]. Some 

simulation results are presented below. 

Figure 3. Simulation scenario


373 

Fig. 3 shows the simulation scenario and the flight trajectory in 2D. The aircraft 

position is in geographical UMT standard. At the top, the positioning accuracy 

results obtained for the point F is shown. This is the point with the best accuracy 

of location because of the most distinguish change of Doppler frequency (Fig. 4). 

Figure 4. Doppler frequency courses 

Figure 5. Localization accuracy 

The r localization accuracy is shown in Fig. 5. The best results are obtained 

during the flight in sector EF, after few changes of flight route. 

V. Summary 

The paper shows the possibility of using the space platform to spatial localization 

of the radio emission sources (both stationary and movable). 

The SDF method in the spatial localization can find wide applications in air and 

sea rescue systems as well as in national security agencies. Installing the localization


stand on helicopters and aircrafts increases the range and accuracy of the method 

in relation to the ground localization. 

The exact characteristics of this method, the many experiments should be 

worked out. At present, the preparations are in progress to develop the new research 

project. Within this grant it is expected to carry out researches and the attempt 

to implement the SDF technology on an aircraft from the Navy Sea Rescue Unit. 

The expected better precision of the Doppler localization method in the case 

of placing the localization stand on aircraft than placed on land vehicles allows 

to infer of high capacities of the method, which can be also used for air navigation, 

particularly within aircraft landing systems in reduced visibility conditions. 

The possibility to use the SDF method to navigation purposes is presented, among 

others, in [2],[13]. 

References 

[1] J. Rafa, C. Ziółkowski, “Influence of transmitter motion on received signal parameters 

– Analysis of the Doppler effect”, Wave Motion, vol. 45, no. 3, pp. 178-190, January 

2008. 

[2] J.M. Kelner, Analiza dopplerowskiej method lokalizacji źródeł emisji fal radiowych 

(Analysis of the Doppler location method of the radio waves emission source), 

Ph.D. thesis, Military University of Technology, Warsaw, Poland, 2010, (in Polish). 

[3] P. Gajewski, J.M. Kelner, C. Ziółkowski, “Subscriber location in radio communication 

nets”, Journal of Telecommunications and Information Technology, no. 2, pp. 88-92, 

2008. 

[4] J.M. Kelner, C. Ziółkowski, L. Kachel, “The empirical verification of the location 

method based on the Doppler effect”, 17th International Conference on Microwaves, 

Radar and Wireless Communications MIKON 2008, Wroclaw, Poland, vol. 3, pp. 755-758, 

May 2008. 

[5] P. Gajewski, C. Ziółkowski, J.M. Kelner, “Mobile location method of radio wave 

emission sources”, Progress in Electromagnetics Research Symposium PIERS 2009, 

Moscow, Russia, August 2009. 

[6] A. Amar, A.J. Weiss, “Localization of narrowband radio emitters based on 

Doppler frequency shifts”, IEEE Transactions on Signal Processing, vol. 56, no. 11, 

pp. 5500-5508, 2008. 

[7] Y. Zhao, “Standardization of mobile phone positioning for 3G systems”, IEEE 

Communications Magazine, vol. 40, no. 7, pp. 108-116, 2002. 

[8] M. Vossiek, L. Wiebking, P. Gulden, J. Wieghardt, C. Hoffmann, P. Heide, 

“Wireless local positioning”, IEEE Microwave Magazine, vol. 4, no. 4, pp. 77-86, 2003. 

[9] I.J. Gupta, “Stray signal source location in far-field antenna/RCS ranges”, IEEE 

Antennas and Propagation Magazine, vol. 46, no. 3, pp. 20-29, 2004. 

[10] A. Küpper, Location-based services. Fundamentals and operation, John Wiley & Sons, 

Chichester, UK, 2005.


375 

[11] K.W. Kołodziej, J. Hjelm, Local positioning system. LBS applications and services, 

CRC Press, Boca Raton, FL, USA, 2006. 

[12] N. Levanon, M. Ben-Zaken, “Random error in ARGOS and SARSAT satellite 

positioning systems”, IEEE Transaction on Aerospace and Electronic System, vol. AES-21, 

no. 6, pp. 783 790, 1985. 

[13] P. Gajewski, C. Ziółkowski, J.M. Kelner, “Influence of length and location 

of the measurement route on location accuracy of the radio waves sources using 

the SDF method”, Przegląd Telekomunikacyjny i Wiadomości Telekomunikacyjne, 

vol. LXXXIII, no. 8-9, pp. 1360-1369, August-September 2010 (in Polish). 

[14] C. Ziółkowski, J.M. Kelner, “Using the Doppler methodology for object location 

estimation in lifeboat service”, 6th International Conference on: Perspectives and 

Development of Rescue, Safety and Defence Systems in the 21st Century RSDS 2008, 

Gdansk, Poland, June 2008.

On the Effect of Tuner Phase Noise 

on TDOA Measurements 

Anders M. Johansson, Patrik Hedström 

Swedish Defence Research Agency, Linköping, Sweden, 

{ajh, pathed}@foi.se 

Abstract: Radio geolocation of unknown signals using time difference of arrival techniques is dependent 

on signal measurements that are recorded using a time and frequency coherent measurement 

system. This paper investigates what effect the phase noise in the reference oscillator inside the tuner 

has on the geolocation accuracy, and proposes a model for the phase noise in addition to a method 

for counteracting its effect. The results include both real world measurements and computer simulations. 

Results from a field experiment indicate that the proposed method outperforms traditional 

time difference of arrival techniques by a factor between 3 and 5.8 depending on SNR. 

1. Introduction 

Time difference of arrival (TDOA) techniques for geolocation of unknown 

signals has been researched within acoustics since the First World War [1]. In later 

years systems for measuring the TDOA on radio signals has been developed and 

a few military systems based on TDOA techniques are in use today. One problem 

when dealing with narrow band signals or signals with poor signal to noise ratio 

(SNR) is that long measurement times are required to reach sufficient TDOA 

accuracy. This can easily be seen by studying the Cramer-Rao Bound (CRB), 

for details see [2]. In particular, narrow band signals with poor SNR requires 

measurement times exceeding one second in order to reach sufficient accuracy. 

In this paper we show that phase noise in the reference oscillator used in the radio 

receivers impacts negatively on the accuracy of TDOA estimation. Furthermore, 

we present a method for modeling the phase noise, and results from a real world 

field trial. The results are presented using both real measurements and computer 

simulations. A method for counteracting the problems arising from the phase 

noise is also presented. 

Section 2 presents a signal model and an oscillator model, followed by a description 

of the TDOA algorithm including modifications to counteract the phase 

noise in Section 3. Measurements and simulations are presented in Section 4, and 

conclusions in Section 5.


2. Signal models 

Assume a band limited signal st (), where denotes time, transmitted from 

a position in space, impinge on a group of receivers placed in the positions 

, m{1, 2, , M} 

, see Figure 1. By comparing the received base band signals x m() 

t , 

the TDOA algorithm can be used to form an estimate ˆ of the transmitter position. 

Figure 1. System model 

The propagation channel from to is denoted hm 

() t and is considered to 

be stationary over the measurement time . At each receiver, the received signal 

is corrupted by noise denoted vm() 

t , which is considered spectrally white, time 

invariant and uncorrelated between the receivers. The modulation frequency is denoted 

and the signal bandwidth . The local oscillator of the receiver is assumed 

to be unstable with the phase noise m() 

t . The signal chain from the transmitter 

to receiver is depicted in Figure 2. The received signal is: 

 

( ( )) 

() () j c 

t m 

t 

m m 

* () 

Bm 

(), 

 

x t h te st v t 

(1) 

where denotes convolution. The above equation shows that the output signal, 

x () m 

t , from the receiver is a demodulated version of the system impulse response 

convolved with the transmitted base band signal. 

The base band noise term, vBm() 

t , in Equation (1) is [3]: 

 

v t v jH v 

(2) 

 

 

( c m( )) 

() e j t t 

Bm m m 

, 

where H () denotes the Hilbert transform.


379 

Figure 2. Signal model 

2.1. Free space propagation 

Assuming free space propagation without attenuation, the propagation channel 

simplifies to a pure delay hm() t ( t m) 

, where is the propagation delay 

in seconds between the transmitter and receiver , and is calculated according to: 

 

m 

m 

q 

p , 

(3) 

c 

where denotes vector norm, and 

the speed of light. 

2.2. Oscillator phase noise 

The model for phase noise is here defined as flicker noise denoted m() 

t 

added to the centre frequency of the reference oscillator. Typically this noise is what 

causes Allan variation [4]. 

() t t (), t 

(4) 

where 

m f m 

is the noise power. The power spectral density (PSD) of the flicker noise 

m() 

t 1N 

f 

. Both and are oscillator specific constants, 

where oscillators with high precision typically have a low value of and a high 

value of and vice versa. Several numerical methods for simulating flicker noise 

have been presented in the literature, the method used here is a fast Fourier transform 

(FFT) based method called “the discrete spectrum method”, and is described 

in [5]. Section 4.1 presents a method for estimating the constants and for 

a real oscillator. 

As the power of the flicker noise is concentrated at very low frequencies, it is 

possible to model the oscillator as stable with a phase offset for short measurement 

times. A simplified model for the noise for short measurement times is 

() t (), t 

(5) 

m 

where 

m( t ) is a constant over tTs 

/2 t t Ts 

/2for small values of the measurement 

time . Note that due to the phase noise, the phase offset can be assumed 

to vary between measurements. 

m


3. The TDOA algorithm 

The TDOA algorithm estimates the position of the transmitter by identifying 

the time difference of arrival between all receiver pairs. This can be done either in the time 

domain or in the frequency domain, by analysing the cross correlation or the cross 

spectral density (CSD) between the received signals. To study the effect of phase noise, 

we have here chosen the frequency domain approach described in [6]. 

Using Equation (1) the CSD between the receivers and for free space 

propagation is derived as 

R ( , t) E x * 

tx() 

t 

(6) 

xm, n 0 

m n 

 

* 

j( m( t0) n( t0)) 

Hm( c) Hn( c) Rs( )e 

R ( ) vm n 

(7) 

( )( ) ( ( 0 ) ( 0 

( )e j c mn m n 

e j t 

R 

t )) 

s 

Rvm, 

n( ), 

(8) 

for short measurement times. In the above equation, 

denotes Fourier transform 

and E 

is the expected value operator, () denotes complex conjugate, 

Hm( ) hm() 

t , Rs 

( ) is the PSD of the transmitted signal st () and Rvm n( ) 

is the CSD between the noise in xm() 

t and xn() 

t . 

To estimate the TDOA we remove the influence of constant phase terms according 

to 

R 

( , t ) 

xm, n 0 

xm, n 

( , t0) 

 

Rxm, n 

(0, t0) 

(9) 

R 

 

j( mn) 

e , 

(10) 

where we have made the assumption that Rs( ) / Rs(0) 1without loss of generality, 

and the measurement time is sufficient such that the approximation Rvm n( ) 0 

can be made. The above equation shows that the phase reveals the TDOA: 

mnm n. By calculating estimates ˆ mn of the TDOA for multiple receiver 

pairs it is possible to solve for the transmitter position using Equation (3). Estimating 

is however not the focus of this paper, the rest of the paper will therefore 

concentrate on the two receiver case to highlight the effect of oscillator instability. 

In a practical application the CSD is estimated by sampling the time signals 

at the sample frequency , xm() t xm( n/ Fs) 

where , and splitting them 

into element segments 

L 

xm n k k 

 

R ( ) X ( , m) X ( , n), 

(11) 

where X ( k 

, m ) is the th FFT bin of sensor and signal segment .


381 

TF 

s s 

By studying Equation (11), we find that for low values of , i.e. when L , 

K 

the approximation made in Equation (5) holds, and the phase is stable over the 

segments. A problem does however arise when the SNR is poor or when very 

high SNR levels are required. In these cases the value of must be large to reduce 

the impact of the noise on the TDOA estimate. From Equation (8) we can see that 

by adding components with different phase ( m( t) n( t ) will be different for different 

values of ) the effective SNR will drop as increases. As a result the TDOA 

estimation accuracy also drops as the measurement time increases. 

The solution to the problem is simply to estimate the TDOA mn() 

t over 

multiple short time samples whereby a simple incoherent data fusion (IDF) can be 

performed using time averaging: 

T 

ˆ 1 

 

() t 

(12) 

mn , mn , 

T 

t 1 

The measurement time used to calculate each TDOA estimate, mn() 

t , is less 

than . The resulting algorithm is here denoted IDF-TDOA. 

4. Measurements 

4.1. Oscillator modeling 

The oscillator parameters and were estimated using the experimental 

setup depicted in Figure 3. The experiment was designed to make it possible to 

measure oscillator parameters without using a stable frequency reference. The input 

for the experimental setup was a signal generator and the signal was received and 

saved to mass storage using a custom built data acquisition system. The signal 

generator was adjusted to a frequency 1 kHz above the tuning frequency for 

the two tuners. 

A simulation of the measurement setup was made in order to determine 

the flicker noise parameters. The PSD of the output signal yn ( ) for the measurement 

and the simulation setup is depicted in Figure 4. The measurement time was 30 s. 

The oscillator parameters were experimentally determined to be 

f 

0.0008 

and N = 3.8. The figure clearly show the low frequency characteristic of the flicker 

noise in the oscillator.


Figure 3. Experimental setup for evaluating oscillator stability 

Figure 4. PSD for simulation and experiment setup


383 

4.2. TDOA accuracy 

In order to investigate the effect of oscillator instability on TDOA estimation 

for varying measurement times the signal generator in Figure 3 was replaced by 

a vector signal generator generating a 10 kHz wide 8-PSK signal. In the experiment, 

independent white noise was added to the signal in the digital domain to study 

the effect of varying SNR. In Figure 5, the standard deviation in micro seconds 

is displayed as a function of measurement time for SNR levels of 65 dB, 25 dB and 

5 dB, along with the CRB and the measurement system accuracy for TDOA and 

IDF-TDOA. The constants and L 32 . The CRB is displayed for 5 dB and 

25 dB SNR levels only. The standard deviation is estimated by calculating an estimate 

of the TDOA on 490 data segments drawn from 400 seconds of recorded data. 

The figure shows that the oscillator noise reduces the accuracy for long measurement 

times for traditional TDOA while the proposed IDF-TDOA is unaffected by 

the oscillator instability for long measurement times. 

Figure 5. Standard deviation in micro seconds versus measurement time for TDOA and IDF-TDOA 

on measured signals. The red line shows the measured accuracy inherent to the system, and the black 

dotted lines the CRB. Note that the CRB is only displayed for 5 dB and 25 dB SNR 

The experiment described above was repeated in a simulated environment 

where the oscillator phase noise parameters estimated in Section 4.1 were used. 

The results are displayed in Figure 6. By comparing the results presented in Figure 5 

and 6 it becomes obvious that it is the phase noise that is causing the observed 

errors, and no other tuner parameter such as linearity or noise.


4.3. Results from a field experiment 

A field experiment was conducted in order to test the IDF-TDOA under real 

world conditions. A transmitter emitting a 10 kHz wide 8-PSK modulated pseudo 

random sequence centered at 306 MHz was recorded for 15.9 s. The transmitter 

output power was varied to achieve four different signal to noise ratios at the receivers. 

The distance between the receivers and the transmitter was around 5 km, 

and the experiment was conducted in an urban environment. The result for TDOA 

and IDF-TDOA is tabulated in Table 1 along with the CRB. The table shows that 

IDF-TDOA improves the accuracy by a factor of 3 to 5.8 depending on the SNR. 

The table also shows that the error is far above the CRB. The reason for this will 

however require further scrutiny and is beyond the scope of this paper. 

Table 1. TDOA standard deviation in micro seconds versus SNR 

SNR TDOA IDF-TDOA CRB 

30 0.6 0.2 0.002 

20 0.7 0.2 0.005 

10 1.5 0.3 0.02 

0 8.7 1.5 0.06 

5. Conclusions 

This paper investigates the effect of oscillators phase noise in tuners used for 

TDOA based radio geolocation of unknown signals. The presented theoretical 

study shows that the phase noise causes an effective decrease in the SNR if the rate 

of frequency change caused by the phase noise exceeds the measurement time. These 

results are used to propose a modification to the TDOA algorithm, and the final 

algorithm is denoted IDF-TDOA. 

Measurements on a tone using two tuners are used to estimate the spectral 

contents of the phase noise. The measured spectrogram is used to calibrate an FFT 

based phase noise model. 

Measurements on noise using the two tuners are used to investigate the effect 

of phase noise on TDOA estimation. The measurements are compared to the simulations 

and to the CRB. A comparison between the simulations and the measurements 

shows that it is the phase noise causing performance degradations. 

The paper is concluded with results from a field trial showing the effectiveness 

of the proposed new IDF-TDOA algorithm under real world conditions. The new 

algorithm outperforms traditional TDOA with respect to accuracy by a factor 

of 3 to 5.8 depending on the SNR.


385 

Figure 6. Standard deviation in micro seconds versus measurement time for TDOA 

and IDF-TDOA on simulated signals. The black dotted lines shows the CRB. Note that the CRB 

is only displayed for 5 dB and 25 dB SNR 

References 

[1] D.H. a. D.D.E. Johnsson, Array Signal Processing: Concepts and Techniques, Upper 

Saddle River, New Jersey 07458: Prentice-Hall Inc., 1993. 

[2] W.R. a. T.S.A. Hahn, ”Optimum Processing for Delay-Vector Estimation in Passive 

Signal Arrays,” IEEE Transactions on Information THeory, vol. 19, nr 5, pp. 608-614, 

1973. 

[3] A. a. C.P.B. a. R.J.C. Carlson, Communication Systems, An introductions to signal 

and noise in electrical communication, Fourth edition, Mc Graw Hill, 2002. 

[4] W.J. Riley, Handbook of frequency stability analysis, Boulder, COlorado U.S. 

Department of Commerce: National Institute od Standards and Technology, 2008. 

[5] C.A. Greenhall, ”FFT-Based Methods for Simulatin Flicker (FM),” 34th Annual 

Precise Time and Time Interval (PTTI) Meeting, pp. 481-491, November 2002. 

[6] C.K. a. G. Carter, ”The generalized correlation method for estimation of time 

delay,” IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 24, nr 4, 

pp. 320-327, 2003.

Aircraft Tracking Using Mobile Devices 

Michał Andrzejewski 1 , Radosław Schoeneich 2 

1 

Institute of Control and Computation Engineering, 2 Institute of Telecommunications, 

Warsaw University of Technology, 

M.Andrzejewski@stud.elka.pw.edu.pl, R.Schoeneich@tele.pw.edu.pl 

Abstract: The following article presents an overview of airship tracking system using mobile devices 

equipped with GPS receiver, running under Android operating system. It describes general approach, 

usage of store, carry and forward paradigm in delay and disruption tolerant networking and 

elements of architecture and implementation of presented applications. Shown results represent tests 

run in real environment. 

Keywords: tracking, monitoring, general aviation, aircraft, disruption tolerant network, DTN 


The dynamic development of mobile devices and navigation systems in recent 

years has introduced many new everyday services. Among the most commonly 

used are location and remote monitoring. Their potential was quickly spotted by 

industries related to the transport and logistics, where knowledge about the current 

position and status of objects such as vehicles and packages is often a key factor 

in the success of the operation. 

The main task of the monitoring system is to provide reliable information 

about the position of tracked objects. This information can be interpreted and 

used in different ways, depending on current needs. In general, however, it serves 

two basic purposes: to increase safety and to improve efficiency. Location data 

are widely used in planning and decision support systems, fleet management and 

emergency assistance. 

Unfortunately, despite the continuous development of systems of various 

complexity, there is still no good solution that could be used to monitor general 

aviation aircraft. For commercial and military aviation, there are several extensive 

technical capabilities, such as primary or secondary surveilance radars or satellite 

systems capable to track objects in real time with very high accuracy. However, 

the cost of such solutions disqualifies them from usage in general aviation flying 

clubs, flight schools and by private users.


Key idea of this article is to consider popular mobile devices (such as smartphones, 

tablets, Personal Digital Assistants) as a reliable source of location data 

and their usage in real application. 

II. Environment 

General aviation aircraft are moving in the sky at various altitudes and with 

different speeds, depending on the type of an airship, airspace category or purpose 

of the flight. However, assumption can be made that the vast majority of flights take 

place at an altitude up to 1000 m above the ground level at speeds not exceeding 

250 km/h. Flight paths at such altitude can be set freely, but near the airports some 

constraints and obligations to accurately proceed between specified points exist. 

General aviation flights may run across areas with different levels of urbanization, 

and at potentially large scale of height above ground level. This means that 

the use of GSM mobile network to monitor the aircraft may be inefficient or even 

impossible. This is due to cellular antennas settings. Additionally, the level of noise 

and interference caused by signals from nearby stations raises with the altitude. 

In practice, however, very often even at the height of up to 1500 meters mobile 

network is available, and quality of connection allows data transmission. Making 

a voice call is however not possible at this altitude. This makes an opportunity 

of creating simple tracking system based on GSM-capable devices and data transmission 

provided by mobile telephony operators. Results obtained in this field are 

presented later in this article. 

The environment in which flights are made, is a challenge for active position 

monitoring systems. Continuity and quality of data transmission becomes a major 

issue. Direct use of GSM network as a way to transmit information related to 

the location may not give expected results in terms of reliability and availability. 

This is an area where algorithms developed for use in incoherent networks may be 

applied. Their essence is to enable operation in the harsh environment of unknown 

characteristics. 

III. Subject of research 

The main task and also the source of the biggest problems of active monitoring 

systems is to develop a suitable method for data transmission. Such method, 

consisting of software and technology, should provide high availability and reliability. 

In the perfect case, monitored object should always have the possibility 

of an effective data transmission. 

The research, which became the basis for this article was focused on: 

• exploring the possibilities of using mobile devices and GSM networks to 

keep track of general aviation aircraft,


389 

• testing and evaluating prototype of aircraft tracking system in a real environment, 

• identification of advantages and disadvantages of the proposed solution, 

• proposing possible directions of development and improvement of the system. 

During the developement of an aircraft monitoring system, the natural 

choice was to provide transmission using wireless communications. The concept 

of the satellite communication solutions was rejected because of their high cost. 

Emphasis has been placed on technologies directly offered by mobile devices with 

data transmission over GSM network in particular. Fig. 1 presents the basic scenario 

where proposed system, in addition to direct communication between the aircraft 

and data collection server, uses also the paradigm of data storage and forwarding. 

This approach, commonly referred to as a store-carry-forward paradigm, is specific 

to delay and disruption tolerant networks. The aircraft marked as SP1 is out 

of GSM coverage, so it can not send data on the position in a direct way. Using 

a wireless network, it communicates and exchanges all necessary information 

with the aircraft marked as SP2. The aircraft SP2 can then forward information 

regarding both objects to the server collecting data. The system has information 

about the positions of both vessels despite the fact that SP1 is not in a GSM range 

to establish a direct connection. 

Figure 1. Wireless data transmission scenario 

Wireless communication channel between objects must meet the following 

conditions for its use in the proposed system to be possible: 

• time required to estabilish two-way communication channel and its effective 

range should allow data exchange between aircraft flying in opposite 

directions at speeds of around 150 km/h, 

• each pair of nodes using appropriate software should be able to connect 

automatically in the same way, 

• estabilishing the connection shouldn't require any additional operations 

from the user.


Lack of any of these features reduces the area of application of wireless communication 

between the aircraft and makes it difficult to achieve given functionality. 

A. Architecture of aircraft tracking system 

The system this research has been based on is divided into two parts, performing 

distinct functions and embedded in different environments, as shown 

in the Fig. 2. It separates two basic areas of functionality – one related to the collection 

and transmission of data carried by the mobile application, and the other responsible 

for their analysis and visualization using an application server. It was a natural approach, 

given the purpose and the use of both items supplied in two applications. 

Figure 2. Aircraft tracking system architecture 

Mobile devices transmit files to an FTP server using the best currently available 

mean of communication. This can be a 2G/3G mobile network, Wi-Fi or another, 

depending on the particular device. If estabilishment of a direct connection 

to a remote server hosting the data is impossible due to lack of network coverage, 

high levels of noise, latency, or other actual problems, the application uses the storecarry-forward 

paradigm and gathers data locally. At a time when it is again possible 

to transfer data to another node, the device acts as discussed before. 

An attempt to transfer data to the server is controlled by an event timer – 

the expiration of the specified period of time since the last attempt to send or other 

parameters associated with the flight. 

Before a file is put on the FTP server, its validity is checked, and the transmission 

is performed only when attempting to send a file containing more data 

than the former. This allows nodes that are willing to send data less accurate 

than those already stored on the server not to overwrite and lose them.


391 

In order to simplify whole process and increase the reliability of data exchange 

between system elements, it was decided to use one common data model. This allows 

the system architecture as a whole to remain open, since all it requires is that 

the source data have a specific format. 

Usage of database management systems was rejected and flat text files in a specified 

comma separated values format were used instead. This reduces the requirements 

for the server, simplifies the implementation and developement of the solution and 

allows further analysis using common tools. Text files can still be imported and 

stored in a database when necessary. 

B. Technical limitations 

Available methods for wireless connectivity between mobile devices are limited 

to two common standards: IEEE 802.15.1 (Bluetooth) and Wi-Fi. 

The first one has too short range in order to be used for communication between 

aircraft in flight. For this purpose it is necessary to estabilish an effective connection 

between the devices in a distance of at least 300 meters, while Bluetooth allows 

efficient transmission at a distance several times smaller. In addition, the process 

of finding neighboring nodes using this method can take tens of seconds. The use 

of Bluetooth in incoherent networks is possible, but requires distances between 

nodes to be small, as well as their relative speeds [1]. 

Use of Wi-Fi networking was a natural choice. Despite the low power of transmitters 

mounted in mobile devices, it was possible to connect two nodes within a distance 

of about 200 meters apart in the open area. One of the devices, the Samsung 

Galaxy tablet, worked as an access point, while the second – LG-GT540 mobile 

phone – as a client. This combination allowed a stable exchange of data at speeds 

of about 100 kbps. Unfortunately, the long time needed to set up an access point and 

then to connect the other device (in this particular case – a total of more than 45 

seconds) and the need for manual adjustment of connection parameters prevented 

this method from the application in the system. 

In the course of the work it was discovered that the mobile device platform 

capabilities are insufficient to meet all the required functionality. The main limitation 

was the lack of ability to automatically establish a wireless connection in adhoc 

mode between devices without user intervention. This feature is essential 

for establishing a connection in an incoherent network and the exchange of data 

between nodes. Application Programming Interface of Android operating system 

used in experiments does not support such mechanism, nor is it possible to emulate. 

There is also no documentation on this issue available. 

Further studies show that the mechanism for supporting automatic wireless 

connections between mobile devices is not available on any other of the popular 

platforms.


Therefore, it was impossible to verify and evaluate inconsistent network 

performance in practice. The actual implementation required from the operating 

system and the mobile device functionality, which it was not able to provide. Perhaps 

in the future, with the development in this field of technology, this will become 

possible. After examining the possibility to connect the devices during the flight 

(discussed further in the next section) it was decided to abandon the functionality 

associated with the use of incoherent network in the proposed system. 

IV. Tests and results 

An important element of the work was to conduct several series of tests, not 

only to check the correctness of the implemented system, but also its usefulness 

in proposed applications. Results were used to verify presented concepts and theoretical 

considerations in actual, real environment. 

Tests in the target environment were divided into three stages. In the first, 

a prototype of mobile application was used. It was able to collect environmental 

data, such as the GSM signal strength, network mode, the height and position 

of the airship, etc. In addition, the aircraft was equipped with a second instrument 

– Garreht Volkslogger – which is a logger device certified by the International Air 

Sports Federation used to record position in gliding competitions. At this stage, 

the ability to perform data transmission using the GSM network at different heights 

was verified. Another income was the examination of GPS receiver accuracy by 

reference to the indications presented by the certified device. This part of the test 

was conducted in May and June 2011. 

The second step was to test the basic, working implementation of aircraft 

monitoring system between July and October 2011. The system was used to monitor 

the position of student pilots and pilots during training in the area of Warsaw- 

Babice airfield. 

The last stage completed in November 2011 was to test the operation of the system 

in the mountainous region. Differences in system’s behavior were identified and 

investigated basing on analysis of data collected during the flights in the vicinity 

of Bezmiechowa. Some additional opportunities for air-to-ground communications 

were diagnosed. 

A. Collecting data about environment 

Fig. 3 shows the altitude above mean sea level and GSM signal strength 

during one of the flights. The value of the signal strength equal to -113 dBm 

is identified as the inability to connect to the network. Data were sampled with 

a period of 30 seconds. 

Obtained results differed significantly from what was expected. It was anticipated 

that the range of the GSM network will be unavailable for much of the flight because


393 

of the altitude and its characteristics. Data collected during the first test flights, however, 

revealed that GSM network was available up to a height of about 1400 meters 

above terrain at urban areas throughout the duration of the flight. This is the result 

of reflection and reinforcement of radio waves from the ground, buildings, etc. 

Figure 3. Altitude and GSM signal strength correlation 

Position samples collected by the mobile device does not differ significantly 

from those provided by a professional logger. The maximum absolute error 

was 8 meters for static measurement and 25 meters for the measurement of motion 

on average. Errors associated with determining the height were greater, however 

they did not exceed 75 meters. 

During one of the flights, an attempt to establish a wireless connection using 

Wi-Fi network between two parallel flying aircraft was made. They were flying 

at an altitude of 500 meters at speeds of 120 km/h. The distance between objects 

was 150 meters. On board one of the aircrafts was a device which acted as an access 

point, while the second tried to connect to that network. An access point was seen 

by the other device, but the connection could not been established. Performed test 

showed that in given conditions it is not possible to create and use Wi-Fi networking. 

Results confirmed however, that the use of mobile devices for determining 

the position is sufficiently accurate for use in most of the monitoring applications. 

The quality of connections over the GSM network was sufficient for efficient transmission 

of data text files within a given height of 0-1000 meters. 

B. Tests in lowland area 

Most of the testing was made in the lowlands, because it is an environment 

in which the highest number of general aviation flights are performed. The mobile 

device was installed on board several aircrafts operating various types of flights 

in the northern area of Mazovia region. Position of the aircraft was observed 

in visualisation system and verified with reports collected via VHF radio. Between 

July and November 2011, 28 such flights were made. Below analysis in Tab. I shows 

12 flights selected of them.


Flight # 

Table I. Results of tests in lowland area 

Time in seconds [s] 

Flight duration MTBT a ATBT b 

1 3716 231 218 

2 6642 212 201 

3 823 180 164 

4 9144 288 234 

5 5323 222 190 

6 6006 265 207 

7 4571 303 253 

8 5701 281 219 

9 3122 201 183 

10 6717 568 305 

11 1559 180 173 

12 1287 180 160 

a) Maximum Time Between air-to-ground Transmissions 

b) Average Time Between air-to-ground Transmissions 

The most important parameter of such system is the location data refresh 

rate, understood as time between air-to-ground communication. Information 

about the successful upload to the server are recorded in the logs, so it is possible 

to calculate the time between successive transmissions. During the test, the interval 

of 180 seconds was set up to make updates of the data as frequent as possible, 

but not to excessively exhaust the battery. Tab. I shows a summary of flight time, 

the value of the maximum interval between transmissions of data and the average. 

Both of these times directly translate to the refresh rate of position data, therefore 

it is desired to minimize both of them. 

Each of the flights presented in a table has taken place in different weather 

conditions, using a different route and at different heights. For this reason it is 

impossible to directly compare obtained results. Obtained average values for each 

flight lead to the conclusion that despite initial concerns related to the use of GSM 

network for data transmission, the frequency of updating information about 

the position of the aircraft is sufficient for the intended application. 

C. Tests in mountainous area 

The usefulness of the system to monitor flights taking place in a mountainous 

environment, due to potential differences in performance, was also tested. The research 

was focused on the influence of the location of GSM network base stations 

on the system effectiveness, compared to the results collected before. In the moun-


395 

tainous terrain antennas are placed on the tops of hills to embrace the largest area 

lying below. It was expected that due to the smaller than in the lowlands height 

difference between GSM transmitter and the aircraft, the average interval between 

transmissions would be smaller. Because of the weather conditions, only 3 flights 

were made, as shown below. 

Flight # 

Table II. Results of tests in mountainous area 

Flight duration 

Time in seconds [s] 

c 

MTBT 

ATBT d 

1 1232 180 176 

2 747 180 172 

3 1338 191 186 

c) Maximum Time Between Transmissions 

d) Average Time Between Transmissions 

During the first flight, it turned out that in close proximity to one of the mountain 

ranges it is possible to connect to public Wi-Fi network. Its access point 

was located near one of the hotels and its range included a substantial portion 

of the ridge. The results of the flights are shown in Tab. II taking both transmission 

methods into account. 

The achieved results confirmed that the air-to-ground communication using 

the Wi-Fi standard during the flight is possible, though available only in specific 

conditions. Shortening the average interval between transmissions is the result 

of virtually uninterrupted stay of the aircraft in a GSM network and usage of availible 

Wi-Fi for data transmission. 


The results of the analyzed system, based on data collected during test flights, 

were better than expected. It was possible to obtain better average refresh interval 

of location data then initially planned. Value of 300 seconds was thought as satisfactory, 

while the average value for 12 flights in a lowland area was 209 seconds. 

Increasing the frequency of data transmissions could further improve this value, 

however, this would be associated with an increased resource consumption. 

Area of testing had a significant impact on the results. Analyzed flights were 

made up of several dozen kilometers from Warsaw agglomeration and mostly 

ran above the urban areas. Execution of test flights over areas characterized by poor 

coverage of GSM network is likely to lead to a deterioration of results. 

Analysis of the system in mountainous terrain allowed the discovery of additional 

options – usage of Wi-Fi networks for data transmission. Thanks to this observation, 

capabilities of mobile applications were increased and results could be improved.


The biggest obstacle in the development of the system is lack of possibility to 

create automatic connections between devices on an ad-hoc basis. This prevents 

direct application of the store-carry-forward paradigm, which could significantly 

improve performance. 

References 

[1] M. Liberatore, B.N. Levine, and C. Barakat, “Maximizing transfer opportunities 

in bluetooth DTNs”, Proc. CoNEXT, 2006.

Index

A 

Adrat Marc 81, 171 

Åkermark Hans 135 

Amanowicz Marek 55, 277, 333 

Andrzejewski Michał 387 

Anton Constantin 107 

Antweiler Markus 81, 161, 171 

Aubrecht Vladimir 67 

B 

Bloebaum Trude H. 9 

Bronk Krzysztof 215 

Bryś Rafał 289 

C 

Caban Przemysław 9 

Chaudhary Muhammad Hafeez 247 

D 

Debbah Mérouane 229 

Dołowski Jerzy 55 

E 

Ebinger Peter 349 

Elders-Boll Harald 171 

G 

Gajewski Piotr 151, 367 

Głowacka Joanna 277 

Goetz Michael 27 

Golan Edward 99 

Grzybkowski Maciej J. 215 

H 

Hedström Patrik 377 

I 

Idzikowska Ewa 117 

Ionescu Laurenţiu 107 

J 

Johansson Anders M. 377 

Johnsen Frank T. 9 

K 

Kaniewski Paweł 151 

Kelner Jan M. 367 

Koutny Tomas 67 

Kraśniewski Adam 99 

Krygier Jarosław 307 

Kuijper Arjan 349 

L 

Leduc Jan 161, 171 

Le Martret Christophe J. 229 

Le Nir Vincent 187, 201 

Liedtke Ferdinand 81 

M 

Mahoney Austin 135 

Marks Michał 319 

Maseng Torleiv 161 

Matyszkiel Robert 151 

Mazăre Alin 107 

N 

Nabrdalik Filip 319 

Niewiadomska-Szynkiewicz Ewa 319 

Niski Rafał 215 

Nissen Ivor 27 

O 

Osten Tobias 171 

P 

Pszczółkowski Jacek 289 

R 

Romanik Janusz 99


Rose Luca 229 

Ruszkowski Mirosław 289 

S 

Saarnisaari Harri 265 

Scheers Bart 135, 187, 201, 247 

Schenkels Léon 9 

Schoeneich Radosław 387 

Şerban Gheorghe 107 

Singh Sarvpreet 81 

Skarżyński Paweł 99 

Suchański Marek 151 

Ś 

Śliwa Joanna 9 

T 

Teguig Djamel 187 

Tschauner Matthias 81 

Turčaník Michal 45 

Tutănescu Ion 107 

U 

Urban Robert 99 

V 

Vanninen Teemu 265 

W 

Wawryszczuk Marcin 333 

Wolthusen Stephen D. 349 

Z 

Ziółkowski Cezary 367 

Ż 

Żurek Jerzy 215

Military Communications and Information Technology: A Trusted ...

Create successful ePaper yourself

Delete template?

Save as template?