Applications of Distributed Architectures in Software Defined Radio ...

ffi.no

Applications of Distributed Architectures in Software Defined Radio ...

Applications of Distributed Architectures in

Software Defined Radio and Dynamic Spectrum

Access

Tore Ulversøy

THESIS IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF

PHILOSOPHIAE DOCTOR

Kjeller, February 10, 2011

Web edition with additional hyperlinks: August 18, 2011


c○Tore Ulversøy, 2011

Series of dissertations submitted to the

Faculty of Mathematics and Natural Sciences, University of Oslo

No. 1046

ISSN 1501-7710

All rights reserved. No part of this publication may be

reproduced or transmitted, in any form or by any means, without permission.


Abstract

This thesis is a collection of five papers preceded by an overview part which summarizes

them, puts the results reported on into perspective and contains some further related results.

The papers relate to three research goals which are all within emerging radio systems

technology: Software Defined Radio (SDR), Cognitive Radio (CR) and Dynamic Spectrum

Access (DSA). The dominant focus is on the intersection between radio technology and

distributed processing technology. The research goals aim to contribute to the evolution

from inflexible transceivers with pre-regulated frequency-use to flexible software-defined

transceivers capable of dynamically exploiting the available radio spectrum.

SDR is frequently viewed as an enabler for Cognitive Radio and Dynamic Spectrum

Access systems, in providing a technology basis for dynamically reconfigurable radio nodes.

However, although it has been an active research topic for nearly two decades, the adoption

of SDR into products has been much slower than anticipated, and many of the initial goals of

SDR have not been met. This situation motivated the first main research goal of analyzing

the conceptual challenges of SDR and the degree to which these still remain challenges.

The results are provided in the first paper. This provides a comprehensive overview of the

challenges and opportunities associated with SDR, along with projections and suggestions

for the way ahead in this field. The paper addresses the architectures, waveform processing,

security, regulations and business models. The paper is the only Software Defined Radio

tutorial in the IEEE Communications Surveys and Tutorials online journal, and could serve

as a starting point for those entering this field of work.

Software Communications Architecture (SCA) is the dominant architecture standard for

SDR and defines a standardized environment for the SDR application components. While

promoting portability and reusability of code components, SCA also causes a processing

overhead. This motivates the second research goal, which investigates this processing overhead

in a particular processing environment. The results are provided in the second paper.

Workload measurements are conducted, and two simple models for the workload overhead

are described. The paper shows that for small intercomponent packet sizes and frequent

package transmissions the main overhead effects are related to the data transmission between

components. The models and measurement results can be used to provide guidance

on issues such as application granularity, intercomponent communication frequency and intercomponent

packet sizes. The paper makes a contribution to part of a research field where

there are few directly related publications.

In follow-up work to this paper, in Part I of this thesis, a walkthrough of various ways of

optimizing SCA-based environments is provided.

The continuing growth in the demand for wireless services, such as for wireless mobile

Internet access, creates an increasing demand for spectrum. Many authors have pointed to

the fact that the radio spectrum can be utilized more efficiently if radio nodes are allowed to

i


access the spectrum in a dynamic manner, rather than just relying on static spectrum assignments.

SDRs contribute to the evolution in this direction by providing flexible platforms and

a processing capacity for local processing of spectrum decision algorithms. The concepts

and algorithms for the spectrum decisions and the architectural context in which they are

made are equally challenging questions. This has motivated the third research goal, which is

to contribute to establish practical ways for dynamically sharing the frequency spectrum between

radio nodes. In particular, compare autonomous, distributed and centralized spectrum

sharing architectures. The results are provided in the third, fourth and fifth papers.

The third paper outlines a concept in which the spectrum decisions are made by a central

infrastructure. It outlines a hierarchical distributed system of Dynamic Frequency Brokers

(DFBs), that handles spectrum inquiries from radio clients and grants time-limited leases.

In simple terms, the paper proposes replacing parts of the manual regulatory processes by

a full hierarchy of brokers, with national or even higher-level authorities at the top. The

communication between radio nodes is proposed to be based on web services. A list of

security challenges is also included. The conceptual suggestions in the paper may be of use

for regulators or organizations interested in pursuing frequency broker systems.

In the fourth paper, centrally computed global optimum and autonomously computed

competitive optimum spectrum sharing are investigated and compared in a system consisting

of simplex radio links. In the analysis it is assumed that the available radio spectrum

can be divided into a number of spectrum segments over which interference and noise are

assumed to have constant power spectrum density. It is further assumed that the link bit rates

in each segment can be calculated as the information capacity of a modified signal-to-noiseand-interference

ratio in the segment. Illustrative map-based deployments of two single

interfering links, sharing two spectrum segments, are made and analyzed in detail. The paper

provides illustrations of cases where the autonomously computed competitive optimum

can become significantly worse, and more unfair between links, than the global optimum

solution. The paper makes suggestions about how to improve this situation. The investigations

and suggestions provide background information and advice for those interested in the

tradeoffs between centralized and autonomous solutions.

In the fifth paper a comparison is made between autonomous, distributed and central

(broker) architectures, in terms of spectrum decisions, computational complexity and communication

of spectrum management data between nodes or between nodes and brokers. An

n-link interference model that includes the same basic assumptions as in the fourth paper

has been used, with a further assumption of ideal administrative communication between

the links. The simulations include a significantly higher number of links and segments. Two

algorithms for distributed spectrum decisions are proposed, both of which target maintaining

a minimum rate for each radio link while maximizing the average rate over all links. It

is shown that for given minimum link data rates, better sum data rates can be achieved with

these distributed algorithms than with autonomous polices. The policies and algorithms can

be useful as candidates for the implementation in Dynamic Spectrum Access systems.

Software Defined Radio may have available computing power that it is attractive to

also use for other purposes than the execution of waveforms. In an example of this, in

the overview part, the integration of a Peer-to-Peer (P2P) agent for spectrum brokering has

been sketched out. In this P2P spectrum broker concept, the centralized broker is reformulated

into a distributed P2P functionality. By the use of a P2P command interface, spectrum

requests are handled jointly by the relevant peers in an overlay network, rather than by a

DFB.

ii


In summing up the different architectures, it is concluded that they all have advantages

and disadvantages. The distributed interaction one, supplemented by spectrum sensing and

aggregated persistent information in the form of centralized or distributed databases, is seen

as a good compromise. There are advantages in having flexible DSA agents that may coordinate

with DFBs or databases where appropriate, but may fall back to distributed coordination

and autonomous decisions when coordination channels are unavailable.

In conclusion, this dissertation has addressed key research issues in the evolvement from

fixed functionality fixed frequency-use radio transceivers to flexible software-defined dynamic

frequency use ones. The first paper makes suggestions for and clarifies the remaining

challenges to progress SDR as the preferred radio transceiver implementation technology.

The second paper and the related discussion in Part I of this dissertation provide guidelines

for granularity and performance optimization of SDR-applications in the dominating standardized

SDR-architecture, the Software Communications Architecture. The third, fourth

and fifth papers provide proposals for architectures and/or algorithms to enable the SDR

transceivers to dynamically exploit available radio spectrum.

iii


Preface

This dissertation is submitted in partial fulfillment of the requirements for the degree of

Philosophiae Doctor (Ph.D.) at the University of Oslo (UiO). My main supervisor has been

Professor and Director of Research Torleiv Maseng at UiO and Forsvarets Forskningsinstitutt

(FFI) respectively, and my co-supervisor has been Professor Frank Eliassen at UiO. The

studies have been carried out full-time between March 2006 and June 2009, and part-time

between July 2009 and October 2010, at FFI and at the University Graduate Center at Kjeller

(UniK).

The studies have been funded by FFI.

The subject areas of the dissertation, Software Defined Radio and Dynamic Spectrum

Access have both been proposed by Torleiv Maseng. Both he and FFI have given me extensive

freedom to select specific research topics both within and beyond these areas.

v


Acknowledgements

First and foremost I would like to thank my supervisors, Professor Torleiv Maseng and

Professor Frank Eliassen. Professor Maseng has enthusiastically followed up the progress

of the work several times per week, contributed with and discussed proposals, and also has

made his wide network of contacts available. When setbacks to the work have occured, his

convincing motto of ’persistence and sincerity’ has had a convincing effect also on me, and

made me continue forward. Professor Eliassen has responded quickly when I have emailed

questions, and has also generously offered work proposals.

I also wish to express my gratitude to FFI, represented by Torleiv Maseng, for funding

the studies.

I also want to thank everyone who has given me input and assistance during my work. In

particular I would like to thank Christian Serra, Marc Adrat, Audun Jøsang, Tor Gjertsen and

Asgeir Nysæter for providing input and advice for Paper A, Synnøve Eifring for helping me

with language-related issues of the same paper, Stewart Clark for helping me with languagerelated

issues of the same paper and also Part I. I would like to thank Lars Bråten and

Walther Åsen for proof-reading and inputs, Juan Pablo Zamora Zapata for sharing some of

his insights on software architecture and Unni Næss for artwork assistance. Further, thanks

to the co-authors of one or more of the other papers: Torleiv Maseng, Jon Olavsson Neset,

Jørn Kårstad and Toan Hoang. I also want to thank my colleague Elin for having volunteered

to be a test audience for conference presentations. Thanks also to Professor Pål Orten for

providing valuable advice during the final stages of the work.

I am grateful to the administration at UniK for providing a good environment for Ph.D.

studies.

Finally I would like to thank my fiancée Svanhild for morally supporting my studies, for

not complaining about weekend work and for helping me with the proof-reading of articles.

vi


List of Publications

Papers Included in Part II

Paper A Tore Ulversøy, ”Software Defined Radio: Challenges and Opportunities,” IEEE

Communications Surveys & Tutorials, Vol. 12, No. 4, 2010

Online journal: http://www.comsoc.org/livepubs/surveys/index.html

Also available from: http://ieeexplore.ieee.org

Paper B Tore Ulversøy and Jon Olavsson Neset, ”On Workload in an SCA-based System,

with Varying Component and Data Packet Sizes,” NATO RTO Symposium on Military

Communications, Prague, Apr. 21-22, 2008.

Available from: http://www.rta.nato.int/Pubs/RDP.asp?RDP=RTO-MP-IST-083

Paper C Torleiv Maseng and Tore Ulversøy, ”Dynamic Frequency Broker and Cognitive

Radio,” The IET Seminar on Cognitive Radio and Software Defined Radios: Technologies

and Techniques, The IET, Savoy Place, London, UK, Sept. 18, 2008.

Available from: http://ieeexplore.ieee.org

Paper D Tore Ulversøy, Torleiv Maseng and Jørn Kårstad, ”On Spectrum Sharing in Autonomous

and Coordinated Dynamic Spectrum Access Systems: A Case Study”,

Wireless VITAE ’09, Ålborg, Denmark, May 17-20, 2009.

Available from: http://ieeexplore.ieee.org

Paper E Tore Ulversøy, Torleiv Maseng, Toan Hoang and Jørn Kårstad, ”A Comparison of

Centralized, Peer-to-Peer and Autonomous Dynamic Spectrum Access in a Tactical

Scenario”, MILCOM 2009, Boston, October 18-21, 2009.

Available from: http://ieeexplore.ieee.org

vii


Other Co-authored Publications

In conjunction with the writing of this thesis, the candidate has also contributed to the following

conference papers:

S. Singh, M. Adrat, M. Antweiler, T. Ulversøy, T.M.O. Mjelde, L. Hanssen, H. Ozer,

A. Zumbul, ”NATO RTO/IST RTG on SDR: Acquiring and sharing knowledge for

developing SCA based Waveforms on SDRs”, Information Systems Technology Panel

Symposium IST-092/RSY-022 on Military Communications and Networks, Wroclaw,

Poland, September 28-29, 2010.

Available from: http://www.rta.nato.int

T. Hoang, M. Skjegstad, T. Maseng, T. Ulversøy, ”FRP: The Frequency Resource

Protocol”, IEEE International Conference on Communication Systems (IEEE ICCSS

2010), Nov 2010.

Available from: http://ieeexplore.ieee.org

Also in conjunction with the thesis, the candidate has contributed to the following multiauthor

Nato Research Task Group reports:

”NATO RTO/IST RTG on SDR Final Report”, Report from IST-080 RTG on SDR,

submitted to NATO RTO/IST Dec. 2010

”Cognitive Radio in NATO”, Report from IST-077 RTG-035 (in preparation)

viii


Contents

Abstract

Preface

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

List of Publications

Papers Included in Part II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Other Co-authored Publications . . . . . . . . . . . . . . . . . . . . . . . . . . .

Contents

Abbreviations

i

v

vi

vii

vii

viii

ix

xiii

I Overview 1

1 Introduction 3

1.1 Thesis Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2 Research Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.3 Research Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.4 Short Summary of Results and Implications . . . . . . . . . . . . . . . . . 14

1.5 Unaddressed Issues / Proposals for Future Work . . . . . . . . . . . . . . . 16

1.6 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2 Background 19

2.1 A Brief Introduction to Software Defined Radio . . . . . . . . . . . . . . . 19

2.2 Introduction to Dynamic Spectrum Access . . . . . . . . . . . . . . . . . . 22

3 Related Work 27

3.1 Related Work to the Results from Research Goal 1, Conceptual Challenges

of Software Defined Radio . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.2 Related Work to the Results from Research Goal 2, SCA Workload Overhead 28

3.3 Related Work to Research Goal 3, DSA Architectures and Algorithms . . . 32

4 Results and Implications 39

4.1 Research Goal 1, Software Defined Radio Challenges . . . . . . . . . . . . 39

4.2 Research Goal 2, SCA Workload Overhead . . . . . . . . . . . . . . . . . 42

4.3 Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms 58

ix


5 Conclusions and Recommendations for Further Work 75

5.1 Revisiting the Research Goals . . . . . . . . . . . . . . . . . . . . . . . . 75

5.2 Major Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

5.3 Critical Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

5.4 Recommendations for Further Work . . . . . . . . . . . . . . . . . . . . . 79

References 83

II Included Papers 95

A Software Defined Radio: Challenges and Opportunities 97

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

2 SDR and the Software Communications Architecture . . . . . . . . . . . . 103

3 SW Architectural Challenges . . . . . . . . . . . . . . . . . . . . . . . . . 104

4 Challenges and Opportunities Related to Computational Requirements of SDR110

5 Security Related Challenges . . . . . . . . . . . . . . . . . . . . . . . . . 115

6 Regulatory and Certification Issues . . . . . . . . . . . . . . . . . . . . . . 122

7 Opportunities Related to Business Models and Military and Commercial

Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

B

On Workload in an SCA-based System, with Varying Component and Data

Packet Sizes 143

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

2 Analysis Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

3 Workload Assessment Through Empirical Analysis . . . . . . . . . . . . . 148

4 Workload Assessment through Low-Complexity Analytical Models . . . . . 152

5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

6 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

C Dynamic Frequency Broker and Cognitive Radio 159

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

3 The Dynamic Frequency Broker . . . . . . . . . . . . . . . . . . . . . . . 165

4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

D On Spectrum Sharing in Autonomous and Coordinated Dynamic Spectrum

Access Systems: A Case Study 173

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

2 Background: DSA Concepts . . . . . . . . . . . . . . . . . . . . . . . . . 177

3 Near-Optimum Spectrum Sharing Study . . . . . . . . . . . . . . . . . . . 179

4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

x


E

A Comparison of Centralized, Peer-to-Peer and Autonomous Dynamic Spectrum

Access in a Tactical Scenario 191

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

2 Background and Calculation Models . . . . . . . . . . . . . . . . . . . . . 195

3 Spectrum Decisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

4 Computational Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . 202

5 Spectrum Coordination Traffic . . . . . . . . . . . . . . . . . . . . . . . . 204

6 DSA in a Hostile Environment . . . . . . . . . . . . . . . . . . . . . . . . 205

7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

xi


Abbreviations

A/D

AEP

API

ASIC

BS

CAB

CCM

CDMA

CDSA

CF

CO

CORBA

CPE

CPU

CR

D/A

DCIE

DCUE

DDS

DFB

DHCP

DIMSUMnet

DPC

DSA

DSAP

DSL

DSP

EAL

EDACC

ESSOR

EVP

FCC

FDMA

FFT

FIR

FPGA

GAO

Analog-to-Digital

Application Environment Profile

Application Programming Interfaces

Application-Specific Integrated Circuit

Base Station

Coordinated Access Band

Configurable Computing Machine, CORBA Component Model

Code Division Multiple Access

Coordinated Dynamic Spectrum Access

Core Framework

Competitive Optimum

Common Object Request Broker Architecture

Consumer Premise Equipment

Central Processing Unit

Cognitive Radio

Digital-to-Analog

DSA Capable Infrastructure Equipment

DSA Capable User Equipment

Data Distribution Service

Dynamic Frequency Broker

Dynamic Host Configuration Protocol

Dynamic Intelligent Management of Spectrum for Ubiquitous Mobile networks

Data Processing Component

Dynamic Spectrum Access

Dynamic Spectrum Access Protocol

Digital Subscriber Line

Digital Signal Processor

Evaluation Assurance Level

Event Driven, Administrative and Control Component

European Secured Software Defined Radio Referential

Embedded Vector Processor

Federal Communications Commission

Frequency Division Multiple Access

Fast Fourier Transform

Finite Impulse Response

Field-Programmable Gate Array

United States Government Accountability Office

xiii


GO

GPP

GPU

GTRS

HF

HTTP

HW

IDL

IF

IP

IST

ITU

IWF

JPEO

JTRS

kbps

MAC

MDD

MHAL

MILS

MIMD

MIMO

MIPS

MMT

MOM

MPMB

NC-OFDM

NCO

NIAG

OMG

ORB

OS

OSA

OSSIE

P2P

PCI

PCS

PDMA

PIM

PR

PSM

QoS

R&TTE

RANMAN

RF

RTG

RTO

Global Optimum

General Purpose Processor

Graphical Processing Unit

Swedish Common Tactical Radio System

High Frequency

Hypertext Transfer Protocol

Hardware

Interface Definition Language

Intermediate Frequency

Internet Protocol

Information Systems Technology Panel

International Telecommunication Union

Iterative Waterfilling

Joint Program Executive Office

Joint Tactical Radio System

Kilo-bits per second

Medium Access Control

Model-Driven Development

Modem Hardware Abstraction Layer

Multiple Independent Levels of Security

Multiple Instruction Multiple Data

Multiple Input Multiple Output

Million Instructions Per Second

Mobile Multi-standard Terminal

Message-Oriented Middleware

Multi-Protocol Multi-Band

Non-Contiguous Orthogonal Frequency Division Multiplexing

Network Centric Operations

NATO’s Industry Advisory Group

Object Management Group

Object Request Broker

Operating System

Opportunistic Spectrum Access

Open-Source SCA Implementation Embedded

Peer-to-Peer

Peripheral Component Interconnect

Partitioning Communication System

Polarization Division Multiple Access

Platform Independent Model

Packet Rate (data packets processed per second)

Platform Specific Model

Quality of Service

Radio and Telecommunications Terminal Equipment Directive

Radio Access Network Manager

Radio Frequency

Research Task Group

NATO Research and Technology Organization

xiv


RX

SCA

SCARI

SDMA

SDR

SIMD

SK

SNR

SOAP

SPIM

STRS

SW

TAO

TC

TCAM

TDMA

TGS

TI

TX

UAV

UDDI

UMTS

USRP

UWB

WF

WRAN

WSDL

XML

Receiver

Software Communications Architecture

Software Communications Architecture - Reference Implementation

Space Division Multiple Access

Software Defined Radio

Single Instruction Multiple Data

Separation Kernel

Signal-to-Noise Ratio

Simple Object Access Protocol

Spectrum Information and Management

Space Telecommunication Radio System

Software

The Ace ORB

Trusted Computing

Telecomm. Conformity Assessment and Market Surveill. Committee of the EU

Time Division Multiple Access

The TCAM Group on SDR

Texas Instruments

Transmitter

Unmanned Aerial Vehicle

Universal Description, Discovery and Integration

Universal Mobile Telecommunications System

Universal Software Radio Peripheral

Ultra Wide Band

Waterfilling

Wireless Regional Area Network

Web Services Description Language

Extensible Markup Language

xv


Part I

Overview

1


Chapter 1

Introduction

The short history of wireless communications, taking Hertz’ 1887-1888 experiments as the

start [1], has shown breathtaking advances in both the efficiency by which the spectrum is

exploited and in the quality and variety of the wireless services offered.

In only the last two to three decades, revolutionizing services have appeared, such as the

digital mobile phone systems, wireless local area networks, personal area data networks, digital

TV broadcasts, broadband data mobile services and global positioning services. These

astonishingly rapid advances have been made possible by technological innovations in many

different areas, ranging from implementation technology through coding inventions and to

advances in computer network and distributed processing technology.

Radio implementation technology has gone through several technology stages, starting

with the spark gap transmitter era from 1887/1888 to approximately the mid 1920s, known

as the ’electric’ era of radio technology. This was overlapped and followed by an ’analog

electronics’ era, characterized first by the vacuum tube-based transceivers and subsequently

supplemented and followed by transistor equipment some years after the discovery

of the transistor in 1947 [2]. Towards the end of the 1950s, the idea to integrate a multitude

of transistors on a single substrate was pursued [3], leading to integrated circuits of

increasingly higher transistor densities, and where the number of components that could be

cost-effectively placed on the integrated circuits were later shown to increase by a yearly

factor [4], now referred to as Moore’s law. The integrated circuits allowed for denser integration

of the analog electronics, but more importantly paved the way for integrated digital

circuitry and for the era of the digital transceiver, in which major parts of the signal processing

are conducted in the digital domain. Of particular importance in the context of this thesis,

was that this development also prepared the grounds for the Digital Signal Processor (DSP),

appearing from 1978/1979 [5] and the Field-Programmable Gate Array (FPGA), invented

in 1985 [6], as well as the general purpose microprocessor that came as early as 1971 [7].

These components gradually allowed increasing parts of the radio transceiver functionality

to be implemented in software, as pointed out by Mitola in 1991 [8, 9], who then coined the

term ’Software Radio’. This leads into the ’software’ era of radio technology and into one

of the research topics of this thesis, the Software Defined Radio (SDR).

The use of software in the implementation of radio transceivers has not only been motivated

by the implementation technologies that have become available, but has also been

influenced by several other technological developments. The evolvement of the computing

discipline itself, including computer architecture, computer networks and distributed processing

technologies, are of course part of this. The evolution from radio links and broad-

3


1. INTRODUCTION

cast transmissions to networked radio communications, where the ALOHA network is an

early example [10], is another factor in that networked radio communications involved an

increased protocol content that favoured a software implementation. The creativity in the

signal processing and information theory disciplines, example innovations being the 1993

Turbo codes [11] and the Multiple Input Multiple Output (MIMO) [12] technology, resulting

in rapid releases and a multitude of wireless standards both within commercial and military

domains, are also motivating software implementations. However, there are still many challenges

related to software-based transceiver implementation and the transition from ’digital

transceivers’ is slower than many expected, as will be discussed in this thesis.

The negative effects of interference between different radio stations were discovered

very early in the history of radio communications, and at that time being a major concern

for safety at sea, e.g. in terms of interference with distress messages from ships. The need

for national and international regulations of the use of the radio spectrum were thus clearly

seen. The first international spectrum allocation regulations were put in place through the

1906 Berlin agreement [13], and the World Radio Conferences have since become and are

an arena for international radio spectrum agreements. The national spectrum management

authorities provide the specific spectrum assignments to radio network operators or users

consistent with these international regulations. While this detailed regulatory approach unquestionably

has been an important factor in the success of wireless communications, it has

been pointed out that the processes involved are slow and the resulting efficiency in the use

of the spectrum is low. From about 1980 and on we have, however, seen a trend of deregulation

of spectrum access, putting more focus on market-based use of spectrum and that the

spectrum users know how to make the best use of the spectrum compared to detailed government

regulation [14]. In recent years, the most remarkable evidence of the deregulations

are the tremendous success of wireless local area network devices in unlicensed open-access

bands, and the in-progress opening-up to secondary access of spectrum assigned for primary

use in TV broadcasting. The deregulation may be seen as a political, liberal trend, but it is

also a result of the increasing needs for wireless communications, due to e.g. the evolution

of the personal mobile phones, the success of the Internet and the connected everywhere

paradigm causing a need for more effective use of the spectrum. As well, it is a result of

the impracticalities of manual regulatory processes in contrast to the ever-increasing number

of transceivers. Nowadays, computerized devices can have several transceivers and are in

networked contact with other such computerized devices, and in the future such computerized

devices are likely to be integrated in humans, pets and a vast variety of the ’things’ in

our surroundings. A number of researchers have in recent years argued for further spectrum

efficiency increases through radio systems that efficiently and dynamically take advantage

of the available spectrum, a lot of activity is focused on this field but still a number of challenges

exist.

The software approach to radio terminals is one of the factors making such a dynamic

and deregulated approach to frequency more and more viable. The use of the SDR for decentralized

dynamic spectrum access decisions, enabled through the pairing of software radio

technology with artificial intelligence which was coined ’Cognitive Radio’ [15], was suggested

by Mitola and Maguire in 1999. Software Defined Radio may run both the processing

of the sensing of the radio spectrum and other forms of ’context awareness’, as well as the

spectrum decision algorithms. For such decentralized spectrum access changes, or for dynamic

changes dictated by other mechanisms as described later in this thesis, the Software

Defined Radio provides the runtime flexibility needed to dynamically change waveforms,

4


1.1. Thesis Motivation

such that the spectrum resources may be dynamically and optimally exploited.

Software Defined Radio, Cognitive Radio and Dynamic Spectrum Access (DSA) radio

terminals are often referred to as emerging radio systems, and are predicted to be major

trends in the next decades of wireless communications. The research goals in this thesis are

within these emerging radio technologies.

In this introductory chapter, the motivations for the research goals are described first, and

illustrated by example wireless scenarios. Then the specific research goals are defined. This

is followed by a description of the research methods used in pursuing them. The research

contributions are then summarized. Lastly, suggestions for further work are provided, as

well as an overview of the remaining parts of the thesis.

1.1 Thesis Motivation

Todays wireless communication systems and scenarios are still dominated by ones where

the waveform-functionality is fixed when the units leave the factory, not allowing waveform

upgrades. Similarly the bulk of the systems are not prepared for flexible use of spectrum,

the use of spectrum instead being dictated by lengthy regulatory or frequency planning processes.

Particularly in military radio communication there is a striking contrast between the long

acquisition and system lifetime (i.e. decades) of the radio communication systems on the one

hand, and on the other hand the need for changes and upgrades to waveforms on these units

that are on a much more rapid timescale, and that are disallowed by these radio platforms

without costly unit hardware upgrades. Similarly there is a striking contrast between the

laboursome frequency pre-planning processes for ad hoc operations, and the need for indeployment

rapid frequency changes e.g. due to additional system deployments, network

mobility or interference from other systems.

The research goals, described in Chapter 1.2, are motivated as individual sub-goals

within an overall vision of moving away from these wireless scenarios and systems where

transceiver functionality, frequency assignments and also the roles of the transceiver are predominantly

fixed or preplanned. The envisioned future scenarios are ones where transceivers

have reconfigurable functionality, dynamic frequency assignments, dynamic transceiver

roles and the networks ultimately are wide-coverage multi-frequency ones.

As a further motivation that leads to the research goals, issues with some presently deployed

wireless systems will be reviewed.

1.1.1 Issues with Presently Deployed Wireless Systems

Consider the scenario depicted in Figure 1.1, which contains a few typical presently deployed

civilian wireless systems: A mobile phone cellular access system, a terrestrial TV

broadcaster and receivers, and a satellite services system.

Typical for the transmitters and receivers in this scenario is that they conform to one or

a fixed small number of waveform standards. The waveform functionality, reflecting each

waveform standard, in the transmitters and receivers, is also fixed and is not programmable.

When new waveform standards are introduced in Figure 1.1, e.g. a higher-bitrate standard

for the cellular access system, a replacement of the whole transmitter/receiver is needed

5


1. INTRODUCTION

UN, FFI 2010

FIGURE 1.1: A civilian wireless scenario with some typical system elements: Cellular access, TV

broadcast and satellite services. The colored circles and lines illustrate different spectrum intervals.

(referred to by market analysts as the ’handset replacement model’ [16]). Unfortunately,

this replacement model is a large waste of resources.

A further characteristic of the systems in Figure 1.1 is that the frequencies that they

use are typically assigned by the relevant National Frequency Authority as long-duration

(normally several year) licenses. Such fixed long-term assignments provide excellent interference

control. However, measurement campaigns have shown [17, 18] that typically only

a few percent of these frequencies are actually occupied at any time and location, i.e. the

frequency use is inefficient. This wasteful use of the frequency resources is in contrast to the

increasing need for wireless communication capacity for existing and new wireless services.

Another characteristic is also that the radio terminals have a specific, factory defined set

of available ’roles’, e.g. the role as a cellular mobile device and the role as a WiFi data access

terminal. They may, however, not be field configured to fulfill new roles, such as enabling

the mobile cellular access terminal to additionally be a satellite terminal.

Figure 1.2 illustrates typical ad hoc deployed systems in the same way, such as in a

peace-keeping mission: The workhorse in the operations is the mobile operation ad hoc network,

which serves as the main voice and data communication link for the mobile units, and

which may work autonomously or as illustrated be linked to a headquarters or wider network.

Since typically several partner nations cooperate, a mobile operation ad hoc network

of a coalition partner in the same mission is also illustrated. Typically many sources of interference

may exist in such a scenario, including unregulated ad hoc networks of uncertain

origin and ownership, as well as pirate broadcasters. Additionally, typical civilian systems

such as in Figure 1.1 are likely to coexist in the same scenario.

In the same way as in the previous figure, the transceivers in the scenario conform to one

or a small number of fixed waveform standards, with factory-defined waveform functionality.

For a peace-keeping mission, this has the effect that when a new coalition partner, typically

using a different set of waveform standards, enters the scenario, transceivers typically

6


1.1. Thesis Motivation

Unregulated

broadcast

Partner network

Unregulated

ad hoc network

Mobile operation

ad hoc network

Partner

network

UN, FFI 2010

FIGURE 1.2: A wireless scenario for an ad hoc operation, e.g. a peace-keeping mission, with some

typical system elements: Ad hoc networks, radio links, satellite services. Colored circles / lines depict

different spectrum intervals. Issues arise due to unregulated electromagnetic activity causing interference,

and due to cooperating parties using radio equipment with hardware-defined incompatible

waveforms.

need to be loaned between the coalition partners in order to be able to inter-communicate.

Alternatively, the coalition partners would both need to replace their transceivers. This

causes both logistical and lead-time issues in the set-up of such an operation, and also involves

additional expenses.

Also, in the same way as in the previous figure, the radio terminals have factory-defined

roles, not allowing the ad hoc terminals to take on new roles (e.g. as maritime communication

devices) if they did not have this functionality when leaving the factory. Again

this leads to logistics and lead-time issues in getting additional equipment in place if the

communication needs in the operation change during its progress.

The frequencies used by the systems in Figure 1.2 are typically planned prior to the actual

mission, based on the knowledge as to what interference will be present in the actual

scenario, and on the available, but imperfect, knowledge about which other networks the

transceivers will have to coexist with. By experience [19], this process does not always yield

good network connectivity in the actual scenario, due to unexpected interference from such

sources as pirate transmitters, unregulated devices, rescue workers fielding their own communication

systems in an uncoordinated manner and other types of unplanned interference

(illustrated by an unregulated ad hoc network and an unregulated broadcaster in Figure 1.2).

This type of complex, partly unregulated and partly hostile electromagnetic environment is

another argument for the need for more agile spectrum behavior in the radio system.

A further typical characteristic of the ad hoc networks in Figure 1.2 is that they are

single-frequency-interval ones. This implies that all the nodes in one network, in a time

divided way, compete for the same part of the wireless medium. This results in frequent

7


1. INTRODUCTION

data packet collisions, and hence low throughput, when the node density is high.

1.1.2 Envisioned Future Wireless Scenarios

Conventional fixed

freq. broadcaster

Conventional

mobile access

DSA mobile access

DSA multihop

multifrequency best

effort access

UN, FFI 2010

FIGURE 1.3: An envisioned future wireless systems scenario with a mixture of conventional fixed

frequency mobile access, dynamic spectrum mobile access and best-effort dynamic spectrum access

multihop ad hoc networks.

The envisioned future version of the Figure 1.1 scenario is illustrated in Figure 1.3 which

illustrates a mixture of the conventional cellular access and broadcast systems we know

today, with new ones utilizing various forms of dynamic access to the radio spectrum.

In the figure, all the radio nodes are assumed to allow loadable waveform functionality,

rather than having fixed, factory-defined functionality. As new communications standards

are released, this means that, within the hardware processing limitations of the radio

platforms, the standards can be implemented as software upgrades in the radio nodes, and

even as automatic over-the-air downloaded upgrades. This also implies that the radio functionality

may be software-adapted to the actual needs in the user scenario. With sufficient

flexibility of the radio platforms, such adaptations may also include dynamic, rapid changes

to waveforms in order to optimally exploit available spectrum resources and dynamically

access new spectrum resources as needed. The waveform loadability and radio platform

flexibilities also enable radio nodes to be software-uploaded to fulfill new roles.

As mentioned, better utilization of the electromagnetic spectrum may be achieved by

allowing the spectrum assignments to be dynamic rather than fixed. It is anticipated that

such dynamic access systems will coexist with systems that will remain fixed licensed ones,

at least for the not-to-distant future. This is illustrated in the figure with (to the left) a

conventional mobile access system and a conventional broadcaster with fixed frequency assignments.

The Dynamic Spectrum Access mobile access network (to the right of these)

on the other hand may benefit from dynamically available spectrum and provide cheaper

mobile access, but with a lesser quality-of-service guarantee.

8


1.1. Thesis Motivation

Another envisioned way of exploiting Dynamic Spectrum Access may be DSA multihop,

multifrequency networks, as illustrated to the right in the figure. Here, the communication

may be forwarded by several radio nodes, and different spectrum intervals may be used at

the different links in the network. The traffic is envisioned to be connected to the backbone

network utilizing very different types of access, depending on what is available at given

locations, ranging from satellite communication through cellular infrastructure access to

low cost or freely available connections to home networks or to networks at universities or

organizations. For the consumer domain, such networks are envisioned to provide cheaper

mobile data access on a best-effort basis.

A further argument for the need of Dynamic Spectrum Access solutions is also new wireless

services that will appear and need communication capacity, and the enormous number

of wireless devices that will be present. Examples of such services are home multimedia

broadcasts, camera to home server video and picture transfers, medical sensors on humans

and pets that communicate with hospital servers or pet clinics, respectively.

The envisioned future version of the scenario in Figure 1.2 is illustrated in Figure 1.4.

Figure 1.4, like Figure 1.2, shows a mobile operation ad hoc network and a coalition ad

hoc network, as well as terrestrial and satellite links, together with sources of interference

that may be expected in such a scenario. In contrast to Figure 1.2, all the radio nodes are

software-defined ones, providing flexibility both in terms of waveform functionality, radio

terminal roles and in the use of spectrum. Additionally it is assumed that the waveform

software and radio platforms are defined in such ways that the software to a large degree

does not rely on one specific platform.

Unregulated

broadcast

Unregulated

ad hoc network

Partner

network

Mobile operation

ad hoc network

UN, FFI 2010

FIGURE 1.4: An envisioned future wireless systems scenario for an ad hoc operation or peacekeeping

mission. The systems may dynamically avoid interference sources. As the radio nodes are

SDR ones, waveforms to enable interoperability between coalition partners may be loaded as needed.

Dynamic use of frequencies throughout the multi-hop network enables flexible, high number of nodes

wide-area coverage.

9


1. INTRODUCTION

In the scenario in Figure 1.4, the waveform loadability properties in combination with

such waveform to platform independence properties, mean that the interoperability issue between

the partner network and the mobile operation network may now be solved by moving

the waveform application from either type of radio platform to the other one. Alternatively,

a special ’coalition’ waveform software may be loaded onto both types of radio platforms to

enable interoperability.

The waveform loadability and radio platform flexibilities enable radio nodes also to be

software-uploaded to fulfill new roles. Thus in the scenario in Figure 1.4, one fulfilling

the mobile ad hoc role may load a waveform in order to simultaneously provide satellite

services.

The DSA ad hoc network, instead of using pre-planned frequencies, may dynamically

avoid the unregulated sources of interference in the scenario, and even simultaneously run

defensive and offsensive counter-efforts towards these unregulated radio nodes. Multihop

DSA, with different spectrum intervals used on different links or subnets [20], enable

scenario-adapted wide-coverage ad hoc networks.

1.1.3 Motivation for the Research Goals

Obviously, a vast number of research challenges are present in order to move from the

present-type wireless communication as depicted in Figure 1.1 and Figure 1.2 to reach the

envisioned scenarios in Figure 1.3 and Figure 1.4. The research goals in this thesis are

motivated from observations of individual challenges that are small pieces in what is needed

for this transformation to occur.

A cornerstone component at the bottom of this vision is the flexible radio platform enabled

by the Software Defined Radio implementation technology. SDR has been a hot research

topic for almost two decades, and with some of the base technologies having matured

far longer than that. At the start of this thesis work in 2006, a picture of SDR emerged

that was full of contrasts and on the one hand a lot of optimism: A number of books, articles

and presentations with very visionary and optimistic thoughts of SDR existed. On the

other hand, there were disappointing observations: A number of optimistic projections for

SDR to take over as a baseline technology, for example the projection in [21], had proved

to be too optimistic. Mobile phones, although they had evolved to multi-standard devices,

were still using conventional digital design and not SDR. Also the prestigeous US Defence

Joint Tactical Radio System (JTRS) SDR program had experienced severe difficulties and

delays [22], and comments were made with respect to severe remaining risks. This cornerstone

importance of Software Defined Radio within the overall vision, and this contrasting

picture of its status and remaining challenges, have motivated the first research goal, related

to SDR conceptual challenges, as described in the next section.

Flexible radio platforms become an even more significant enabler in the above envisioned

future scenarios, if the platforms are standardized such that individual waveform

applications may have a larger market, or at least such that waveform applications can be

moved from platform to platform with moderate effort. The dominant SDR architectural

standard is the Software Communications Architecture (SCA). SCA promotes a platformto-application

separation by specifying an SCA platform environment for SCA-based application

components. Each application component may be executed on one of the individual

processing elements within a distributed-processing platform. A fine granularity of these

applications promotes code reuse and thereby the efficiency of building new waveform ap-

10


1.2. Research Goals

plications, in that individual components more easily may be reused in new applications.

However, the finer the granularity, the more processing cycle overhead is generated. A wish

to find out more about the negative effects of the Software Communications Architecture approach

in the form of these overhead losses and their implications on application granularity,

has motivated the second research goal, as described in the next section.

Dynamic radio spectrum assignments are a key factor in the future scenarios in both

providing a higher spectrum utilization and also for ad hoc wireless network deployment

in complex environments with unknown electromagnetic activity. Although in principle

dynamic sharing of spectrum may seem straightforward, it triggers numerous technological

questions.

In particular the author observed in the early research on Dynamic Spectrum Access, that

there was an unrealistically high reliance on spectrum sensing of other radio transceivers in

the suggested approaches. Issues such as silent receivers that are not detected by sensing,

and symmetry issues with low-power transceivers that are difficult to detect yet may still be

disturbed by higher-power DSA transceivers, pointed in the direction that spectrum sensing

alone is likely to be inadequate as the single source of input for doing DSA. This influenced

the research in the direction of other ways of doing DSA than just sensing.

With a base focus on the intersection between distributed processing and radio technology,

and coming from the distributed processing environment of the Software Defined Radio,

it was natural to take the distributed processing view also towards DSA. Together with

the above observations on sensing feasibility, this motivated the study of DSA architectures

part of the third research goal. The ’algorithms’ part was motivated from the challenging

nature of the associated issues.

Some further motivations for the third research goal were the interaction with the NATO

research group RTO/IST RTG on Cognitive Radio in NATO and also reading of Mitola’s

dissertation on Cognitive Radio [23].

1.2 Research Goals

Motivated from the challenges described above, this work has had the following research

goals:

1.2.1 Research Goal 1: The Conceptual Challenges of Software

Defined Radio

The concept of implementing radio waveform functionality in software deployed on a nonwaveform-specific

platform has many associated challenges. Some of these have been

solved, partly or wholly, but some still remain challenges at the time of this writing. The

first research goal targets this issue:

• To assess the degree to which the conceptual challenges of SDR have been solved

and what challenges still remain in the fields of software architecture, computational

requirements, security, regulations and business structure.

11


1. INTRODUCTION

1.2.2 Research Goal 2: SCA Workload Overhead

The second research goal explores the negative effects of the Software Communications

Architecture on processing efficiency. The goal is:

• To analyze the workload overhead of Software Communications Architecture-based

SDR in a specific processing configuration.

The details of the specific processing configuration are described in Paper B in Part II.

1.2.3 Research Goal 3: Dynamic Spectrum Access Architectures and

Algorithms

The third research goal in this thesis is:

• To contribute to establishing practical ways of dynamically sharing the frequency

spectrum and in particular compare autonomous, distributed and centralized spectrum

sharing architectures.

The scope of the algorithms 1 part of this research goal is limited to using the Waterfilling

algorithm as described in [24] as a starting point for the work.

1.3 Research Methodology

1.3.1 Methodology for Research Goal 1: Conceptual Challenges of

Software Defined Radio

The main research method used in this part of the work is a thorough study of the published

literature. Knowledge has been gathered by investigating sources such as IEEE Explore [25],

SpringerLink [26] and the ACM Digital Library [27], and by doing broader searches using

Google Scholar [28] and Google [29]. Further, the electronic proceedings from the major

yearly conference in the SDR arena, the Software Defined Radio Technical Conference, have

been searched. In order to get the broader overview, relevant chapters of recent textbooks

[21, 30–32] on the subject have been studied.

In order to get a deeper understanding of the SDR challenges and allow for in-depth

discussions with an international group of experts, the author has attended and contributed

to the work of a research task group on SDR, the RTO-IST-080 ”Software Defined Radio”.

The task group has facilitated information sharing between its members, and has performed

experiments and demonstrations [33, 34] using Stanag 4285 [35].

Design experiments with SDR applications, mainly using the Stanag 4285 code, have

also been used as a method to get hands-on experience, both with SDR application development

in general, and also to get experience with various development tools. Two different

types of SDR platforms have been used. The first is a commercial platform, the flexComm

Waveform Design Studio made by Spectrum Signal Processing. The second is a Linux PC

1 In this thesis, the term algorithm has been used both when describing centralized calculations as well

as distributed and autonomous calculation processes. While each autonomous or distributed agent may do its

decision calculation according to an algorithm, a better term for the total process of calculating a resulting state

between a number of autonomous or distributed radio agents is interactive computation.

12


1.3. Research Methodology

based platform, using the OSSIE [36] SCA-based Core Framework made by Virginia Tech,

and optionally an Universal Software Radio Peripheral (USRP) for transforming the signal

samples to/from the radio frequency domain.

The paper has been reviewed by IEEE-appointed experts. During the process, the reviewers

have suggested improvements and requested expansion in certain parts. These improvements

and expansions have been included in the final version of Paper A that is included in

Part II of this thesis.

1.3.2 Methodology for Research Goal 2: SCA Workload Overhead

Research goal 2 is entirely within the computer science discipline. As cited in [37], Denning

et al. have proposed three major paradigms for this discipline; theory, abstraction and design

[38]. Denning et al. comment that the three paradigms are intertwined [38].

TABLE 1.1: Stages in the theory, abstraction and design paradigms, quoted from [38].

Stage theory abstraction design

(1) characterize objects of form a hypothesis state requirements

study

(2) hypothesize possible construct a model and state specifications

relationships among make a prediction

them

(3) determine whether the

relationships are true

design an experiment

and collect data

design and implement

the system

(4) interpret results analyze results test the system

Other authors have suggested finer grained division of scientific research methods. For

example, Glass [39] promotes a taxonomy of four methods, referred to as the scientific, engineering,

empirical and analytical methods. In another example [40], 22 different research

methods are listed.

Here, the targeted methodology for research goal 2, following the terms by Denning

et al., was the abstraction paradigm, forming a hypothesis and creating a model, then attempting

to verify this model with measured data. In setting up the system in which the

experiment is conducted, elements of the design paradigm are mixed in.

The specific choices as to which type of models, techniques and measurements that were

used, and the reasoning behind these choices, are described in Chapter 2 of paper B and in

Section 4.2.1.1 in Part I.

The experimental work was preceded by a literature study. The same sources of literature

as cited in 1.3.1 were used in order to find relevant literature. This work benefitted from

reading Fortier and Michel’s book on systems performance evaluation [41] as preparation.

1.3.3 Methodology for Research Goal 3: Dynamic Spectrum Access

Architectures and Algorithms

Research goal 3 is rooted in a broader scientific context, including information theory, mathematics,

radio technology and computer science.

13


1. INTRODUCTION

In relation to the spectrum sharing part of the goal, the methodology is mainly an engineering

one in Glass’ terms (and which resembles the design paradigm in the terms in [38]):

observe existing solutions, propose better solutions, measure and analyze.

In observing existing solutions, a literature study was conducted, again with the same

sources as used above for the search. The study also included reading some of the recent

textbooks on Cognitive Radio, including Fette’s [42] work.

The next step is propose better solutions, an example being the suggested distributed

interaction algorithm in Paper E in Part II of this thesis.

Measure in the strict sense means measuring the actual performance of the actual proposed

system in the actual real world surroundings. Mainly due to time limitations, Paper

C only has a qualitative discussion of the proposed concepts, and in Papers D and E, Matlab

computer calculations and simulations with idealistic assumptions are presented. Such

simulations provide information on how the proposed solutions work in the simulated environment.

Depending on the accuracy of the simulated environment, the simulations can

provide a variable degree of support for the proposed solutions. Such simulations, however,

cannot be regarded as scientific evidence that the proposed better solutions work with the

real system in the real world. Hence, further work is needed in this area in order to verify

the performance in real-world surroundings. The simulation assumptions and their validity

are discussed in Section 4.3.2.2.

The analysis step is based mainly on the simulation results, for which the same comments

as above apply.

In order to discuss the results in this research topic with an international group of experts,

the author participated in the research task group RTO-IST-077 ”Cognitive Radio in NATO”.

1.4 Short Summary of Results and Implications

1.4.1 Research Goal 1: Conceptual Challenges of Software Defined

Radio

Research goal one is answered in Paper A in Part II. The paper provides a comprehensive

summary of SDR challenges and opportunities, and makes suggestions for the future path

of SDR.

The paper contains a brief summary of the Software Communications Architecture.

While one of the goals of SCA is to make portability easier, there are still remaining challenges

in this area, as detailed in the ’SW Architectural Challenges’ section. This section

also reviews the present state of SCA tools, and discusses alternative architectures to SCA.

In the ’Challenges and Opportunities Related to Computational Requirements of SDR’

section, computational requirements of SDR as well as available processing elements are

reviewed. The most interesting trend in this area, for handheld units, is seen to be multiple

Single Instruction Multiple Data (SIMD) core processors.

There is a lot of research literature on aspects of security such as secure downloading of

applications and protection of the SDR platform against attacks. The portability of the security

components of SDR applications between different platform domains is still a challenge,

however. Ways of dealing with this issue are suggested.

14


1.4. Short Summary of Results and Implications

Regulatory changes are already introduced, and further regulatory changes that are considered

still needed, are described. The need for an SCA certification authority and procedure

for the European domain is pointed to.

Product opportunities in both the military and civilian domains are reviewed. Such

opportunities include military cognitive communication systems and military waveform libraries.

Examples of civilian opportunities are multiprotocol multiband base stations, mobile

multi-standard terminals and satellite devices.

The paper has a comprehensive reference list with 151 references including a number of

landmark SDR publications.

This paper is the only SDR tutorial in the IEEE Communications Surveys and Tutorials

and may serve as an introduction for those entering this area of research or engineering. The

paper also provides suggestions to the SDR community as to how SDR could be further

developed.

1.4.2 Research Goal 2: SCA Workload Overhead

This research goal is treated in Paper B, as well as in the follow-up work to Paper B, in Part

I of this thesis. Paper B investigates the effects of varying the granularity of an SCA-based

application on the total workload of a processor. Empirical analysis is made. Also, two

simple analytical models are formulated for the total workload, one that does not include the

effects of context switches, and another that includes these effects. Comparing results from

the first model with measurements shows that it includes the dominant part of the workload

overhead but underestimates the total workload of the system. This underestimation is

greater the larger the packet size. A likely major contributor to the underestimation is that

the first model does not include the indirect losses due to the required updating of caches

following a context switch.

Table 2 in the paper also illustrates the large startup cost for pushing a minimum packet

from one component to another, suggesting that for workload optimization it is advantageous

to avoid the frequent intercomponent transmission of small packets.

In the follow-up work in Part I, further optimizations of SCA-based environments and

applications are described and further measurements provided.

The results provide guidelines for application granularity of SCA-based applications for

designers of such applications, when several components run on the same CORBA-enabled

processor. The results may also be used to provide guidelines for packet size versus packet

frequency for such applications. The results also provide guidance as to other factors in

SCA-environments and applications that can be optimized.

1.4.3 Research Goal 3: Dynamic Spectrum Access Architectures and

Algorithms

The results related to this research goal are provided in Papers C through E, as well as in

Section 4.3.1.3 in Part I.

The architectural part of the research goal is mainly treated in Paper E, which provides

a comparison between DSA architectural concepts. The main conclusion drawn is that the

distributed (decentralized) architecture, preferably with combinations of local information

exchange and indirect information in the form of databases and sensed spectrum, is the

15


1. INTRODUCTION

favored one, but that centralized, distributed and autonomous architectures all have their advantages

and disadvantages. A particular architecture, the hierarchical central coordination

concept named the Dynamic Frequency Broker (DFB), is outlined in Paper C. Also, Section

4.3.1.3 in Part I provides a description of a Peer-to-Peer (P2P) DSA architecture, where the

main idea is that of a distributed spectrum broker functionality and interaction between the

radio peers on a P2P overlay rather than bound to specific physical coordination channels.

The algorithm part of the research goal is treated in Papers D and E. Paper D compares

autonomously calculated power versus frequency assignments in a link interference model to

centrally calculated global optimums. Paper E proposes three autonomous policies and two

distributed interaction algorithms for power versus frequency assignments. It also provides

Matlab simulation results for these proposed policies and algorithms. The simulations, under

the specific assumptions made for the simulated environment, show that when requiring

all links to meet the same minimum rate, the distributed interaction algorithms provided a

higher average rate over all the links, than the autonomous policy-governed cases.

1.5 Unaddressed Issues / Proposals for Future Work

1.5.1 Research Goal 1: Conceptual Challenges of Software Defined

Radio

This research goal does not include the analog hardware and hardware-component-related issues

of SDR. To some extent, analog-to-digital conversion issues, however, are dealt with in

Section 2.1. Furthermore, Paper A and Section 3.1 provide advice about additional sources

of information on these aspects. As is evident from Section 3.1, there is already a rich variety

of publications available dealing with hardware issues of SDR, and as such there is not

an imminent need to extend the work of Paper A in this direction.

Software Defined Radio in relation to the new trend of Green Telecommunications is

also a topic which is not dealt with in Paper A. In light of the societal importance of the

environmental aspects of telecommunications, this is a topic that deserves further work.

1.5.2 Research Goal 2: SCA Workload Overhead

The work on this research topic did not comprise verification of modeled versus measured

data of the second derived model in Paper B, as accurate estimates of the parameters in

the second model were not within the scope of the work. Further research on the second

model including improved methodology for estimation of the parameters is suggested as an

extension of this work.

The research covers the single-processor implications of application granularity, as well

as multiprocessor environments that are such that more than one component needs to be

run on each CORBA-capable General Purpose Processor (GPP). A natural extension of the

work, however, is towards workload optimization in a general heterogeneous multiprocessor

environment. A qualitative overview of considerations for such environments is given in

Section 4.2.2 and a more detailed, quantitative analysis is suggested as future work in this

area.

16


1.6. Thesis Organization

1.5.3 Research Goal 3: Dynamic Spectrum Access Architectures and

Algorithms

The results from this research goal are based largely on numerical calculations and simulations,

as well as references to the literature. Due to the fact that the time required and other

resources needed to build experimental networks exceeded what was available, the plan of

the work did not include measurements of experimental systems. The design of experimental

networks for further verification and experimentation with the suggested algorithms is

suggested as a future work topic.

1.6 Thesis Organization

This thesis is organized in two parts. Part I provides an overview of the research work and a

summary of the resulting published papers and Part II contains these papers.

In Part I, Chapter 2 provides some further introduction to the research areas of this thesis,

Software Defined Radio and Dynamic Spectrum Access, for the benefit of the reader who

does not already have detailed knowledge about these topics. Chapter 3 contains a review of

research related to the three research goals. Chapter 4 summarizes the contributions related

to the same three research goals. Chapter 5 contains conclusions and suggestions for further

work.

In the papers in Part II where there are co-authors, the contributions of this author is

detailed on the front page of each paper.

17


Chapter 2

Background

2.1 A Brief Introduction to Software Defined Radio

The term ’Software Radio’ was coined by Joseph Mitola III in 1991 [9], who also described

the concept in [8]. The term ’Software Defined Radio’ was adopted [32] by the SDR Forum,

now the Wireless Innovation Forum 1 , a membership organization for the different actors

with an interest in SDR.

While there is no globally unique definition of Software Defined Radio, many definitions

exist, as proposed by different authors and organizations. The Wireless Innovation Forum

defines SDR as a ”Radio in which some or all of the physical layer functions are software

defined” [43]. The Federal Communications Commission (FCC) similarly has its own definition

[44], that also includes the software that affects the compliance with FCC rules: ”A

radio that includes a transmitter in which the operating parameters of frequency range, modulation

type or maximum output power (either radiated or conducted), or the circumstances

under which the transmitter operates in accordance with Commission rules, can be altered

by making a change in software without making any changes to hardware components that

affect the radio frequency emissions.” Some sources define alternative terms (e.g. ’Flexible

Architecture Radio’ [32]) depending on how much of the radio functionality is implemented

in software. A main parameter in the various definitions and alternative terms, is how flexibly

the radio waveform can be changed through changing software without modifying the

SDR Platform. The ’waveform’ term in SDR includes the physical layer of a standard, but

may also include higher levels of the relevant wireless standard.

The trend to include more and more software as part of radio designs had, however,

started well before 1991, as also pointed to by Mitola. A key influencing factor was the

introduction of the Digital Signal Processor (DSP), with the TMS320 [45] from Texas Instruments,

introduced in 1982, being an important example, enabling parts of the signal

processing to be implemented in software. A further influencing factor was packet radio

networks [46] which caused protocol software to be integrated in radio sets.

The ’Idealized Software Radio’ as it was described by Mitola [8] is illustrated in Figure

2.1.

In the book by Kenington [32] it is shown that Figure 2.1, with the addition of the filtering

needed to avoid aliazing in the resulting digital spectrum, is not a near-future realistic

1 SDR Forum was renamed to Wireless Innovation Forum in December 2009. The SDR Forum name is

used throughout Paper A in Part II, as this paper was filed prior to the name change

19


2. BACKGROUND

A/D

D/A

Software

A/D

D/A

FIGURE 2.1: The ’Idealized Software Radio’, simplified version of the figure in [8].

architecture when the receiver is to cover a range up to 2.2 GHz, to include standards such

as GSM, UMTS and mobile satellite communication. The main assumption made in this assessment

is that the Analog-to-Digital converter (A/D) contains a sample-and-hold element,

as is typical for high-resolution A/Ds. An optimistic estimate of the A/D power consumption

is made, considering only the power that goes into charging this sample-and-hold, and

neglecting other features of the A/D. A further assumption is that a dynamic range of approx.

122 dB (20 bits) is needed in order to be able to receive a weak signal in the presence

of other strong signals. The resulting power consumption is found as [32]

P = kT

t s

10 (6n+1.76)/10 (2.1)

where P is the power consumption of the Analog-to-Digital converter’s sample-and-hold, k

is Boltzmann’s constant= 1.38·10 −23 , T is the device temperature in Kelvin, t s the sampling

interval and n the number of bits in the A/D. A resolution of 20 bits and the Nyquist sampling

rate of 4.4 Gsamples/second yields a power consumption of approximately 30 W, which

certainly is too high for a handheld device. Additionally, practically achieved A/D power

consumption is several orders of magnitude above this theoretical best-case number [32].

Theoretically it is possible to overcome the above power limits with technologies that

either do not rely on the charging up of the sample-and-hold element, or technologies using

superconductive devices [32]. In reviewing typical currently available A/D-converters,

Figure 2.2, they all consume significantly more power than the above estimate. As for the

superconductive devices, these are still impractical to use as they need to be cooled to very

low temperatures.

Less demanding Mitola receivers, in terms of bandwidth and dynamic range, are practical

to implement. As an example, a 3-30 MHz High Frequency (HF) SDR receiver of the

Mitola type, with 12 bit A/D, has been built [49].

While the ideal Mitola receiver may receive every frequency channel concurrently [49],

practical receivers for higher frequencies must sacrifice some of this flexibility to achieve

more practical requirements for A/D conversion and for the processing power. As an example

of practical modification, switchable bandpass filters may be added prior to the A/D

in Figure 2.1 and used together with the undersampling technique [49] to relax the requirements

on the A/D converter. Another example is one where it is the Intermediate Frequency

(IF) processing and onwards that is performed in software or reconfigurable digital hardware,

see Figure 2.3.

20


2.1. A Brief Introduction to Software Defined Radio

1,00E+02

1,00E+01

Power consumption [W]

1,00E+00

1,00E-01

1,00E-02

1,00E-03

12 Bits Analog Devices

16 Bits Analog Devices

12 Bits Texas Instruments

16 Bits Texas Instruments

12 Bits Optimistic Calculation

16 Bits Optimistic Calculation

20 Bits Optimistic Calculation

1,00E-04

1,00E-05

1,00E+08 1,00E+09 1,00E+10

Sampling frequency [Samples/sec]

FIGURE 2.2: The optimistic power consumption estimates from Equation 2.1 at device temperature

of +80 deg C plotted for 12 bits, 16 bits and 20 bits resolution. The figure gives the listed (per

channel) power consumption of available 12 bits and 16 bits A/D converters with sample rates from

100 MegaSamples/second from Analog Devices [47] and Texas Instruments [48] (A/D data compiled

December 2009).

LNA

A/D

Software/

Reconfig.

Digital

processing

Bandselect

IF

Variable LO

FIGURE 2.3: Software / reconfigurable Intermediate Frequency (IF) processing receiver example.

LNA = Low Noise Amplifier. LO = Local Oscillator.

SDR has been driven partly by the continuous improvements of programmable processing

circuitry such as DSPs and Field-Programmable Gate Arrays (FPGAs). An important

driver has also been the military’s vision to achieve interoperability between systems by

being able to easily port waveforms onto platforms and thus of having flexible, waveformupgradeable

platforms for the Network Centric military tactical environment.

Many commercial-domain motivations also exist, including the increasing number of

waveform standards needed to be supported by cellular terminals and base stations [50],

the remote upgrade possibilities for satellite communicatations, the predicted future reconfigurable

cellular networks and the future Cognitive Radio and Dynamic Spectrum Access

(DSA) nodes and networks.

While some manufacturers may want to maintain proprietary couplings between the software

waveform application and the radio platform, a vision for SDR is a separation between

the two. Such a separation enables portability of waveform applications from platform to

platform. It also enables a business model where platforms and waveform software may be

made by different companies, and where customers may buy additional waveform software

21


2. BACKGROUND

from third parties. This may promote platform volume benefits through the standardization

of platforms with fewer and more specialized platform manufacturers. It also will lead to

more competition in waveform software design, where there may be true competition also

in waveform additions to existing platforms.

The separation of applications from its operating environment is also one of the goals of

the Software Communications Architecture (SCA). SCA was developed for and is managed

by the Joint Program Executive Office (JPEO) of the Joint Tactical Radio System (JTRS)

program. SCA is an architecture that enables SDR applications to be built as compositions

of components, and enables them to connect to platform devices. SCA defines interfaces for

managing and deploying the software components. The Software Communications Architecture

also defines the domain profile, a set of Extensible Markup Language (XML) files,

which is used when deploying components, logical devices and services onto the system.

Paper A has a further description of SCA, and also dicusses to what extent SCA has

provided real platform-to-application separation and to what extent SCA-based applications

are portable.

While SDR is still a field of active research, it is also in a phase where SDR-implemented

radio systems are becoming available. Some military radio unit types meeting some of

the basic JTRS requirements [51] have been fielded [52], while production start for other

JTRS types are still in the future. In the civilian domain, commercial research platforms are

available [53, 54], as are also some cellular base stations [55, 56].

A further and more detailed introduction to Software Defined Radio is provided in Paper

A.

2.2 Introduction to Dynamic Spectrum Access

2.2.1 General

The dominating frequency assignment regime at present is one where the frequencies or

spectrum sub-bands are typically assigned for a significant amount of time. A frequency

assignment is defined as an ”Authorization given by an administration for a radio station to

use a radio frequency or radio frequency channel under specific conditions” [57]. In Norway,

as an example, such an assignment may last up to 20 years for some services [58].

This ’static’ frequency assignment regime in many cases leads to low efficiency in utilization

of the spectrum. While most of the spectrum is fully assigned to users, measurements

show that many frequency bands ’are not in use or are only used part of the time’ [17].

Even for urban areas such as Bern this is true, as shown by measurements in [18].

At the same time as very limited additional spectrum can be made available under this

static regime, the demand for communication capacity and hence additional spectrum, particularly

for mobile two-way data communication is continuing to grow. In the military

domain, the information superiority and mobility demanded by the Network Centric Operations

(NCO) doctrine causes additional spectrum demands for example for the full-motion

video from Unmanned Aerial Vehicles (UAVs) and for high-capacity ad hoc networks.

A number of more dynamic approaches to frequency assignment have been put forward

by a number of authors to improve the utilization of the spectrum resource.

The approach that is attracting most attention is Cognitive Radio. Cognitive Radio was

introduced by Joseph Mitola III and Gerald Q. Maguire Jr. [15] as an extension of software

22


2.2. Introduction to Dynamic Spectrum Access

radio with ’radio-domain model-based reasoning’. The ’intelligent agent’ [15] perspective

of the radios implies that systems of such radios are multiagent systems. Cognitive Radio

was further detailed in Mitola’s dissertation [23] where one characterization is that of a

’trainable’ radio rather than just ’programmable’ [23], and where it is discussed as a widecontext

environmentally aware radio. The Cognitive Radio is aware of both the user’s needs,

the present location, available networks and so on, and uses this knowledge to tailor both

radio resources and services [23]. The spectrum access part of Cognitive Radio has since

dominated in publications, to such an extent that many publications only refer to the spectrum

access properties of Cognitive Radio. As an example, on page 27 of [18] it is stated that

’Cognitive Radios are radio systems that autonomously coordinate the usage of spectrum’.

One of the goals of this work is to analyze the advantages and disadvantages of different

Dynamic Spectrum Access architectures. To avoid the ambiguities in the Cognitive Radio

term in this context, it will be used to a lesser degree, and the terms ’Autonomous radio

node’, ’Distributed Coordination’ 2 and ’Centralized Coordination’ will be used instead, as

will be explained in the following.

Before discussing DSA architectures, spectrum sharing principles will be reviewed.

2.2.2 Spectrum Sharing Principles

A recommended categorization of spectrum sharing principles is found in the work by Zhao

and Sadler [60], as presented in Figure 2.4.

Dynamic Spectrum Access

Dynamic Exclusive Use

Model

Open Sharing Model

Hierarchical Access Model

Spectrum Property Rights

Dynamic Spectrum

Allocation

Spectrum Underlay

Spectrum Overlay

FIGURE 2.4: A categorization of Dynamic Spectrum Access sharing principles (taxonomy of Dynamic

Spectrum Access [60]).

The Dynamic Exclusive Use model, Figure 2.4, has two categories, Spectrum Property

Rights (licensees sell and trade spectrum) and Dynamic Spectrum Allocation. In Dynamic

Spectrum Allocation, a chunk of spectrum is dynamically provided for a service, in a defined

region. Efficient exploitation of the spectrum chunk is left to the service provider. Restrictions

must ensure that the use of the spectrum does not interfere with other services in other

regions.

In the Open Sharing model, the spectrum is shared between peer users on an equal basis.

In the Hierarchical Access model, the primary users are the spectrum owners that have

prioritized rights to the use of the spectrum. Secondary users may use the spectrum when

not interfering with the primary users. In what is referred to as an underlay mode, secondary

users may use the spectrum, for example with low power and a very wide bandwidth, when

2 In this thesis the term ’Distributed Coordination’ has been used to be consistent with e.g. the survey by

Akyildiz et al. [59]. An alternative term is ’Decentralized Coordination’.

23


2. BACKGROUND

the resulting interference level is neglectable and does not degrade the communication links

of the primary user. Ultra Wide Band (UWB), which is based on short-duration low-power

pulses, is an example of such underlay use. In the overlay mode, spectrum opportunities

in the form of vacant combinations of time and frequency slots in a geographical area are

identified by the secondary users. The secondary user uses the spectrum opportunity until

the primary user re-enters, at which time the secondary user needs to move to a new spectrum

opportunity.

As is natural since DSA is still an area where the terminology is young, some parallel

terms exist. Hierarchical Access is also referred to as ’Vertical Spectrum Sharing’ [18].

Another term, ’Horizontal Spectrum Sharing’ refers to ’sharing between radio systems with

a similar regulatory priority’ [18], which is similar to Open Sharing.

The dynamic sharing of spectrum has a clear parallel in the Medium Access Control

(MAC) protocols [59], that are responsible for the access to the shared medium of existing

wireless systems. Channel-based medium access may use combinations of Frequency Division

Multiple Access (FDMA), Time Division Multiple Access (TDMA) and Code Division

Multiple Access (CDMA) [10]. Space Division Multiple Access (SDMA) such as through

beamforming antennas or MIMO technologies, may also be included, as may Polarization

Division Multiple Access (PDMA). Packet-based medium access typically use TDMA, with

the 1971 ALOHA protocol as the first example [10].

Dynamic spectrum sharing, however, has differences from and has challenges that extend

beyond those of these MAC protocols [59]. In Hierarchical Access, a difference is that

the spectrum sharing needs to take into account various types of licensed users. In both

Hierarchical Access and Open Sharing a challenge is that many different types of wireless

systems may need to coexist, rather than just radio nodes from a single radio system with a

single MAC protocol. Typically DSA also aims to share wider spectrum ranges.

In the enclosed paper C, all of the three models Dynamic Spectrum Allocation, Hierarchical

Access and Open Sharing will be relevant. In Papers D and E, an Open Sharing model

is assumed.

2.2.3 Architectures

Different DSA approaches have been suggested in the literature. An architectural classification

of these approaches is shown in Figure 2.5.

Dynamic Spectrum Access

Coordinated

Autonomous

Centralized Coordination

Distributed Coordination

FIGURE 2.5: Dynamic Spectrum Access Architectures.

The coordinated architectures are relevant for all of the access sharing principles in Figure

2.4. The Autonomous architecture is applicable for the Open Sharing and Hierarchical

Access models.

24


2.2. Introduction to Dynamic Spectrum Access

In Autonomous architecture, each radio node, or a co-working group of such nodes,

autonomously decides on the use of spectrum based on spectral sensing and is restricted and

governed by predefined policies. This is often referred to as Opportunistic Spectrum Access.

A Cognitive Radio which does not have functionality that coordinates its spectrum decisions

with other spectrum users or with any spectrum management entity, belongs to this category.

In a Centralized Coordination architecture, the coordination is enabled through specific

infrastructure, for example in the form of spectrum servers that provide spectrum leases to

radio nodes. Administrative communication channels are required between the infrastructure

decision elements and the radio nodes. This could be in the form of direct wireless

connections, in the form of wireless Internet or in the form of fixed Internet links.

In a Distributed Coordination architecture, the coordination is between the radio nodes

themselves. This has the advantage of avoiding the dependency on additional spectrum

decision infrastructure. Administrative communication channels are required between the

nodes, either as direct wireless connections or Internet ones as in the previous case.

Paper C outlines a concept that belongs to the Centralized category. Papers D and E

discuss all of the three architectures.

2.2.4 Transmitter- and Receiver-centric Models

Another view of dynamic spectrum sharing is a division into transmitter-centric sharing and

receiver-centric sharing.

Transmitter-centric sharing is the more conservative approach. Here the spectrum used

by a transmitter is considered occupied in a given region around the transmitter even if there

are no receivers in (parts of) this region. This occupancy region is defined as the region

where the transmitter field strength is above some defined minimum level.

In receiver-centric sharing, decisions about whether a spectrum interval may be reused

takes into account the accumulated interference at all relevant receivers. Examples of acceptance

criteria may be minimum signal-to-noise-and-interference at the influenced receivers,

or minimum communication bit rates of influenced communication links. Receiver-centric

sharing may permit a denser reuse of spectrum objects, since regions where there are no

receivers may permit spectrum reuse even if they are within transmitter-centric occupancy

regions.

For the centralized concept outlined in Paper C, both transmitter-centric and receivercentric

sharing will be relevant. In Papers D and E receiver-centric sharing is assumed.

25


Chapter 3

Related Work

3.1 Related Work to the Results from Research Goal 1,

Conceptual Challenges of Software Defined Radio

The main results from research goal 1 are in Paper A. In addition to the author’s own experience

with SDR, Paper A is built on a number of references, as listed in the paper. These

include foundational papers by Mitola [8] and others, Tuttlebee’s book on enabling technology

for Software Defined Radio [21], Kenington’s book [32] as well as a number of classical

and recent papers.

The literature on SDR is vast, and hence a review of all the work that is related in

some way to Paper A will certainly lead too far. The review is therefore restricted to the

most important other articles of a tutorial or survey type that deal with challenges and/or

opportunities of SDR:

Naturally such a review needs to start with Mitola, who named and in some ways initiated

the research field with his ’Software Radios Survey, Critical Evaluation and Future

Directions’ 1992 article [8]. This visionary work suggested and predicted many groundbreaking

features of SDR, such as open software architectures, meta-level models and radio

simulation and Computer-Aided Design environments that permitted integrated work from

many different disciplines. It also reviewed the analog-to-digital conversion and processing

resources and requirements of that time. This work was followed up by a number of articles

in the years to come, many of which further elaborated on software architecture for SDRs,

and in the 1999 article [61] promoted a mathematical perspective of SDR. What Paper A

possibly has to offer in a complementary way to these very visionary and forward-looking

papers, is an update with recent progress in the field, as well as putting the practical challenges

with SDR (such as portability, security, regulations), and the opportunities, clearly on

the table.

Jondral has written an excellent (2005) tutorial [62] which covers both Software Defined

Radio and Cognitive Radio, with a focus on reconfigurability, mobile communication

standards and the Software Communications Architecture. It also goes in detail into SDR

receiver front-ends, which is not a topic in Paper A. However, Jondral’s tutorial does not

sum up remaining SDR challenges and it does not go into business models and product

opportunities, regulations, security or alternative SDR architectures as does Paper A.

Several of the tutorials and surveys have subtopics that overlap with Paper A, but deal

only with one or a few subtopics of SDR. Shamsizadeh [63] deals in a thorough way with

27


3. RELATED WORK

security issues in Software Defined Radio and Cognitive Radio, with a particular focus on

security threats. Giacomoni and Sicker (2005) [64] go into great detail about how the transition

from ’static’ radio systems to flexible software defined ones challenge the certification

and assurance processes. Lehr et al. [65] go into business and regulations, discussing implications

of SDR on the wireless value chain, wireless services and public policy. González

et al. [66] have provided a very thorough tutorial on the Software Communciations Architecture,

which also promotes the Open-Source SCA Implementation Embedded (OSSIE).

Zhang et al. [67] convey insights into SCA-compliant waveform development.

Some of the tutorials and surveys have a different approach and view SDR from a different

angle than Paper A, but on the other hand do not provide the same broad view of issues

and benefits: Cummings 2007 tutorial [68] (only an abstract is available on IEEE Xplore)

contains an interesting historical view on the path from wireless communications in 1890 via

’hardware radios’ through to SDR. Tribble’s 2008 [69] ’... Fact and Fiction’ article discusses

some of the postulated benefits of SDR: The reduced logistics cost, the possibilities for rapid

capability upgrades and SDR as an enabler for Network Centric Operations (NCO). Pearson

(2001) [70] emphasizes the system engineering challenges resulting from SDR, including

requirements management and the complex and evolving design as well as verification. He

also emphasizes the opportunities by ’spiral’ (i.e. incremental) design and user interaction

during the development process.

Several publications discuss the front-end and analog-to-digital conversion sub-topics,

and that are not within the scope of research goal 1 and Paper A: Included in this group is

Svensson and Andersson’s ’... Visions, Challenges and Solutions’ [71] that focuses on frontends,

A/D converter and filtering issues. Rusu and Ismail [72] deal with A/D converters in

SDR. Haghighat [73] focuses on hardware aspects of SDR, including antenna technology,

radio frequency modules, conversion between the digital and analog domains, digital signal

processing and interconnection technology.

Although, as evident from this review of related papers, there are a number of tutorial and

survey papers on SDR, there is not any updated paper that summarizes the challenges and

opportunities over the same width of SDR aspects as Paper A. It should also be mentioned

that Paper A is the only SDR tutorial/survey paper in the IEEE Communications Survey and

Tutorials online journal.

3.2 Related Work to the Results from Research Goal 2,

SCA Workload Overhead

In contrast to the first research goal and Paper A where there were many related works in

the literature, there are very few papers discussing workload overhead on CORBA-enabled

processors in SCA. This fact is also confirmed in a paper by the Communications Research

Centre Canada’s SCARI development team [74] where it is explicitly stated that ”The performance

of CORBA in the context of SCA is nearly absent from the research community

literature.” It should be added though, that in addition to the research on CORBA in the

context of SCA, there is also earlier research on the performance of CORBA by itself, some

of which is mentioned in Paper A.

In addition to the few papers discussing specifically workload overhead on CORBAenabled

processors in SCA, as in Paper B, there are a few more papers going into inter-

28


3.2. Related Work to the Results from Research Goal 2, SCA Workload Overhead

component latency and/or latency variation and/or data throughput. After reviewing the

specific workload overhead research, this second group of papers will also be commented

on. The reason for also including these papers in this review of related research, is that

latency, throughput and processor workload are in most cases dependent parameters. E.g. if

the latency is caused by data marshalling inside the processor workspace, the latency at the

same time gives rise to a workload overhead.

The review is divided into research that was published prior to Paper B, and research that

was published at a later time than Paper B.

3.2.1 CORBA in SCA: Workload Overhead

3.2.1.1 Published prior to Paper B

The paper that is most directly related to Paper B and was published prior to it, is one by Balister

et al. [75] which similarly investigates inter-component communication using OSSIE

and omniORB. The waveform application that is used in this case is a different one, an FM

waveform application. Similar to Paper B, the statistical sampling profiler OProfile is used to

aquire processor workload results. The analysis is in some ways less comprehensive than the

one in Paper B: The paper does not vary the granularity of the applications as does Paper B,

but reports profiling results for only a single 3-component application configuration. Also,

it does not vary the inter-component packet sizes and packet rates. Further, it does not propose

any models for the overhead as does Paper B. However, a breakdown of the processor

workload is included. With the specific waveform application and application granularity

used, Balister et al. reaches a somewhat different conclusion than Paper B and the paper

from Singh et al. [76] (see below), the paper concludes that the processing overhead from

CORBA was significantly lower than the required signal processing of the tested waveform.

A conclusion that is consistent with Paper B is that the copying of data into CORBA sequences

in the application components, an operation required by the StandardInterfaces in

OSSIE, creates a certain amount of overhead.

The coauthor of Paper B, Jon Neset, presented some initial SCA-based application workload

measurements in a University project thesis [77] in 2007.

3.2.1.2 Published later than Paper B

The paper that is the closest related to Paper B is the one by Singh et al. [76], also having

been written in the context of work in the RTO IST-080 SDR group, and that was released a

few months later than Paper B. As in paper B, a Stanag 4285 waveform application is used to

investigate the effects of SCA application granularity. The 4285 transmitter is split into two

different granularities, 2 components and 4 components. Compared to Paper B, a different

SCA Core Framework (SCARI) and ORB (TAO) is used. However, as in Paper B it is shown

that the ORB and SCA give significant overheads relative to the functional processing of

the components, and it is concluded that the components should perform significant signal

processing in order to avoid being dominated by the overheads. The paper has a useful

breakdown of the overheads from the ORB, SCA and libraries.

The profiler used in this paper, Valgrind [78], is significantly different from the ones used

in Paper B. Valgrind instruments executable files by running them on a synthetic Central

Processing Unit (CPU). It is capable of profiling the activity of the executable, and also

29


3. RELATED WORK

all dynamically linked-in libraries. It does this very accurately, and can provide answers

in the number of clock cycles on the specific processor. The downsides of this approach

are, however, that the executables run slower than real time, and also that the profiling is

limited to the executable and the dynamically linked in libraries, it is not system wide.

Further differences relative to Paper B are that [76] does not investigate high application

granularities and different packet sizes, does not attempt to formulate any models for the

workload overhead, and does not investigate the contribution of context switches.

In a paper by Abgrall et al. [79], both CPU load and latency are studied and compared

between the SCA-based OSSIE environment (which is also used in Paper B) and GNU radio,

which defines a single-processor non-CORBA architecture. Somewhat in line with the

conclusions in Paper B, the CPU load is found to be five times higher for the OSSIE system

than for GNU radio, when running the same waveform functionality. An FM receiver waveform

is used for the comparison, using six different waveform granularities, the maximum

being nine components and the minimum three components. With OSSIE, as in Paper B a

dependency on granularity is observed, with the three component version using 22% CPU

and the nine component version using 36%. The non-signal-processing CPU load is also

discussed, conluding that there should be significant processing inside an SCA-based component,

which is the same conclusion as made in Paper B. Memory use of the waveform

implementations in OSSIE, as well as intercomponent latencies, are also discussed, and

which is outside the scope of Paper B. The paper has, later than and independently of Paper

B, reached some of the same conclusions. Abgrall et al. do not, however, propose models

for the CPU overhead caused by SCA/CORBA, and do not go into the relation between the

CPU load caused be the inter-component communication itself versus the context-switching,

as does Paper B.

The review of related research clearly shows that there are few related papers to Paper

B, and that there are none that do all of the investigations and to the same extent as in Paper

B. Paper B hence clearly fills a space in the research literature.

3.2.2 CORBA in SCA: Latency and/or Throughput

The review of the latency and throughput research is included here for completeness as latency,

throughput and workload overhead are in many cases related. Also, the added latency

and throughput reduction due to CORBA affect SCA-based systems in the same way as

workload overhead, i.e. as a drawback and an argument against SCA.

3.2.2.1 Published prior to Paper B

In research by Bertrand et al. in 2002 [80], latency is compared between the SCA-defined

packet transfer, transfer by local (same-address-space) method invocations and also transfer

by CORBA remote one-way and two-way methods. The authors find that 256-byte packets

transferred by the SCA-defined pushPacket method use 1300 µsecs on average, compared to

a local call using just 9 µsecs. The paper still concludes that, given the advantages provided

by the SCA in other areas, the latencies are still acceptable.

30


3.2. Related Work to the Results from Research Goal 2, SCA Workload Overhead

3.2.2.2 Published later than Paper B

Using OSSIE and omniORB, Tsou et al. [81] make an interesting comparison of latency on

a single-processor system using Unix domain sockets as CORBA transport, relative to using

the default TCP/IP. Roughly a factor of two reduction of latency is found when using Unix

domain sockets. By comparison, Paper B employs the default TCP/IP endpoint settings

of omniORB. The paper [81] does not provide processing overhead numbers for the two

different cases.

Navarro et al. [82] compare throughput performance in the OSSIE and GNU Radio

frameworks, as in [79]. The comparison used the same base waveform application, a binary

phase shift keying modulation waveform, adapted to each of the frameworks, in both cases.

It is concluded that the maximum throughput in both cases was surprisingly low, around

700 kbps, the performance in OSSIE being better than in GNU radio by a small margin.

The paper, however, does not contain any analysis of why the throughput was surprisingly

low. In particular, the paper does not contain any profiling of the OSSIE-implemented application,

and does not contain any CORBA-related investigations. The conclusion that the

OSSIE-implemented application was only narrowly better is somewhat contradictory to [79]

and raises some questions.

In Bernier et al. (2009) [74], round-trip measurements have been made of buffers of data

between a PerformanceApp SCA resource, a PassThrough resource and back to the PerformanceApp.

Here, measurements for four data packet (buffer) sizes were made: 1024 and

2048 octets, as well as 1024 and 2048 double-precision (64-bit) numbers. The measurements

were done for both an Intel/Linux system and a PowerPC/Integrity system. While measurements

were done for both systems using TCP/IP transport, for the PowerPC/Integrity system

a different transport named INTCONN was also tested, as well as bypassing the transport

by running in a single address space. The article demonstrates round-trip-time performance

dependencies on the underlying transport, where the INTCONN is shown to provide better

performance than TCP/IP. The SCARI development tool that was used in the experiment

had the possibility to automatically generate a ResourceFactory that enabled it to run the

SCA application Resources in a single address space. Also, the ORB used was able to recognize

this and substitute remote invocations with local ones. Running the SCA Resources

in a single address space in this way enabled even better latency performance. These results

are very relevant in terms of possible workarounds of the issues that are observed in

Paper B, and further details as well as advantages and disadvantages of this procedure will

be discussed further in Section 4.2.2.

Abgrall et al. (2010) [83] measure latency and latency distributions. They make comparisons

between a single-processor OSSIE system and a single-processor system that in

the same way as the OSSIE system is using omniORB to send the same data between the

components but that does not include the OSSIE Core Framework. The paper concludes that

the latency with and without the OSSIE Core Framework are very similar, and that hence

the OSSIE framework does not add significant latency beyond that introduced by omniORB.

The paper leaves some uncertainty, however, due to insufficient details being provided for

the reader to know that the comparison is done with the exact same set of parameters. The

paper interestingly proposes closed-form models for the latency-distributions and demonstrates

how they fit the measured data.

31


3. RELATED WORK

3.3 Related Work to Research Goal 3, DSA Architectures

and Algorithms

3.3.1 Work Related to the Dynamic Frequency Broker

3.3.1.1 Timeline

The Dynamic Frequency Broker (DFB) concept, that is presented in Paper C in Part II,

belongs to a group of ’spectrum broker’ concepts, and is classified as a Centralized Coordination

DSA architecture. Paper C was written in 2007 and presented in 2008.

The spectrum broker term in itself is not very new. It can be traced back at least to being

mentioned in an article on spectrum management tools in 1990 [84], to presentations and

publications from the European ’Drive’ project in 2000-2002 [85, 86], and there is a patent

application that was filed on a particular broker system in 2002 (patent dated 2008) [87].

In ’Drive’, the spectrum broker was used for Dynamic Spectrum Allocation, i.e. to allocate

chunks of spectrum to different radio networks. A frequently cited concept, and possibly

the first to be thoroughly described in an open publication on a detailed conceptual level,

is DIMSUMnet [88] in 2005. This was followed by DSAP [89] and a few others [90, 91]

the same year. In the following years, central contributions have come from followup articles

involving one or more of the DIMSUMnet authors, providing spectrum measurements

in existing cellular networks to demonstrate the usefulness of broker solutions [92] and

providing approximate algorithms for broker spectrum assignment [93, 94]. Approximate

algorithms have also been suggested by Kovács et al. [95, 96] by splitting the spectrum assignment

problem into a spatial and a temporal part. Contracted by the British regulator

Ofcom, Roke Manor Research has prepared a thorough description of the Ofcom Candidate

Architecture [97] that appeared in 2007.

The spectrum broker is one possible mechanism that may enable an efficient market for

spectrum. Accordingly there are a number of related papers that discuss spectrum auctions,

these are skipped in this overview, however, as this topic is not central in Paper C.

The spectrum broker and centralized coordination research area still generates activity,

with several contributions made also in 2009. An example is Ge et al. [98] that describe a

laboratory prototype broker that gathers spectrum information through cooperative sensing,

presents a table of vacant channels to secondary users and may send alarms to secondary

users in case primary users need them to vacate the spectrum.

A current trend is to use the centralized element as a spectrum database but not a decision

maker, and allow the decisions as to which part of the spectrum to use, to be made

decentralized instead of in the centralized element, this is discussed further in 4.3.1.4.

3.3.1.2 Three Important Concepts

The most important conceptual contributions in this author’s view, and the ones cited in

in Paper C and/or Paper E, are DSAP [89], DIMSUMnet [88] and the Ofcom Candidate

Architecture [97]. These will be briefly described here as background for the DFB. For the

description of DFB, refer to Paper C and the summary in Section 4.3.1.1.

Dynamic Spectrum Access Protocol (DSAP) [89] is a concept where spectrum is managed

through a time limited lease management mechanism. The concept is aimed at brokering

in limited geographic areas, at small timescales and on a per LAN basis. The archi-

32


3.3. Related Work to Research Goal 3, DSA Architectures and Algorithms

tectural elements of DSAP [89], see Figure 3.1, are the DSAP server, the DSAP clients and

the DSAP relay. The client is defined as any wireless device that uses DSAP for spectrum

coordination. The server takes into account current assignments and grants leases. The relay

allows multihop contact between clients and server. DSAP contains a description of the

messages used between the different architectural elements.

A

Radio Map

+

Policy DB

?

DSAP Server

Channel lease

negotiation

B

C

DSAP relay

?

Channel lease

negotiation

802.11b Ch 11

FIGURE 3.1: The architectural elements in DSAP are a DSAP server, DSAP relays and DSAP clients.

Reprinted from [89] with permission

Fig.

from

1. Components

IEEE, c○IEEE.

of DSAP.

The DFB has similarities to DSAP, including the lease-based mechanism, the lower level

broker inof DFB network is similar nodes, to thetakes DSAPon server, the the role DSAP of the clients spectrum to DFBarbitrator.

radio nodes, and the

DSAP relay Thetoserver DFB slave stores stations. information A difference about between its DFB clients and and DSAP channel is that DSAP is

aimed atconditions local brokering throughout with the clients the being network wireless in adevices, database whilethat DFBwe is aimed call at a full

hierarchical a RadioMap. computerizedBased replacement on ongoing of traditional client frequency communications, allocation andthe

assignment,

includingset both

ofsmall administrator-defined

scale and large scalepolicies brokering.

and

Here

the

theRadioMap, clients can bethe

both radio

devices, aggregations of devices or radio infrastructure entities. DFB additionally includes

DSAP server determines an “optimal” distribution of radio

ideas concerning the use of Web Services for communication and universal discovery of

brokers. spectrum DSAP on the among other hand, the clients includesin a definition the network of protocol and reconfigures

messages, which is not

includedthe DFB. clients accordingly. We envision DSAP as a very dynamic

DIMSUMnet protocol: [88] some also manages configuration spectrumparameters through a time of network limited lease nodes management

mechanism. mayThere be reconfigured are two modes several that are times described, a second, in thewhile one mode others it ismay

the network

operatorsremain that request unchanged spectrum. for In extended the second periods mode theof endtime.

user clients request spectrum

for communication with other endusers or base stations.

The DIMSUMnet architecture has the architectural elements as illustrated in Figure 3.2.

The Spectrum Information and Management (SPIM) broker provides spectrum leases, has a

complete topographical map and maintains a ’spectrum snapshot’ of the spectrum use. The

Radio Access Network Manager (RANMAN) controls spectrum leases for several DIM-

SUM base stations. In mode 1 the clients listen to spectrum snapshot broadcasts from the

DIMSUMnet base stations and configure themselves, in terms of services, network operators

and spectrum and waveform parameters, accordingly. In mode 2, the client can additionally

request spectrum and optionally may send information about the sensed spectrum to the

DIMSUMnet base stations. DIMSUMnet as an option describes a hierarchy of SPIM servers

when there are many cells in a region, and additionally describes redundancy in servers.

A. Protocol Entities

DSAP defines the following entities (see Figure 1):

DSAP client: any wireless device that uses DSAP for

coordinated spectrum access. Before communicating a DSAP

client will request a channel from the DSAP server.

DSAP server: the centralized entity that coordinates spectrum

access requests. It accepts spectrum lease requests from

clients, considers the current spectrum assignments, the RadioMap

and the policy database and responds back with a

time-bound spectrum allocation.

DSAP relay: an entity that allows multi-hop communication

between DSAP server and clients that are not in direct range

of each other.

F

E

D

database that hol

bly including geo

throughout the ne

odic updates from

in their vicinity a

This information

optimal spectrum

accordingly.

The RadioMap

low the server t

ment under “poli

administrator-defi

tribution of lease

quality of service

identifiers (MAC

C. DSAP Messag

A ChannelDisc

that wish to obt

The parameters o

identifier, locatio

ported wireless M

the desired lease

ChannelOffer m

client either in re

quest (described

server’s choice of

from what the cli

ChannelReques

parameters and is

edge the terms o

renegotiate certai

renew a lease.

ChannelACK i

Request. This me

request for a leas

33

ChannelReclai

fully reassign or

message can be p


3. RELATED WORK

SPIM

Spectrum

Broker

Internet

DIMSUMnet: Dynamic Intelligent

Management of Spectrum for

Ubiquitous Mobile networks

DIMSUM

IP Core

Network

DIMSUM

RANMAN

SPEL

Protocol

dio tech

operatin

mation c

where se

sible for

3.2 R

MN

DIMSUM

IP BS

MN

DIMSUM

IP BS

MN

DIMSUM

IP BS

FIGURE 3.2: The architectural elements in DIMSUMnet are the SPIM spectrum broker, the RAN-

MAN, DIMSUM Figure base stations6. andDIMSUMnet clients. Reprinted architecture

from [88] with permission from

IEEE, c○IEEE.

terested in using the CAB spectrum in a given geographical

region Ê. It manages all dimensions of the CAB spectrum,

namely frequency, time, space (location, direction),

signal (polarization, coding/modulation) and power. The basic

tenet of DIMSUMnet spectrum usage is that any use

of CAB spectrum not approved by SPIM server is a noncompliant

usage.

The SPIM server maintains a complete topographical map

of the region Ê which records position of all base stations

and approximate extent of cell coverage associated with each

BS. For every cell, it maintains a spectrum snapshot [10]

that records: (1) spectrum used for SPI channels and (2) a

spectrum allocation map (SAM). Each SAM entry records

spectrum parameters such as (a) start and end of spectrum

band, (b) network service provider (NSP) it is allocated to,

(c) the current waveform or network access method (e.g:

GSM) Architectures used in the spectrum, (d) time duration of lease, (e)

maximum transmission power allowed, and (f) optionally, information

about interference due to thermal noise and other

sources such as secondary users. The SAM changes with

time as the various parameters in SAM entries change. For

example, when the SPIM server de-allocates a spectrum, the

DFB is different from the DIMSUMnet architecture in that DIMSUMnet is centered

around radio clients in access and mesh networks in its description, while DFB does not

make any particular assumptions on the type of radio clients. DFB also describes a full

hierarchy, and Web Services oriented communication and discovery mechanisms. DIM-

SUMnet on the other hand has more architectural elements and mechanisms than DFB and

is described in more detail than DFB.

The Ofcom DSA Candidate Architecture, see Figure 3.3, has slightly different architectural

elements and spectrum management principles compared to DSAP, DIMSUMnet

and also to DFB. Rather than using the broader spectrum lease term, the DSA Service

Provider manages call requests. The end users, here termed DSA Capable User Equipment

(DCUE), receive price quotes per call through the Price Quote Generator at the DSA

Network Provider.

A clear difference to DFB is that the Ofcom DSA Candidate Architecture is focused on

calls through networks, with the DSA Network Provider as a central element, whereas DFB

aims at all wireless communication, including ad hoc and push-to-talk type communication.

The Ofcom DSA Candidate Architecture has also more focus on the business side of spectrum

management including billing and price quotes, as DFB does not address these areas.

The architecture does not describe a hierarchy of brokers or the use of Web Services for

communication and discovery, as does DFB.

3.3.2 Related Work, Comparison of Dynamic Spectrum Access

This section describes work related to the comparison of DSA architectures, presented in

Paper E in Part II. While there are many articles that touch on DSA architectures in one

34

The

new net

several

characte

and soft

dio freq

form (C

ties, ma

characte

power e

tion cha

for serv

quired t

spectrum

stations

specific

access c

minates

sons suc

band, an

3.3 D


2 DSA NETWORK ARCHITECTURE

2.1 INTRODUCTION

3.3. Related Work to Research Goal 3, DSA Architectures and Algorithms

The candidate architecture for DSA is shown in Figure 1. The items shown in blue are the

key network elements. The Billing Agents, shown in red, are also required although the

protocol associated with billing has not been a major focus for this project.

Billing

Agent

Billing

Agent

PQG-Proxy

User

Interface

DCUE

Air

Interface

DCIE

Price Quote

Generator

(PQG)

Internet

User Agent

(PRSA)

Location

Register

(DCUE-LR)

Service

Gateway

DSA Network Provider

Address

Register

(DCIE-AR)

DCUE - DSA Capable User Equipment

DCIE - DSA Capable Infrastructure Equipment

DSA Service

Provider

Figure 1: DSA Candidate Architecture

DSA is best described as a ‘network of networks’ since a number of participating access

networks operated by DSA Network Providers (DNPs) are affiliated to an overlay or

‘meta-‘ network operated by a DSA Service Provider (DSP). The DSA concept is

technology agnostic so that the access technologies of DNPs can range from, say, cellular

Providers. Reprinted

through

from

WiFi to

[97]

Private

withMobile permission

Radio (PMR)

from Roke

systems.

Manor

DNPs

Research

operate DSA

Ltd.

Capable

Infrastructure Equipment (DCIEs), for example, cellular base stations or WiFi Access

Points. In fact, the DSA architecture has been designed so that only minor modifications are

required to any legacy network choosing to participate in DSA. The common denominator is

that the DNP is willing and able to support the roaming onto its network of DSA Capable

User Equipments (DCUEs). The Internet is used to provide the inter-connect between

DNPs and DSPs as well as access to services for End Users.

not ended, and The where role and there functionality is still of each active network research element in within each DSA of is now thedescribed.

directions.

FIGURE 3.3: The main architectural elements in the Ofcom DSA Candidate Architecture [97] are

the DSA Capable User Equipments (DCUE), the DSA Capable Infrastructure Equipment (DCIEs) at

the network provider, agents for billing and pricing of spectrum and the registries at the DSA Service

way or another, there are few that make a systematic comparison of autonomous, distributed

(decentralized) and centralized architectures. Also, the topic is one where the discussion has

Use, duplication or disclosure of data contained on this sheet is subject to the restrictions on the title page of this document

72/07/R/028/U Page 9 of 30

The review here will center on the most directly related publications, those that directly

compare DSA architectures, prior to and after Paper E was published.

An inspiration for paper E was the ’Architectures and Implications’ article from Tran

[99]. This draws the architectural lines from manual static spectrum access through centralized

dynamic access and to opportunistic dynamic access. The article presents a good

overview of vulnerabilities in the opportunistic and centrally coordinated architecture, also

in hostile environments. It does not include distributed coordination as a specific case. The

DSA comparison in Paper E expands on this paper by including the distributed coordination

architecture and by also relating the architectural comparison to a model and presenting

simulation results.

Nekovee in two similar articles [100, 101] discusses the three types of architectures,

associated research challenges and reviews examples from the research literature. These do

not provide quantification, however, nor go into specifics for hostile environments.

Chen et al. [102] also review the three types of architectures by going through a number

of example concepts from the literature. There are also several DSA surveys that briefly

discuss the three forms of architectures [103–105]. A publication from the European IST

Winner Project [106] has a one-page comparison of centralized and distributed architectures,

but the discussion is limited to the particular spectrum management system proposed in

Winner.

The article that by its title is closest to Paper E is one that appeared after Paper E was

published (it appeared in IEEE Early Access recently in 2010, and was accepted for a future

issue of IEEE Communications Surveys & Tutorials). It is entitled ’A comparison Between

the Centralized and Distributed Approaches for Spectrum Management’ [107]. The focus

35


3. RELATED WORK

is different from Paper E, though, with its main line of comparison being between Dynamic

Spectrum Allocation (i.e. the allocation of spectrum chunks to network operators)

and what the authors term Dynamic Spectrum Selection (Opportunistic Spectrum Access /

Dynamic Spectrum Access). It compares distributed (ad hoc) Dynamic Spectrum Selection

and network-assisted Dynamic Spectrum Selection. The paper is comprehensive, however,

and reviews many examples.

Unlike the above publications, Paper E mainly bases the comparison of the architectures

on a link-interference model and provides simulation results that serve to illustrate the differences.

Compared to all other papers but the one by Tran [99], Paper E extends the discussion

by also including hostile environments.

3.3.3 Related Work to Algorithms Based on the Iterative Waterfilling

(Paper D and Paper E)

The starting point for the work on algorithms and calculation models based on the Iterative

Waterfilling has been the comprehensive and very frequently cited Cognitive Radio article

by Haykin [108], as well as the doctoral dissertation by Yu (2002) [24] and the textbook by

Barry et al. [109].

Haykin’s article is a broad tutorial on a number of aspects of cognitive radio, including

spectrum sensing and interference, channel state-estimation and cooperation and competition.

It highlights two types of spectrum decision approaches for cognitive networks, one

being game-theoretic learning exemplified by ’No-regret algorithms’, the other one being

the Iterative Waterfilling approach. The article formulates the achievable user-rates for a

multilink scenario, and introduces the signal-to-noise-ratio gap factor that is also used in

Papers D and E. The Iterative Waterfilling algorithm is explained for a multiuser scenario,

and a small example is illustrated. The performance maximization of each transceiver is

subject to a constraint that a ’interference temperature limit’ is not exceeded. The concept

ofinterference temperature limit’ has since become less relevant and it is not included in

Papers D and E. Paper D expands on subtopics in this paper by studying the Iterative Waterfilling

solutions relative to centralized optimum ones, by case-studying the concept of a

priori target rates that is briefly mentioned in [108], and by making suggestions about what

to do when the deployments are such that the Iterative Waterfilling does not provide good

solutions for the radio links. Paper E goes deeper by investigating random multilink deployments

and policy-based autonomous modifications as well as distributed interaction-based

modifications.

Haykin [108] cites Yu [24] as the source for the ’competitive optimality’, i.e. the Iterative

Waterfilling converged solution. As part of this work, Yu outlines the Iterative Waterfilling

algorithm for an interference channel in a Digital Subscriber Line (DSL) environment. This

work has been the reference for the design of the Iterative Waterfilling part of the Matlab

calculation code for Papers D and E.

According to Google Scholar [28], the Iterative Waterfilling term first appeared in a 2001

work by Yu et al [110]. However, the basic Waterfilling solution by itself, the well-known

optimal power allocation for fixed interference power, is not of new origin. It can be traced

back to being illustrated in a 1964 work by Holsinger [111] which further references a work

by Fano in 1961.

36


3.3. Related Work to Research Goal 3, DSA Architectures and Algorithms

In the review of related work, no papers have been found that present the same results

as in Papers D and E, and in particular no paper has been found that presents the distributed

interaction algorithm in Paper E.

There are several related papers, however, that discuss a wide range of aspects of the

Gaussian interference channels and the Iterative Waterfilling, of which the following papers

are highlighted here as particularly relevant: Hayashi and Luo [112] discuss for which

crosstalk levels Frequency Division Multiple Access (FDMA) is optimal, relative to sharing

the frequency. Shum et al. [113] provide general convergence criteria for the Iterative Waterfilling.

Zhang et al. in a very recent publication [114] provide a broad overview of convex

optimization in Cognitive Radio Networks.

A candidate waveform for employing the Waterfilling or modified versions of waterfilling

principles, with a segmented frequency model as in Papers D and E, is the Non-

Contiguous Orthogonal Frequency Division Multiplexing (NC-OFDM) [115], in which the

individual OFDM carriers may be set up with different power levels and bit rates, or turned

completely off, as required to accomodate the electromagnetic environment.

37


Chapter 4

Results and Implications

4.1 Research Goal 1, Software Defined Radio Challenges

4.1.1 Paper A: Software Defined Radio: Challenges and

Opportunities

Tore Ulversøy, ”Software Defined Radio: Challenges and Opportunities,” IEEE Communications

Surveys & Tutorials, Vol. 12, No. 4, 2010

4.1.1.1 Summary of Contributions

The scientific contribution of this paper consists of the comprehensive summary of challenges

and opportunities, the assessment of the state of SDR, the projections for the way

ahead and the suggestions for improvements.

The following is a walk-through of the sections of the paper, highlighting the main conclusions

and providing supplementary comments:

SDR and the Software Communications Architecture

This section in the paper provides a summary of the Software Communications Architecture,

and serves as background information for the ’SW Architectural Challenges’ section.

SW Architectural Challenges

This section summarizes architectural challenges, specifically in these areas:

• Portability of SCA-based applications: While SCA has made portability easier

through defining an environment and a protocol for running and controlling SCAbased

applications, and by defining their allowed operating-system access, important

challenges still remain, as pointed out in detail in the paper.

• SCA application development: The present state of SCA-based application development

and tools are reviewed, concluding that a further improvement of the efficiency

of the development of SCA-based applications is needed.

• Common Object Request Broker Architecture (CORBA)-related challenges: The issues

with the use of CORBA as a middleware in SCA are pointed out, along with

recent developments. Though a leaner and even more efficient middleware for SDR

than CORBA is very much desired, a clear path to such a middleware is missing.

39


4. RESULTS AND IMPLICATIONS

• SCA challenges and alternative architectures: Alternative architectures to SCA are

reviewed and an architectural outlook is provided.

Challenges and Opportunities Related to Computational Requirements of SDR

This section reviews the computational requirements of SDR as well as available computational

elements. For handheld units, multiple SIMD type processors are seen as a particularly

interesting trend. While the increases in processor performance each year will make

it gradually easier to meet the requirements of existing standards, meeting the requirements

of new high-capacity standards will still be challenging for SDR even in the years to come.

Security-Related Challenges

The security related challenges include secure software download and authorization,

SDR high-assurance platforms and portability issues of the security modules. There are

a number of publications on secure download and software authorization. However, selecting

the right solutions for a particular platform and ensuring that they resist all possible

threats is still a challenging task in each case. For high-assurance platforms, the ’Multiple

Independent Levels of Security’ (MILS) is a particularly promising approach. Portability

of security related code is difficult between different domains of SDR development because

security architectures and Application Programming Interfaces (APIs) to a large degree are

not published or interchanged between domains. MILS has the potential of also reducing

such obstacles.

On the issue of portability it should additionally be noted that it is the ’security by obscurity’

attitude of many governmental institutions, i.e. the unwillingless to publish any

security-architecture related information, including just the APIs, that make portability of

security related modules difficult for SDR.

Regulatory and Certification Issues

Regulatory changes in the USA and Europe to accomodate SDR, are reviewed and projections

made for the further progress in this area. The need for a Software Communications

Architecture certification authority in Europe is highlighted.

Opportunities Related to Business Models and Military and Commercial Markets

The emerging business opportunities and the possible business restructuring that will

occur due to the separation of product development into waveform software development

and platform development are pointed to. Furthermore, product opportunities in both the

military and civilian domains are reviewed.

4.1.1.2 Topics not Included

Software Defined Radio requires base technology from a number of different areas and may

be viewed from a number of different perspectives. While Paper A covers a wide range of

aspects, considerations about the length of the paper meant that it could not cover every one,

as mentioned earlier. In particular, SDR front-end technology, including Digital-to-Analog

(D/A) and Analog-to-Digital (A/D) converters, wide-band radio frequency amplifiers and

up- and down-converters and other central issues are not included. Readers interested in

this topic are advised to consult [21, 32] or some of the related references listed in Paper A.

Section 2.1 in this thesis also provides some introduction to A/D fundamentals.

Another such topic is the environmental aspect of SDR. SDR has the potential of prolonging

the life cycle of radio units due to the enabled waveform software upgrade model

40


4.1. Research Goal 1, Software Defined Radio Challenges

rather than a unit replacement model, and thus from a life cycle point of view this change

will be an improvement. The increased power consumption of SDRs, however, opposes the

recent trend of ’Green Telecommunications’ where the goal is energy-efficient telecommunications.

4.1.1.3 Implications

Paper A is aimed at researchers and development engineers who want to have an introductory

overview of the field of SDR and what the remaining challenges are. It is also aimed

at all other disciplines and parties taking part in the SDR evolution, including business developers,

regulators, security organizations, software and hardware manufacturers, cellular

infrastructure governmental purchasing offices and more. The paper is the only Software

Defined Radio tutorial in the IEEE Communications Surveys and Tutorials online journal.

Paper A also reflects the candidate’s projections and ideas for the way ahead for SDR, in

the hope that these will have a positive influence on the evolution of SDR.

4.1.2 Other Work Related to Challenges of Software Defined Radio

During this doctoral work the author has been a member of ”The Regular Task Group on

SDR, RTO-IST-080 RTG-038 Software Defined Radio”. This NATO Research and Technology

Organization group has the goals to ’Share knowledge & experience of (multi)national

SDR/SCA developments’ and to ’Investigate portability and interoperability’. The group

has members from a number of European countries, as well as Canada and the USA.

The interoperability and portability work in the group uses Stanag 4285, which is a rather

low-complexity High Frequency (HF) modem waveform with a maximum bit rate of 3600

bits / second. Base code written in C was provided by one of the industrial partners in the

group, TELEFUNKEN RACOMS. The low-complexity HF waveform was chosen due to the

availability of code, and because the primary focus was on SCA-related issues of portability

and interoperability, rather than performance.

In connection with work in this group, the author ported one version of the transmitter

base code into an SCA-based application running with the OSSIE Core Framework from

Virginia Tech. This was used together with a receiver part of the same system that had

been converted by a master’s thesis student. Similar exercises were done in Turkey, using

a proprietary Core Framework, and in Germany, using Communications Research Centre

Canada’s SCARI Core Framework. Subsequently, files with baseband and passband data

samples and at various bit rates and interleaver settings were exchanged between the three

nations. All TX-generated files were interpreted successfully at all the different SCA-based

receivers. From this exercise it was concluded that a simple basecode, such as Stanag 4285,

can be converted to an SCA-based form with low to medium complexity effort, and that the

resulting applications readily produced interoperable transmitters and receivers.

As a part of an investigation into portability of Digital Signal Processor (DSP) deployed

components, the author studied and reported on the porting of one of the initially General

Purpose Processor (GPP) deployed components (the ’coder’) of Stanag 4285 TX into a DSPdeployed

component, using FFI’s Spectrum Waveform Development Station [53] platform.

This involved replacing the GPP coder component with a GPP coder proxy and a DSP

coder worker component, with the associated changes to both the SCA-based application

and the functional code.

41


4. RESULTS AND IMPLICATIONS

Coder

(proxy on

GPP/CORBA)

Coder

component

(GPP/CORBA)

coder_worker

(on DSP)

FIGURE 4.1: The GPP coder component was replaced by a coder adapter component and a DSP

coder worker component, on the Spectrum Waveform Development Station platform.

The conclusions from this work were:

• Porting the code from the GPP implementation to the DSP implementation was little

work in itself, but getting the SCA related changes correct in every detail consumed

quite a lot of time

• The data communication from/to the GPP adapter component to/from the DSP component

required programming at a fairly low abstraction level, involving the Spectrum

proprietary quicComm API, and the definition of specific memory locations. If

subsequently we wanted to port this application to a platform with a different API

or different processors, this would require reprogramming. As mentioned in Paper A,

standardized API (e.g. the Modem Hardware Abstraction Layer (MHAL)) or CORBA

on DSP would be preferable for such platform to platform porting

• The Software Communications Architecture was of limited benefit in providing portability

in this case: It helped by concentrating the platform dependent code, it provided

a means of obtaining a handle to the processor where the target code was deployed,

and provided a defined way to load the target code to the DSP

The author has co-authored a paper detailing the activities and conclusions of the Task

Group [33]. The author is also a co-author of the group’s report [34].

4.2 Research Goal 2, SCA Workload Overhead

4.2.1 Paper B: ”On Workload in an SCA-based System, with Varying

Component and Data Packet Sizes”

Tore Ulversøy and Jon Olavsson Neset, ”On Workload in an SCA-based System, with Varying

Component and Data Packet Sizes,” in NATO RTO Symposium on Military Communications,

Prague, Apr. 21-22, 2008.

42


4.2. Research Goal 2, SCA Workload Overhead

4.2.1.1 Background and Context

The Software Communications Architecture (SCA) allows applications to be built as compositions

of components. The software components implement the SCA Resource interface,

and are hence also termed ’Resources’. In the applications, each component is an entity

with defined interfaces to its run-time environment, and each component communicates with

other components through defined ports.

In addition to its functional code, the component realization also implements the interfaces

prescribed by SCA, i.e. the Resource interface and the interfaces inherited by Resource.

The realization also includes a set of descriptive XML files that are defined by SCA.

For CORBA-enabled processors, the communication between the ports is facilitated through

the CORBA middleware.

This component approach has the advantage that code can be reused in other applications

by the reuse of components. As SCA is a distributed architecture, it also defines a scalable

system where more processors can be added when more processing power is needed.

Disadvantages of this component approach include the workload overhead due to the

CORBA middleware as well as the overhead introduced by separating code that could have

been run in a single process and thread into multiple processes and threads.

Having fine application granularity, i.e. the application is split into a number of components,

promotes reusability of the components in that entities that are common to several

waveforms, such as an interleaver or modulator used in several waveforms, may be put into

separate components.

However, the finer the granularity, the greater the overhead. Paper B investigates these

workload overhead effects when the granularity is such that more than one SCA-based component

needs to run on one CORBA-capable GPP.

As reviewed in Chapter 3, the literature prior to Paper B on workload overhead in SCAbased

systems was very limited and there was no research exploring the combination of

granularity and intercomponent packet size variation in depth. As such, Paper B clearly fills

a gap in the literature.

The investigations were carried out on a Linux system using the OSSIE [36] open source

Core Framework and toolsets from Virginia Tech (the specific version numbers are listed in

Table B.1 in the paper) and the required omniORB [116] Object Request Broker. Prior to

selecting this platform, two other platforms had also been considered: FFI had an available

Waveform Development Station platform from Spectrum Signal Processing. While this had

the advantage of being a heterogeneous environment for applications, having both DSPs

and FPGAs in addition to the GPP, it had several disadvantages: The operating system on

the GPP was Windows XP, which was considered less convenient than Linux for workload

type measurements due to many unpredictable background tasks. The proprietary Core

Framework and toolset software on the platform and the lack of source code availability,

were also negative factors, as was also the fact that it was a single unit shared between

many researchers. Another open source Core Framework and toolset, the SCARI Open

[117], was also considered, but its Java implementation made it less attractive for workload

measurements than OSSIE’s C++ implementation. OSSIE was and has continued to be,

a popular tool for research on SCA-based SDR, with currently 58 publications (including

Paper A, Paper B, and the coauthored paper [33]) listed on its website [118].

In the measurements done as part of the investigations, the application software components

(Resources) are, as is the normal and default way on a GPP, instantiated as separate

43


4. RESULTS AND IMPLICATIONS

processes. This was also the default and only way of instantiatiating components in the

OSSIE Core Framework and toolset that was used for the analysis in Paper B, and was consistent

with related work, e.g. [75, 76, 79]. The application instantiation is initiated from the

OSSIE tool ’alf’, which invocates the ’create’ method of the SCA ApplicationFactory. This

creates the application instance, allocates capacities, instantiates the Resources through the

execute method of the ExecutableDevice, connects the ports and registers the application, as

illustrated in Figure 4.2.

ApplicationFactory::Create()

ALF

Call

ApplicationFactory::Create()

Request profile

Create

application

instance

Allocate device

capacities

Instantiate

components

(Resources)

Connect

Resources

Inform

DomainManager

FIGURE 4.2: Simplified sequence [42] for the creation of a waveform application in OSSIE.

The workload test applications are run with a programmed packet size, and a programmed

number of packets per second, the ’packet rate’. The packet rate is set using

blocking reads of samples from an audio-card, e.g. 40 samples at an audio sample rate of

1600 samples per second to get a packet rate of 40 packets per second.

The empirical analysis used the CPU workload measurement tools OProfile [119] and

SYSSTAT sar [120]. Other alternatives here would have been manual instrumentation of the

code, i.e. insertion of time measurements in the code, or using a profiler tool that would automatically

insert instrumentation instructions in the code. OProfile and SYSSTAT sar were

preferred to the latter alternatives due to concerns with the instrumentation code possibly

adding workload and becoming a source of error, and also because OProfile and SYSSTAT

sar are simple tools to use and they monitor all activity on the processor.

In the SYSSTAT sar measurements, it is assumed that the workload background activity

from other processes than the ones related to the running of the SDR application in the SCA

environment, is negligible. Background activity, with just four idle terminal windows open

in Linux, was measured with SYSSTAT sar over 5 periods of 40 seconds to be on average

less than 0.1 % workload combined for the user and system CPU space, implying that this

assumption is good in the average case.

4.2.1.2 Results

The paper provides results concerning the workload effects of varying the application granularity

of an SCA-based application on the total workload of a processor. All the components

are deployed on a single General Purpose Processor (GPP), simulating that the granularity

is such that many components need to be run on the same CORBA-enabled GPP.

The results provided in the paper consist of empirical analysis, the formulation of two

simple analytical models, estimates of parameters of the models and comparison of modeled

versus measured data.

For the empirical analysis, both a ”real” waveform application, the transmitter part of the

HF standard Stanag 4285, and a ”synthetic” application, consisting merely of Finite Impulse

Response (FIR) filters, are used.

44


4.2. Research Goal 2, SCA Workload Overhead

For the Stanag 4285 application, versions with 2,7, and 11 components were compared,

as well as a single-process ’C’-implementation. The comparison was made both at approximately

the original symbol rate of the waveform (2400 symb/sec), but also at higher symbol

rates in order to get more significant CPU readings (above approximately 4 %) and thus

better accuracy of the results. In general, it was observed that the more components in the

implementation, the higher workload for the processor. As an example, when compared at

a symbol rate of 25600 symb/sec, it was found that while the ’C’ implementation lead to

a total user+system CPU load of 4.4%, the corresponding CPU load of the 11-component

SCA-based implementation was 16.6%, see Figure 4.3.

Stanag 4285 TX

Non-SCA CPU: 4.4%

Stanag

4285 TX

Data

Sink

CPU: 5.5%

Data

Source

FEC

Encoder

Interleaver

Symbol

Mapper &

Scrambler

Symbol to

I/Q & TX

Filter

Float to

fixed

converter

Data

Sink

CPU: 11.2%

Data

Source

FEC

Encoder

Interleaver

Symbol

Mapper &

Scrambler

Symbol to

I/Q & TX

Filter

Float to

fixed

converter

Forwarder Forwarder Forwarder Forwarder

Data

Sink

CPU: 16.6%

FIGURE 4.3: User+system CPU % for the non-SCA (’C’) 4285 TX version, compared to three

SCA-based versions with different granularity. Measured at a symbol rate of 25600 symb/sec.

Further measurements were done with the synthetic waveform applications, using 2, 3,

5, and 11 components, and again compared to a standalone ’C’ implementation. Again it

was confirmed that splitting the application into more components led to more workload

overhead. The relative size of the overhead to the total workload, however, depended on the

workload of the functional useful work.

Also with the synthetic workload, the relation between the CPU workload and the packet

size, i.e. the size in number of floating point numbers sent from component to component

through CORBA, was studied. A clear and significant workload overhead dependency on

the packet size was found.

Two simple analytical models were formulated for the total workload of the applications.

The first is a lower-bound model, taking into account the workload of the useful functional

work in the components and the workload of the communication through CORBA and the

data conversions associated with that, but not taking into account any increase in overhead

from context switches, i.e. the process of saving and restoring state as the active thread or

process changes. The second model additionally takes into account the direct and indirect

costs [121] of the context switches.

Using a test application with only one source and one sink component, the parameters

in the lower-bound model were estimated. A comparison with measured data showed, as

expected, that the lower bound model underestimates the actual workload, particularly at

large packet sizes. Still, it predicts the major part of the workload, particularly at low packet

45


4. RESULTS AND IMPLICATIONS

sizes. The underestimation is greater the larger the packet size. A likely contributor to

the underestimation is that the first model does not include the indirect losses due to the

required updating of the caches following a context switch. Since the memory footprint of

each component contains several buffers of the same size as the packet, these losses logically

increase with the packet size.

For the model including context switches, only rough estimates of the parameters were

made, and the numerical output in Figure B.10 should therefore only be considered as an

example and an illustration of the model output. It is expected, however, that this second

model will give better estimates than the lower-bound given the necessary tools to estimate

these parameters accurately.

Table 2 in the paper also illustrates the large startup cost when pushing a minimum packet

from one component to the other, implying that for workload optimization it is advantageous

to avoid frequent intercomponent transmission of small packets.

4.2.1.3 Comments on Measurement Accuracies

In general the chosen measurement tool for measuring processor workload, SYSSTAT sar

[120], provided consistent readings. A standard deviation of between 0.01% and 0.05%

of the presented average values was found when evaluating the variation in the different

measurement series. At low workload numbers, however, small offsets due to factors such

as background activity have a more significant influence than at higher workload numbers.

Therefore, in setting the packet rate in the experiments, typically rates that provided higher

workload values than 5-10% were chosen in order to minimize this effect.

The author is aware that the paper could have been more consistent in the presentation

of numerical data, and that in many cases more decimals than are statistically significant are

presented.

As mentioned, the estimates of the parameters in the second model are very coarse ones,

and are only meant for the purposes of illustrating this model. In particular, t CSD is in the

paper coarsely estimated at 5µsec, which, however, overestimates t CSD . t CSD has since

been measured, using the procedure and available software from [121], to be 1.6µsec. The

exhaustion of the L2 cache size, refer to the edge of the black graph in Figure B.10, is also

not modeled in detail. Also, the value of the fraction ’c’ parameter is selected only for

the purposes of illustrating the model. The reason for not proceeding with more accurate

parameter estimates of this model was a combination of available time and software tool

issues, with difficulties arising due to some features not being available with the particular

operating system kernel that was used.

4.2.1.4 Discussion and Implications of the Results

The results are of relevance for SCA-based SDR where more than one component is deployed

on one of the CORBA-enabled processors in the system. While the measurements

and parameter estimates apply for the specific hardware, software and middleware environment

investigated in the paper, the trends and conclusions are expected with a high probability

to be also valid for other SCA-based platforms, since the sources for the observed

workload overhead (data copying, CORBA and transport overhead, context switching) will

also be present on other SCA-based platforms. As an example supporting this statement,

and as mentioned in Section 3.2, similar conclusions regarding component granularity are

46


4.2. Research Goal 2, SCA Workload Overhead

reached in a later paper [76] using a different Core Framework and toolset (the commercial

version of SCARI), and a different ORB (TAO). The results will be relevant also for heterogeneous

platforms, with multiple types of processing elements, as the CORBA-enabled

processors on these typically will run multiple components. However, it will vary with the

specific design principles applied, whether these components run performance-critical parts

of the application or not. Further insights into heterogeneous platforms, and further insights

into parameters that can be optimized on CORBA-enabled processors beyond that examined

in Paper B, will be provided in the next section.

Based on the results in Paper B, there are several conclusions that can be drawn that

have implications for the design of SCA-based applications for the above mentioned environments:

The results show that in designing an SCA-based component, care should be taken such

that the functional processing in the component is significant relative to the CORBA-related

overhead, such that the overhead does not dominate the CPU workload. In other words,

better computational efficiency is achieved by having components with significant packet

processing workload.

The above results imply that the computational efficiency is enhanced by having course

application granularity with few components per processor. In this way, there is a compromise

between reusability of the code, which is made easier by having a fine application

granularity, and application computing efficiency (or low computing overhead).

Computing efficiency is one of the factors that needs to be taken into account when determining

the block sizes communicated through the CORBA connections. Table 2 in the

paper shows the large startup cost for pushing a small data packet from one component to the

other, and illustrates that for computing efficiency it is better to send the same amount of data

in few large packets rather than in many small ones. There are however several other considerations

that need to be taken into account when determining packet sizes. Data throughput

considerations point in the same direction to larger packets, it has been shown [122] that

block sizes are a major factor influencing the throughput, with throughput increasing with

block size in an asymptotic way. Latency, in this context defined as the delay between the

time when a packet is sent in one component until it is received in the target component, is

another factor that needs to be taken into account, with latency increasing with packet size.

Another factor is the frame size of the actual waveform. Further, local waveform requirements

may dictate frequent feedback from one component to another, in which case it may

be difficult to increase the packet size.

The two models in paper B can be used to predict the total workload and workload

overheads for a given system, with a varying number of components. By estimating the

parameters of the two models for the specific SCA-based system with the specific processor,

Core Framework, toolset and ORB being used, workload estimates can be made, that can

provide guidance as to the application granularity.

4.2.2 Beyond Paper B: An Overview of Optimization of Processing

Efficiency of SCA-based Software Defined Radios

We have learnt from the results in Paper B that in order to keep workload overheads at

satisfactory low levels, it is beneficial to have a low application granularity. Also, intercomponent

communication should preferably have low packet rates and instead accumulate

47


4. RESULTS AND IMPLICATIONS

data into larger packet sizes, if allowed by the functional restrictions of the SDR waveform

application.

But how can the workload overheads be reduced, at a given granularity, and when packet

rates cannot be further reduced? In this section, a more holistic view is taken, and a wider

overview of the various ways of optimizing SCA-based SDRs is given.

The overview is divided into considerations for optimizations of SCA-based GPP environments,

and considerations for SCA-based heterogeneous processing environments,

where in the latter case the platform has different types of processors, typically including

GPPs, DSPs and FPGAs.

For the SCA-based GPP environment case, some further workload and also latency measurements

are provided. These have been conducted on exactly the same platform with exactly

the same software (RedHat Linux operating system, OSSIE [36] Core Framework and

toolset, and omniORB [116]) with exactly the same revisions as the ones in Paper B (refer

to the righthand column of Table B.1 in Paper B for further details).

For the workload measurements, the same measurement tool SYSSTAT sar has also been

used, as well as the same synthetic applications with FIR-filter processing componenents,

having 2,3,5, or 11 components (W2, W3, W5, W11). The measurements are average values

over 5 periods of 40 seconds. For latency measurements, very careful instrumentation of the

CORBATSTSRC (a source component sending packets) and CORBATSTSINK (a packet

receiver component) components of the CORBATEST application was implemented. The

time measurements used a high resolution timer method, relying on counting the cycles that

the CPU has gone through since startup, with source code from [121]. In order to add minimally

to the CPU workload, the measurement samples were accumulated in arrays while

running, then dumped to file when the measurement sequence was finished. The time measurements

were positioned in the following way: One prior to converting the floating point

data to be sent to the FloatSequences required by the OSSIE-defined realFloat interface, one

before the invocation of pushPacket to send the data to the CORBATSTSINK component,

one immediately after in order to monitor when control is returned to the CORBATSTSRC

component, one when the data is received in the CORBATSTSINK and the final one when

the data have been converted back to the native array of float type, as illustrated in pseudocode

in Figure 4.4.

CORBATSTSRC

while (1)

{

blockUsingAudioDevice(); //correct packet rate

beforeFloat[packetNumber]=measureTime();

copyFloattoFloatSequence(); //copy to FloatSequence

beforePush[packetNumber]=measureTime();

pushPacket(); //remote invocation to transfer data

return[packetNumber]=measureTime();

}

CORBATSTSINK

while (1)

{

getData(); //wait until pushed data available, get it

afterGet[packetNumber]=measureTime();

measureTime();

copyFloatSequencetoFloat(); //copy to float

afterFloat[packetNumber]=measureTime();

}

FIGURE 4.4: Illustration of the CORBATEST test application for latency measurements, with pseudocode

showing the placement of the time measurements in the components’ processing loops. The

packet size used was 5000 floating point numbers.

48


4.2. Research Goal 2, SCA Workload Overhead

4.2.2.1 Optimization of SCA-based General Purpose Processor Environments

ORB Selection

Selecting an efficient ORB is a prioritized optimization consideration needed for an SCA

GPP environment. A number of ORBs exist, ranging from free general-purpose ones to commercial

highly optimized ones. An example of the latter category is ORBexpress RT [123],

which is a Real-Time CORBA compliant ORB, implying that it has Real-Time CORBA

quality-of-service control features. In this present work, omniORB has been used, due to it

being required by OSSIE. On its website, omniORB is promoted as a free high-performance

ORB [116].

ORBs are tunable in that they have a number of configuration parameters, such that

the ORB may be optimized by further tweeking these adjustments. In omniORB, the

configuration parameters are in the file omniORB.cfg. In Paper B, the file is used with the

default settings as when installed with OSSIE.

The Contribution from the SCA Core Framework

According to [83] and referring to OSSIE and to latency measurements, the contribution

from the SCA Core Framework itself, relative to the contribution from CORBA, is small.

This is natural as the Core Framework is basically a ’Deployment and Configuration

engine’ [124], implying that when all the components are deployed and running, it has a

memory footprint overhead but is in idle mode. However, again referring to OSSIE, there

is a contribution to latency and workload overhead from the OSSIE-defined StandardInterfaces

in that an additional data copy and thread switch is made, relative to a direct CORBA

invocation [81]. These effects are also evident in Paper B. Hence, optimizing the way one

uses and interfaces with CORBA may have some limited potential for reducing memory,

latency and workload overhead.

Optimization of Data Converting and Formatting Operations

A significant contributor to workload overhead and latency are all the data-converting

and copying operations, all the way from the native data representation through the CORBA

Common Data Representation format and to the native format in the receiving component.

The operations are simple copying and formatting operations that put the data into different

representations. When processed serially on a processor, these are time consuming. Obviously,

though, there is a lot of parallellism in these operations (the same instruction may be

applied to many data items in parallel). In this respect, it is very interesting to note some

results from subsets of ORB functionality having been implemented at native gate level on

FPGAs. Such implementations make it possible to create very parallel implementations of

data formatting operations. When sending packets between such FPGA ORBs, on very high

throughput and low latency transports, impressively low latencies have been achieved, in the

order of a few hundred nanoseconds 1 [125]. A comparison of packet transmission between

two GPP ORBs and between an FPGA ORB and a GPP ORB [126], points in the same

direction of the FPGA ORB being faster than the GPP one. This illustrates the potential

effectiveness of a parallel implementation of the formatting and decoding of the CORBA

packets to/from the transport. If, in the same way as carried out on these FPGA ORBs, a

dedicated CORBA packet formatter and deformatter could be an integrated hardware-part

1 The specific packet size was not specified in this reference.

49


4. RESULTS AND IMPLICATIONS

of specific GPPs, this would have a potential of significantly lowering latency and processor

workload overhead. This would come, however, at the expense of having to manufacture

specialized CORBA processors (the integration of a GPP and CORBA accelerator circuitry)

that would have a smaller niche market than fully general processors.

Transport

Another important performance factor is the underlying transport. While the default

transport typically is TCP/IP, CORBA may be configured to use other transports, e.g. faster

shared memory ones or circuit switched rather than packet switched ones. Using a faster

transport reduces the transport induced part of the latency.

Optimizing the choice of transport is also important in the specific case discussed in

Paper B, i.e. when having such a high granularity that several SCA-based components run

on the same processor. Here, one may take advantage of intra-processor transports, such

as a Unix domain transport. It has been shown in [81] that for a single-processor system

using OSSIE and omniORB, latencies may be significantly reduced when switching to a

Unix transport, rather than using the default TCP/IP.

These results have been confirmed in this work here, by using the instrumented COR-

BATEST application illustrated in Figure 4.4 with a packet size of 5000 floats. Since the

measurements are taken before the full packet is sent and after the full packet has been received,

the latencies are packet latencies representing the duration between the first bit sent

and the last one received (rather than a bit oriented latency that would be between the first

bit sent and first bit received).

Figure 4.5 shows a histogram (101 measurements, 10 microsecond bins) of the packet

latency between before pushPacket and after getData (i.e. not including the conversion

from/to native floating point data), for the default TCP/IP transport and a Unix transport,

respectively. The outlier point is the first packet that comes through, and that takes considerable

more time than the next ones, probably due to ORB activity carried out when handling

the first message. It is observed that the latency is considerably lower when using the Unix

transport. It is worth noting that the change of transport is done by only changing three lines

of parameter settings in the omniORB.cfg file, no other changes are needed!

Figure 4.6 in the same way shows the total latency from native float data to native float

data in SINK, again for TCP/IP and Unix transport. It is observed that the absolute difference

in latency is about the same, but that the relative difference is less since the converting to

and from native float data is the same in both the TCP/IP and Unix cases.

It is to be expected that the improvement from using the Unix transport rather than

TCP/IP will be visible also on SYSSTAT sar workload measurements. To verify this,

the workload measurements from the left part of Figure B.6 in Paper B were repeated,

first with the original omniORB settings with the default TCP/IP, then using the Unix

transport. The results are shown in Figure 4.7. It is seen that a small workload overhead

improvement is achieved from using the Unix transport, and that the dominant part of the

improvement is in the system space. That the improvement is in the systems space is natural

since both the TCP/IP and the Unix sockets processing take place inside the Linux kernel.

Twoway versus Oneway

Another variable is that CORBA defines different levels of quality-of-service in terms of

the reliability of the CORBA remote invocations. The default type of invocation is the two-

50


4.2. Research Goal 2, SCA Workload Overhead

Latency Comparison, TCP and Unix Transport (pushPacket‐>getData)

Number of packets

70

60

50

40

30

20

10

0

TCP average: 201 µs

Unix average: 107 µs

TCP

Unix

0

30

60

90

120

150

180

210

240

270

300

330

360

390

420

450

480

510

540

570

600

630

660

690

720

750

780

810

840

870

900

930

960

990

1 020

1 050

1 080

1 110

1 140

Latency (microseconds)

FIGURE 4.5: Packet latency from prior to pushPacket in source component to after getData in sink

component, for two different transports, TCP/IP and Unix. Packet size 5000 floating point numbers.

90

Latency Comparison, TCP and Unix Transport (float‐>float)

Number of packets

80

70

60

50

40

30

20

10

0

TCP average: 269 µs

Unix average: 174 µs

TCP

Unix

30

60

90

120

150

180

210

240

270

300

330

360

390

420

450

480

510

540

570

600

630

660

690

720

750

780

810

840

870

900

930

960

990

1 020

1 050

1 080

1 110

1 140

0

Latency (microseconds)

FIGURE 4.6: Packet latency from prior to copying float data to FloatSequences to after all the data is

in floating point data representation in the sink component, for two different transports, TCP/IP and

Unix. Packet size 5000 floating point numbers.

way one, where the remote invocation blocks until it completes and the client gets a reply

back. The reply optionally may return data in out or inout parameters, but it is worth noting

that the behavior is a blocking one even if the invocation is not to return data. The advantage

with two-way invocations is reliability; the client will have information about whether the

invocation was successful or not. The disadvantages with two-way are (1) that the client

does not resume control before the signalling from the remote object is received, and (2)

that the return-signalling consumes additional overhead in terms of transport bandwidth and

processing. OSSIE, in its ’StandardInterfaces’ of which the ’realFloat’ interface is used in

Paper B, uses two-way invocations.

One-way invocations are ’ORB best effort’ or also referred to as ’at most once’ invocations.

Here, the client performs the remote invocation (in our case pushes out the packet of

51


4. RESULTS AND IMPLICATIONS

CPU Workload versus Configuration & Transport

B=2000, PR=40

16

14

12

CP PU WL [%]

10

8

6

System

User

4

2

0

FUNC N=10

W2 N=10

TCP

W2 N=10

Unix

W3 N=10

TCP

W3 N= 10

Unix

W5 N=10

TCP

W5 N=10

Unix

W11 N=10

TCP

W11 N=10

Unix

FIGURE 4.7: Workload comparison, Unix and TCP.

processed data) and resumes control directly afterwards 2 .

In a real geographically distributed environment with varying quality-of-service in the

communication network, reliable invocation mechanisms are a must in many applications.

In a stable environment inside the chassis of a Software Defined Radio, it is not so obvious

that two-way invocations are needed.

In order to see if oneway versus twoway have significant implications on the latency, the

time of return of control to the client (the ’sender’ component) as well as workload overhead,

another experiment was performed, again using the same environment and application components

as in Paper B. Here, the pushPacket method of the realFloat interface was modified

into a oneway one by simply adding the oneway keyword into the relevant CORBA Interface

Definition Language (IDL) file, i.e. from

to

void pushPacket(in PortTypes::FloatSequence I);

oneway void pushPacket(in PortTypes::FloatSequence I);

then remaking the ’StandardInterfaces’ library as well as the CORBATEST application

and the W2, W3, W5, and W11 applications.

The CORBATEST application was run, in the same way as for twoway invocations, to

measure latency of 101 packets of 5000 floats, as well as the time of return of control to the

2 The latest CORBA specifications additionally allow defining quality-of-service for the one-way more

specifically, but omniORB does not have this feature and it is therefore not described here.

52


4.2. Research Goal 2, SCA Workload Overhead

CORBATSTSRC component, for both the TCP and Unix transport. The measurements are

presented in Figures 4.8 and 4.9, and compared to the twoway ones. Here, time 0 is the time

before the packet was converted from the original type float data in the CORBATSTSRC

component. Twoway is shown on the positive vertical scale, and oneway on the negative

one, for purpose of easier comparison. The first packet is not included in these figures.

Latency Comparison, Twoway versus Oneway, TCP

oneway Number of packets

twoway

100

80

60

40

20

0

‐20

‐40

‐60

‐80

‐100

0

8

16

24

32

40

48

56

64

72

80

88

96

104

112

120

128

136

144

152

160

168

176

184

192

200

208

216

224

232

240

248

256

264

272

280

288

296

Latency (microseconds)

Twoway before push

Twoway return

Twoway after get

Twoway after float

Oneway before push

Oneway return

Oneway after get

Oneway after float

FIGURE 4.8: Histogram of packet latency for oneway and twoway invocations, TCP/IP transport.

The time sample just prior to converting from the float data type in the SRC component is taken as

time 0 for each packet transmission. The oneway measurements are shown on the negative vertical

scale. The packet size is 5000 floating point numbers. Measurements of 100 of the 101 packets are

shown (the measurements of the first packet skipped due to the outlier effect). Bin size 2 µsec.

Latency Comparison, Twoway versus Oneway, Unix

twoway

oneway Number of packets

100

80

60

40

20

0

‐20

‐40

‐60

‐80

‐100

0

8

16

24

32

40

48

56

64

72

80

88

96

104

112

120

128

136

144

152

160

168

176

184

192

200

208

216

224

232

240

248

256

264

272

280

288

296

Latency (microseconds)

Twoway before push

Twoway return

Twoway after get

Twoway after float

Oneway before push

Oneway return

Oneway after get

Oneway after float

FIGURE 4.9: Histogram of packet latency for oneway and twoway invocations, Unix transport.

53


4. RESULTS AND IMPLICATIONS

The figures confirm the earlier return of control in the oneway cases. For TCP, an average

improvement of the float → float latency from 269 µsec to 240 µsec is observed. For the

Unix transport, the measured float → float latency is actually worsened by 6 µsec to 180

µsec, possibly influenced by the additional context switches due to the early return of control

to the CORBATSTSRC component. The total time to the latest measurement point in each

packet transmission is higher in the twoway case, though.

SYSSTAT sar workload measurements, again comparable to the left part of Figure

B.6 in Paper B, are provided in Figure 4.10. The measurements show a small but evident

workload reduction by using oneway invocations, and as expected the reduction is larger for

the higher-granularity application versions.

CPU Workload versus Configuration, Transport and type of Remote Invocation

B=2000, PR=40

16

14

12

CPU WL [%]

10

8

6

System

User

4

2

0

FUNC N=10

W2 N=10 TCP

TWOW.

W2 N=10 Unix

TWOW.

W2 N=10 TCP

ONEW.

W2 N=10 Unix

ONEW.

W3 N=10 TCP

TWOW.

W3 N=10 Unix

TWOW.

W3 N=10 TCP

ONEW.

W3 N=10 Unix

ONEW.

W5 N=10 TCP

TWOW.

W5 N=10 Unix

TWOW.

W5 N=10 TCP

ONEW.

W5 N=10 Unix

ONEW.

W11 N=10 TCP

TWOW.

W11 N=10 Unix

TWOW.

W11 N=10 TCP

ONEW.

W11 N=10 Unix

ONEW.

FIGURE 4.10: Workload comparison, oneway and twoway invocations. Block size 2000 floating

point numbers.

Forcing Same-Address Space

With advanced SCA modeling tools, another way of optimizing performance for the

particular case discussed in Paper B, is forcing a same-address space of several SCA Resources

[74], and using an ORB that is able to take advantage of this by substituting remote

invocations by local calls. According to the SCARI team of authors in a 2009 publication

[74], ”with proper modeling tools, an SCA ResourceFactory can be automatically generated

to allow the instantiation of pre-existing SCA Resource into a single address space ”,

and the authors have demonstrated the latency benefits of this approach. Such a Resource-

Factory, in their implementation, is compiled and linked together with all of the SCA Resources

that it is to instantiate, into a single executable [124]. The ResourceFactory is then

actually the entity which is instantiated by the SCA ApplicationFactory (through a call to the

Execute method of the ExecutableDevice), and the ResourceFactory then creates servant in-

54


4.2. Research Goal 2, SCA Workload Overhead

stances for the Resources as threads in the ResourceFactory’s address space. This contrasts

with the normal case where each SCA Resource is directly instantiated as a process.

The possibility of generating and deploying such a ResourceFactory was not present

in OSSIE. In order to perform an equivalent test of such a deployment without having the

necessary tool support, the CORBATEST and W11 applications were modified into having

same-space instantiations of all their components. Using W11 as an example, the procedure

was as follows:

• The first component in W11, SRC, was modified to include all the source code of

all the components, including both the SCA administrative code and the processing

thread code.

• The ’main’ file, that originally was set up to create just the SRC CORBA servant,

was carefully modified to setting up all the SRC, F1, F2 etc. CORBA servants, all as

separate threads.

• The ’main’ file was set up to use the original CORBA IDs and labels, such that the

XML files for the components and for the port connections could be reused exactly as

they were.

• All the binary files, except the SRC one, were replaced by Linux batch files that merely

echoed their name to the terminal window (just as a check to see that they had been

executed), then terminated.

In this way, the whole original application, including all of the CORBA related code,

could run inside the process space of the SRC component. For the OSSIE ’alf’ tool and the

ApplicationFactory, however, the application appeared exactly as in the multiprocess case.

Figure 4.11 shows a histogram of the latency of this same-address-space case compared

to TCP oneway. The latency is seen to be much lower in the same space case, with average

measurements of float→ float of 105 versus 240 µsec, and pushPacket→ getData of 37

versus 172 µsec.

The SYSSTAT sar workload measurements, ref. Figure 4.12, also show a very significant

improvement from the cases where the SCA-based Resources are instantiated directly

in separate processes. The workload is still seen to be higher than the reference

’C’-implementation, though. The same space case allows higher granularity for the same

overhead, e.g. the workload of the 11-component same space case is seen to be approximately

equal to a 5-component implementation using Unix oneway.

The ResourceFactory solution appears very attractive for resource-constrained GPPs,

particularly if there is tool support such that the ResourceFactory can be automatically generated.

A major disadvantage with this single address space approach is, however, that implicitly

some of the component isolation properties of the SCA Resources are lost, in that one

Resource may accidentally access the memory space of others. Another side effect is that

the invocations, being converted by the ORB to local calls (i.e. normal function calls) are

automatically twoway ones (even if oneway is specified). Thus, on the one hand the local

calls have low latency, on the other hand they have the normal function call blocking behavior

that forces the calling object to have to wait [124], which according to one of the authors

of the above cited paper in some cases may have a slowing effect on execution [124]. This

was not observed as a problem in the OSSIE system measurements here, though.

55


4. RESULTS AND IMPLICATIONS

TCP oneway

100

80

60

Latency Comparison, TCP Oneway versus Same Address Space

Number of packets

Same space

40

20

0

‐20

‐40

‐60

‐80

‐100

0

8

16

24

32

40

48

56

64

72

80

88

96

104

112

120

128

136

144

152

160

168

176

184

192

200

208

216

224

232

240

248

256

264

272

280

288

296

Latency (microseconds)

TCP oneway before push

TCP oneway return

TCP oneway after get

TCP oneway after float

Same space before push

Same space return

Same space after get

Same space after float

FIGURE 4.11: Packet latency, TCP and same space. Packet size 5000 floating point numbers. Bin

size 2 µsec.

16

CPU Workload, W11 Application, Forced Local Invocations

Compared to TCP/Unix/Twoway/Oneway

B=2000, PR=40

14

12

CPU WL [%]

10

8

6

4

System

User

2

0

FUNC N=10

W5 N=10

Unix ONEW.

W11 1 N=10

TCP TWOW.

W11 1 N=10

Unix TWOW.

W11 1 N=10

TCP ONEW.

W11 1 N=10

Unix ONEW.

W11 1 N=10

LOC. TWOW.

W11 1 N=10

LOC. ONEW.

FIGURE 4.12: Workload comparison, same-address space instantiated W11 compared to normal

instantiation W11 using combinations of oneway/twoway/Unix/TCP. In the same figure, W5 Unix

oneway and the single-thread reference, FUNC.

4.2.2.2 SCA-based Heterogeneous Processing Environments

CORBA for the Control Path

By far the most typical way of trying to avoid the negative effects of CORBA in SCA

on workload, latency and throughput of the SDR-application, is using CORBA mostly

for the control data path, where the volume of data is less and the timing is less critical.

56


4.2. Research Goal 2, SCA Workload Overhead

The most computing-intensive signal processing components in this case have ’worker’

implementations that typically are deployed on DSPs or FPGAs. The ’workers’ have

SCA-component proxy implementations that reside on CORBA-enabled GPPs. [127]. The

signal processing data exchange typically in this approach is through high-speed transports,

such as RapidIO [127], with standardized or platform-defined APIs for the communication

calls. Spectrum Signal Processing’s platforms are typically designed with this intercomponent

communication approach in mind, and with Spectrum’s own quickComm API being

used for the non-SCA communication calls. This type of solution represents a practical

compromise that reduces the negative effects of CORBA and to a large extent concentrates

the platform-dependent application code. It still represents a less standardized approach,

with several communication API standards around and more issues with portability. It is

also a design pattern that complicates the application design relative to one consisting of

components for CORBA-enabled processors only.

Offloading the General Purpose Processor

Another proxy design pattern is that of keeping both the control and data communication

between components on CORBA-enabled GPPs, but ’offloading’ the GPP by running the

most processing-intensive functionality on specialized non-CORBA-enabled processors,

such as DSPs or FPGAs. Just as described for the approach above, and as described for the

experiment in Section 4.1.2, proxy components on the GPP are responsible for the ’worker’

component implementations on the special processors. A difference from the approach

above, is that the proxy component now receives its input data and sends out the processed

data through CORBA. The data is communicated to / from the worker through platformspecific

transports and APIs (for the Waveform Development Station platform: a PCI bus

and the quickComm API). A positive aspect of this approach is that the design model

very closely resembles one where all the components reside on fully CORBA-enabled

processors. A further positive aspect is that with the processing residing in the special

processors, the footprints of the GPP components become small, reducing the indirect

costs of context switches. A negative aspect is that the through-CORBA communication

becomes critical to the signal processing path, such that the factors previously discussed for

optimizing CORBA on GPP environments now become very important again. The timing

of the data relaying to/from the special processor also becomes critical. In the experiment in

Section 4.1.2, this was accomplished successfully, and the communication to the worker, the

processing in the worker, and the data return, were all completed in time for the application

to maintain its processing speed.

All-CORBA

In recent years, lean ORBs for DSPs, and as mentioned even native-gatelevel ORBs for

FPGAs have become available, making it possible to extend CORBA in SCA onto these

special processors. An approach where CORBA communication is also used on the special

processors, is referred to as an ”all-CORBA” approach. The obvious advantage is consistent

communication between all the components in a waveform application. It also promotes

portability as platform-dependent proxy code for special processor communication is eliminated.

Also, as mentioned previously, good performance with remarkably low latencies have

been reported for FPGA ORBs.

The disadvantages with the ”all-CORBA” approach include that CORBA still creates

57


4. RESULTS AND IMPLICATIONS

workload overhead latency, compared to not using CORBA, particularly for DSPs and GPPs.

Another disadvantage is that it is not a full-feature CORBA that is present on the DSPs

and ORBs, only selected features are included. CORBA also creates significant memory

consumption overhead in the DSPs, and adds real estate overhead in the FPGAs.

4.3 Research Goal 3, Dynamic Spectrum Access

Architectures and Algorithms

4.3.1 Dynamic Spectrum Access Architectures

The architectural part of research goal 3 is treated mainly in Papers C and E. Paper C is a

proposal for a particular broker architecture concept, while Paper E provides a comparison

of DSA architectural concepts.

4.3.1.1 Paper C: Dynamic Frequency Broker and Cognitive Radio

T. Maseng and T. Ulversøy, ”Dynamic Frequency Broker and Cognitive Radio,” in The IET

Seminar on Cognitive Radio and Software Defined Radios: Technologies and Techniques,

The IET, Savoy Place, London, Sept. 18, 2008.

Background

Paper C outlines a Centralized Coordination DSA concept. Centralized Coordination

has the advantage that the use of the spectrum is managed through an infrastructure, such

that the interests of all spectrum users, including users in the form of silent receivers and

other ”hidden” nodes, may be taken care of according to predefined priorities and spectrum

decision algorithms. The paper is a conceptual proposal only, it does not include simulations

or prototyping.

Summary of Contributions

Paper C points to the issues associated with traditional frequency allocation and assignment,

and also to the problems associated with using Cognitive Radio as a dynamic alternative

to increase spectrum utilization. A particular Centralized Coordination concept, the

Dynamic Frequency Broker (DFB), is suggested. The DFB is a regional frequency coordination

authority, which keeps a complete list of frequency assignments within an area and

maintains an updated terrain propagation path loss model of its area. The regional DFBs are

organized in a hierarchy, with the national frequency authorities or even higher-level authorities

at the top. It assigns time limited spectrum leases to requesting radio nodes. A radio

node in this context may be an individual radio transceiver, a base station or any aggregation

of transceivers that negotiate as a group with the broker. The communication between radio

nodes and the broker, and the broker discovery mechanisms, are based on Web Services.

It is assumed that the Dynamic Frequency Brokers are connected to the Internet, and that

radio nodes that do not have a fixed connection, connect into the Internet through dedicated

control channels, or get their leases through proxies. Challenges related to a DFB solution,

such as security challenges, are also pointed to.

Figure 4.13 illustrates a typical scenario showing the lower level DFB that is responsible

for a geographical region, and including an Internet-attached DFB slave station that aids in

58


4.3. Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms

—Fixed line Internet access continuously

—Internet access via radio service channels

—Prior service negotiation via the Internet

FIGURE 4.13: A DFB scenario illustrating different radio nodes and different ways of communicating

with the DFB.

monitoring the spectrum as well as in providing a wireless gateway for control communication.

The figure illustrates different types of radio nodes that receive spectrum leases from

the DFB, ranging from cellular base stations to ad hoc hunting team radio equipment and

wireless microphones. It also illustrates the various ways of communicating with the DFB,

ranging from online wideband Internet connections through wireless dedicated channels to

this Internet connection and to software proxies running on Internet-connected computers.

Policies

Coordination

Conflict

Conflict resolution

FIGURE 4.14: Illustration of a hierarchy of coordinating DFBs.

59


DFB, Communication Middleware

4. RESULTS AND IMPLICATIONS

Radio

Node

UDDI or Catalogue of DFBs

WSDL

e.g. SOAP

messages

WSDL

DFB

FIGURE 4.15: The use of Web Services in DFB.

• The DFB is implemented as a We

Service

• XML-based messages over SOAP

and HTTP

• Nearest DFB found in UDDI or DF

Catalogue

• Web Service interface downloade

as WSDL file

Figure 4.14 illustrates the hierarchy of coordinating brokers in DFB, where the DFBs

at the lower level each have responsibility for the radio nodes in a particular geographical

region. Spectrum decisions are coordinated with neighbor DFBs. Spectrum policies migrate

from the upper levels, e.g. national levels, to the lower level DFBs. Spectrum conflicts are

pushed upwards for resolution.

Figure 4.15 illustrates the use of Web Services in DFB. The information about a DFBs

service is stored in a discovery registry or a catalogue of DFBs. This allows a radio node

to easily discover its nearest DFB, and allows the correct Web Services interface to be retrieved

as a Web Services Description Language (WSDL) file. The communication between

nodes and DFBs is proposed to use SOAP (originally an acronym for Simple Object Access

Protocol) messages, using XML.

In Section 3.3.1.2, the three other most important conceptual spectrum broker contributions

were reviewed. Compared to one of these, DIMSUMnet [88], the DFB defines a

full hierarchy of brokers for the management of all radio nodes, as well as the use of Web

Services for convenient communication and discovery of brokers. Seen in relation to the

Ofcom DSA Candidate Architecture [97], in addition to the same differences as for DIM-

SUMnet, the DFB is node or network lease oriented rather than call oriented. Compared

to DSAP [89], in addition to the same points as for DIMSUMnet, the DFB concept has an

ambition to manage all the use of spectrum, rather than brokering on a more local level as

does DSAP.

4.3.1.2 Paper E: A Comparison of Centralized, Peer-to-Peer and Autonomous

Dynamic Spectrum Access in a Tactical Scenario

Tore Ulversøy, Torleiv Maseng, Toan Hoang and Jørn Kårstad, ”A Comparison of Centralized,

Peer-to-Peer and Autonomous Dynamic Spectrum Access in a Tactical Scenario”,

MILCOM 2009, Boston, October 18-21, 2009

Note: This section reviews the DSA architecture comparison part of Paper E. See also

Section 4.3.2.

60


4.3. Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms

Introduction and Assumptions

The centralized, distributed and autonomous DSA architectures all have their advantages

and disadvantages. This is the case for commercial wireless communication, and even more

so for military communication where there are even more factors that need to be taken

into account. A goal of Paper E is to provide a comparison of these architectures. For

the comparison of decisions, complexity and traffic, an n-link interference model has been

used. It is further assumed that the available radio spectrum may be divided into a number

of spectrum segments over which the interference and noise are assumed to have constant

power spectrum density. Furthermore, it is assumed that link bit rates in each segment can be

calculated as the information capacity based on a modified signal-to-noise-and-interference

ratio in the segment.

Noise

TX1

Signal

+

RX1

Noise

TX2

Signal

+

RX2

FIGURE 4.16: A simplified illustration of the link-interference model, here with only two radio links.

Each of the receivers, in addition to the useful signal from the link transmitter, also sees background

noise and interference from the other link. The interference is treated as noise.

For the simulation of the Peer-to-Peer interaction algorithms, ideal 0-loss and 0-delay

information interchange between the links is assumed. In other words, the simulation model

does not include transmission delays due to the coordination messages, and also does not include

any effects of capacity limitations or transmission errors of the coordination channels.

The comparison in hostile environments is generic and does not make assumptions as to

a specific model.

Summary of Contributions

This paper compares centralized, distributed and autonomous architecture in terms of

spectrum decisions, computational complexity in the radio nodes, the need for spectrum

coordination traffic between nodes as well as in terms of the special considerations that

apply in hostile environments.

For the spectrum decisions comparison, three different autonomous policies are analyzed

and the number of operational links above a minimum bit rate and the resulting average bit

rate per link are plotted. These are compared to two different distributed algorithms that

were developed and simulated, and it is shown that by allowing administrative communication

between the links in this way, higher average bit rates per link can be achieved at a

requirement of 100% operational links, than in the autonomous policy governed cases. For

the centralized architecture, theoretically optimum spectrum decisions can be found (as re-

61


4. RESULTS AND IMPLICATIONS

ferred to the model) and hence better (or equal) decisions than than in the distributed and

autonomous architectures can be made. Due to the very high computational complexity,

however, finding these optimum decisions becomes impossible as the number of links increases.

As to the comparison of coordination traffic, the autonomous architecture per definition

manages without coordination traffic. When doing exact computations in the centralized

model, this on the other hand requires new spectrum assignments to be communicated to

a number of links whenever a change is made or a link added, which means the traffic

demand will be prohibitively high if there are frequent changes. To reduce such traffic in the

centralized architecture, spectrum decisions need to be suboptimal such that the spectrum

assignment of only one or a few links are changed at a time.

For one of the proposed distributed algorithms, and in the simulated scenario, it is shown

that the coordination traffic decays towards zero traffic. Such a convergence towards zero

coordination traffic of course assumes that the node density relative to the available spectrum

allows all nodes to eventually reach at least their minimum data rate.

As to the comparison of computational complexity, the autonomous links make their

own spectrum decisions based on their sensed spectrum readings, and hence this provides a

solution that scales fully with the number of links. In the distributed case, for the described

distributed algorithms, each node additionally will need to process the arriving messages

from other nodes.

When compared in a hostile environment, coordinated approaches will have an advantage

in their ability to do collaborative surveillance, defensive and offensive actions. The

coordination channels represent a vulnerability though, and proper protection measures are

needed. It is beneficial if DSA systems have graceful degradation, such that they can perform

as autonomous nodes if the coordination channels are no longer available.

Relative to related work of making comparisons of DSA architectures, see Section

3.3.2, Paper E bases the comparison on the link-interference model, makes the comparison

using the decision, traffic and complexity categories, and extends the discussion, compared

to most papers, by including hostile environments.

Discussion

The comparison of the architectures mainly uses the simplex interference link model,

and from time and space considerations the paper does not go into comparisons for other

types of network models. From the same considerations, the paper goes in more detail on

spectrum decisions rather than on node-to-node communication, computational complexity,

or DSA considerations for a tactical environment. The work may naturally be extended

along these lines.

The conclusions for the autonomous and distributed cases are based on computer simulations

in Matlab, with the model and idealized assumptions as discussed earlier. The

validity of this simulation model and the validity of the idealized assumptions are discussed

in Section 4.3.2, along with the general discussion on the algorithms developed in Paper E.

62


4.3. Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms

4.3.1.3 Additional Dynamic Spectrum Access Work: The Peer-to-Peer Spectrum

Broker

Introduction

From the discussion of DFB and also that in Paper E, it is clear that a disadvantage

of centralized approaches is the computational complexity of optimum or near-optimum

spectrum decisions. For a system attempting to manage large parts of the spectrum with

national or global coverage, a lot of infrastructure is required such as spectrum servers.

At the same time, since the coming generations of wireless equipment will increasingly

use SDR technology, the equipment itself will have computational power that it is tempting

to use for the spectrum decision operations. Wireless equipment will increasingly also be

connected to the Internet, which means that IP communication may be used as a standard

for spectrum decision administrative messaging.

In recent years, Peer-to-Peer (P2P) distributed systems have become increasingly popular,

and well-known for distributed data storage and file sharing purposes. In P2P systems,

the nodes in the system are equal peers instead of being clients or servers. In the third

generation P2P systems, the functionality and the distributed data are entirely in the peers

themselves, without the need for index servers or other infrastructure assistance elements.

Inspired by such third generation P2P data storage systems, an attempt has been made to

reformulate the centralized spectrum broker into a distributed P2P functionality. An initial

formulation of the P2P Spectrum Broker concept has been compiled in [128] (unpublished

work), of which a short summary is provided here. Development of a simulator for the

P2P broker was also started, but this was found to be too time consuming relative to the

remaining time available for this Ph.D. work and hence deferred to future work.

Short Description of the Concept

Radio node

Network

Device

P2P SB

agent

Radio

Application

Radio

Devices

A radio node is defined in the same way as in Paper C. The spectrum broker functionality

is distributed onto all the radio nodes, by having each radio node contain a P2P agent, see

Figure 4.17. Each agent upon activation obtains a unique address in the P2P network, in a

P2P overlay address space.

Radio Node

Network

P2P Spectrum

Radio

Radio

Device Broker Agent Application Devices

FIGURE 4.17: Each radio node contains a P2P Spectrum Broker agent. Each agent communicates

via wired Internet connection (solid green line) or wireless (red dash-dot) network connection.

A P2P command interface is defined, such that defined spectrum request and release

actions are RNsent to the other peers in the network. From each participating peer, the idea is

Network P2P SB Radio Radio

that a spectrum coordination action will appear as if the coordination is done by a central

Device agent Application Devices

entity.

The permission to use a portion of the spectrum for a given time period and in a given

region, is termed a Spectrum Object. Each Spectrum Object also holds an address in the

63


4. RESULTS AND IMPLICATIONS

overlay space, and is stored in the relevant radio node, or replicated onto more radio nodes

for redundancy.

In order for the coordination communication to avoid flooding the entire network and

causing a communication breakdown, the coordination is limited to the relevant set of peers

holding the relevant set of Spectrum Objects. Two ways of addressing these relevant sets are

described:

TABLE 4.1: An example of a multidimensional overlay address construction for the structured P2P

approach.

Longitude

minimum:

20 bits

Longitude

maximum:

20 bits

Latitude

minimum:

20 bits

Latitude

maximum:

20 bits

Spectrum

interval,

low frequency:

40 bits

Spectrum

interval,

high frequency:

40 bits

Time,

start:

32 bits

Time,

stop:

32 bits

Node

hash:

40 bits

If mainly fixed position nodes are assumed, an overlay address that includes the geographical

position may be used. An example of such a multidimensional address 3 is provided

in Table 4.1. The relevant set may in this case be addressed by a multidimensional

search [130] in the P2P overlay space.

In the second alternative, the overlay address is of a conventional distributed hash table

type. In this case, each peer maintains a lookup table with the addresses of all P2P radio node

agents that are relevant to coordinate with. Inspired from [131], the lookup table is updated

by listening to advertisments from peers announcing a successful ’claim’ of an Spectrum

Object, and storing such relevant announcements.

A main advantage of the P2P Spectrum Broker is that it may quickly be put in action, also

as add-ons to radio-systems. Another advantage is that it provides for computing workload

scalability since spectrum decision calculations may be distributed on all the radio nodes.

The P2P distributed spectrum broker concept introduces new challenges relative to

those of a centralized spectrum broker, including: Updating spectrum policies becomes

more difficult, as does also obtaining fairness in spectrum leases. The approach also puts

more strain on network communication capacity, since in general the spectrum inquiries

need to be communicated to a number of peers. Further the dynamics of appearing and

disappearing nodes need to be handled.

Summary

A benefit of a P2P Spectrum Broker system is the avoidance of the cost of a dedicated

infrastructure. Further, the sharing of the workload of spectrum decision calculations among

a number of equal-peer radio nodes creates a system that scales well in terms of workload.

Another advantage is seamless transitions between spectrum occupancy calculations in different

geographical regions.

Spectrum authority control of the spectrum usage is obviously more difficult than in the

central broker case. Further challenges are the required network communication capacity

for P2P spectrum handling and administrative traffic, and reliability concerns due to failing,

disappearing and disconnected nodes.

3 With reference to the time columns in the address, a possible 32 bit time format is the standard Unix time

format. This uses a signed 32 bit representation with a zero reference time of January 1 1970 and a resolution

of one second. The software will need to handle the wraparound that will occur in year 2038 [129].

64


4.3. Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms

The P2P broker system additionally to the peer interaction may benefit from sources of

indirect information, as explained in the next section.

A P2P distributed broker appears to be an attractive concept to pursue further. The

development of a P2P spectrum decision and communication overhead simulator, as well as

a prototype implementation, are suggested as future projects.

4.3.1.4 Sidenotes on Hybrid Architectures and Recent Developments

In the Dynamic Frequency Broker concept, the administrative communication in the system

is between the radio nodes and the broker, with radio nodes sending requests to and receiving

leases from the broker. In the P2P Spectrum Broker concept, on the other hand, the

administrative communication is directly between the peers, through direction interaction in

the form of message passing.

Making a simple parallel to interaction in human societies, the first case is equivalent

to the workers always asking the manager, or the governing authorities, for the necessary

resources and permission whenever a new task is to be carried out. Common known disadvantages

are that the manager tends to be busy or unavailable so the workers have to wait

idle now and then. While efficiency may suffer, there is system and order.

The second case is equivalent to negotiating with colleagues to get the necessary resources

or acceptances for carrying out the new task, without any intervention by management.

The disadvantages here are that there could be many people that one would need to

talk to, information could be uncertain or fuzzy and it is difficult to know if all the relevant

people have been consulted. An obvious improvement is that there are no managerial

bottlenecks, but it requires more worker skills to negotiate and interpret the coordination

information correctly, and sometimes due to the fuzzy information the coordination effort

will need to be repeated.

For general multiagent systems, and inspired by biology, the benefits of having indirect

interaction in addition to direct interaction between agents [132], has been pointed out. Here,

direct interaction is defined as ”interaction via messages, where the identifier of the recipient

is specified in a message” [132]. Indirect interaction is defined as ”interaction via persistent,

observable changes to a common environment; recipients are any agents that will observe

these changes” [132]. A commonly used biology example of indirect interaction is ant

colonies, where the ants leave pheromones on their way to find food. With ants following

a combination of pheromones and the scent of food, trails leading to good food will be

reinforced.

Translating from a general multiagent system environment to radio agents in an electromagnetic

environment, we can view the interference from the other radio links, sensed in

the autonomous architecture and in the distributed architecture in Paper E, as such indirect

interaction. The spectrum decisions in the radio links causes an aggregated interference that

is observed by each link. A property of this indirect interaction is, however, that the flow of

information has a directional character; it informs the receivers what the aggregated behavior

of the transmitters are, but the aggregated performance of the receivers (or links) are not

revealed. The resulting issues, e.g. weak links or hidden receivers, are discussed in several

places in this dissertation. Another issue is that it is difficult to reveal the spectrum priority

of the various users only through interference sensing.

In the distributed interaction algorithms in Paper E, see Section 4.3.2.2, information flow

in the other direction, from receiver (or link), has been taken care of. Here, there are broad-

65


4. RESULTS AND IMPLICATIONS

cast messages from links not reaching their minimum rates, within their geographical neighborhood,

signalling to other links to reduce their use of spectrum (and optionally power).

The algorithm in Paper E restricts itself to discussing Open Sharing scenarios, however, and

not Hierarchical Sharing or priority of certain links.

An obvious and simple way of creating an environment with persistence and observability

for indirect interaction, also taking into account receivers and weak transmission links as

well as Hierarchical Sharing, is establishing databases for spectrum use. This is the recent

path taken by the U.S. Federal Communications Commission (FCC) when allowing reuse

of the TV bands for secondary priority devices. The FCC requires that each secondary device

has geolocation capability, and that it checks with a database about which channels can

be used at its location [133]. All services entitled to protection are to be registered in this

database, as are all fixed TV band secondary devices. FCC invited private companies to

administrate such a database [133]. Google is among the nine companies who have applied

to be a designated database manager.

With the database approach, the database is the element for indirect interaction, but

the actual decisions as to which spectrum to use are taken in a distributed manner in the

individual radio agent (or associated group of such agents). The database element thus

performs an easier task than the DFB does, for example.

Returning to the P2P broker system, this also may exploit the information from the

central database, in addition to its local coordination. Another hybrid option is for P2P radio

agents to carry their redundant pieces of such a database, such that the database is distributed

in the network of nodes, and only the relevant parts of it replicated to other regions.

In a recently co-authored paper [134], an architecture that incorporates both a centralized

database as described above, as well as negotiations in a P2P network, has been sketched.

The paper also contains a proposal for a frequency resource protocol for the interactions in

this architecture.

When the protection of primary or prioritized users is included, the currently most likely

way forward for DSA architectures are those that incorporate all the three elements: Distributed

direct interaction, indirect interaction through interference sensing, as well as indirect

interaction through centralized or distributed databases.

4.3.2 Dynamic Spectrum Access Algorithms

The DSA algorithms 4 part of the third research goal is treated in Paper D and part of Paper

E.

4.3.2.1 Paper D: On Spectrum Sharing in Autonomous and Coordinated Dynamic

Spectrum Access Systems: A Case Study

Tore Ulversøy, Torleiv Maseng and Jørn Kårstad, ”On Spectrum Sharing in Autonomous

and Coordinated Dynamic Spectrum Access Systems: A Case Study”, Wireless VITAE ’09,

Ålborg, Denmark, May 17-20, 2009

4 As commented previously, the term algorithm has been used both when describing centralized calculations

as well as distributed and autonomous calculation processes. While each autonomous or distributed agent

may do its decision calculation according to an algorithm, a better term for the total process of calculating a

resulting state between a number of autonomous or distributed radio agents is interactive computation.

66


4.3. Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms

Introduction and Assumptions

In a centralized DSA architecture, all spectrum-decision relevant information about individual

nodes and links is aggregated in a central infrastructure database. This database

could be on a single central server, or a distributed database that from the perspectives of

nodes and links appears to be a central database. The centralized information makes it possible

to do an optimization in this central infrastructure element to find optimum spectrum

power density assignments. In paper D, maximum sum capacity is used as the optimization

criterion. The centrally calculated optimum is termed the Global Optimum.

A drawback of the Global Optimum is that exact solutions have prohibitively high computational

complexity. Additionally, in autonomous cases where the link condition information

is not aggregated in a central infrastructure, and assuming that only the link’s own state

and the accumulated noise plus interference are known by each link, global optimization is

not possible. Treating the accumulated noise plus interference as a constant, the local optimization

of each link may be found by the method of Lagrange multipliers. This leads to the

well known Waterfilling solution [24, 108, 109]. The spectrum power density assignment

solution achieved by running the Waterfilling iteratively (Iterative Waterfilling) for each link,

is referred to here as the Competitive Optimum [24].

The Competitive Optimum is often referred to as a computationally tractable approximation

for the Global Optimum. Paper D investigates the relation between the Competitive

Optimum and Global Optimum for illustrative link deployments, and discusses the implications

on DSA architectures.

It is assumed that the transmitters and receivers can be modeled as simplex, interfering

links. The radio spectrum is divided into spectrum segments over which the interference

and noise are assumed to have constant power spectrum density. Furthermore, it is assumed

that the link bit rates in each segment can be calculated as the theoretical information

capacity based on the modified signal-to-noise-and-interference ratio in the segment, where

the modification of the signal-to-noise-and-interference is to account for the difference

between theoretical information capacity and practically achievable bit rates.

Summary of Contributions

Based on the receiver-centric n-link interference model, and using the informationtheoretic

view, the paper compares ideal centralized power density assignments (Global

Optimum assignments), with optimum selfish autonomous power density assignments (referred

to as Competitive Optimum assignments). Two different variants of the Competitive

Optimum solution are studied, one in which optimal target bit rates are known beforehand,

such that these can be used as target rates for the Iterative Waterfilling in each individual

link. For the other variant, it is assumed that optimal rates are not known beforehand, thus

the autonomous links run with aggressive target bit rates.

A detailed case study is conducted for a numerical example with two links and two

spectrum segments, comparing the Global Optimum and Competitive Optimum solutions,

for a low interference, high interference and asymmetrical interference deployment. It is

shown that for conflicting links with high symmetric or asymmetric interference, the Competitive

Optimum solution may be significantly worse than the Global Optimum one. It

is also shown, in the asymmetric interference case, that using an a priori known optimal

Competitive Optimum target rate made the Competitive Optimum solution equal the Global

Optimum one.

67


4. RESULTS AND IMPLICATIONS

The results are related to the different DSA architectures, and suggestions for practical

policies and how to improve on the Competitive Optimum solution, are made.

Discussion

The conclusions in Paper D are based on a calculation model using information theory

and the assumptions listed previously. From a research methodology point of view, such

calculated model-based results of course do not strictly fulfill the ’measure’ step, refer to

Section 1.3.3. Thus, measurements on a real-world system would have been an improvement,

but were outside the ambitions of this work. An advantage with the simple model is

that it serves to provide a clear understanding between the starting point and the results.

A fundamental assumption for the treatment of the autonomous links in Paper D is that

each receiver observes the accumulation of the background noise and the interference from

other transmitters. The power density versus frequency is calculated in each step by the Waterfilling

solution, assuming that the background noise plus interference is constant. Each

link does not get any coordination messages from other links, and does not attempt to analyze

the measured interference to gain knowledge about other individual links.

It should be noted that uncorrelated noise levels betwen the frequency segments of different

links will tend to result in fairer Competitive Optimum solutions between different

links.

Also, if the basic assumptions are changed, better power density allocations may in some

cases be found than the pure Competitive Optimum solution. Such changes of assumptions

can be that the links are allowed to learn from their power density allocations, draw conclusions

about the influence of their power density allocations on other links, analyze the

interference to gain knowledge about other links in the vicinity or receive coordinating communication

from other links. Game theory [18, 42, 135] can be used in such cases to analyze

the problem further.

Game theory is a set of models and analytical tools for the analysis of interacting

decision-making entities. Examples of such entities are players in a poker game and the

market participants in an economic system.

Game theory has become popular in recent years in modeling interacting agents in wireless

systems, and including in the modeling of cognitive spectrum access agents.

A game has three basic components: A set of players, a set of actions and a set of

preferences [135]. The actions are the alternatives available to each player. In a Dynamic

Spectrum Access example, such actions may be transmit power levels and spectrum intervals.

The preferences represent the player’s evaluation of the action outcomes. Relationships

between preferences are typically expressed as utility functions.

One goal of game theory is to predict what will happen when the game is played [135].

One such common prediction is the Nash Equilibrium, which is an action profile at which

no player has an incentive for deviating unilaterally.

The ’high interference’ scenario from Paper D is revisited as an example. As in the paper,

two segments and two links are taken into consideration. Each of the links has four action

alternatives, either to not transmit (00), transmit in segment 1 (10), transmit in segment 2

(01) or transmit in both segments (11) with the available power shared between the two

segments. The utility function is taken as the calculated achievable data rate according to

Equation D.4. The game is illustrated in what is termed ’strategic form’ [135] in Table 4.2.

The utility function values are presented in (player 1, player 2) pairs in the table. It is seen

68


4.3. Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms

in this example that there are multiple Nash Equilibriums. Two Nash Equilibrium examples,

where it is readily seen from the table that there is no incentive for any of the players to

unilaterally deviate, are the points (10,01) and (01,10). These two points give reasonably

fair solutions between the two radio links. However, it is seen that the inefficient point

(11,11) is also a Nash Equilibrium and a possible end point of the game.

TABLE 4.2: The ’high interference’ game with two players each capable of using two frequency

segments, in strategic form. The actions of player one (see text) are in the rows, and those of player

two in the columns. The table cells are the utility function values of player one and player two

respectively. The utility function in this example is the calculated data rate according to Equation

D.4.

Player 1

Player 2

00 01 10 11

00 (0,0) (0, 87.48k) (0, 87.48k) (0,131.06k)

01 (77.04k,0) (0.43k,0.40k) (77.04k,87.48k) (0.85k,65.7k)

10 (77.04k,0) (77.04k,87.48k) (0.43k,0.40k) (0.85k,65.7k)

11 (112.14k,0) (56.29k,0.798k) (56.29k,0.798k) (0.86k,0.80k)

Playing the game repeatedly and including punishment strategies is a way of potentially

avoiding the game ending in such inefficient points. If player 1 selects action (11), player 2

may punish player 1 by also selecting (11). In the next round of the play, player 1 will then

know that playing (11) is unwise, and instead select an action that it believes it will not be

punished for, e.g. (10).

This repeated playing line of thought may be extended into longer-term strategies and is

applicable also for a general scenario with mobility and users entering and leaving. Here,

although unreasonable occupancy of spectrum resources may be individually rewarding and

not lead to punishment on a short time scale, players may take into account the longerterm

negative effects of such actions. Such ’foresighted users’ [136] are ones that seek to

maximize their rewards long-term.

4.3.2.2 Paper E: A Comparison of Centralized, Peer-to-Peer and Autonomous

Dynamic Spectrum Access in a Tactical Scenario (continued)

Tore Ulversøy, Torleiv Maseng, Toan Hoang and Jørn Kårstad, ”A Comparison of Centralized,

Peer-to-Peer and Autonomous Dynamic Spectrum Access in a Tactical Scenario”,

MILCOM 2009, Boston, October 18-21, 2009

Introduction and Assumptions

This section deals with the DSA-algorithms related part of Paper E (the architecture

comparison part was discussed in Section 4.3.1.2).

The same n-link interference model and information-theoretical view as in Paper D is

employed, but simulations with a greater number of randomly deployed links and greater

number of frequency segments are used.

As explained in Section 4.3.1.2, for simulation of Peer-to-Peer interaction algorithms,

ideal 0-loss and 0-delay information interchange between the links is assumed. That is, the

effects of latency and throughput limitations of the coordination channels are not included

in the model.

69


4. RESULTS AND IMPLICATIONS

The Ideas

Core contributions in Paper E are the distributed interactive computation algorithms.

The starting point for the computation is the solution that is reached collectively by the

system of links when they each optimize their power allocation in each segment in order to

selfishly reach their optimum data rate, the already mentioned Iterative Waterfilling solution.

The ’Waterfilling’ term refer to the nature of the solution, a ’filling level’ which the transmission

power, added to the noise and interference translated to the transmitter side, should

be filled up to.

The possible problems with these power density allocations, when not having any feedback

in the system of links, are discussed both in Paper D and Paper E. In short, some of the

links have their performances destroyed by interference from other links, without being able

to improve this situation by changing their own power allocation. The broad nature of the Iterative

Waterfilling allocation, with power typically allocated in many segments, contributes

to making this problem worse.

The links that do not achieve their minimum data rate within their maximum transmit

power limitation, need access to more lower-interference spectrum in order to manage to

increase their performance. The idea behind the D1 interactive algorithm in the paper is to

create more room for the poor-performing links by having the other links not use as many

segments as they would normally do in the Iterative Waterfilling solution, but rather limit

themselves to using fewer segments, in a dynamic manner and as needed.

This principle is sketched in Figure 4.18, where link 2 does not reach its minimum

data rate and therefore, within a coordination distance, broadcasts a reduce message. If not

already at maximum, link 2 also increases its own allowed number of segments. If link 1

determines it is a significant contributor to the interference seen at the poor-performance

link (further details in the paper), link 1 reduces the maximum number of segments it can

use in the Iterative Waterfilling by 1.

In the second interactive computation algorithm D2, more room for poor-performing

links is created in the same way, but by both reducing segments and also target rate of the

interfering links.

Summary of Contributions

In the paper, three different autonomous policies are defined and analyzed, and the number

of operational links above a minimum bit rate and the resulting average bit rate per link

are plotted. The first of these policies sets a uniform target bit rate for the links, the second

sets a maximum number of frequency segments that the links are allowed to use, and

the third is one where the power is reduced proportionally to the maximum number of frequency

segments. The one with a limitation on the number of frequency segments is shown

to be the best performing.

Two interactive algorithms where the links on an on-demand basis, as explained in the

section above, have administrative communication, are developed and simulated. It is concluded

that by allowing administrative communication between the links in this way, higher

average bit rates per link can be achieved at a requirement of 100% operational links, than

in the autonomous policy governed cases.

For the centralized case, it is suggested to run the same distributed algorithms, but in the

centralized server instead.

Particularly considering their simplicity, the algorithms may be attractive candidates for

implementation.

70


4.3. Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms

a)

Link 2

Link 1

Reduce

P. Dens.

segment

P.Dens.

segment

b)

Link 2

Link 1

P. Dens.

PDens P. Dens.

segment

segment

FIGURE 4.18: Principle for D1 distributed computation: Poor-performing links request other links

within a coordination radius to reduce the number of spectrum segments used in the Iterative Waterfilling.

Discussion Concerning Computational Scalability

In this section the scalability of the D1/D2 algorithms is discussed, from a computational

complexity viewpoint of the required calculations.

In a Dynamic Spectrum Access system implementing the D1/D2 algorithms, a high-level

breakdown of the required DSA radio client functionality is illustrated in Figure 4.19. The

D1/D2 functionality is illustrated inside the rectangle, and the interfacing modules outside

the rectangle. Here, the Send block sends the reduce messages when the data rate is too

low. The Receive and process block continuously monitors for incoming ’reduce’ messages,

processes these and once per waterfilling iteration feeds the resulting maximum segment

number (and in case of D2, target rate) to the Waterfilling block. The Waterfilling block at

this iteration rate uses the sensed spectrum data and segment limits (and in case of D2, target

rate) and runs the waterfilling calculation, then feeds the transceiver with the new settings.

With respect to the computational complexity of the D1/D2 DSA client, the dominant

parts in terms of peak workload will be the Waterfilling and the Receive and process. The

Send block will only occasionally send messages (a maximum of one per iteration).

The Waterfilling has a computational complexity dependency on the number of segments,

M, as well as on the maximum allowed number of segments m i . It does not depend

on the number of links N, however, as each link does is own computation.

Figure 4.20 presents measurements of computation time per iteration as a function of

the number of segments (M) for a Waterfilling test implementation. The test implementation

71


4. RESULTS AND IMPLICATIONS

D1/D2:

Send

Receive and

process

Coord.

channel

inter-

face

Spectrum

sensing

m i

R target

Waterfilling

TX & RX

FIGURE 4.19: A breakdown of the required blocks in a radio client implementation of D1/D2 DSA,

together with interfacing blocks.

is run as a Matlab program on a computer with a P8600 2.4GHz processor. The maximum

allowed number of used segments in the Waterfilling is set to m i = M/2. The computation

time is seen to approximately scale linearly for small M and to the second order for large M.

Hence, from a computational point of view, the Waterfilling part scales well to large numbers

of segments.

Comp. time per waterfilling‐iteration [msec]

20

18

16

14

12

10

8

6

4

2

0

Computation Time per Waterfilling‐Iteration Versus Number of

Segments

y = 3,1E‐06x 2 + 1,4E‐02x + 3,9E‐02

0 200 400 600 800 1000 1200

M

Computation time per waterfillingiteration

[msec]

Second order polynomial

l

approximation

FIGURE 4.20: Measurements of the computation time of one iteration of a test implementation of

the Waterfilling block, as a function of the number of spectrum segments, M. The maximum allowed

number of used segments in the algorithm is set to m i = M/2 for all the measurements. Secondorder

fitted line also shown.

The Receive and process block receives and processes the messages from other links,

and thus has a computational complexity dependency on the number of received messages.

For each message, a calculation of the interference to the link that sent the reduce message,

is needed. Assuming only one permitted message per Iterative Waterfilling interval per link,

72


4.3. Research Goal 3, Dynamic Spectrum Access Architectures and Algorithms

the computational complexity worst-case scales linearly with the number of links within the

coordination radius. Typically, though, the number of messages that needs to be handled

will be much less than the number of links within the coordination radius. From both these

arguments, this computation will scale well to large numbers of links.

Discussion Concerning Coordination Messages

The infinite capacity assumption of the coordination channels, made in the simulations,

is likely to be a good approximation in cases with low link densities, where few coordination

messages are necessary.

With high link densities, the effects of the volume of coordination messages exceeding

the capacity of the coordination channels is not accounted for in the simulations, and in a

real system under these conditions may cause messages to get lost and links to not achieve

their minimum data rates. Several parameters can be adjusted to minimize the probability of

such loss of coordination messages: An obvious one is the dimensioning of the capacity of

the coordination channel, as is also the density of links sharing the spectrum band and the

value of the minimum link data rate. By applying the maximum number of segments policy

(A2 in Paper E) at the same time as running the distributed algorithm, this will cause fewer

cases where coordination messages are needed.

Coordination messages are assumed to have zero delay in the model. With slow rates

of recalculation of spectrum assignments, say recalculation every minute, actual message

delays will be small relative to the recalculation period and the effects of real message delays

are likely to be negligible.

For fast recalculation rates in a real system, there is a risk that coordination messages

caused by the last recalculation have not arrived at the time of the next recalculation. Such

delayed messaging under conditions of frequent spectrum assignment recalculation will

need further study. A conservative approach will be to maintain a recalculation period

which is well above the maximum coordination message propagation delay.

Further Discussion Points and Recommendations for Further Work

The results and comparisons of the algorithms are based on Matlab computer simulations,

using the model and theoretical assumptions as listed above. Simulation does not

suffice as scientific evidence that an algorithm works in a real-world environment, but it

may provide support for the design and suggest that further investigations in real-world

environments are recommendable. Hence a further recommended step is prototyping and

measurements in real environments. Such prototyping was, however, outside the scope of

the project in terms of available time and equipment. The results from the simulations indicate

optimism, however, that the algorithms, when adapted to practical systems, will work

in a real environment.

For the centralized case, the paper suggests using the distributed algorithm, running

internally in the spectrum broker. A suggestion for further work on the centralized case

is to apply the autonomous case Competitive Optimum and/or the result from running the

distributed algorithm in the broker as starting points for further optimization using a genetic

algorithm. Genetic algorithms are inspired by biological evolution and use elements such as

inheritance, crossover, selection and mutation.

73


Chapter 5

Conclusions and Recommendations for

Further Work

This chapter first revisits the research goals, followed by a review of the major contributions

of this thesis. Some critical remarks are then discussed, and finally ideas for further research

are provided.

5.1 Revisiting the Research Goals

The overall motivation for the work presented in this thesis has been to contribute to the

evolution of radio systems from fixed waveform functionality + static frequency use to ones

with loadable waveform functionality + dynamic spectrum access. Three main areas were

focused on, and three research goals formulated:

• To assess the degree to which the conceptual challenges of Software Defined Radio

have been solved and what challenges still remain in the fields of software architecture,

computational requirements, security, regulations and business structure.

• To analyze the workload overhead of Software Communications Architecture-based

Software Defined Radio in a specific processing configuration.

• To contribute to establishing practical ways of dynamically sharing the frequency

spectrum and in particular compare autonomous, distributed and centralized spectrum

sharing architectures.

5.2 Major Contributions

The contributions presented in this thesis have been published in five research papers, that

are included in Part II of the thesis. Some further follow-up work has been included in Part

I of the thesis. The work in the thesis has also provided part input to two other co-authored

research papers, as well as to two (multi-authored) NATO Research Technical Organization

reports, one within Software Defined Radio and the other on Cognitive Radio, as listed under

’Other Co-authored Publications’.

75


5. CONCLUSIONS AND RECOMMENDATIONS FOR FURTHER WORK

5.2.1 Conceptual Challenges of Software Defined Radio

This work is presented in Paper A, which is published in IEEE Communications Surveys

& Tutorials, and is available for open access download. The publication is the only SDR

tutorial in this journal, and may serve as a broad introduction to SDR and the remaining

challenges of SDR, for engineers and scientists in this field of work.

These are a subset of the conclusions made in the paper:

Within SDR software architecture, the paper points to the still remaining challenges

in the application portability of Software Communications Architecture-based applications,

with the major issue being that of the standardization of application interfaces between the

waveform application and the platform services and devices, and another issue being the

different approaches to the communication with special processors. The paper also discusses

alternative architectures to SCA, as well as alternative middleware to CORBA.

The paper points to the size, weight and power consumption challenges of providing

sufficient computational processing capacity for future handheld multi-waveform terminals.

Some recent multiple-SIMD processors are pointed to as particularly promising for handheld

devices.

While security of SDR systems is a well-studied subject, it is still one which will require

attention from security organizations and engineers in the years to come, particularly for

military systems. Standardization of security architectures, or at least the application interfaces

for them, in order to also have portability for the security related code, are challenging

issues.

In terms of regulatory certification of SDR equipment, changes have already taken place,

but the paper points to further evolvement of certification rules that are needed.

In the near term, the largest potential for SDR systems are seen, in addition to the military

radio communication domain where there are ongoing programs, in the cellular infrastructure

market, where the rapid upgrades of standards and the fewer restrictions on power, size

and weight point in favour of SDR. Satellite communications is another area which appears

particularly promising for SDR.

A change in business structure is predicted, into SDR platform providers and third-party

software providers, which may lower the threshold for companies entering the SDR market

as software providers, and hence provide further competition in the SDR software applications

area.

5.2.2 Workload Overhead of SCA-based Software Defined Radio

This work is presented in Paper B in Part II, and which is available for public download

from the Nato Research Technical Organization. Additional follow-up work is presented in

Section 4.2.2 in Part I.

The paper investigates the workload effects of varying the granularity of an SCA-based

application, when the application components are deployed in the default manner as independent

components on a CORBA-enabled GPP. Workload measurements are carried out

using two different statistical profilers. Two different analytical models are also formulated,

where the first one does not include context switches but the second one does. When comparing

the results from the first model with the measurements, it was found that this includes

the major part of the workload overhead. It underestimates the total workload of the system

though, and more so the larger the packets are. This is natural since it does not include the

76


5.2. Major Contributions

context switch losses. The larger the packet size, the higher the indirect cost related to the

context switch losses due to cache sharing between processes.

The paper demonstrates that the application granularity, as well as the frequency of the

inter-component communication needs to be taken into account when optimizing a GPPbased

SCA application. In the review of related research it is pointed out that the literature

on workload overhead in SCA is sparse, and that the scope of this analysis provides a useful

contribution to the research literature on the subject. The results are beneficial as guidelines

on how to minimize workload overhead in SCA-based applications. In the follow-up work,

additional ways of optimizing SCA-based applications are discussed, both for GPP environments

and for heterogeneous environments. This includes changing transports, such as

from TCP/IP to a Unix transport, and deploying components in a common address space in

a ResourceFactory and using an ORB that is able to exploit these opportunities.

5.2.3 Dynamic Sharing of Spectrum

This work is presented in Papers C through E in Part II, which are all available through IEEE

Xplore, as well as in Sections 4.3.1.3 and 4.3.1.4 of Part I.

5.2.3.1 The Dynamic Frequency Broker

A centralized approach to DSA, the Dynamic Frequency Broker, is presented in Paper C. The

idea here is a hierarchy of frequency brokers, where the lower level ones are responsible for

a geographic region and provide leases to radio nodes. Here, each DFB has a database of

radio nodes and spectrum assignments as well as a propagation calculation model based on

a topographical map. The DFBs are proposed to communicate with each other and the radio

nodes using Web Services, thus the nearest DFB may be found from a service registry. The

Dynamic Frequency Broker proposal complements other prior work on spectrum brokers.

Related to the DFB, Section 4.3.1.3 in Part I provides a conceptual proposal for a P2P

broker, where the broker functionality is distributed among the peers instead of requiring

central infrastructure.

5.2.3.2 Comparison of Dynamic Spectrum Access Architectural Concepts

The comparison of DSA architectural concepts is presented in Paper E. The comparison is

done along the following categories: Spectrum decisions, computational complexity, coordination

traffic and considerations for hostile environments. For the first three categories,

the comparison uses a link interference model with assumptions as described in the next

section.

The analysis shows that the DSA architectural concepts all have advantages and disadvantages,

but that much is in favor of an architecture that has distributed (decentralized)

interaction between the radio nodes. Vulnerability and security issues exist for the coordination

channels that are needed for such interaction to take place, and these channels must

be protected. Also, it is favorable if systems can degrade gracefully from coordinated ones

to autonomous ones when the coordination channels are for some reason blocked.

In Section 4.3.1.4 further remarks are attached to the coordination communication in

the distributed architecture, pointing to the preferability of having combinations of local

77


5. CONCLUSIONS AND RECOMMENDATIONS FOR FURTHER WORK

information exchange and indirect information such as in the form of aggregated information

from spectrum databases as well as sensed spectrum occupancy.

5.2.3.3 Dynamic Spectrum Access Algorithms based on Iterative Waterfilling

The work on DSA algorithms is based on a link interference model. The bandwidth B is

divided into M segments within which both the noise and interference are assumed to have

constant power densities. The algorithms and policies discussed are all modifications and

additions to the Iterative Waterfilling algorithm.

In Paper D, comparisons are made between the Competitive Optimum Iterative Waterfilling

solution, both with aggressive target rates and optimum precalculated target rates,

and the Global Optimum solution, which is the optimum centrally calculated solution. The

comparisons are made in three different typical deployments. It is shown that the Competitive

Optimum solution may deviate substantially from the Global Optimum in certain

deployments. Suggestions about how to improve from the Competitive Optimum solution

are discussed.

In Paper E, three autonomous policies and two distributed interaction DSA algorithms

are suggested and simulated for a simplex link interference model and under idealized conditions.

The simulations show that the distributed interaction algorithms provide higher average

bit rates over all the links compared to the autonomous policy ones, while at the same

time wanting to assure that each link obtains at least a minimum data rate. The policies and

algorithms may be useful as candidates for implementation in DSA systems.

5.3 Critical Remarks

5.3.1 Conceptual Challenges of Software Defined Radio

For completeness of the review of remaining SDR challenges, it would have been advantageous

if Paper A had also covered analog hardware and front-end issues. Due to the extent of

this topic, however, this was not included in the research goal. The interested reader should

consult some of the references provided in Paper A and in Section 3.1.

5.3.2 Workload Overhead of SCA-based Software Defined Radio

An improvement of this work would have been a better verification of modeled versus measured

data for the second derived model in Paper B, as only course estimates for the second

model were carried out, partly due to time limitations and partly due to some software tools

not being compatible with the operating system kernel on the particular computer platform.

The work could also have been improved by having more consistency in the presentation of

the results in relation to the accuracy of the measured data. There are several ways the work

can be further extended, as outlined further below.

5.3.3 Dynamic Sharing of Spectrum

The results of this research goal are based largely on numerical calculations and simulations,

as well as conceptual discussions referencing supporting literature. While simulation results

can provide support for the hypotheses that the simulation model is built on, and also be a

78


5.4. Recommendations for Further Work

way of finding new hypotheses, simulations of course cannot be considered as scientific evidence.

Thus, going back to and as commented on in the relevant methodology section 1.3.3,

the simulation results and the analysis of those are not scientifically sufficient to check off

the ’measure and analyze’ steps in the engineering paradigm methodology, real-world measurements

on realistic systems are needed. However, the resources needed to build such

experimental systems, in terms of available time and equipment, exceeded what was available,

and such implementation was deferred to future work. An intermediary level, between

the idealized model and the real-world systems, would be finer scale simulations, with realistic

waveforms and taking into account limited capacity and finite latency coordination

channels.

Related to the issue of real-world measurements, it should be added that there are some

voices in the DSA field of research that argue that one of the largest weaknesses in general

with current DSA research, is that too few realistic size DSA systems are being built and

tested.

5.4 Recommendations for Further Work

5.4.1 Workload Overhead of SCA-based Software Defined Radio

There are several possible extensions to the work on workload overhead of SCA-based SDR:

As mentioned above, further work is desireable on the model that includes the context

switching, both in terms of establishing methodology for estimating the parameters in the

model, and also in order to have further verification of the model.

The results in Paper B show that significant workload overhead is caused by mere data

copying and formatting, which are very parallel data operations. It is therefore very likely

that part of this workload overhead could have been eliminated if these operations were

carried out by dedicated hardware functionality. Investigation of such hardware acceleration

of these operations is suggested.

One of the most interesting activities in the area of SCA-based development, at the time

of writing, is the ongoing work on SCA Next, i.e. the next revision of the SCA specification.

One of the goals of this coming revision of SCA is to make it middleware agnostic, such

that other middleware can be used instead of CORBA. This makes research into alternative

middleware of great interest.

The measurements in Paper B were conducted on a single-core processor, while multicore

is now the standard. The topic of optimal reduction of workload overhead on multicore

processors may be worth studying.

A further natural extension is towards workload optimization in a general heterogeneous

multiprocessor environment, including optimization of modules for special processors such

as FPGAs and DSPs, and investigations into suitable middleware for such processors.

5.4.2 Dynamic Spectrum Access Architectures and Algorithms

5.4.2.1 Experimental System

From previous discussions, it is beneficial to carry out further verification measurements of

the algorithms in Paper E on real prototype systems, which is suggested as further work in

DSA. The prototype systems need to have coordination channels such that the coordination

79


5. CONCLUSIONS AND RECOMMENDATIONS FOR FURTHER WORK

messages of the distributed algorithms may flow in the system. Such prototype systems

should have a payload waveform that allows the segmented frequency model to be used.

One candidate waveform which conforms well to this model is Non-Contiguous Orthogonal

Frequency Division Multiplexing (NC-OFDM), where the individual carriers may be

turned off, coded with different modulations and excited with different power levels. An

advantage with NC-OFDM and OFDM in general is a very efficient implementation. At

a penalty of a higher implementation complexity, several narrowband waveform channels

with individually varying power, coding and modulation may be used instead.

5.4.2.2 Evolutionary Computation

In the discussion in Section 4.3.1.2 it is suggested to use the autonomous case Competitive

Optimum and/or the result from running the proposed distributed algorithm in the broker as

seeds for a genetic algorithm in the broker. A genetic algorithm is a heuristic search type algorithm

that works through evolving populations of chromosomes, where the operations on

the chromosomes typically are selection, mutation, inheritance and crossover. The genetic

algorithm is one form of evolutionary algorithms, which again is a subset of evolutionary

computation. Evolutionary computation comprises computational processes where there is

iterative progress, such as a refinement of a population over time, and is typically inspired

by biological processes.

Comparisons of seeded genetic algorithms versus non-seeded ones, in terms of the number

of generations to reach the same average data rate, is suggested as further work in the

centralized broker case.

In the autonomous and distributed cases in Paper D and E, the distributed computation is

based on convex optimization principles. When analyzing more complex cases, optimization

problems often have a non-convex character, however. Also, it is often beneficial to include

several objectives in the optimization, such as e.g. data rate, delay, transmission power

and spectrum occupancy. In these cases evolutionary computation may be attractive as an

alternative approach also in the autonomous and distributed cases.

The approach of using evolutionary computation in the form of a genetic algorithm as

decision- and learning-aid in decentralized radio agents or Cognitive Radios has been pursued

for some time [137], and recent publications [138, 139] has demonstrated that it is a

promising candidate technology for Cognitive Radios. A problem with evolutionary computation

in portable devices, however, is that the algorithms are computationally demanding

and hence may be unacceptably slow [137], requiring focus both on careful selection of

algorithms and search spaces as well as on platform implementation issues.

For the autonomous and distributed cases, suggested research areas are the reduction

of computation time and parallel implementations of evolutionary algorithms on radio platforms,

the seeding of the algorithms with populations believed to be close to the desired

working points [137], system convergence and stepwise refinement of the computations.

5.4.2.3 Other Suggestions

Several other suggestions for expansion of this work have been provided in the preceding

chapters:

The concept a distributed broker using a P2P overlay addressing layer, is interesting to

pursue with a simulation model and/or prototype implementation.

80


5.4. Recommendations for Further Work

There are several interesting areas related to the distributed interaction algorithm in Paper

E that deserves further study, e.g. a detailed study on the effects of message propagation

delays and message congestion in limited-capacity coordination channels. There are many

directions in which the work in Paper E can be expanded. One example is towards models

other than the link interference one. Another example is an expansion to include Hierarchical

Sharing and taking into account prioritized communication, e.g. by calculating a maximum

power density restriction for each of the frequency segments, based on database information.

81


References

[1] D. Cichon and W. Wiesbeck, “The Heinrich Hertz wireless experiments at Karlsruhe

in the view of modern communication,” in International Conference on 100 Years of

Radio, 1995, pp. 1 –6, Sep. 1995.

[2] W. Brinkman, D. Haggan, and W. Troutman, “A history of the invention of the transistor

and where it will lead us,” IEEE Journal of Solid-State Circuits, vol. 32, no. 12,

pp. 1858 –1865, Dec. 1997.

[3] J. Kilby, “The integrated circuit’s early history,” Proceedings of the IEEE, vol. 88,

no. 1, pp. 109 –111, Jan. 2000.

[4] G. E. Moore, “Cramming more components onto integrated circuits,” Electronics,

vol. 38, no. 8, Apr. 1965.

[5] E. Lee, “Programmable DSPs: A brief overview,” IEEE Micro, vol. 10, no. 5, pp. 14

–16, Oct. 1990.

[6] M. Trope, “FPGAs: Past, Present, Future,” Department of Electrical Engineering and

Computer Science, University of Kansas, Research paper, May 2004.

[7] F. Faggin, “The making of the first microprocessor,” IEEE Solid-State Circuits Magazine,

vol. 1, no. 1, pp. 8 –21, 2009.

[8] J. Mitola III, “Software radios - survey, critical evaluation and future directions,” in

National Telesystems Conference, 1992. NTC-92, pp. 13/15–13/23, May 1992.

[9] J. Mitola III. HomePage of Joseph Mitola III, KTH Teleinformatics. (June 2007).

[Online]. Available: http://web.it.kth.se/ ∼ maguire/jmitola/

[10] B. Tavli and W. Heinzelman, Mobile Ad Hoc Networks. Springer Netherlands, 2006,

pp. 31–46.

[11] C. Berrou, A. Glavieux, and P. Thitimajshima, “Near Shannon limit error-correcting

coding and decoding: Turbo-codes. 1,” in IEEE International Conference on Communications,

1993. ICC 93. Geneva. Technical Program, Conference Record, vol. 2,

pp. 1064 –1070 vol.2, May 1993.

[12] A. Kaye and D. George, “Transmission of multiplexed PAM signals over multiple

channel and diversity systems,” IEEE Transactions on Communication Technology,

vol. 18, no. 5, pp. 520 –526, Oct. 1970.

83


REFERENCES

[13] R. Kirby, “History and trends in international radio regulation,” in International Conference

on 100 Years of Radio, 1995, pp. 141 –146, Sep. 1995.

[14] D. W. Webbink, “Frequency spectrum deregulation alternatives,” Oct. 1980, FCC

working paper.

[15] J. Mitola III and G. Q. Maguire, Jr., “Cognitive radio: Making software radios more

personal,” IEEE Personal Communications, vol. 6, no. 4, pp. 13–18, Aug. 1999.

[16] A. Kaul, “Software defined radio: The transition from defense to commercial markets,”

in 2007 Software Defined Radio Technical Conference and Product Exposition,

Nov. 2007.

[17] FCC Spectrum Policy Task Force, “Report of the spectrum efficiency working group,”

Tech. Rep., Nov. 2002.

[18] L. Berlemann and S. Mangold, Cognitive Radio and Dynamic Spectrum Access. Wiley,

2009.

[19] D. P. Johnson, “Dismounted Urban Tactical Communications Assessment / Urban

Spectrum Management,” in RTO-MP-IST-083 Military Communications with a Special

Focus on Tactical Communications for Network Centric Operations, Apr. 2008.

[Online]. Available: ftp://ftp.rta.nato.int/PubFullText/RTO/MP/RTO-MP-IST-083/

Supporting%20Documents/MP-IST-083-10.pps

[20] P. Marshall, “DARPA progress towards affordable, dense, and content focused tactical

edge networks,” in Military Communications Conference, 2008. MILCOM 2008., pp.

1 –7. IEEE, Nov. 2008.

[21] W. Tuttlebee, Software Defined Radio: Enabling Technologies. Chichester: Wiley,

2002.

[22] United States Government Accountability Office, “Restructured JTRS Program Reduces

Risk, but Significant Challenges Remain,” Sep. 2006.

[23] J. Mitola III, “Cognitive Radio An Integrated Agent Architecture for Software Defined

Radio,” Doctoral dissertation, Royal Institute of Technology (KTH), SE-164 40

Kista, Sweden, May 2000.

[24] W. Yu, “Competition and Cooperation in Multi-User Communication Environments,”

Ph.D. dissertation, Stanford University, June 2002.

[25] IEEE Xplore Digital Library. [Online]. Available: http://ieeexplore.ieee.org/

[26] SpringerLink. [Online]. Available: http://www.springerlink.com

[27] The ACM Digital Library. [Online]. Available: http://portal.acm.org/dl.cfm

[28] Google scholar beta. [Online]. Available: http://scholar.google.no/

[29] Google. [Online]. Available: http://www.google.no/

84


References

[30] W. Tuttlebee, Software Defined Radio: Origins, Drivers and International Perspectives.

Chichester: Wiley, 2002.

[31] ——, Software defined radio: Baseband Technologies for 3G Handsets and Basestations.

Hoboken, N.J.: Wiley, 2004.

[32] P. B. Kenington, RF and Baseband Techniques for Software Defined Radio. Boston:

Artech House, 2005.

[33] S. Singh, M. Adrat, M. Antweiler, T. Ulversøy, T. M. O. Mjelde, L. Hanssen, H. Özer,

and A. Zümbül, “NATO RTO/IST RTG on SDR: Acquiring and sharing knowledge

for developing SCA-based waveforms on SDRs.” Information Systems Technology

Panel Symposium IST-092/RSY-022 on Military Communications and Networks,

Sep. 2010.

[34] RTG on SDR, “NATO RTO/IST RTG on SDR Final Report,” Tech. Rep., Dec. 2010,

submitted to NATO RTO/IST.

[35] NATO Military Agency For Standardization, “Stanag 4285 (edition 1),” Feb. 1989.

[36] VirginiaTech. OSSIE development site for software-defined radio. (Dec. 2007).

[Online]. Available: http://ossie.wireless.vt.edu/trac

[37] V. S. W. Eide, “Exploiting event-based communication for real-time distributed and

parallel video content analysis,” Ph.D. dissertation, University of Oslo, June 2005.

[38] P. J. Denning, D. E. Comer, D. Gries, M. C. Mulder, A. Tucker, A. J. Turner, and P. R.

Young, “Computing as a discipline,” Communications of the ACM, vol. 32, no. 1, pp.

9–23, 1989.

[39] R. L. Glass, “A structure-based critique of contemporary computing research,”

Journal of Systems and Software, vol. 28, no. 1, pp. 3 – 7, 1995. [Online].

Available: http://www.sciencedirect.com/science/article/B6V0N-3YGV297-K/

2/8e9216c77711e3d85d79fd235b399cf1

[40] V. Ramesh, R. L. Glass, and I. Vessey, “Research in computer science: An

empirical study,” Journal of Systems and Software, vol. 70, no. 1-2, pp. 165

– 176, 2004. [Online]. Available: http://www.sciencedirect.com/science/article/

B6V0N-49N06GS-K/2/d914c450d030acabbdb8ef9e219ef63e

[41] P. J. Fortier and H. E. Michel, Computer Systems Performance Evaluation and Prediction.

Amsterdam: Digital Press, 2003.

[42] B. A. Fette, Cognitive Radio Technology. Amsterdam: Newnes/Elsevier, 2006.

[43] Wireless Innovation Forum. Wireless Innovation Forum. (2010). [Online]. Available:

http://www.wirelessinnovation.org/

[44] Federal Communications Commission, “In the Matter of Facilitating Opportunities

for Flexible, Efficient, and Reliable Spectrum Use Employing Cognitive Radio Technologies,

Report and Order,” Mar. 2005, FCC 05-57, ET Docket No. 03-108.

85


REFERENCES

[45] Texas Instruments. Single-chip digital signal processor (DSP) announced.

(Nov. 2009). [Online]. Available: http://www.ti.com/corp/docs/company/history/

interactivetimeline.shtml

[46] B. Davies and T. Davies, “The application of packet switching techniques to Combat

Net Radio,” Proceedings of the IEEE, vol. 75, no. 1, pp. 43–55, Jan. 1987.

[47] Analog Devices. A/D Converters. (Dec. 2009). [Online]. Available: http://www.

analog.com/en/analog-to-digital-converters/ad-converters/products/index.html

[48] Texas Instruments. (Dec. 2009). [Online]. Available: http://www.ti.com/

[49] A. A. Abidi, “The path to the software-defined radio receiver,” IEEE Journal of Solid-

State Circuits, vol. 42, no. 5, pp. 954–966, May 2007.

[50] A. Gatherer, “Market and technology drivers for cellular infrastructure modem development,”

Wireless Personal Communications, vol. 38, no. 1, pp. 43–53, June 2006.

[51] Joint Program Executive Office Joint Tactical Radio System, “JPEO JTRS Announces

’Business Model’ for Certifying JTRS Radios,” Jan. 2007.

[52] Defence Update. Harris, Thales Compete Multi-Billion JTRS Radio Procurement.

(June 2007). [Online]. Available: http://www.defense-update.com/newscast/0607/

news/240607 jtrs.htm

[53] Vecima Networks Inc. Spectrum Signal Processing. (2009). [Online]. Available:

http://www.spectrumsignal.com

[54] Lyrtech Inc. (2009). [Online]. Available: http://www.lyrtech.com

[55] Vanu Inc. Where SOFTWARE Meets the Spectrum. (2010). [Online]. Available:

http://www.vanu.com

[56] Huawei. Huawei Industry First with 3G/2G Software Defined Radio (SDR) Single

RAN Product Launch. (Sep. 2008). [Online]. Available: http://www.huawei.com/

news/view.do?id=5587&cid=42

[57] International Telecommunication Union, “Radio Regulations, Edition of 1998,” Tech.

Rep., 1998.

[58] Norwegian Post and Telecommunications Authority. Post- og teletilsynet. (Oct.

2006). [Online]. Available: http:///www.npt.no

[59] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “Next generation/dynamic

spectrum access/cognitive radio wireless networks: A survey,” Computer Networks,

vol. 50, no. 13, pp. 2127–2159, May 2006.

[60] Q. Zhao and B. Sadler, “A survey of dynamic spectrum access,” IEEE Signal Processing

Magazine, vol. 24, no. 3, pp. 79–89, May 2007.

[61] J. Mitola, “Software radio architecture: A mathematical perspective,” IEEE Journal

on Selected Areas in Communications, vol. 17, no. 4, pp. 514–538, Apr. 1999.

[Online]. Available: ISI:000080068400002

86


References

[62] F. K. Jondral, “Software-defined radio – basics and evolution to cognitive radio,”

EURASIP Journal on Wireless Communications and Networking, vol. 2005, no. 3,

pp. 275–283, Apr. 2005.

[63] A. Shamsizadeh, “A survey of security issues in software defined radio and cognitive

radio,” Isfahan University of Technology, 2009, Graduate student research paper.

[Online]. Available: http://omidi.iut.ac.ir/SDR/2009/88 Ali%20Shamsizadeh SDR

Security%20in%20SDR%20and%20CR/index.htm

[64] J. Giacomoni and D. C. Sicker, “Difficulties in providing certification and assurance

for software defined radios,” in First IEEE International Symposium on New Frontiers

in Dynamic Spectrum Access Networks, 2005. DySPAN 2005., pp. 526–538, Nov.

2005.

[65] W. Lehr, F. Merino, and S. E. Gillett, “Software radio: Implications for wireless

services, industry structure, and public policy,” Massachusetts Institute of Technology

Program on Internet and Telecoms Convergence (http://itc.mit.edu), 2002.

[66] C. R. A. González, C. B. Dietrich, and J. H. Reed, “Understanding the software communications

architecture,” IEEE Communications Magazine, vol. 47, no. 9, pp. 50–

57, Sep. 2009.

[67] Y. Zhang, S. Dyer, and N. Bulat, “Strategies and insights into SCA-compliant waveform

application development,” in Military Communications Conference, 2006. MIL-

COM 2006., pp. 1 –7. IEEE, Oct. 2006.

[68] M. Cummings and T. Cooklev, “Tutorial: Software-defined radio technology,” in 25th

International Conference on Computer Design, 2007. ICCD 2007., pp. 103 –104, Oct.

2007.

[69] A. Tribble, “The software defined radio: Fact and fiction,” in 2008 IEEE Radio and

Wireless Symposium, pp. 5 –8, Jan. 2008.

[70] D. Pearson, “SDR (systems defined radio): How do we get there from here?” in

Military Communications Conference, 2001. MILCOM 2001. Communications for

Network-Centric Operations: Creating the Information Force., vol. 1, pp. 571 – 575

vol.1. IEEE, Oct. 2001.

[71] C. Svensson and S. Andersson, “Software defined radio - visions, challenges and

solutions,” in Radio Design in Nanometer Technologies, M. Ismail and D. R.

de Llera González, Eds. Springer, 2006, ch. 3.

[72] A. Rusu and M. Ismail, “Analog-to-digital conversion technologies for software defined

radios,” in Radio Design in Nanometer Technologies, M. Ismail and D. R.

de Llera González, Eds. Springer, 2006, ch. 6.

[73] A. Haghighat, “A review on essentials and technical challenges of software defined

radio,” in Military Communications Conference, 2002. MILCOM 2002., vol. 1, pp.

377 – 382. IEEE, Oct. 2002.

87


REFERENCES

[74] S. Bernier, C. Auger, J. Zapata, H. Latour, and M. Michaud-Rancourt, “SCA advanced

features - optimizing boot time, memory usage, and middleware communications,”

in SDR ’09 Technical Conference and Product Exposition, Dec. 2009.

[75] P. Balister, M. Robert, and J. Reed, “Impact of the use of CORBA for inter-component

communication in SCA based radio,” in SDR ’06 Technical Conference and Product

Exposition, Nov. 2006.

[76] S. Singh, M. Adrat, S. Couturier, M. Antweiler, M. Phisel, and S. Bernier, “SCA

based implementation of STANAG 4285 in a joint effort under the NATO RTO/IST

panel,” in SDR ’08 Technical Conference and Product Exposition, Oct. 2008.

[77] J. O. Neset, “Software defined radio - evaluation of software communications architecture

components using open source SCA implementation - embedded,” NTNU -

Norwegian University of Science and Technology, Student report, Dec. 2007.

[78] Valgrind Developers. Valgrind. [Online]. Available: http://valgrind.org/

[79] G. Abgrall, F. Le Roy, J.-P. Delahaye, J.-F. Diguet, and G. Gogniat, “A comparative

study of two software defined radio platforms,” in SDR ’08 Technical Conference and

Product Exposition, Oct. 2008.

[80] J. Bertrand, J. W. Cruz, B. Majkrzak, and T. Rossano, “CORBA delays in a softwaredefined

radio,” IEEE Communications Magazine, vol. 40, no. 2, pp. 152–155, Feb.

2002.

[81] T. Tsou, P. Balister, and J. Reed, “Latency profiling for SCA software radio,” in SDR

’07 Technical Conference and Product Exposition, Nov. 2007.

[82] P. Navarro, R. Villing, and R. Farrell, “Software defined radio architectures evaluation,”

in SDR ’08 Technical Conference and Product Exposition, Oct. 2008.

[83] G. Abgrall, F. Le Roy, J.-P. Diguet, G. Gogniat, and J.-P. Delahaye, “Predictibility

of inter-component latency in a software communications architecture operating environment,”

in IEEE International Symposium on Parallel Distributed Processing,

Workshops and Phd Forum (IPDPSW), 2010, pp. 1 –8, Apr. 2010.

[84] R. Haines, “Quantification of spectrum use: Spectrum management tools for the

twenty-first century,” in IEEE International Symposium on Electromagnetic Compatibility,

1990, pp. 390 –396, Aug. 1990.

[85] S. Ghaheri-Niri and P. Leaves, “Traffic control & dynamic spectrum allocation in

DRiVE,” Nov. 2000, presentation slides.

[86] P. Leaves, J. Huschke, and R. Tafazolli, “A summary of dynamic spectrum allocation

results from DRiVE,” in IST Mobile and Wireless Telecommunications, pp. 245–250,

June 2002.

[87] J. Huschke and R. Keller, “Dynamic frequency spectrum re-allocation US 7,436,788

b2,” 2008, United States Patent, (PCT filed 2002).

88


References

[88] M. Buddhikot, P. Kolodzy, S. Miller, K. Ryan, and J. Evans, “DIMSUMnet: New

directions in wireless networking using coordinated dynamic spectrum,” in World of

Wireless Mobile and Multimedia Networks, 2005. WoWMoM 2005., pp. 78–85, June

2005.

[89] V. Brik, E. Rozner, S. Banerjee, and P. Bahl, “DSAP: A protocol for coordinated spectrum

access,” in First IEEE International Symposium on New Frontiers in Dynamic

Spectrum Access Networks, 2005. DySPAN 2005., pp. 611–614, Nov. 2005.

[90] O. Ileri, D. Samardzija, and N. Mandayam, “Demand responsive pricing and competitive

spectrum allocation via a spectrum server,” in First IEEE International Symposium

on New Frontiers in Dynamic Spectrum Access Networks, 2005. DySPAN 2005.,

pp. 194 –202, Nov. 2005.

[91] H. Kim, Y. Lee, and S. Yun, “A dynamic spectrum allocation between network operators

with priority-based sharing and negotiation,” in IEEE 16th International Symposium

on Personal, Indoor and Mobile Radio Communications, 2005. PIMRC 2005.,

vol. 2, pp. 1004 –1008 Vol. 2, Sep. 2005.

[92] T. Kamakaris, M. M. Buddhikot, and R. Iyer, “A case for coordinated dynamic spectrum

access in cellular networks,” in First IEEE International Symposium on New

Frontiers in Dynamic Spectrum Access Networks, 2005. DySPAN 2005., pp. 289–298,

Nov. 2005.

[93] A. P. Subramanian, H. Gupta, S. R. Das, and M. M. Buddhikot, “Fast spectrum allocation

in coordinated dynamic spectrum access based cellular networks,” in 2nd IEEE

International Symposium on New Frontiers in Dynamic Spectrum Access Networks,

2007. DySPAN 2007., pp. 320–330, 17-20 April 2007.

[94] A. Subramanian, M. Al-Ayyoub, H. Gupta, S. Das, and M. Buddhikot, “Near-optimal

dynamic spectrum allocation in cellular networks,” in 3rd IEEE Symposium on New

Frontiers in Dynamic Spectrum Access Networks, 2008. DySPAN 2008., pp. 1 –11,

Oct. 2008.

[95] L. Kovács and A. Vidács, “Spatio-temporal spectrum management model for dynamic

spectrum access networks,” in TAPAS ’06: Proceedings of the first international

workshop on Technology and policy for accessing spectrum, p. 10. New

York,USA: ACM, Aug. 2006.

[96] L. Kovacs, A. Vidacs, and J. Tapolcai, “Spatio-temporal dynamic spectrum allocation

with interference handling,” in IEEE International Conference on Communications,

2007. ICC ’07., pp. 5575 –5580, June 2007.

[97] Roke Manor Research Ltd et al, for Ofcom, “Study into Dynamic Spectrum Access

Summary Report,” Ofcom, Tech. Rep., Mar. 2007.

[98] F. Ge, R. Rangnekar, A. Radhakrishnan, S. Nair, Q. Chen, A. Fayez, Y. Wang, and

C. Bostian, “A cooperative sensing based spectrum broker for dynamic spectrum

access,” in Military Communications Conference, 2009. MILCOM 2009., pp. 1 –7.

IEEE, Oct. 2009.

89


REFERENCES

[99] C. Tran, R. Lu, A. Ramirez, C. Phillips, and S. Thai, “Dynamic Spectrum Access: Architectures

and implications,” in Military Communications Conference, 2008. MIL-

COM 2008., pp. 1–7. IEEE, Nov. 2008.

[100] M. Nekovee, “Dynamic spectrum access - concepts and future architectures,” BT

Technology Journal, vol. 24, no. 2, pp. 111–116, Apr. 2006. [Online]. Available:

http://dx.doi.org/10.1007/s10550-006-0047-4

[101] ——, “Dynamic spectrum access with cognitive radios: Future architectures and

research challenges,” in 1st International Conference on Cognitive Radio Oriented

Wireless Networks and Communications, 2006. CROWNCOM 2006., pp. 1 –5. IEEE,

June 2006.

[102] B. Chen, H. Huang, Z. Chao, X. Yang, and L. Lei, “Future Architecture

with Dynamic Spectrum Access and Cross-layer Design: An Overview,” 2008.

[Online]. Available: http://www.wireless-world-research.org/fileadmin/sites/default/

files/meetings/Past%20Meetings/2008/WWRF21/Presentations%20and%20Papers/

WG3/papers/WWRF21-WG3-Doc3-FutureArch-DSA.pdf

[103] I. F. Akyildiz, L. Won-Yeol, M. C. Vuran, and M. Shantidev, “A survey on spectrum

management in cognitive radio networks,” IEEE Communications Magazine, vol. 46,

no. 4, pp. 40–48, Apr. 2008.

[104] O. Holland, A. Attar, M. Sooriyabandara, T. Farnham, H. Aghvami, M. Muck,

V. Ivanov, and K. Nolte, “Architectures and protocols for dynamic spectrum sharing

in heterogeneous wireless access networks,” in Heterogeneous Wireless Access

Networks. Springer US, 2009, pp. 1–35, 10.1007/978-0-387-09777-0 3. [Online].

Available: http://dx.doi.org/10.1007/978-0-387-09777-0 3

[105] B. Wang and K. Liu, “Advances in Cognitive Radio Networks: A Survey,” to

appear in IEEE Journal of Selected Topics on Signal Processing. [Online]. Available:

http://terpconnect.umd.edu/ ∼ bebewang/Wang JSTSP 2010.pdf

[106] M. Bennis, J.-P. Kermoal, P. Ojanen, J. Lara, S. Abedi, R. Pintenet,

S. Thilakawardana, and R. Tafazolli, “Advanced spectrum functionalities for future

radio networks,” Wireless Personal Communications, vol. 48, pp. 175–191, Jan.

2009, 10.1007/s11277-007-9423-8. [Online]. Available: http://dx.doi.org/10.1007/

s11277-007-9423-8

[107] G. Salami, O. Durowoju, A. Attar, O. Holland, R. Tafazolli, and H. Aghvami, “A

comparison between the centralized and distributed approaches for spectrum management,”

IEEE Communications Surveys & Tutorials, April 2010, available from IEEE

Xplore (early access), to appear in the second issue 2011 of IEEE Communications

Surveys & Tutorials.

[108] S. Haykin, “Cognitive Radio: Brain-Empowered Wireless Communications,” IEEE

Journal on Selected Areas in Communications, vol. 23, no. 2, pp. 201–220, Feb.

2005.

[109] J. R. Barry, D. G. Messerschmitt, and E. A. Lee, Digital Communication: Third

Edition. USA: Springer, 2003.

90


References

[110] W. Yu, W. Rhee, S. Boyd, and J. M. Cioffi, “Iterative water-filling for Gaussian vector

multiple access channels,” in IEEE International Symposium on Information Theory

(ISIT), p. 322, June 2001.

[111] J. L. Holsinger, “Digital communication over fixed time-continuous channels

with memory - with special application to telephone channels,” MIT Research

Laboratory of Electronics, Tech. Rep., Oct. 1964. [Online]. Available: http:

//hdl.handle.net/1721.1/4395

[112] S. Hayashi and Z.-Q. Luo, “Dynamic spectrum management: When is FDMA sumrate

optimal?” in IEEE International Conference on Acoustics, Speech and Signal

Processing, 2007. ICASSP 2007., vol. 3, pp. III–609 –III–612, Apr. 2007.

[113] K. Shum, K.-K. Leung, and C. W. Sung, “Convergence of iterative waterfilling algorithm

for Gaussian interference channels,” IEEE Journal on Selected Areas in Communications,

vol. 25, no. 6, pp. 1091 –1100, Aug. 2007.

[114] R. Zhang, Ying-chang Liang, and S. Cui, “Dynamic resource allocation in cognitive

radio networks,” IEEE Signal Processing Magazine, vol. 27, no. 3, pp. 102 –114, May

2010.

[115] R. Rajbanshi, A. Wyglinski, and G. Minden, “OFDM-Based cognitive radios

for dynamic spectrum access networks,” in Cognitive Wireless Communication

Networks, E. Hossain and V. Bhargava, Eds. Springer US, Oct. 2007, pp.

165–188, 10.1007/978-0-387-68832-9 6. [Online]. Available: http://dx.doi.org/10.

1007/978-0-387-68832-9 6

[116] omniORB. (Feb. 2008). [Online]. Available: http://omniorb.sourceforge.net

[117] Communications Research Canada. SCARI-Open Downloads. (Oct. 2010). [Online].

Available: http://www.crc.gc.ca/en/html/crc/home/research/satcom/rars/sdr/

products/scari open/downloads

[118] VirginiaTech. OSSIE SCA-based open source software defined radio. (Oct. 2010).

[Online]. Available: http://ossie.wireless.vt.edu

[119] OProfile - A System Profiler for Linux. (Feb. 2008). [Online]. Available:

http://oprofile.sourceforge.net

[120] SYSSTAT. (Feb. 2008). [Online]. Available: http://pagesperso-orange.fr/sebastien.

godard/

[121] C. Li, C. Ding, and K. Shen, “Quantifying the cost of context switch,” in ExpCS ’07:

Proceedings of the 2007 workshop on Experimental computer science, June 2007.

[122] Z. Jianfan, D. Levy, and A. Liu, “Evaluating overhead and predictability of a real-time

CORBA system,” in Proceedings of the 37th Annual Hawaii International Conference

on System Sciences - 2004, p. 8, Jan. 2004.

[123] Objective Interface Systems, Inc. ORBexpress RT. (2010). [Online]. Available:

http://www.ois.com/Products/orbexpress-rt-for-c.html

91


REFERENCES

[124] J. P. Z. Zapata, “SCA ResourceFactory,” Oct. 2010, Email correspondence.

[125] D. Paniscotti and J. Bickle, “SDR signal processing distributive-development approaches,”

in 2007 Software Defined Radio Technical Conference and Product Exposition,

Nov. 2007.

[126] F. Casalino, G. Middioni, and D. Paniscotti, “Experience report on the use of CORBA

as the sole middleware solution in SCA-based SDR environments,” in SDR’08 Technical

Conference and Product Exposition, Oct. 2008.

[127] L. Pucker and J. Holt, “Extending the SCA core framework inside the modem architecture

of a software defined radio,” IEEE Communications Magazine, vol. 42, no. 3,

pp. 21–25, Mar. 2004.

[128] T. Ulversøy and T. Maseng, “Coordinated dynamic spectrum access employing peerto-peer

architectures,” Jan. 2009, unpublished (4 pages).

[129] C. Jones, “Bad days for software,” IEEE Spectrum, vol. 35, no. 9, pp. 47 –52, Sep.

1998.

[130] T. Schūtt, F. Schintke, and A. Reinefeld, “Range queries on structured

overlay networks,” Computer Communications, vol. 31, no. 2, pp. 280–

291, Feb. 2008. [Online]. Available: http://www.sciencedirect.com/science/article/

B6TYP-4PJCYCY-4/1/9c59d2509d6e7bb6da433e81b92c6c62

[131] P. Gu, J. Wang, and H. Cai, “ASAP: An advertisement-based search algorithm for unstructured

peer-to-peer systems,” in International Conference on Parallel Processing,

2007.ICPP 2007., p. 8, Sep. 2007.

[132] D. Keil and D. Goldin, “Indirect interaction in environments for multi-agent systems,”

in Environments for Multi-Agent Systems II. Springer Berlin / Heidelberg, Feb. 2006,

pp. 68–87.

[133] Federal Communications Commission, “Office of Engineering and Technology Invites

Proposals from Entities Seeking to be Designated TV Band Device Database

Managers,” Nov. 2009, ET Docket No. 04-186.

[134] T. Hoang, M. Skjegstad, T. Maseng, and T. Ulversøy, “FRP: The frequency resource

protocol,” in IEEE International Conference on Communication Systems

(IEEE ICCSS 2010), Nov. 2010.

[135] A. B. MacKenzie and L. A. DaSilva, Game Theory for Wireless Engineers. USA:

Morgan & Claypool, 2006.

[136] M. van der Schaar and F. Fu, “Spectrum access games and strategic learning in cognitive

radio networks for delay-critical applications,” Proceedings of the IEEE, vol. 97,

no. 4, pp. 720–740, Apr. 2009.

[137] A. MacKenzie, J. Reed, P. Athanas, C. Bostian, R. Buehrer, L. DaSilva, S. Ellingson,

Y. Hou, M. Hsiao, J.-M. Park, C. Patterson, S. Raman, and C. da Silva, “Cognitive

radio and networking research at Virginia Tech,” Proceedings of the IEEE, vol. 97,

no. 4, pp. 660 –688, Apr. 2009.

92


References

[138] S. Chen and A. M. Wyglinski, “Efficient spectrum utilization via cross-layer optimization

in distributed cognitive radio networks,” Computer Communications,

vol. 32, no. 18, pp. 1931–1943, Dec. 2009.

[139] A. He, K. K. Bae, T. Newman, J. Gaeddert, K. Kim, R. Menon, L. Morales-Tirado,

J. Neel, Y. Zhao, J. Reed, and W. Tranter, “A survey of artificial intelligence for

cognitive radios,” IEEE Transactions on Vehicular Technology, vol. 59, no. 4, pp.

1578 –1592, May 2010.

93


Part II

Included Papers

95


Paper A

Software Defined Radio: Challenges and Opportunities

Tore Ulversøy

IEEE Communications Surveys & Tutorials

Vol. 12, No. 4, 2010

(IEEE online journal)

Will be available online at: http://www.comsoc.org/livepubs/surveys/index.html

Available online at: http://ieeexplore.ieee.org

97


Abstract

Software Defined Radio (SDR) may provide flexible, upgradeable and longer lifetime

radio equipment for the military and for civilian wireless communications infrastructure.

SDR may also provide more flexible and possibly cheaper multi-standardterminals

for end users. It is also important as a convenient base technology for the

future context-sensitive, adaptive and learning radio units referred to as cognitive radios.

SDR also poses many challenges, however, some of them causing SDR to evolve

slower than otherwise anticipated. Transceiver development challenges include size,

weight and power issues such as the required computing capacity, but also SW architectural

challenges such as waveform application portability. SDR has demanding

implications for regulators, security organizations and business developers.

c○2010 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained

for all other uses, in any current or future media, including reprinting/republishing this material for

advertising or promotional purposes, creating new collective works, for resale or redistribution to

servers or lists, or reuse of any copyrighted component of this work in other works.

DOI: 10.1109/SURV.2010.032910.00019

99


1. Introduction

1 Introduction

The term ’Software Radio’ was coined by Joseph Mitola III to signal the shift from HW

design dominated radio systems to systems where the major part of the functionality is

defined in software. He described the concept in [1].

Software Defined Radio (SDR) has no single, unified, globally recognized definition.

Slightly different interpretations exist among the actors in the field. Adding to this, a variety

of related terms has been proposed and is used to variable degrees. These include Software

Based Radio [2], Reconfigurable Radio, Flexible Architecture Radio [3]. The main parameter

in the various interpretations and the related terms, is how flexibly the radio waveform

can be changed through changing software (SW) and without modifying the SDR Platform

(the combination of hardware and operating environment where the waveform application

is running). Obviously the ideal however unrealizable goal is to be able to communicate

at any desirable frequency, bandwidth, modulation and data rate by simply loading the appropriate

SW. Usually SDR is given a more practical interpretation, implying simply that

large parts of the waveform are defined in SW, giving the flexibility to change the waveform

within certain bounds as given by the actual system. The flexibility is commonly assumed

to extend at least to multi-band and multi-modulation. Examples of specific definitions are

those provided by the Software Defined Radio Forum (SDR Forum) [4] and by the Federal

Communications Commission (FCC) [5].

The evolution towards SDR systems has been driven in part by the evolution of the enabling

technologies, first and foremost the DA and AD converters and the Digital Signal

Processors (DSPs), but also that of the General Purpose Processors (GPPs) and the Field

Programmable Gate Arrays (FPGAs). A major driving force has also been the demand

for more flexible and reconfigurable radio communication solutions, in particular from the

military sector. This demand has resulted in several major governmental development programmes,

e.g. in the US the SpeakEasy I, the SpeakEasy II [6] and the ongoing Joint Tactical

Radio System (JTRS) [7–9] programme. Examples from Europe are the ”Software Radio

Architecture” Plan d’Etude Amont (PEA) [10, 11] in France, the Terminal Radio Software

(TERSO) programme [12] in Spain, the Finnish Software Radio Programme (FSRP) [13],

the Swedish Common Tactical Radio System (GTRS) [14, 15] and the European Secured

Software Defined Radio Referential (ESSOR) [16] programme. The latter is the result of

cooperation between several European countries under the banner of the European Defence

Agency (EDA).

There are many motivations for utilizing SDR solutions. For the military sector, where

communication systems need to have a longer service life time than in the commercial sector,

SDR helps to protect investments by prolonging the useful service life of communication

systems. This is facilitated through SDR allowing the possibility to change waveforms

and/or load new waveforms on already acquired SDR equipment. It also allows SDR applications

(waveforms) that are already invested in to be ported to new and more capable SDR

platforms. SDR furthermore provides the flexible asset suited for the changing environments

of coalition and Network Centric Operations (NCO).

A major motivation within the commercial communications arena, is the rapid evolvement

of communications standards, making SW upgrades of base stations a more attractive

solution than the costly replacement of base stations.

Common for both the military and the commercial sector, is that SDR opens up a range

of possibilities by making existing types of radio applications easier to implement, and by

101


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

allowing new types of applications. In particular the computing capacity and the flexibility

of the SDR may be exploited to develop Cognitive Radios (CR), context-sensitive and

adaptive units that may also learn from their adaptations. As an example, the SDR unit may

adapt to harsh interference and noise conditions by instantly changing parts of the waveform

processing through loading different SW modules, in order to still maintain adequate bit error

rates. The cognitive functionality may also be used for improved spectrum utilization,

e.g. through the coexistence of cognitive systems with legacy systems.

SDR is also beneficial for space applications as it provides the flexibility that will allow

deployed satellite communication equipment to be SW upgraded according to advances

in algorithms and communication standards. This will allow communication functionality

changes and multiple uses during the lifetime of the satellite [17].

The exploitation of SDR technology in actual products has evolved more slowly than

what was anticipated some years ago. As late as in 2002 [2] it was predicted that by 2006

the adoption of SDR in commercial mobile terminals would have ”widespread adoption

and movement to SDR as baseline design”, something which certainly has not happened.

Also the US government JTRS programme has at various times since its initialization in

1997 ”experienced cost and schedule overruns and performance shortfalls” [7]. However,

lately there have also been several positive signs and accomplishments within SDR, which

indicates that we are getting closer to a larger-scale adoption of SDR in commercial products.

Examples are Vanu Inc.’s ’Anywave’ base station approved by the FCC [18], the first

’SCA approved with waivers’ military communications product Thales AN/PRC-148 JEM,

and the first ’SCA approved with no waivers’ Harris Falcon III(TM) AN/PRC-152(C). Also

an increasing number of SDR research and prototyping platforms are being offered on the

market, along with SDR development tools.

This tutorial will review the fundamental challenges that SDR imposes on the various

actors within the field, i.e. developers, security organizations, regulators, business managers,

and users. It will further review part of the important past and ongoing work, when available,

that has contributed to deal with these challenges. During the discussion of each topic there

will be a summary of the remaining open items and/or the projections.

The tutorial will start with SDR SW architecture. As a background to this discussion,

Software Communications Architecture (SCA) is briefly reviewed. Then there is a review

of the challenges and existing/ongoing work within application portability, application development,

the underlying middleware platform and alternative architectures.

A fundamental challenge of SDR is to provide the necessary computational capacity to

process the waveform applications, in particular the complex and high data rate waveforms

and especially for units with strict power- and size limitations. The computational requirements

and the available computing elements required to handle them will be reviewed.

Further the implications for security, regulations and for the radio manufacturer business

structure will be discussed and the remaining challenges and/or future projections will be

commented on.

SDR poses severe challenges also in analogue RF hardware design and the conversion

between the analogue and digital domains, particularly in wideband implementations. In

order to limit the scope of this tutorial, and as these topics are not unique to SDR alone, these

topics have not been discussed here. A recommended source of information on these topics

is the recent work by Kenington [3]. The aspects of AD conversion are also excellently

treated in [19] which occurs in what is considered a landmark special issue on Software

102


2. SDR and the Software Communications Architecture

Radio [20]. Additional recommended sources on SDR receiver front-end technology are

[17, 21–24].

2 SDR and the Software Communications Architecture

The most widely used software architecture for SDR is the Software Communications Architecture

(SCA). SCA is published by the Joint Program Executive Office (JPEO) for JTRS,

with SDR Forum having assisted the JPEO with the development of SCA. SCA is considered

a ’de facto’ standard for the military domain, and has been implemented within the

SDR industry, research organizations and at universities.

SCA together with available SCA-based tools allow designers to build component-based

SDR applications, as assemblies of components and logical devices. The components are

SW processing modules with input and output ports, context dependencies in the form of

processing element dependencies, and with settable properties. The logical devices are abstractions

for HW modules that in some way process the information stream.

Data Source

FEC

Encoder

Interleaver

Symbol

Mapper

and

Scrambler

Symbol to I/Q

and

Transmission

Filter

FIGURE A.1: An example application component structure, the TX baseband processing of a relatively

simple Stanag 4285 waveform. The processed data is sent in packets from the output port of

one component to the input port of the next component.

The component-based approach makes the reuse of parts of applications easier as the

components have clearly defined inputs, outputs and context requirements, and are deployable

units. The component-based approach also promotes a separation of roles in the development.

Thus, a radio systems engineer could assemble an SDR application based on preprogrammed

components without having to be a SW specialist, whereas a signal processing

implementation specialist could concentrate on the processing code of a component.

The SCA is a distributed systems architecture, allowing the various parts of applications

to run on different processing elements, i.e. each component is deployed on one of

a set of available processors. The communication between the components, and between

components and devices, is based on using the Common Object Request Broker Architecture

(CORBA) middleware. For communication with specialized processors, e.g. DSPs or

FPGAs, SCA advises the use of adapters between CORBA and these units.

System

Component

FC

FS

A

P

I

CORBA

Application

Component

FC

AEP

FS

A

P

I

CORBA

Application

Component

FC

AEP

FS

A

P

I

CORBA

System

Component

FC

FS

OS

FIGURE A.2: A visualization of the layers of the SCA. ’FC’ is an implementation of the Framework

Control Interfaces, and ’FS’ an implementation of the Framework Services Interfaces. ’AEP’ is the

Application Environment Profile, that limits the applications access to the Operating System (OS).

103


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

The SCA defines a protocol and an environment for the application components. SCA

does this by defining a set of interfaces and services, referred to as the Core Framework (CF)

[25], by specifying the information requirements and formats for the Extensible Markup

Language (XML) descriptions for components and applications, termed the ”Domain Profile”,

and by specifying underlying middleware and standards. By providing a standard

for instantiation, management, connection of and communication between components, the

SCA contributes to providing portability and reusability of components.

The CF interfaces are grouped in four sets:

• The Base Application Interfaces provide management and control interfaces for the

application components, and they are implemented by these application components.

• The Base Device Interfaces allow management and control of hardware devices

through their software interface, and are implemented by the logical device components.

• The Framework Control Interfaces control the instantiation, management and removal

of software from the system, and these interfaces are implemented by software modules

that form part of the system platform.

• The Framework Services Interfaces provide file functions, and are implemented by

software modules that form part of the system platform.

SCA is a general architecture, targeted for but not limited to SDR systems. SCA has similarities

with other distributed component architectures, e.g. the CORBA Component Model

(CCM). A conceptual difference relative to CCM is the support for system components or

’devices’.

3 SW Architectural Challenges

Since the SCA is the dominant SDR architecture, the SCA-related challenges will be focused

on first. Then the more general SW architectural challenges and alternatives to SCA will be

discussed. The remaining open issues will be highlighted.

3.1 Portability of SDR SCA-Based Applications

A fundamental challenge of SDR is to provide an ideal platform to application separation,

such that waveform applications can be moved from one SDR platform to be rebuilt on

another one without having to change or rewrite the application. Such waveform portability

is highly desirable, particularly in the military sector, for example in order to achieve

interoperability in coalitions by exchanging waveforms.

SCA contributes to such application portability by providing a standard for deployment

and management of SCA-based applications [25]. It also standardizes the interconnection

and intercommunication both between the components of the application, and between components

and system devices. Using the SCA Application Environment Profile (AEP), SCA

also standardizes the minimum subset of operating system capabilities that must be available

for the applications, and hence the limited subset that applications may use.

104


3. SW Architectural Challenges

The SCA compliance of an application is not sufficient to cover all aspects of portability.

Significant pieces that are not standardized by the SCA itself are the APIs to the services

and devices of the system platform (see Figure A.2). Since these are linked to the actual

implementation of the system platform, they are supposed to be standardized per system or

domain, as is clearly pointed out in the SCA 2.2.2 specification [25].

Within JTRS, a number of such APIs have been developed [8]. Although previously not

publicly accessible from the SCA website, 14 APIs were made available in April 2007 [26]

and as of February 2009, 18 were available. Presumably this is not a complete set of APIs.

Also for some APIs there may be strategic reasons for not wanting to release them from a

particular domain, examples are the security-related APIs.

In order for portability to extend across domains, the APIs to the services and devices

will need to be standardized across domains as well. With the JTRS APIs now being available,

these may be one option for such standardization, particularly for military domains.

There are also several other initiatives in this area, including one from the Object Management

Group (OMG) Software Based Communications Domain Task Force [27]. Another

example is the ESSOR project, which aims at giving ”European industry the capability to

develop interoperable SDR” [16]. It remains to be seen if ESSOR will develop standards to

be used by a European military domain only, or whether this initiative could also contribute

to providing inter-domain waveform portability.

An alternative and equivalent approach to that of standardizing APIs to system components

is providing abstraction layers between the platform and the application components.

An example is a proposal for a ”Transceiver Facility Platform Independent Model” [28].

Another related portability issue is the various alternatives for transport mechanisms for

the communication with components deployed on DSPs and FPGAs. SCA 2.2.2 prescribes

adapters between CORBA and the DSP and FPGA components as the primary means of

communication with these elements. The JTRS has standardized a specific adapter referred

to as the Modem Hardware Abstraction layer (MHAL) for this purpose [8, 26]. Other similar

solutions exist, e.g. Spectrum Signal Processing’s ’QuicComm’ [29]. In recent years,

Object Request Broker (ORB) implementations have also been made available on DSPs and

FPGAs, making CORBA communication possible also to these components [30]. The fact

that various messaging protocols are currently used implies, however, that communication

with DSPs and FPGAs will remain a portability issue until one standard or another has

become the de facto standard.

Furthermore there are some minor portability issues related to differences in ORB implementations

[31].

Lastly, portability obviously requires that the component code is interpreted correctly on

the platform. This again has two aspects, language compatibility issues and target processor

functionality compatibility. Since SCA is based on CORBA which has support for several

programming languages, using different code languages will be possible as long as the appropriate

compilers and libraries are available. However, different processing elements, in

particular different types of DSPs and FPGAs, support different functionalities and features.

This either requires several component implementations, one for each family of processing

elements, resulting in an overhead of work-hours used. The other approach is to have the

component functionality defined in a high-level language, which is compiled to create a correct

code image for the actual processing element to be used. Obviously such a compiler

may become very complicated. The resulting target code or image may also become less

optimal than a target code written specifically for the target processor.

105


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

Further information on portability issues may be found in recent publications, including

API standardization [32], lessons learned from porting a soldier radio waveform [33], SCA

aspects in heterogeneous environments [34] and the trade-off between portability and energy

efficiency of the processing [35].

The portability issues with SCA-based applications are summarized in Table A.1. As is

evident from the table, important challenges remain in this area.

TABLE A.1: A Summary of Portability Issues for SCA-Based Applications

Portability aspect: Standardized through

SCA?

Environment and protocol for

the installation, instantiation,

Yes. (But: SCA Security requirements

not public)

control, connection and intercommunication

of application

components

Defined allowed Operating- Yes, through AEP

System access

APIs to system units (devices)

and system services

No. (SCA states this is to be

handled per domain.)

Communication (message No. Multiple solutions available.

transport) with specialized

JTRS has standardized

processors (DSPs, FPGAs) on MHAL.

ORB SCA specifies CORBA.

There are however some

minor differences between

ORB-implementations.

Programming language No, but this merely presumes

availability of compilers, li-

Target processor compatibility

of the code

braries etc.

No. Different DSPs and FP-

GAs may support different

features.

3.2 Challenges related to SCA Application Development

For the traditional communications equipment design engineer, with a communications or

radio engineering and less of a SW distributed systems background, SCA may appear challenging

to learn and understand.

Even for embedded systems engineers without a CORBA and Object Oriented Programming

background, according to [36] ’it could take several months to fully understand SCA’.

SCA tools help abstract away some of the difficulties in SCA. Commercial tools are

available from various sources, e.g. Zeligsoft [37], PrismTech [38] and Communications

Research Canada (CRC) [39], and there is also a University Open Source initiative, the

Ossie Waveform Developer [40] from VirginiaTech. While defining SCA components manually

is tedious work involving a lot of XML and CORBA details, the tools allow the SDR

106


3. SW Architectural Challenges

designers to define the components through user-friendly tool interfaces. The tools also allow

applications to be formed by making connections between the various components, and

between the components and the devices of the platform. Still, even with the tools being

of significant help in the development process, concluding for example from SCA-based

development efforts within own organization, detailed SCA knowledge is still needed and

in a starting phase a lot of time is spent on non-signal-processing issues, particularly on a

heterogeneous platform.

The tools typically generate the necessary XML and the SCA infrastructure part of the

components [41], while functional processing code needs to be added by the designer, either

coded manually or using her/his favourite tools. A more unified higher-level design

approach possibly could improve productivity. An approach where the functional skeleton

code is imported into a Unified Modelling Language (UML) to allow higher abstraction level

modelling of the functional behaviour, is described in [41]. It is envisioned that SDR-design

will increasingly be performed at higher abstraction levels, eventually using fully integrated

Model-Driven Development (MDD) [42] tools with automatic transformation from model

level to any specific heterogeneous platform.

In summary, a further enhancement of the efficiency of designing SCA-based applications,

as well as a general availability of MDD tools with fully automated conversion to code

level for any given HW platform are important remaining challenges.

3.3 CORBA Related Challenges

CORBA is demanded by SCA as a middleware platform. The use of CORBA, however, has

known challenges in the form of implications on communication throughput, latency and

latency variation, as well as an overhead of consumed computation and memory resources.

Another issue is that CORBA has lost its popularity in some application domains, which

naturally raises the question of whether an alternative middleware is also needed for SDR.

Throughput is a factor of both CORBA and the underlying transport used by CORBA. In

[43] throughput of CORBA one-way invocations has been measured using a TAO 1.2 ORB,

TCP/IP and 100MBit/sec Ethernet, and compared to using TCP/IP socket programming

directly. The results show that the CORBA throughput is highly dependent on message size.

With a message size of 64 bytes the CORBA TAO throughput was only a few MBit/sec,

whereas with TCP/IP socket programming above 90 Mbit/sec. For message sizes above 8K

bytes the throughput came close to that of using socket programming directly. These results

show that avoiding too small packet sizes in SCA-based applications is important in order

to keep the throughput optimized.

Where the throughput is limited mainly by the underlying transport, other transport

mechanisms than TCP/IP may be used with CORBA. ”CORBA over RapidIO” [44] and

CORBA over a PCI bus [45] are recently described examples.

Latency implies processing delays in the SDR, and tolerable level is application and

waveform dependent. As with throughput, latency is a function of both CORBA and the

underlying transport. Average latency tends to show a linear relation with message size

[44, 46], and with a significant non-zero latency for message size 0. As an example, latency

with CORBA over a RapidIO transport between GPPs [44] is measured at 114 µs for size 32

bytes and 180 µs for 4096 bytes. Latency variation may be reduced [47] by using Real-Time

CORBA features.

107


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

Measurements of CORBA computation overhead for a GPP-system with the Ossie CF

[40] and OmniORB have been provided in [48]. Figure A.3 shows an example of how the

computation overhead is increased when an application is split into more and more components,

and which in this case is dominated by a CORBA-related overhead [48]. The

application waveform was a Stanag 4285 TX base band waveform running at an increased

symbol rate of 25600 symbols / sec, with frames of 256 symbols being processed at a time.

Stanag

4285 TX

Data

Sink

CPU: 5.5%

Data

Source

FEC

Encoder

Interleaver

Symbol

Mapper &

Scrambler

Symbol to

I/Q & TX

Filter

Float to

fixed

converter

Data

Sink

CPU: 11.2%

Data

Source

FEC

Encoder

Interleaver

Symbol

Mapper &

Scrambler

Symbol to

I/Q & TX

Filter

Float to

fixed

converter

Forwarder Forwarder Forwarder Forwarder

Data

Sink

CPU: 16.6%

FIGURE A.3: Processor workload (user applications + system) in % when running a Stanag 4285 TX

waveform, at an increased symbol rate of 25600 symbols /sec, using the Ossie CF on a GPP, and with

having the waveform application implemented as one SCA component (top), 6 components (middle)

and 6 components with 4 additional (no-functionality) forwarding components (bottom). With the

useful signal processing being identical in all three cases, the processor workload is seen to increase

significantly as the processing is split into more components.

The memory footprint may be significant on a general full-feature ORB [49] but slim

implementations have been made needing less than 100 kBytes of memory [44, 50].

In order to eliminate CORBA latency, throughput and footprint implications, and in

particular when the processing is done on DSPs and FPGAs, data transfers are done on many

current SDR systems on specific high-speed connections with low-level formatting [50, 51]

instead of using CORBA. A downside of this approach is that it makes portability more

difficult, unless the high-speed connections and formatting are standardized. Adapters may

be used for control and data from CORBA-enabled processors.

While CORBA used to be limited to GPP processors only, ORBs for specialized processing

elements such as DSPs [50] and FPGAs [30, 44, 50, 52] have been developed in

recent years. The ORB on an FPGA may be put in an embedded microprocessor [52], or (as

a CORBA subset) directly implemented at native gate level [44, 50], where the latter has a

processing speed advantage. This theoretically facilitates CORBA communication between

all typical processing elements on an SDR platform, which is excellent for portability. The

downside of the approach is the amount of resources occupied by these ORBs on the processing

elements, and the latency and throughput implications. As for FPGA ones, latency

numbers published in [44] and statements in [50] that an FPGA native level ORB may process

a message in ’a few hundred nanoseconds’ indicate that FPGA native level ORBs can

108


3. SW Architectural Challenges

now be made very effective in terms of performance, but further public domain results are

needed on this subject.

CORBA has its popularity in embedded and real-time applications, but has become

less popular in general business-oriented applications. This naturally raises the question of

whether there are other likely candidates to take over from CORBA also in SCA and SDR

applications. In [53], CORBA has been compared to two main competitors in the businessoriented

domain, Enterprise Java Beans (EJB) and Web Services. CORBA was found to be

the most mature and better performing technology, 7 times faster than Web Services in a

specific single-client evaluation test, but also by far the most complex one. Web Services

was the worst performer but the simplest one and the one that tackled Internet applications

best. Web Services is popular in this domain, which illustrates that it is not because it is being

outcompeted on performance that CORBAs popularity has diminished, but rather due to

other technologies meeting a weighted set of requirements for this type of application better.

CORBA’s standing in this domain is thus not directly transferrable to the SDR domain, as

the SDR domain has more focus on performance issues.

The Data Distribution Service (DDS) has been suggested as an alternative middleware in

an SCA context. DDS is an OMG-managed standard for publish-subscribe communication

for real-time and embedded systems [54]. DDS belongs to the group of Message-Oriented

Middleware (MOM). MOM provides a looser coupling between senders and receivers than

in CORBA, messages are sent to an intermediary layer from which the message is received

by the addressee [55].

In summary, there are exciting ongoing CORBA activities, such as enabling ORBs to

work with fast transports and new ORBs for FPGAs. In the near term, it is anticipated that

this migration of CORBA onto specialized processors and faster transports will continue, but

that low-level non-CORBA data connections will still be used where they are advantageous.

So far there is not a clear path for a middleware that is less complex yet better performing,

to potentially take over the role of CORBA in SDR.

3.4 SCA Challenges and Alternative Architectures

Several technical- and complexity-related SCA challenges have been reviewed in the previous

subsections. A further political argument against SCA is that it is also not an open

standard as it is directly managed under the supervision of the JPEO. With these issues as a

background, it is interesting to explore the alternatives.

A closely related alternative architecture specification for SDR, and derived from the

SCA, is OMG’s ”PIM and PSM for Software Radio Components Specification” [56]. Its

current Platform Specific Model (PSM) utilizes CORBA-interfaces [56], but the division in

a Platform Independent Model (PIM) and a PSM makes it easier to substitute CORBA with

some other middleware, if more suitable middleware platforms were to emerge in the future.

OMG’s standards are open ones also in the sense that all members have an equal vote on the

final content of a standard. OMG’s specification is used by the WINTSEC project [57] in

Europe. It has been put forward as a promising candidate for future use in the commercial

domain [58].

Of particular interest for resource-constrained systems is NASA’s ’Space Telecommunication

Radio System’ (STRS) [59–61] architecture. Electronic devices used in space require

radiation hardening [59], and processors are hence slower than terrestrial equivalents, which

places further requirements on reduced resource consumption on the application and runtime

109


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

environment. STRS has many characteristics in common with SCA, such as the separation of

waveform applications from hardware, but there are also differences. No particular communication

mechanism is described, i.e. CORBA is neither mandated nor precluded. Likewise,

an XML parser is not part of the STRS infrastructure, XML files may be pre-processed prior

to deployment [59]. STRS does not have the notion of ports but rather optional source and

sink APIs [60]. The standard is a NASA managed one, but it is influenced through collaboration

with OMG and SDR Forum [61]. An open-source implementation, ”Open Space

Radio” [62], has been made available by Virginia Tech.

The GNU radio architecture [63] is an open-source initiative, where the signal processing

is carried out on GPP computers. GNU radio is adapted to the Universal Software Radio

Peripheral (USRP) which converts between base band and RF signals. Radio applications

are formed as graphs where the vertices represent signal processing blocks and the edges

are the data flow, and where the blocks have input and output ports. The signal processing

blocks are written in C++ and the graph is connected using the Python programming

language. A comparison between GNU radio and a OSSIE SCA-based system has been

recently published [64].

Since part of the features of MOM fit well with SDR’s needs, there could be a potential

for MOM-based architectures as future alternatives for SDR, however this has to be

demonstrated.

With the evolution towards cognitive radios which requires the radio to have reasoning

capability and adaptivity there will probably be a need for architectural features beyond

the present SCA. As an example, with cognitive radios it is beneficial to have convenient

framework functions to be able to swap a component from/to a running application in close

to real-time. Also, although the cognitive functionality itself, e.g. adaptation of the waveform

application to external conditions, may be implemented as application components, it

may be beneficial to partly support this functionality through middleware. An example of a

middleware-based approach to system self-adaptation is provided in [65].

Concluding on the outlook for SDR architectures, it is expected that the SCA will remain

a dominating architecture in the military segment, due to its momentum and the high

importance of portability in this domain. In the commercial civilian segment, where there

is less focus on portability and more on hardware cost and low power consumption, it is expected

that a significant portion of designs will use dedicated and proprietary lighter-weight

architectures. In a longer time perspective, with decreasing hardware cost and increasing

performance, it is expected that open and standardized architectures such as the OMG one

will gain wider acceptance in this sector.

4 Challenges and Opportunities Related to

Computational Requirements of SDR

4.1 Computational Requirements

A fundamental challenge with SDR is how to achieve sufficient computational capacity,

in particular for processing wide-band high bit rate waveforms, within acceptable size and

weight factors, within acceptable unit costs, and with acceptable power consumption.

This is particularly challenging for small handheld units, e.g. multi mode terminals. The

power consumption must be below certain limits to keep the battery discharge time within

110


4. Challenges and Opportunities Related to Computational Requirements of SDR

acceptable limits, and with the smallest handheld units it will also be an issue of not causing

the surface temperature of the device to become unpleasantly high for the user.

For base stations like cellular network infrastructure stations, and for vehicular mounted

stations, the power, size and weight factors are easier to accommodate, however performance

versus these parameters and cost may still be challenging for complex high bit rate

waveforms.

SDR applications perform processing of the various stages of receive and transmit signals,

but they also perform protocol handling, application control activities, user interaction

and more. Conceptually, as an abstraction, we can consider SDR applications to consist of

two main groups of components, (1) Data Processing Components (DPCs) and (2) Event

Driven, Administrative and Control Components (EDACCs). DPCs typically have deterministic

behaviour, in the form of processing a package of data according to its defined

algorithm. DPCs typically also have a high degree of inherent parallelism that may be

exploited, an example being a Finite Impulse Response (FIR) filter where a number of additions

and multiplications may be carried out in parallel, and the deterministic behaviour also

allows low-level optimization of implementations, see, for example [66]. EDACCs depend

on events, on data content or user interaction, or perform various administrative and control

tasks and are less predictable in their path of execution. Also they typically have far less

inherent parallelism.

The SDR components may to a large degree run in parallel, e.g. a decimator component

may run in parallel with a filter component and a Fast Fourier Transform (FFT) component,

since they work on different stages of the processed data. The Software Communications

Architecture (SCA) facilitates this type of parallel processing as it is a distributed systems

architecture, where processes may run on several processing elements while exchanging

processed data. A good exploitation of this parallelism depends on a well devised component

structure of the waveform application, along with optimized deployment of the components

on the available processing elements.

With complex waveforms, the DPCs will be the components that require the most computing

capacity, and the ones that drive the processing requirements of the SDR computing

platform. Table A.2 lists some computational complexity numbers for some key algorithms

of the 802.11a waveform at 24 Mbps, as provided in [67]. The calculated complexity numbers

are in the form of gigacycles per second referenced to an Alpha GPP. The complexity

is seen to be overwhelming for such a single GPP, as the required cycle rate is far above

achievable rates.

TABLE A.2: Computational complexity for some key algorithms of 802.11a

802.11a signal processing

operation at 24

Mbps

FFT 15.6

FIR (Rx) 6.08

Viterbi decoder 35.0

Gigacycles per second,

Alpha GPP

111


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

4.2 Processing Options

There is a large variety of available processing elements, each with their associated strong

and weak points. For the DPCs, processing elements that are able to exploit regular structures,

deterministic flows of instructions and internal parallelism will be beneficial from a

performance point of view. In the following some of the most important processing alternatives

will be reviewed. The alternatives will be listed according to reconfiguration time,

starting with the least configurable options and proceeding to the real-time configurable,

as the ability of SDRs to be reconfigured or reloaded with new waveform components or

applications is one of its most essential properties. For some SDR applications, it will be

sufficient to be able to reconfigure the unit at a maintenance site, and it will not be a problem

if it takes a few minutes to load a new application. For other applications, reconfiguration

will need to happen while switching from one service, network or waveform standard to

another, e.g. while switching from GSM to WiFi, and the reconfiguration should then typically

be done in less than a few tenths of a second. For other applications, e.g. a fully

context-adaptive SDR, reconfiguration will need to be done in real-time without disturbing

any operation of the radio system.

4.2.1 Static Processing Elements and Tailored Functional Arrays

In a non-SDR device, the computationally demanding and often highly parallel parts of an

algorithm would typically be implemented as logical circuitry in an Application-Specific

Integrated Circuit (ASIC). This is often regarded as the optimum solution for computation

efficiency and power consumption, and is typically regarded as a reference solution, which

the reconfigurable solutions may be compared against. ASIC solutions may provide low unit

costs for very high production volumes, but with high development costs.

ASIC implementations are static and as such not usable in an SDR. SDR approaches

that focus on the advantages of ASIC implementations, and on the fact that waveforms tend

to have common modules, have however been suggested. It has been pointed out [68] that

CMOS ASIC devices with more total logic than alternative Field Programmable Gate Arrays

(FPGAs), have significantly less quiescent power and dynamic power consumption. Taking

this into account, it is further suggested [68] it is beneficial for any design to evaluate the

waveforms and determine which functions are common across waveforms, which then could

be hosted in an ASIC to allow a low power implementation.

Related suggestions are also discussed in [69], where the processing platform is constructed

of interconnected application-domain tailored functional units.

Along the same line, some commercial signal processors are having coprocessors specifically

tailored for specific functions [70].

The functional units may be parameterized to adapt to the specific waveform application.

By changing parameters and switching functional units in and out, fast reconfiguration

within the solution space provided by the functional units is available.

It can be argued that such ASIC-hosted modules or tailored functional units will have

a negative impact on portability of waveform applications, limiting portability to platforms

that have the necessary ASIC implementations or functional units. Against this it can be

argued that having alternative software implementations of the same functionality for more

general processing elements, and by allowing the deployment manager to make intelligent

decisions on whether to utilize the ASIC-hosted modules, tailored functional units, or more

112


4. Challenges and Opportunities Related to Computational Requirements of SDR

general processing elements, portability will still be achieved. Still, for part of the SDR

community, SDR platforms with ASIC-hosted processing modules will not be considered

true SDR ones.

4.2.2 Reconfigurable Processing Elements

The Field Programmable Gate Array (FPGA) is the reconfigurable alternative to the ASIC.

At the expense of higher power consumption and circuit area than the corresponding ASIC

solution, an FPGA can be field-programmed with the specific code needed for the specific

waveform application. Reconfiguration times may become as low as fractions of a second

or just some milliseconds, and hence may allow reconfiguration of the SDR unit to connect

to a network via a different waveform standard, for example.

FPGAs have become computationally powerful circuits, and come in many variants. An

additional advantage of the FPGAs is the rich availability of toolsets. Further, the amount of

designs done for FPGAs and the amount of know-how about this type of designs in typical

electronics development organizations, make designs for FPGAs easily planned development

tasks with low or moderate risk. Also, compilers are being promoted that make it

possible to generate FPGA code directly from Matlab or Simulink, or directly from c-code.

This type of compilers can be used for applications such as accelerating bottleneck pieces of

c-code running on a GPP or DSP, by converting them to FPGA code. An example is Altera’s

Nios II C2H compiler which is described in [71].

4.2.3 Fast Reconfigurable Units

Configurable Computing Machines (CCM) offer shorter reconfiguration times than FPGAs,

and for some types close-to real-time reconfiguration. CCMs are ’customizable FPGAs with

a coarser granularity in its fundamental composition that is better suited for signal processing

or wireless applications’ [72]. CCMs have application-domain tailored processing units,

connected via a highly flexible and fast reconfigurable fabric.

There is a huge variety of proposed CCMs, both academic initiatives and commercial

products. [72] and [3] provide overviews of different CCMs.

CCMs may seem ideally suited for SDRs that need high performance and fast reconfiguration.

A disadvantage, however, is the diversity in approaches, which makes efforts to use

them very much a unique effort for each type. This also reduces the availability of SW tools

for programming them.

4.2.4 Real-Time Reconfigurable Units

Microprocessor systems are processing alternatives that provide full real-time programmability.

Year after year, processors have shown remarkable performance increases. Whereas

cycle clock increases have become more difficult due to technological barriers and also less

desirably because of power consumption and power density concerns, the average number

of instructions processed per clock cycle has increased by various means. This is often the

result of having more features operating in parallel within the processor. This has advanced

into multiple-core processors, e.g. multi-core GPPs, multi-core DSPs and Massively Parallel

Processor Arrays. Notably, the trend towards parallel processing within microprocessor

technology fits well with the characteristics of DPC SDR components.

113


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

While GPPs are designed for good average performance for a wide variety of program

sources, DSPs have features specifically targeted for digital signal processing, e.g. combined

multiply-accumulate operations, and including features to exploit parallelism [73]. There

are DSPs that are optimized for performance, and others that are optimized for low power

consumption, e.g. for battery-driven applications.

Most multi-core processors may be classified as Single Instruction Multiple Data

(SIMD), Multiple Instruction Multiple Data (MIMD), or as a combination of the two.

SIMD units have a single instruction stream, i.e. they execute a single program and each

processor is constrained to executing the same instruction. They operate on multiple data

streams at a time, i.e. each processor may operate on a different set of data. They are very

well suited for algorithms where there is high data parallelism, i.e. a number of identical

operations that can or need to be performed at the same point in time, on multiple data sets.

An example of a unit that has several SIMD processing elements is the Cell processor

[74]. The peak processing power is quoted at an impressive > 256 GFlops [75], however the

corresponding power consumption is not stated, in a chart comparison with other processors

[67] it is located at approximately 50W. Although lower-power versions have been released

[76, 77], these presumably still have a somewhat high power consumption when considering

battery powered applications.

Examples of units that have SIMD processing elements and that are targeted for lowpower

consuming units, are the NXP Embedded Vector Processor (EVP) [78, 79], the Sandbridge

SB3011 [80] and the Icera Livanto [81]. An example university approach is the

SODA architecture [67].

It is argued that a solution to the performance/power challenge of the fourth generation

communication standards, is an increased number of cores, with each core including a very

wide SIMD processor, hence exploiting the parallelism of the algorithms [82].

MIMD units have multiple instruction streams, and operate on multiple data streams at

a time. This is hence a more general and flexible architecture than SIMD, allowing separate

programs to be executed on each core. This allows problems that exhibit some parallelism,

but where the parallelism does not have a regular structure in the form mentioned for SIMD

above, to be speeded up through parallel execution.

Graphical Processing Units (GPUs) may also be used for accelerating or running signal

processing algorithms [83]. Recent GPUs for the gaming market are powerful computing

units with a number of parallel processors, e.g. the nVidia 8800 GTX has 128 floating

point units running at 1.35GHz, and an observed 330GFlop/sec [83]. As the processors

are targeted specifically at graphics related processing, they have a higher user threshold

for general purpose or signal related processing than a GPP. However, due to the attractive

prices and high computational capacity of GPUs, they may become attractive in low-budgettype

PC-based SDR solutions, as accelerators for processing-intensive blocks. GPUs may

also have importance for SDR as high-volume massively parallel computation technology

that may be specifically tailored to SDR applications.

4.2.5 Processing Elements, Concluding Comments

With the wide variety of types of processing elements available, how do they compare and

which ones should be preferred?

The answer depends on the individual weighting of a number of factors, and hence no

easy answer is available. In addition to the processing performance the factors of reconfig-

114


5. Security Related Challenges

urability time, power consumption, size, weight, cost, suitability for the actual processing

load and probably many more factors need to be taken into account to make proper choices.

Unfortunately there is no unified scale by which the processing capacity of the different

types of processing elements can be judged. For the various processing elements there

are different figures of merit that are promoted, examples being MIPS, MOPS, MMACS,

MFLOPS and so on. An interesting approach is described in [84] where various elements

(ASIC, FPGA, DSP, GPP) are evaluated and comparative charts provided. A specific FFT

is used as a benchmark algorithm and Real-Time Bandwidth (RTBW) as a scale, defined as

the maximum equivalent analogue bandwidth that the unit is able to process with the given

algorithm, without losing any input information. Still these types of comparisons are only

valid for the particular benchmark algorithm. Also they do not indicate the capacity of the

unit for running other algorithms in parallel, e.g. in an FPGA case.

Guidance may also be obtained from commercially available benchmark analysis reports

from Berkeley Design Technology, Inc. (BDTI) [85].

In order to give an indication of what may be achieved with the different types of elements,

Table A.3 provides some examples of implementation results from various published

work [79, 80, 86, 87].

4.3 Requirements versus Capacity, the Way Ahead

With the continuous improvement of processing elements, will having adequate processing

power in handheld SDR terminals soon be a non-issue? To make some projections, and

inspired by [88], the data rate evolution of mobile cellular systems [89] has been plotted

against the DSP performance evolution of Texas Instruments (TI) DSPs [70, 90], see Figure

A.4. Estimates of the single channel processing requirements of 2G/3G mobile systems [2],

and for a particular 4G case (conditions in [91]) are also plotted.

For this example, the throughput download data rate in mobile cellular terminals increases

at a higher exponential rate than the exponential rate of the DSP processing capacity

in MIPS. The required processing rate increases at an even higher pace, this being due to the

algorithmic complexity increasing with the generations.

While the progress of processing element capacity continuously makes it easier to meet

the capacity requirements of today’s existing waveforms, the rapid system evolution particularly

in the civilian mobile communications sector indicates that providing adequate processing

power at target power consumption will remain a challenge in the years to come, and

there will be an increasing need for data processing elements that further exploit parallelism.

5 Security Related Challenges

The flexibility benefits of SDR at the same time causes challenges in the security area, both

for developers and security certification organizations. In the following the most important

of these security related challenges will be reviewed along with important research contributions

in these areas, and a summary of the remaining difficulties.

115


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

TABLE A.3: Waveform Implementation Examples for Different Processing Elements (references,

see text)

Proc.

element

class

FPGAs

and

CPUs

SIMD

SIMD

CCM

32-bit

VLIW

DSP

Source

type

and

2x XiLinx

XC2V4000,

6000, 8000

2x CPUs (unspecified)

430

MIPS

Sandbridge

Sandblaster

SB3011

NXP EVP

Stallion

(VirginiaTech)

TI C6201

Implementation(s)

and result(s)

802.11a

W-CDMA

WiMAX, 2.9Mbps:

80% utilization

802.11b 11Mbps: 55%

utilization

W-CDMA 2.4Mbps:

87% utilization

@0.5W

Estimated approx 45%

utilization for HSDPA

@ few hundred milliwatts

A benchmark suite of

W-CDMA algorithms

Computations/sec:

7118 @0.7W

A benchmark suite of

WCDMA algorithms

Computations/second:

10293 @1.94W

Pub.

year

2004

2007

2005

2004

2004

5.1 Software Load and Protection against Unauthorized SW

A major security challenge is introduced through the possibility to load and install new SW

on an SDR unit [92], possibly also over-the-air [23, 92–94] or via a fixed network [94]

connection, and the consequent threat of having unauthorized and potentially malicious SW

installed on the platform. This problem domain is very similar to that of maintaining SW

installations on personal computers, and avoiding unintended or malicious functions to be

installed. With SDRs, the consequences of unauthorized code can be even more far-reaching,

from compromising threats to the user’s assets, e.g. his confidential items, via threats to the

communication ability of the equipment, to threats to other users and networks, e.g. by the

SDR jamming other radio activity [95]. In the USA, SDRs are required by the FCC to have

the means to avoid unauthorized SW [96], the specifics of these means are however left to

the manufacturer.

If the SW is downloaded over the air, this also exposes the system for someone illegally

obtaining the SW (privacy violation) or altering the SW while in transport (integrity

violation).

116


5. Security Related Challenges

Data Rate and MIPS Requirements for Mobile Cellular Systems vs DSP

MIPS Evolution

10000000

4G

1000000

100000

kpbs and MIPS

10000

1000

3G

Data Rate (kbps)

TI DSP (MIPS)

Comp. req. (MIPS)

Exp. Trend DSP

100

10

2G

1

1980 1985 1990 1995 2000 2005 2010 2015

Year

FIGURE A.4: Data rate in kbps of the download channel of mobile cellular systems, and approximate

MIPS requirements, plotted against the performance evolution of TI DSPs.

Several publications describe the preventing of unauthorized code by using Digital Signatures

[97–104]. The manufacturer (or any other party authorizing the code) computes a

one-way hash of the code module, then encrypts this hash code using their private key of

a private-public asymmetric key pair. This encrypted hash is the digital signature which is

added to the code module before it is sent to the SDR platform. A verification application

on the SDR platform then verifies the signature by decrypting the signature using the manufacturer’s

public key, and checks that the decrypted signature equals the one-way hash of

the code module. A Digital Certificate is a way of assuring that a public key is actually from

the correct source. The Digital Certificate is digitally signed by a trusted third-party. The

trusted third-party is verified through a chain of trust to a root certificate on the platform.

A critical issue with the above approach is that root certificates must be distributed to

all terminals through a secure out-of-band channel. With a new platform this is easy as root

certificates may be factory installed or installed through physically delivered SW, however

the issue becomes important when a certificate has expired when the terminal is in the field.

A further issue is that of revocation of certificates. A certificate can be revoked at any point

in time, and in order for the terminal to know if this is the case or not, it needs to check

against a revocation registry, for each and every download operation. It is well known from

general computing that this pattern of action is not always obeyed.

Another published way of providing code authorization relies on the sharing of a secret

between the SDR platform and the manufacturer. The manufacturer may then make a oneway

hash of the SW code and the shared secret and send this hash to the SDR platform

together with the SW code [102]. The SDR platform may then verify that the code is from

the manufacturer by doing the same one-way hash and compare. A negative implication of

this approach is that if the secret is a unique key for each SDR platform, there will be a high

number of keys to administrate for the manufacturer. On the other hand, if it is a single key

with a wide distribution, this makes it more susceptible to be compromised.

Trusted Computing (TC) functionality [95] is also an optional way to address threats

against an SDR platform and against downloaded software. A trusted platform subsystem

117


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

has a ’trusted component’ integrated into the platform, which is immutable, i.e. the replacement

or modification is under the control of the platform manufacturer. The trusted part may

be used for integrity measurements of a program, and for creating certified asymmetric key

pairs for the software downloading [95]. TC is further commented in Section 5.2.

A suggested further barrier against potentially malicious code is the pre-running of the

new SDR component in a ”sandbox” [99, 104], a sheltered environment where it can be evaluated

without posing threats to the actual system. Ordinary personal computer protection

means as virus protection [92, 103] and memory surveillance [103] may be further barriers,

as may also radio emission monitoring [99]. The efficiency of the sandbox pre-running may

be debated, as there is no guarantee that the malicious code will expose its behaviour in this

test.

The authorization schemes described above also provide integrity protection of the code

while in transit. Privacy protection, i.e. protecting the code in transit from being disclosed

to a third party, may be achieved through encrypting [98, 105, 106] the code and including

the digital signature.

An SDR ideally should have exchangeable cryptographic algorithms too. A motivation

for exchangeability of cryptographic components is that even if a current security evaluation

does not reveal any weaknesses of some cryptographic approach, cryptanalysis techniques

developed later may render it insecure [98, 103, 105]. The cryptographic components are

in [98, 103, 105] viewed as a matrix with columns for hash algorithm, digital signature primitive,

crypto cipher, secret key and public key, and with rows for the entries for alternative

cryptographic components. It is assumed that there is a minimum of two alternatives for

each of the crypto components, such that even if there is one that is compromised, there is

one that is secure that can be used in the downloading process. Any weak cryptographic

component, e.g. a crypto algorithm, is downloaded using trusted crypto components from

the matrix, and in an automatic manner.

A solution utilizing Altera Stratix III FPGAs has been described [107]. The FPGA

configuration bit stream is transferred to the FPGA in Advanced Encryption Standard (AES)

encrypted form, with the FPGA containing a crypto key and a decryption module that allows

it to decrypt the configuration bitstream when it is loaded into the circuit. Once inside the

circuit, the configuration file cannot be read back [107].

In summary, many research contributions have been made in the area of software download,

providing a menu of technological options that can be used. Still, there is increased

potential for security threats to SDR systems versus non-reconfigurable ones. With the anticipated

creativity of attackers, the specific solutions for the specific SDR systems, and ensuring

these resist potential threats, will remain a challenging area for both developers and

security organizations. Any allowed downloading of security components will be particularly

challenging. Also, the question of who will authorize the SW remains. Should it be the

hardware manufacturer, the SW company, a third-party certifier, a government institution,

or all the above mentioned. This remains an issue that needs to be further matured.

5.2 Trusted and High-Assurance Systems

Many communication systems, in particular military ones, have high-assurance security requirements.

Demonstrating such high-assurance security on a fully flexible and general

computing platform is a very difficult task. This contrasts that the fully flexible platform,

118


5. Security Related Challenges

where all the functionality is defined in SW applications only, is the ideal computing platform

for an SDR in terms of portability.

The high-assurance SDR system will have certain assets, like crypto keys, the user’s

plain text messages, his/her personal information and more, that need to be protected e.g.

for confidentiality and integrity. Hence practical strong security solutions typically employ

combinations of hardware solutions and software solutions [103]. Examples of modules that

have impact on the hardware structure are the protected storage for crypto keys, the protected

storage for the crypto-algorithm, and the separation between black (encrypted) data and red

(plain-text) data (Figure A.5). Such architectures are typically custom and dedicated.

Black data

Black side application

A

P

I

Crypto

A

P

I

Red side application

Red data

Control

Control

A

P

I

Security Control

A

P

I

Control

FIGURE A.5: An illustration of a red (plain-text) to black separation barrier, and the data and control

interfaces to the security related modules.

As mentioned, TC, standardized by the Trusted Computing Group [108], incorporates

dedicated immutable hardware elements. The immutable elements are in this case the Root

of Trust for Measurement (RTM), Root of Trust for Storage (RTS) and the Root of Trust for

Reporting (RTR) [109]. These enable measurements of SW on the platform to be made so

that changes can be detected, confidentiality-protection of data, and allow an outside challenger

to assess whether the platform is trustworthy. TC also facilitates isolation between

different SW components on the platform.

While being developed originally for general computing platforms and PCs, the benefits

of TC on SDR platforms have been pointed to [95], and there is also a TCG Mobile

Phone Work group in existence [108]. For SDR, TC can provide authenticated download,

ensure access to SW only by the intended recipient, detection of malicious or accidental

modification or removal, verification of the state that the platform boots into, and isolation

of security-critical software [95]. TC functionality does not ensure the integrity of securitycritical

software while in storage, and does not prevent denial-of-service attacks in the form

of deletion of the downloaded SDR software [95]. While TC has many beneficial properties

for use in SDR, it does not cover every relevant aspect of security for SDR, and hence in

itself is not sufficient for a high-assurance SDR system.

Modern military information and communication systems often need to handle information

at different security classification levels simultaneously. With conventional security

architectures, this requires multiple sets of security HW and processors, which implies both

high cost and high power consumption. Examples [110] have also shown that with the conventional

security architectures it is demanding to certify the SW security core to the highest

assurance levels, as this part of the SW and hence the certification task often grows too large

to handle at the highest certification levels.

119


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

The Multiple Independent Levels of Security (MILS) architecture offers solutions to

these issues. According to [111], the earliest references to MILS are NSA internal papers

by Vanfleet and others from 1996. At the basis of the architecture is a Separation Kernel

(SK), as outlined by John Rushby in 1981 [112]. The SK allows several independent partitions,

e.g. partitions handling different security levels (Figure A.6), to run on the same

microprocessor, through separating them in space and time [110]. The SK provides data

separation, in ensuring that the different partitions operate on physically separated memory

areas. It also provides authorized communication channels between the partitions. Further

it provides for sanitization, i.e. the cleaning of shared resources (e.g. registers) before a new

partition can use them. Finally it provides damage limitation, in that an error in one partition

is not able to affect the processes in the other partitions. The SK schedules processing

resources to each partition in such a way that the partition will always be able to process its

tasks, independently of any behaviour in the other partitions. The SK is the only piece of

SW in the architecture that is allowed to run in supervisor mode, all the partitions run in user

mode, which means under no circumstance can they alter the SK.

Application

Application

Application

Secret

Confidential

Restricted

Middleware

Middleware

Middleware

PCS

Separation Kernel (SK)

FIGURE A.6: The basic building blocks of the MILS architecture.

Since the SK only includes the limited functionality as described above, it can be made

fairly small, about 4K lines of code has been quoted [113]. This makes it manageable to

certify the SK to the highest Evaluation Assurance Levels (EAL) levels in the Common

Criteria [114], EAL 6 and EAL 7. Green Hills Inc recently announced the completion of

the certification of their Integrity-178B SK-based OS as the worlds first certified at EAL6+

[115]. It should be noted also that real-time OSes using SK principles have been used for

some years already, probably with some more functionality than the minimum-size SK, in

the aviation industry, in other embedded applications and in SDR applications.

The SK requires a certain amount of support from specific HW [116], the most important

function being the Memory Management Unit (MMU), needed for physical memory

separation. A remarkable advantage with the architecture is that the specific HW support

needed is already available in many commercial microprocessors [116].

120


5. Security Related Challenges

The individual partitions include application code and middleware. Middleware in MILS

has a wider interpretation than the conventional one, it includes both traditional communication

oriented middleware such as CORBA, and may also include more OS-related functions

[110]. Some partitions may be Single-Level-Secure (handles data at one security level)

while others may be Multi-Level Secure (for example a module that downgrades information

from one security level to a lower one, through filtering or encryption) [116].

Application and middleware in a partition are to be security evaluated at the appropriate

level for that partition, and without needing to take into account the other partitions in the

system. This separates the problem of evaluation and certification, and assures that no part

of the system needs to be certified at a higher level than what is needed for the security level

of the information it is to handle.

Obviously the control of the information flow between the partitions is an important

part of the security in a MILS system. The allowed information flow may be planned as

a directed graph [110], specifying which partitions are allowed to exchange information in

which direction. Within a single processor the information flow is moderated by the SK. In

a distributed system with multiple processors involved, the information flow moderation is

facilitated through the Partitioning Communication System [113] (PCS).

A vision for MILS, and one that potentially could significantly reduce the amount of

time spent on evaluations, is that of a ”compositional approach to assurance, evaluation

and certification” [117], enabling security evaluation of a system to be based on previous

evaluations of the components that it is composed of.

The disadvantages with the MILS architecture include the higher memory consumption

due to the partition allocated memory areas, the less dynamic processing performance exploitation

due to the need to guarantee processing resources to all partitions, and the higher

cost of context switches due to the increased number of separate processes and due to the

sanitization operations needed. With the advances in processors and memory devices, these

disadvantages have become less important over time.

MILS is a very promising security architecture for SDR, offering a security-domain flexibility

that goes hand-in-hand with the waveform application flexibility desired for SDR.

MILS for SDR is still a work-in-progress, for example in terms of certifications of components,

and it will be very interesting to follow the advances in this area.

In summary, it is challenging for an SDR system to obtain the best possible compromise

between high-assurance security and having a computing platform that is as flexible and

general as possible. The MILS architecture is very promising in this context, and additionally

offers cost-effective handling of multiple security levels and a compositional approach

to certification. The further availability of certified MILS components will strengthen its

position.

5.3 Portability of Security Related Modules

As a way to achieve interoperability between secure radio systems, for example in military

coalition operations, portability of security SW modules is a highly desired feature.

Code portability between platforms with conventional security architectures requires that

the APIs to the security related devices and services are standardized. In many cases each

user domain (e.g. a nation) will be reluctant to disclose security features and APIs, due to the

fear that the information may be useful for organizations that wish to develop threats against

the type of SDR platform. The issue of whether security features should be disclosed or kept

121


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

secret (”security by obscurity”) is however a debated one. FCC stated in a final rule [118]

that ”manufacturers should not intentionally make the distinctive elements that implement

that manufacturer’s particular security measures in a software defined radio public...”. SDR

Forum in their response [119] pointed to that ”History repeatedly has shown that ”security

through obscurity” often fails, typically because it precludes a broad and rigorous review

that would uncover its flaws”.

A possible way ahead is to design the security features and the security APIs in such a

way that making the security APIs public does not increase the vulnerability of the platform.

Another potential solution is having dual security APIs, an intra-domain API and another

inter-domain API, where only the inter-domain API is disclosed outside of its own domain.

With the current (2.2.2) [25] version of the SCA, neither the security requirements nor

the security APIs have been openly published, making portability between different development

domains difficult. The ESSOR project aims at providing what it terms a ’common

security basis to increase interoperability between European forces as well as with the United

States’ [16]. It remains to be seen though, if ESSOR will define the needed security parts

for the whole of the European domain or which part, and whether this will contribute in any

way to portability with US platforms.

MILS-type architectures are the most promising developments for providing drastic reductions

in technical obstacles for portability of security code. Since the MILS-specific HW

requirements are already present in many commercial microprocessors, this enables different

platforms to provide compatible environments for MILS-type security code. It should be

noted though that in the case of implementation-specific additional bindings to non-standard

devices this would give similar concerns as discussed earlier.

In summary, the lack of interdomain security APIs and security feature documentation

is presently a major challenge and obstacle for SDR application portability. Ongoing initiatives,

e.g. ESSOR, are likely to improve this situation by providing complementing standards.

MILS-type security architectures have the potential of greatly reducing technical

obstacles for portability of security code, such that the dominant issue will be that of trust

between organizations and the willingness to share crypto algorithms or having available

coalition algorithms (such as the ”Suite B” initiative [120]) and security related code. Thus

MILS potentially forms an important part of the solution for exchanging secure operational

waveforms between nations and thereby achieving multination interoperability in the battlefield.

6 Regulatory and Certification Issues

In the following, certification challenges with SDR equipment are reviewed. The remaining

issues are pointed out.

6.1 SDR Certification

Traditionally, radio equipment has been approved with the specific frequencies, bandwidths,

modulations and with specific, and fixed, versions of functionality. This certification regime

is challenged when the future application waveforms of the equipment are not known at the

time of shipment, and it must be expected that a specific radio platform is updated with

122


6. Regulatory and Certification Issues

new SW versions several times during its lifetime, or even, in future systems, reconfigured

dynamically according to communications needs.

6.1.1 SDR Certification in the USA

In the USA, significant steps have already been taken in changing certification rules to accommodate

SDR equipment. The FCC in the USA adopted rule changes on 13 September

2001 [96] that defined SDRs as a new class of equipment. With the previous rules any

changes to output power, frequency or type of modulation implied that a new application

form and a new approval would be needed and the equipment be re-labelled with a new identification

number. With the changed rules, updates to the software that affected the output

power, frequency or type of modulation could be handled in a more streamlined approval

process referred to as a ”Class III permissive change”, provided the equipment originally

had been approved as an SDR. An important requirement introduced was that ”manufacturers

must take steps to prevent unauthorized software changes”. The concept of electronic

labelling of equipment was also introduced.

The certification rules were further updated on 10 March 2005 [5]. Under these rules,

radio equipment that has SW that affects the RF operating parameters, and where this SW

is designed to or expected to be modified by a party other than the manufacturer, is required

to be certified as an SDR. One of the reasons for this change was a fear that third-party SW

modifiable radio equipment could otherwise be declared non-SDR, and hence would not be

required to have protection against unauthorized software.

These updated rules also define SDR radios as where ”the circumstances under which

the transmitter operates in accordance with Commission rules, can be altered by making a

change in software”, which points to the conditional use of spectrum, modulation or output

power, as given in FCC regulations.

Certification is required to be carried out at FCC labs, no self-certification or certification

by Telecommunications Certification Bodies (TCBs) is allowed.

Further information on SDR certification may be found in [121]. A related standardization

effort is IEEE 1900.3 [122].

6.1.2 SDR Certification in Europe

In Europe, steps have also been taken that allow faster certification of reconfigured radio

equipment. Whereas previous processes demanded independent type approval processes in

test houses, the Radio and Telecommunications Terminal Equipment Directive (R&TTE)

that came into force in April 2000 in Europe allows self-certifications for telecommunications

equipment, yielding faster update cycles when reconfiguration of equipment is

needed [123]. Work on making specific adaptations to the R&TTE directive to accommodate

SDR products has been carried out by the TCAM Group on SDR (TGS), where TCAM

is the Telecommunications Conformity Assessment and Market Surveillance Committee of

the European Union. A final report was presented from TGS in 2004. Based on this and

further discussion in TCAM, the European Commission has drawn current preliminary conclusions

[124]. Further conclusions have been expected from TCAM, but as of November

2008, none have been drawn up. It is expected that this process will continue. It has been

suggested that a specific harmonized standard for SDR in Europe is to be developed [124].

123


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

The TGS report identifies ’responsibility for the product’ as a key issue, such as when

third-party SW is installed on the equipment. The need for a more flexible marking, e.g.

digital, is concluded. The need for safeguarding the equipment against unauthorized SW is

a discussion point, but currently, unlike in the USA, the manufacturer is not responsible for

unauthorized code installation [124].

Further perspectives on regulatory aspects in Europe may be found in [123–125].

6.1.3 Remaining Issues and Projected Evolution of SDR Regulatory Certification

It is clear from the above that the regulatory certification aspects of SDR need to be further

matured, both in the USA and in Europe.

The FCC lab certification approach in the USA is likely to become an overwhelming

task as the number of products increase and with the product complexity and the amount of

functionality in an SDR. Even the Class III procedure for SW updates is likely to saturate

with future highly reconfigurable equipment with a vast amount of possible SW combinations.

Self-certification in the form of manufacturer Declaration of Conformity, combined

with process and organizational certification of the manufacturers to ensure their capability

of self-certification, will both allow the tasks to be manageable and give a shorter time to

market.

In Europe, more final conclusions on SDR certification need to be drawn and standards

updated. The concept of a single party being responsible for the equipment is likely to

become increasingly difficult and will need reconsideration.

In both the USA and in Europe, since future dynamic reconfigurable equipment is likely

to have a very large number of possible software application combinations, it is unlikely

that each and every possible combination of software components can be tested on each

and every platform type. An alternative way is to establish trust by testing the components

themselves.

A way of creating a further barrier against non-conformity with applicable regulations

is to include in each product a mandatory regulation policy enforcement SW module that

defines the opportunity window of frequencies, modulations and output power. Additionally,

a policy monitor component may be included, that monitors the spectrum from the SDR in

a periodic manner and issues alarms or closes down transmission when a policy breach is

detected (see Figure A.7).

6.2 SCA Compliance and Domain Certification

In market domains where the SCA specification is used, certification of compliance to the

SCA specification is likely to be a market demand, and important for application portability.

The same applies to other features that are particular to the domain, for example domain

APIs. It is a challenge to establish such certification also outside the JTRS.

JPEO states that it is the Certification Authority for the SCA, and that it will assign one

or more test organizations as the Test and Evaluation Authority [25]. An overview of the

model for certification of JTRS products is provided in [126].

There is a need for architectural certification authorities also in other domains than the

JTRS domain, and in other parts of the world than the USA. For example, since it is likely

that there will be some differences between European standards and the US one, Europe

124


7. Opportunities Related to Business Models and Military and Commercial Markets

Data input

Waveform

application

RF

Policy input

Policy

enforcement

module

Policy

monitor

FIGURE A.7: A way of creating a further barrier against non-conformance to regulations: A policy

enforcement module communicates to the waveform application which regulatory policies apply at

the present location and time, e.g. which frequency range and radiated power is allowed to use. A

policy monitor periodically checks the actual generated waveform for non-conformance, in which

case it will instruct the Policy enforcement module to shut down the transmitter.

will need its own certification facilities. Platform and waveform certification needs as defined

in the WINTSEC project in Europe are discussed in [57]. Referring to the view of

the Finnish Software Radio Programme, a ’European certification network must be operational

about 2013-2014’ [127]. How this will happen and which organizations will have the

responsibility are open issues.

7 Opportunities Related to Business Models and Military

and Commercial Markets

SDR provides new product and market opportunities, and has the potential of changing the

business models in the radio communication industry. Here, these opportunities are reviewed

along with the present status in pursuing these in the military and commercial domains, and

with references to recent publications on this subject. Lastly, projections for the further

development in this area are provided.

7.1 Opportunities in the Military Domain

7.1.1 SW Upgradeable and Reconfigurable Military Radio Communications

Equipment

SDR provides opportunities for having military radio communications equipment which is

SW upgradeable and reconfigurable, possibly even field reconfigurable and reconfigurable

in space deployment [128]. This represents both benefits for users and a considerable market

opportunity for manufacturers.

Since the military domain is characterized by long-lifetime acquisitions, while missions

and technical requirements vary at a faster scale, SW upgradability and reconfigurability

is very much in demand. Additionally, the possibility of contracting the SW updates from

third-party providers may provide more competition and contribute to reduced lifetime-costs

for the military.

125


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

The military SDR domain in the USA is dominated by the JTRS programme [7–9, 129].

JTRS has great focus on standardization and portability, with the JPEO managing the SCA.

The first production contracts for JTRS ’interim’ products were awarded in June 2007 [130],

these are variants that meet some basic JTRS requirements [126] but not all requirements as

laid out in [126]. According to [131], low-rate production start of ’Handheld, Manpack and

Small Form Fit’ and ’Ground Mobile’ radios are scheduled for 2010 and 2011 respectively.

JTRS has evolved to also be a programme that delivers tactical wireless networking [9]

for the US military and thus is an important part of the NCO transformation.

In recent reports [7, 132] the United States Government Accountability Office (GAO)

raises its concern about JTRS, pointing to the risks due to the technology challenges [7, 132]

and the cost of each unit being significantly higher (up to 10x) than the legacy units they

replace [132].

In the military domain in Europe, as in the USA, there is support for and focus on

the SCA. The military domain is dominated by several national [12, 13, 15] projects and

demonstration platform developments and cooperative [16, 133, 134] projects. Sweden has

received its first vehicular mount SCA-based GTRS units [14]. Apart from this it is expected

that major development efforts for volume products will await the architectural outcomes of

the ESSOR [134] project.

A discussion on SDR processing in satellites is provided in [128].

In summary, the opportunities provided by SDR in the military domain are starting to be

exploited in the form of deployed SDR units. Some interim radio types meeting basic JTRS

requirements have been contracted and production and deliveries have started. In Europe,

several projects are ongoing. Still, taking into consideration the challenges discussed elsewhere

in this paper and the concerns raised by GAO, the pace where fixed military radios

are replaced by SCA-certified high-flexibility SDR ones is expected to be slow in the next

few years.

7.1.2 Waveform Library

SDR also provides a possibility for building up libraries of waveform applications. In this

way, SDR platforms may be loaded with the specific applications needed in the scenarios

and operations they are to be deployed in. Libraries can be national ones or coalitional ones.

Library waveforms represent market opportunities for SW companies, as well as units that

will be tradable between organizations.

JTRS is building up a repository of waveform applications for porting onto the various

platforms. In 2006 it was reported that JTRS code had accumulated to 3.5 million lines with

Government Purpose Rights [8].

In Europe the political and business issues of building up a waveform inventory are

difficult, as manufacturers in some cases are the owners of the waveforms and as there are

also many national interests. NATO’s Industry Advisory Group (NIAG) has investigated

”the dynamics and Business Models behind Industrial Contribution of Waveform Standards

and how these may and could change with the advent of SDR technology” [133]. NIAG has

issued its report but it is unfortunately not publicly available.

Availability of advanced communication waveforms for exchange of video and data in

coalitions is a critical issue, which is hampered both by the above-mentioned political and

business issues and due to the JTRS ones being classified as ”US ONLY” [131]. A way of

getting round such political and business issues for coalition waveforms is to develop new

126


7. Opportunities Related to Business Models and Military and Commercial Markets

waveforms through collaborative contributions from nations. COALWNW is an example of

such a multinational cooperative effort [131].

Due to the reasons presented and discussed in Section III, porting efforts are likely to be

required for putting library software onto a specific platform.

It is projected that the trend towards building up libraries of waveform applications for

national and coalitional use will remain an active one, and that this will be a growing market

opportunity for third-party SW companies.

7.1.3 Military Cognitive Radio

SDR, as a base implementation technology, provides opportunities for providing the future

military versions of Cognitive Radio (CR).

Cognitive radio in the military domain is a highly active research field, that generates

considerable interest both as a means for reconfiguring waveforms according to sensed electromagnetic

conditions, and as a means to provide increased spectrum utilization through

dynamic use of spectrum [135]. It is also foreseen that as more and more radios in battlefield

environments will have cognition, and as the cognitive abilities are likely to be used for

both offensive and defensive purposes, cognitive abilities in military radio equipment will

become mandatory. CR will thus be another major driver for the transition to SDR in the

military domain.

The literature on CR and potential use of SDR for CR purposes is overwhelming, a few

recommended sources of information are [135–140].

It is expected that the replacement of military radio systems with smarter CR ones will

represent a continued SDR opportunity for many years forward.

7.2 Opportunities in the Commercial Domain

7.2.1 Multiprotocol Multiband Base Stations

SDR provides an opportunity to switch from conventionally designed cellular base stations

to Software Defined Multi-Protocol Multi-Band (MPMB) base stations [58].

The reconfiguration possibilities provided by SDR MPMBs accommodate future cellular

base station needs, for example:

• the possibility to dynamically add services

• the rapid introduction of new communications standards [141]

• the trend that new communications standards are put into service in a less mature

state than previous standards, implying an increased risk of post-deployment changes

needed [141]

• context-related reconfigurability and the accommodation of the future cognitive terminals

[125, 142]

SDR MPMBs also allow standardization of hardware platforms, which reduces the

amount of capital tied up in hardware inventory. Since the total lifetime cost of the system

is more important than the initial cost, the SDR solution may be preferred even if the initial

127


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

cost of the SDR platform is higher. Also with base stations, increased power consumption

over a conventional design can be tolerated.

At present, cellular base stations are dominated by the traditional non-SDR ones. The

VANU [18] base stations, as well as some recent announcements from Huawei [143] and

ZTE [144], are SDR examples.

Further background on SDR base station opportunities is provided in [58, 142].

The share of SDR base stations compared to the overall number is predicted to grow

significantly from 2010, to become an approximate equal share of the overall number of

base stations in 2016 and continue rising [145].

7.2.2 Mobile Multi-standard Terminals

Mobile Multi-standard Terminals (MMTs) represent another large market opportunity for

SDR. As the number of standards needing to be served [141] by the MMT grows, SDR

will at some point provide a cost advantage relative to a conventionally designed MMT.

Further it provides opportunities for future mobile wireless users to change and personalize

their units by installing additional pieces of waveform software, and upgrade their units

as new standards emerge or as standards are updated. More importantly, with the future

reconfigurable and cognitive radio networks it will be a necessity for the units to be able to

add waveform applications or components dynamically.

MMTs are still almost exclusively using traditional non-SDR designs, utilizing waveform

standard specific integrated HW [58] even if the terminals serve a high number of

waveform standards (e.g. GSM, EDGE, W-CDMA, HSDPA, Bluetooth, WiFi). A multimode

mobile phone with ’software-defined modem’ processing up to 2.8 Mbps has been

demoed [146]. This is possibly an important milestone in the SDR direction. Technical

details about its SW flexibility and power-consumption are however unknown.

MMTs presently are also characterized by a relatively small amount of dominant manufacturers

having a high degree of vertical integration and proprietary solutions, e.g. being

responsible both for the hardware platform and the waveform software, and which have

an interest in maintaining this business model. There are, however, some signs of interfaces

being opened up and value chain restructuring. An obvious observation is the trend

of employing third-party operating systems (e.g. the Symbian OS) allowing third-party user

applications to be loaded. This has an effect in making end users accustomed to adding SW

applications to their units.

So far, the user demand for field upgrading waveform application software on mobile

handsets has been limited, simply due to the fact that the handsets are frequently replaced

(the ’handset replacement’ model [58], since the market is currently driven also by a lot of

other factors than the waveform standards, e.g. improved platform devices such as cameras

and displays. Several authors, however, predict a change to the ’handset service upgrade’

[58] and ’personalization’ [147] model where a ’naked handset’ [58] is uploaded to suit the

user’s needs.

The mainstream MMT evolution into SDR-based design is dependent on several factors,

the most important being power consumption and cost, with the cost trade-off being highly

dependent on the number of waveform standards that the terminal is intended to serve. Due

to these factors being more significant for MMTs than for base stations, the MMTs are predicted

to come after the base station development in the transition to SDR design. However,

since the MMT market has high price pressure, this implies that as soon as the SDR ap-

128


8. Conclusions

proach gives a cost advantage, and assuming acceptable power consumption, there will be a

very significant drive in this direction.

7.2.3 Cognitive Radio

The projected evolution into CR capable MPMBs and MMTs represents a large future market

opportunity and driver for SDR technology.

CRs may both provide context-aware services for the user [148] and improve spectrum

utilization through dynamic spectrum access [135, 137, 149]. In order to continuously take

advantage of spectrum opportunities and adapt to the specific context, CR requires platforms

that have fast dynamic reconfiguration abilities. Recommended sources of information on

CR and the application of SDR in CR systems are [136, 138].

While the first commercial-domain CR standard is already drafted [150], more advanced

CRs are viewed as being further into the future.

7.2.4 Other Commercial Domain Opportunities

Commercial Satellite Communications has already been mentioned as a segment that will

benefit from SDR, where SDR enables remote upgrades and possibly multiple uses during

the lifetime of a satellite [17]. Equipment to be located in remote and poorly accessible

locations on earth is another similar opportunity.

The Femtocell or Home Base Station has been put forward as another market segment

with great opportunities for SDR [58, 151]. The reconfigurability and flexibility provided by

SDR support the multiple bands, multiple standards and simultaneous ’sniffing’ functionality

needed in the Femtocells [151]. Other mentioned market opportunities include devices for

laptops, automobiles, home entertainment and the medical and public safety segments [58].

8 Conclusions

Although SDR technology has evolved more slowly than anticipated some years ago, there

are now many positive signs, the clearest ones being in the form of SDR products entering

the market. Several major initiatives, at national and cooperative levels between nations and

the industry are paving the way for SDR.

The increasing availability of SCA SW tools and development platforms is contributing

to reducing the learning threshold of the SCA and also increase the productivity of SDR

development. Developments within Model Driven Design may further increase this productivity.

The SCA eases portability by providing a standard for deploying and managing applications.

Even so the portability of SCA-based applications between different platforms is not

straightforward. One major issue is the standardization of the APIs between the application

and the system devices and services. Although a subset of the needed APIs have been published

on the JTRS website, parts of the APIs will be difficult to standardize across domains,

for example the security-related APIs. Another major issue is the instruction code compatibility

between different processing elements, which at present requires porting efforts in

terms of rewriting code to fit the processing elements of the target platform. It is expected

in the long term that design in higher abstraction languages will reduce this type of porting

effort.

129


A. SOFTWARE DEFINED RADIO: CHALLENGES AND OPPORTUNITIES

Alternatives to the SCA include OMG’s specification, NASA’s STRS architecture and

the GNU Radio architecture. MOM-based architectures have a potential of becoming alternatives,

but the maturity and the acceptance of these specifications have to be demonstrated.

For cognitive radio systems, additions to the SCA, such as middleware that supports

adaptation, will be beneficial. Such middleware will increase productivity and standardize

solutions when making adaptive and cognitive systems.

It is expected that the SCA will remain the dominating architecture in the military sector

where waveform application portability and reuse are major priorities, especially through

cooperative programmes. On the other hand, a significant portion of designs for the civilian

commercial market, where hardware cost is a major factor, are likely to utilize dedicated

and proprietary lighter-weight architectures. In a longer (∼10 years) perspective, and as

hardware cost progressively becomes a smaller part of the total system cost, standardized

open architectures are likely to become more popular also on the civilian commercial market.

A fundamental challenge for SDR designs is that of providing sufficient computational

performance for the signal processing tasks and within the relevant size weight and power

requirements. This is particularly challenging for small handheld units, and for ubiquitous

units. Parallel computation enhancements and the rapid evolvement of DSP and FPGA performance

help to provide this computational performance. Processing units having multiple

SIMD processing elements appear to be very promising for low-power SDR units. Also, as

waveforms typically have many common functions, it may be sensible to make parameterized,

optimal low-power-consumption dedicated hardware blocks for these common functions,

and run alternative source code on a more general processing element if they do not

exist.

The reconfigurability of SDR systems has security challenges as a side effect. One such

security challenge is that the system must be protected from loading unauthorized and/or

malicious code. Also, the rigidity of conventional security architectures in many ways contrast

the desired flexibility and portability ideally required for SDR. The MILS architecture

provides for larger flexibility and easier portability of security related modules, while offering

multiple security-levels without the need for multiple sets of HW.

SDR has forced regulators to rethink the certification of radio equipment. While traditional

equipment has a fixed number of functional modes and a more or less fixed design that

may be fully characterized, SDRs may be SW loaded to function in a large variety of modes

and hence may not be tested in every possible mode at the time of the initial certification.

Changes in certification rules to deal with SDRs have taken place, but it is likely that as more

SDR products approach the market, there will be a further evolvement of these rules.

The multitude of waveform standards and their rapid progress make it beneficial and

economical to be able to easily update wireless network infrastructure equipment, such as

cellular base stations. Also, base stations are less sensitive to the power consumption of the

SDR processing platforms than the mobile devices. Thus SDR has promising potential in

commercial wireless network infrastructure equipment.

SDR has the potential to increase the productivity of radio communication development

and lower the lifecycle costs of radio communication. This will partly come through a

change in the business models in the radio communication industry, allowing a separation

into SDR platform providers and third-party SW providers. This again will provide volume

benefits for the platforms and lower the threshold for companies entering the market as SW

providers, and hence provide further competition in the SDR SW applications area.

SDR will have continued focus as a highly flexible platform to meet the demands from

130


References

military organizations facing the requirements from network centric and coalitional operations.

SDR will also have continued focus as a convenient platform for future cognitive radio

networks, enabling more information capacity for a given amount of spectrum and have the

ability to adapt on-demand to waveform standards.

Acknowledgment

The author would like to thank Christian Serra at Thales Communications, Marc Adrat at

FGAN (the Research Establishment for Applied Science), Jon Olavsson Neset and Stewart

Clark at NTNU (Norwegian University of Science and Technology), Audun Jøsang at UniK

(University Graduate Center at Kjeller), Frank Eliassen at UiO (University of Oslo) and

Torleiv Maseng, Tor Gjertsen, Asgeir Nysæter and Synnøve Eifring at FFI (the Norwegian

Defence Research Establishment) for their valuable input and advice. The author would also

like to thank the anonymous reviewers for their helpful comments.

References

[1] J. Mitola III, “Software radios - survey, critical evaluation and future directions,” in

National Telesystems Conference, 1992. NTC-92, pp. 13/15–13/23, May 1992.

[2] W. Tuttlebee, Software Defined Radio: Enabling Technologies. Chichester: Wiley,

2002.

[3] P. B. Kenington, RF and Baseband Techniques for Software Defined Radio. Boston:

Artech House, 2005.

[4] Software Defined Radio Forum. SDR Forum. (Oct. 2007). [Online]. Available:

http://www.sdrforum.org

[5] Federal Communications Commission, “In the Matter of Facilitating Opportunities

for Flexible, Efficient, and Reliable Spectrum Use Employing Cognitive Radio Technologies,

Report and Order,” Mar. 2005, FCC 05-57, ET Docket No. 03-108.

[6] P. G. Cook and W. Bonser, “Architectural overview of the SPEAKeasy system,” IEEE

Journal on Selected Areas in Communications, vol. 17, no. 4, pp. 650–661, 1999.

[7] United States Government Accountability Office, “Restructured JTRS Program Reduces

Risk, but Significant Challenges Remain,” Sep. 2006.

[8] D. R. Stephens, B. Salisbury, and K. Richardson, “JTRS infrastructure architecture

and standards,” in Military Communications Conference, 2006. MILCOM 2006, pp.

1–5. IEEE, Oct. 2006.

[9] R. North, N. Browne, and L. Schiavone, “Joint tactical radio system - connecting the

GIG to the tactical edge,” in Military Communications Conference, 2006. MILCOM

2006, pp. 1–6. IEEE, Oct. 2006.

131


REFERENCES

[10] G. Gailliard, E. Nicollet, M. Sarlotte, and F. Verdier, “Transaction level modelling

of SCA compliant software defined radio waveforms and platforms PIM/PSM,” in

Design, Automation & Test in Europe Conference & Exhibition, 2007.DATE ’07, pp.

1–6, Apr. 2007.

[11] C. Serra, B. Sourdillat, and E. Nicollet, “SDR key factors - advanced studies & results

related to the implementations of the SCA,” in 2006 Software Defined Radio Technical

Conference and Product Exposition, Nov. 2006.

[12] H. S. Kenyon. Spanish Research Platform Ready For Service. (Sep.

2007). [Online]. Available: http://www.afcea.org/signal/articles/templates/

SIGNAL Article Template.asp?articleid=1383&zoneid=47

[13] T. Tuukkanen, A. Pouttu, and P. Leppȧnen. Finnish Software Radio Programme.

(June 2007). [Online]. Available: http://www.mil.fi/paaesikunta/materiaaliosasto/

liitteet/finnish software radio programme.pdf

[14] Swedish Defence Materiel Organization. A New Generation Radio. (Apr. 2008).

[Online]. Available: http://www.fmv.se/WmTemplates/news.aspx?id=4104

[15] A. Baddeley. Sweden Seeks Military Communications Flexibility. (May

2006). [Online]. Available: http://www.afcea.org/signal/articles/templates/

SIGNAL Article Template.asp?articleid=1129&zoneid=7

[16] European Defence Agency, “Background on Software Defined Radio,” Nov. 2007.

[17] F. Daneshgaran and M. Laddomada, “Transceiver front-end technology for software

radio implementation of wideband satellite communication systems,” Wireless Personal

Communications, vol. 24, no. 2, pp. 99–121, 2003.

[18] Vanu Inc. Vanu’s Software Radio. (2007). [Online]. Available: http://www.vanu.com

[19] R. H. Walden, “Analog-to-digital converter survey and analysis,” IEEE Journal on

Selected Areas in Communications, vol. 17, no. 4, pp. 539–550, Apr. 1999.

[20] J. Mitola, V. Bose, B. M. Leiner, T. Turletti, and D. Tennenhouse, “Special issue on

software radio,” IEEE Journal on Selected Areas in Communications, vol. 17, no. 4,

Apr. 1999.

[21] A. A. Abidi, “The path to the software-defined radio receiver,” IEEE Journal of Solid-

State Circuits, vol. 42, no. 5, pp. 954–966, May 2007.

[22] M. Laddomada, F. Daneshgaran, M. Mondin, and R. M. Hickling, “A PC-based software

receiver using a novel front-end technology,” IEEE Communications Magazine,

vol. 39, no. 8, pp. 136–145, Aug. 2001.

[23] E. Buracchini, “The software radio concept,” IEEE Communications Magazine,

vol. 38, no. 9, pp. 138–143, Sep. 2000.

[24] M. Laddomada, “Generalized comb decimation filters for sigma-delta A/D converters:

Analysis and design,” IEEE Transactions on Circuits and Systems I: Regular

Papers, vol. 54, no. 5, pp. 994–1005, May 2007.

132


References

[25] (JTRS), JTRS Standards Joint Program Executive Office (JPEO) Joint Tactical Radio

System, “Software communications architecture specification,” May 2007.

[26] JTRS Standards Joint Program Executive Office (JPEO) Joint Tactical Radio System

(JTRS). Software Communications Architecture SCA Document Downloads. (Sep.

2007). [Online]. Available: http://sca.jpeojtrs.mil/downloads.asp?ID=2.2.2

[27] Object Management Group Inc. The Software-Based Communication (SBC) Domain

Task Force (DTF). (Sep. 2007). [Online]. Available: http://sbc.omg.org

[28] E. Nicollet, “The transceiver facility platform independent model (PIM): Definition,

content and usage,” in 2006 Software Defined Radio Technical Conference and Product

Exposition, Nov. 2006.

[29] Vecima Networks Inc. Spectrum Signal Processing. (2007). [Online]. Available:

http://www.spectrumsignal.com

[30] F. Humcke, “Making FPGAs “first class” SCA citizens,” in 2006 Software Defined

Radio Technical Conference and Product Exposition, Nov. 2006.

[31] C. McHale. Corba explained simply. (Feb. 2007). [Online]. Available: http:

//www.ciaranmchale.com/

[32] C. Magsombol, C. Jimenez, and D. R. Stephens, “Joint tactical radio system - application

programming interfaces,” in Military Communications Conference, 2007.

MILCOM 2007., pp. 1–7. IEEE, Oct. 2007.

[33] A. Blair, T. Brown, J. Cromwell, S. Kim, and R. Milne, “Porting lessons learned

from soldier radio waveform (SRW),” in Military Communications Conference, 2007.

MILCOM 2007., pp. 1–6. IEEE, Oct. 2007.

[34] S. Bernier and J. P. Z. Zapata, “The deployment of software components into heterogeneous

SCA platforms,” in SDR’08 Technical Conference and Product Exposition,

Oct. 2008.

[35] T. Kempf, E. M. Witte, V. Ramakrishnan, G. Ascheid, M. Adrat, and M. Antweiler,

“A practical view on SDR baseband processing portability,” in SDR’08 Technical

Conference and Product Exposition, Oct. 2008.

[36] Y. Zhang, S. Dyer, and N. Bulat, “Strategies and insights into SCA-compliant waveform

application development,” in Military Communications Conference, 2006.MIL-

COM 2006., pp. 1–7. IEEE, Oct. 2006.

[37] Zeligsoft Inc. Zeligsoft. (2007). [Online]. Available: http://www.zeligsoft.com

[38] PrismTech. PrismTech Productivity Tools and Middleware. (2007). [Online].

Available: http://www.prismtech.com/

[39] Communications Research Canada. SCARI Software Suite. (Aug. 2008). [Online].

Available: http://www.crc.gc.ca/en/html/crc/home/research/satcom/rars/sdr/

products/scari suite/scari suite

133


REFERENCES

[40] VirginiaTech. OSSIE development site for software-defined radio. (Dec. 2007).

[Online]. Available: http://ossie.wireless.vt.edu/trac

[41] S.-P. Lee, M. Hermeling, and C.-M. Poh, “Experience report: Rapid model-driven

waveform development with UML,” in SDR’08 Technical Conference and Product

Exposition, Oct. 2008.

[42] B. Hailpern and P. Tarr, “Model-driven development: The good, the bad, and the

ugly,” IBM Systems Journal, vol. 45, no. 3, pp. 451–461, July 2006.

[43] Z. Jianfan, D. Levy, and A. Liu, “Evaluating overhead and predictability of a real-time

CORBA system,” in Proceedings of the 37th Annual Hawaii International Conference

on System Sciences - 2004, p. 8, Jan. 2004.

[44] F. Casalino, G. Middioni, and D. Paniscotti, “Experience report on the use of CORBA

as the sole middleware solution in SCA-based SDR environments,” in SDR’08 Technical

Conference and Product Exposition, Oct. 2008.

[45] J. Kim, S. Hyeon, and S. Choi, “Design and implementation of high-speed data transfer

protocol in CORBA enviroment,” in SDR’08 Technical Conference and Product

Exposition, Oct. 2008.

[46] A. S. Gokhale and D. C. Schmidt, “Evaluating CORBA latency and scalability over

high-speed ATM networks,” in Proceedings of the 17th International Conference on

Distributed Computing Systems, 1997, pp. 401–410, May 1997.

[47] J. Bertrand, J. W. Cruz, B. Majkrzak, and T. Rossano, “CORBA delays in a softwaredefined

radio,” IEEE Communications Magazine, vol. 40, no. 2, pp. 152–155, Feb.

2002.

[48] T. Ulversøy and J. O. Neset, “On workload in an SCA-based system, with varying

component and data packet sizes,” in RTO-MP-IST-083 Military Communications

with a Special Focus on Tactical Communications for Network Centric Operations,

Apr. 2008.

[49] P. J. Balister, C. Dietrich, and J. H. Reed, “Memory usage of a software communication

architecture waveform,” in 2007 Software Defined Radio Technical Conference

and Product Exposition, Nov. 2007.

[50] D. Paniscotti and J. Bickle, “SDR signal processing distributive-development approaches,”

in 2007 Software Defined Radio Technical Conference and Product Exposition,

Nov. 2007.

[51] D. R. Stephens, C. Magsombol, and C. Jimenez, “Design patterns of the JTRS infrastructure,”

in Military Communications Conference, 2007. MILCOM 2007., pp. 1–5.

IEEE, Oct. 2007.

[52] C. Lee, J. Kim, S. Hyeon, and S. Choi, “FPGA design to support a Corba component,”

in SDR’08 Technical Conference and Product Exposition, Oct. 2008.

134


References

[53] D. Vassilopoulos, T. Pilioura, and A. Tsalgatidou, “Distributed technologies CORBA,

Enterprise JavaBeans, Web services: A comparative presentation,” in 14th Euromicro

International Conference on Parallel, Distributed, and Network-Based Processing,

2006. PDP 2006., p. 5, Feb. 2006.

[54] Object Management Group Inc. The Object Management Group (OMG). (Oct.

2006). [Online]. Available: http://www.omg.org

[55] Q. H. Mahmoud, Middleware for Communications. Chichester: Wiley, 2004.

[56] OMG. Documents associated with PIM and PSM for Software Radio Components

(SDRP), v1.0. (Mar. 2007). [Online]. Available: http://www.omg.org/spec/SDRP/1.0/

[57] S. Nagel, V. Blaschke, J. Elsner, F. K. Jondral, and D. Symeonidis, “Certification

of SDRs in new public and governmental security systems,” in SDR’08 Technical

Conference and Product Exposition, Oct. 2008.

[58] A. Kaul, “Software defined radio: The transition from defense to commercial markets,”

in 2007 Software Defined Radio Technical Conference and Product Exposition,

Nov. 2007.

[59] T. Quinn and T. Kacpura, “Strategic adaptation of SCA for STRS,” in 2006 Software

Defined Radio Technical Conference and Product Exposition, Nov. 2006.

[60] T. J. Kacpura, L. M. Handler, J. C. Briones, and C. S. Hall, “Updates to the NASA

space telecommunications radio system (STRS) architecture,” in 2007 Software Defined

Radio Technical Conference and Product Exposition, Nov. 2007.

[61] J. C. Briones, L. M. Handler, C. S. Hall, R. C. Reinhart, and T. J. Kacpura, “Case

study: Using the OMG SWRadio profile and SDR Forum input for NASA’s space

telecommunications radio system,” in SDR’08 Technical Conference and Product Exposition,

Oct. 2008.

[62] S. B. Raghunandan, D. Kumaraswamy, L. Le, C. B. Dietrich, and J. H. Reed, “Open

space radio: An open source implementation of STRS 1.01,” in SDR’08 Technical

Conference and Product Exposition, Oct. 2008.

[63] Free Software Foundation Inc. GNU Radio - The GNU Software Radio. (Mar. 2007).

[Online]. Available: http://www.gnu.org/software/gnuradio/index.html

[64] G. Abgrall, F. L. Roy, J.-P. Delahaye, J.-P. Diguet, and G. Gogniat, “A comparative

study of two software defined radio platforms,” in SDR’08 Technical Conference and

Product Exposition, Oct. 2008.

[65] E. Gjorven, F. Eliassen, K. Lund, V. S. W. Eide, and R. Staehli, “Self-adaptive systems:

A middleware managed approach,” Self-Managed Networks, Systems, and Services,

Proceedings, vol. 3996, pp. 15–27, 2006.

[66] A. P. Vinod and E. M. K. Lai, “Low power and high-speed implementation of FIR

filters for software defined radio receivers,” IEEE Transactions on Wireless Communications,

vol. 5, no. 7, pp. 1669–1675, 2006.

135


REFERENCES

[67] L. Yuan, L. Hyunseok, M. Woh, Y. Harel, S. Mahlke, T. Mudge, C. Chakrabarti, and

K. Flautner, “SODA: A low-power architecture for software radio,” in 33rd International

Symposium on Computer Architecture, 2006. ISCA ’06., pp. 89–101, 2006.

[68] John L. Shanton III and H. Wang, “Design considerations for size, weight and power

(SWAP) constrained radios,” in 2006 Software Defined Radio Technical Conference

and Product Exposition, Nov. 2006.

[69] J. Bjorkqvist and S. Virtanen, “Convergence of hardware and software in platforms

for radio technologies,” IEEE Communications Magazine, vol. 44, no. 11, pp. 52–57,

Nov. 2006.

[70] Texas Instruments Incorporated. Texas Instruments. (Oct. 2008). [Online]. Available:

http://www.ti.com

[71] D. Lau, J. Blackburn, and C. Jenkins, “Using c-to-hardware acceleration in FPGAs

for waveform baseband processing,” in 2006 Software Defined Radio Technical Conference

and Product Exposition, Nov. 2006.

[72] S. Srikanteswara, R. C. Palat, J. H. Reed, and P. Athanas, “An overview of configurable

computing machines for software radio handsets,” IEEE Communications

Magazine, vol. 41, no. 7, pp. 134–141, July 2003.

[73] D. Efstathiou, L. Fridman, and Z. Zvonar, “Recent developments in enabling technologies

for software defined radio,” IEEE Communications Magazine, vol. 37, no. 8,

pp. 112–117, Aug. 1999.

[74] J. A. Kahle, M. N. Day, H. P. Hofstee, C. R. Johns, T. R. Maeurer, and

D. Shippy, “Introduction to the Cell microprocessor,” IBM Journal of Research

and Development, vol. 49, no. 4/5, pp. 589–604, Sep. 2005. [Online]. Available:

http://www.research.ibm.com/journal/rd/494/kahle.pdf

[75] The Cell project at IBM Research, The Cell Chip. (Oct. 2007). [Online]. Available:

http://www.research.ibm.com/cell/cell chip.html

[76] J. Pille, C. Adams, T. Christensen, S. Cottier, S. Ehrenreich, T. Kono, D. Nelson,

O. Takahashi, S. Tokito, O. Torreiter, O. Wagner, and D. Wendel, “Implementation of

the CELL broadband engine in a 65nm SOI technology featuring dual-supply SRAM

arrays supporting 6GHz at 1.3V,” in IEEE International Solid-State Circuits Conference,

2007. ISSCC 2007. Digest of Technical Papers., p. 322, Feb. 2007.

[77] O. Takahashi, C. Adams, D. Ault, E. Behnen, O. Chiang, S. R. Cottier, P. Coulman,

J. Culp, G. Gervais, M. S. Gray, Y. Itaka, C. J. Johnson, F. Kono, L. Maurice, K. W.

McCullen, L. Nguyen, Y. Nishino, H. Noro, J. Pille, M. Riley, M. Shen, C. Takano,

S. Tokito, T. Wagner, and H. Yoshihara, “Migration of Cell broadband engine from

65nm SOI to 45nm SOI,” in IEEE International Solid-State Circuits Conference,

2008. ISSCC 2008. Digest of Technical Papers., p. 86, Feb. 2008.

[78] P. Westermann, G. Beier, H. it Harma, and L. Schwoerer, “Performance analysis of

W-CDMA algorithms on a vector DSP,” in 4th European Conference on Circuits and

Systems for Communications, 2008. ECCSC 2008., pp. 307–311, July 2008.

136


References

[79] K. van Berkel, F. Heinle, P. P. E. Meuwissen, K. Moerman, and M. Weiss, “Vector

processing as an enabler for software-defined radio in handheld devices,” EURASIP

Journal on Applied Signal Processing, vol. 2005, no. 16, pp. 2613–2625, Jan. 2005.

[80] J. Glossner, D. Iancu, M. Moudgill, G. Nacer, S. Jinturkar, S. Stanley, and M. Schulte,

“The Sandbridge SB3011 platform,” EURASIP Journal on Embedded Systems, vol.

2007,, Jan 2007.

[81] ICERA. Livanto c○ chipsets. (2008). [Online]. Available: http://www.icerasemi.com/

livanto-chipsets.php

[82] M. Woh, S. Seo, H. Lee, Y. Lin, S. Mahlke, T. Mudge, C. Chakrabarti,

and K. Flautner, “The next generation challenge for software defined radio,”

in Embedded Computer Systems: Architectures, Modeling, and Simulation,

pp. 343–354. Springer Berlin / Heidelberg, July 2007. [Online]. Available:

http://dx.doi.org/10.1007/978-3-540-73625-7 36

[83] M. D. McCool, “Signal processing and general-purpose computing and GPUs,” IEEE

Signal Processing Magazine, vol. 24, no. 3, pp. 110–115, 2007.

[84] F. Stefani, A. Moschitta, D. Macii, and D. Petri, “FFT Benchmarking for Digital

Signal Processing Technologies,” University of Trento, Department of Information

and Communication Technology, Tech. Rep., Mar. 2004.

[85] BDTI. BDTI - Insight, Analysis, and Advice on Signal Processing Technology.

(2008). [Online]. Available: http://www.bdti.com/

[86] H. Harada, “Software defined radio prototype for W-CDMA and IEEE802.11a wireless

LAN,” in IEEE 60th Vehicular Technology Conference, 2004.VTC2004-Fall.,

vol. 6, pp. 3919–3924, Sep. 2004.

[87] J. Neel, S. Srikanteswara, J. H. Reed, and P. M. Athanas, “A comparative study of

the suitability of a custom computing machine and a VLIW DSP for use in 3G applications,”

in IEEE Workshop on Signal Processing Systems, 2004. SIPS 2004., pp.

188–193, Oct. 2004.

[88] S. Rajagopal and J. R. Cavallaro. Communication Processors. (July 2005). [Online].

Available: http://hdl.handle.net/1911/20241

[89] GSM Association. GSM World. (2008). [Online]. Available: http://www.gsmworld.

com

[90] K. Keutzer, “Digital Signal Processors: A TI Architectural History,” 2000. [Online].

Available: http://bwrc.eecs.berkeley.edu/classes/cs252/Notes/Lec10a-DSP1.ppt

[91] M. Etelȧperȧ and J.-P. Soininen, “4G Mobile Terminal Architectures,” Technical

Research Centre of Finland (VTT), Tech. Rep., Sep. 2007. [Online]. Available:

rooster.oulu.fi/materiaalit/4G%20terminal%20architectures.pdf

[92] D. Babb, C. Bishop, and T. E. Dodgson, “Security issues for downloaded code in

mobile phones,” Electronics & Communication Engineering Journal, vol. 14, no. 5,

pp. 219–227, Oct. 2002.

137


REFERENCES

[93] M. Cummings and S. Heath, “Mode switching and software download for software

defined radio: the SDR Forum approach,” IEEE Communications Magazine, vol. 37,

no. 8, pp. 104–106, Aug. 1999.

[94] M. Laddomada, “Reconfiguration issues of future mobile software radio platforms,”

Wireless Communications & Mobile Computing, vol. 2, no. 8, pp. 815–826, Dec.

2002.

[95] E. Gallery and C. Mitchell, “Trusted computing technologies and their use in the provision

of high assurance SDR platforms,” in 2006 Software Defined Radio Technical

Conference and Product Exposition, Nov. 2006.

[96] Federal Communications Commission, “In the matter of Authorization and Use of

Software Defined Radios, First Report and Order,” Sep. 2001, FCC 01-264, ET

Docket No. 00-47.

[97] R. Falk, F. Haettel, U. Lūcking, and E. Mohyeldin, “Authorization of SDR software

using common security technology,” in 2005 Software Defined Radio Technical Conference

and Product Exposition, Nov. 2005.

[98] L. B. Michael, M. J. Mihaljevic, S. Haruyama, and R. Kohno, “A framework for secure

download for software-defined radio,” IEEE Communications Magazine, vol. 40,

no. 7, pp. 88–96, July 2002.

[99] R. Falk, P. Bender, N. J. Drew, and J. Faroughi-Esfahani, “Conformance and security

challenges for personal communications in the reconfigurable era,” in 4th International

Conference on 3G Mobile Communication Technologies, 2003. 3G 2003., pp.

7–12, June 2003.

[100] H. Shiba, K. Uehara, and K. Araki, “Proposal and evaluation of security schemes for

software-defined radio,” in 14th IEEE Proceedings on Personal, Indoor and Mobile

Radio Communications, 2003. PIMRC 2003., vol. 1, pp. 114–118, Sep. 2003.

[101] M. Kurdziel, J. Beane, and J. J. Fitton, “An SCA security supplement compliant radio

architecture,” in Military Communications Conference, 2005.MILCOM 2005., pp.

2244–2250. IEEE, Oct. 2005.

[102] W. T. Scott, A. C. Houle, and A. Martin, “Information assurance issues for an SDR

operating in a manet network,” in 2006 Software Defined Radio Technical Conference

and Product Exposition, Nov. 2006.

[103] D. S. Dawoud, “A proposal for secure software download in SDR,” in AFRICON,

2004.7th AFRICON Conference in Africa, pp. 77–82, Sep. 2004.

[104] R. Falk and M. Dillinger, “Approaches for secure SDR software download,” in 2004

Software Defined Radio Technical Conference and Product Exposition, Nov. 2004.

[105] M. J. Mihaljevic and R. Kohno, “On a framework for employment of cryptographic

components in software defined radio,” in The 5th International Symposium on Wireless

Personal Multimedia Communications, 2002., vol. 2, pp. 835–839, Oct. 2002.

138


References

[106] C. Y. Yeun and T. Farnham, “Secure software download for programmable mobile

user equipment,” in Third International Conference on 3G Mobile Communication

Technologies, 2002., pp. 505–510, May 2002.

[107] C. Jenkins and C. Plante, “Military anti-tampering solutions using programmable

logic,” in 2006 Software Defined Radio Technical Conference and Product Exposition,

Nov. 2006.

[108] TCG. Trusted Computing Group. (2007). [Online]. Available: https://www.

trustedcomputinggroup.org/home

[109] C. Mitchell, Trusted Computing. London: Institution of Electrical Engineers, 2005.

[110] J. , C. Taylor, and P. Oman, “A multi-layered approach to security in high assurance

systems,” in Proceedings of the 37th Annual Hawaii International Conference on

System Sciences (HICSS’04) - Track 9, 2004., vol. 9, Jan. 2004.

[111] B. Randell and J. Rushby, “Distributed secure systems: Then and now,” in 2007 Annual

Computer Security Applications Conference (ACSAC 23), Dec. 2007.

[112] J. Rushby, “Design and verification of secure systems,” in Proceedings of the Eighth

Symposium on Operating System Principles, Dec. 1981.

[113] C. Taylor, “MILS, multiple independent levels of security: A high assurance

architecture,” May 2006. [Online]. Available: http://cisr.nps.navy.mil/guests/

taylor06.html

[114] Common Criteria portal, The official website of the Common Criteria Project.

(2007). [Online]. Available: http://www.commoncriteriaportal.org/

[115] Green Hills Software Inc. Green Hills Software Announces World’s First

EAL6+ Operating System Security Certification. (Nov. 2008). [Online]. Available:

http://www.ghs.com/news/20081117 integrity EAL6plus security.html

[116] J. Alves-Foss, P. W. Oman, C. Taylor, and W. S. Harrison, “The MILS architecture

for high-assurance embedded systems,” International Journal of Embedded Systems,

vol. 2, no. 3/4, pp. 239–247, 2006.

[117] C. Boettcher, R. DeLong, J. Rushby, and W. Sifre, “The MILS component integration

approach to secure information sharing,” in The 27th IEEE/AIAA Digital Avionics

Systems Conference (DASC), Oct. 2008.

[118] Federal Communications Commission, “Cognitive Radio Technologies and Software

Defined Radios,” Jun. 2007, 47 CFR Part 2 [ET Docket No. 03-108; FCC 07-66].

[119] SDR Forum, “SDR Forum Response to FCC MOO,” June 2007.

[120] T. Moore, “Developing an interoperable coalition communication strategy using suite

B,” in Military Communications Conference, 2007. MILCOM 2007., pp. 1–4. IEEE,

Oct. 2007.

139


REFERENCES

[121] J. Giacomoni and D. C. Sicker, “Difficulties in providing certification and assurance

for software defined radios,” in First IEEE International Symposium on New Frontiers

in Dynamic Spectrum Access Networks, 2005. DySPAN 2005., pp. 526–538, Nov.

2005.

[122] IEEE Standards Association. IEEE Standards Coordinating Committee 41 (Dynamic

Spectrum Access Networks). (Sep. 2007). [Online]. Available: http://www.scc41.org/

[123] P. Martigne, B. Deschamps, K. Moessner, G. de Brito, K. El-Khazen, and D. Bourse,

“Regulatory trends for reconfigurable end-to-end systems,” in IST Summit 06, June

2006.

[124] P. Bender, “Regulatory Aspects of Software Defined Radio,” Feb. 2007. [Online].

Available: http://www.etsi.org/website/document/workshop/softwaredefinedradio/

sdrworkshop4-1paulbender.pdf

[125] D. Bourse, W. Koenig, M. Doubrava, J.-M. Temerson, K. Nolte, I. Gaspard, K. Kalliojarvi,

E. Buracchini, and D. Bateman, “FP7 EU project: Technical, business, standardization

and regulatory perspectives,” in SDR’08 Technical Conference and Product

Exposition, Oct. 2008.

[126] Joint Program Executive Office Joint Tactical Radio System, “JPEO JTRS Announces

’Business Model’ for Certifying JTRS Radios,” Jan. 2007.

[127] H. Rantanen, “SCA test, evaluation and certification - the perspective of the Finnish

software radio programme,” Apr. 2008.

[128] T. A. Sturman, M. D. J. Bowyer, and N. R. Petfield, “Skynet 5: MILSATCOM using

SDR,” in Military Communications Conference, 2007. MILCOM 2007., pp. 1–7.

IEEE, Oct. 2007.

[129] M. S. Hasan, M. LaMacchia, L. Muzzelo, R. Gunsaulis, L. R. Housewright, and

J. Miller, “Designing the joint tactical radio system (JTRS) handheld, manpack, and

small form fit (HMS) radios for interoperable networking and waveform applications,”

in Military Communications Conference, 2007. MILCOM 2007., pp. 1–6.

IEEE, Oct. 2007.

[130] Defence Update. Harris, Thales Compete Multi-Billion JTRS Radio Procurement.

(June 2007). [Online]. Available: http://www.defense-update.com/newscast/0607/

news/240607 jtrs.htm

[131] K. Bailey, “JTRS and COALWNW Overview for MILCIS 2008,” Nov. 2008.

[132] United States Government Accountability Office, “Department of Defense Needs

Framework for Balancing Investments in Tactical Radios,” Aug. 2008.

[133] C. Serra, “Evolution of SDR Standards - The SDR Forum Perspective,” Nov. 2007,

Key note at 2007 Software Defined Radio Technical Conference and Product Exposition

(presentation slides).

[134] European Defence Agency and OCCAR, “European Secure Software Defined Radio:

Contract Launced,” Dec. 2008.

140


References

[135] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “Next generation/dynamic

spectrum access/cognitive radio wireless networks: A survey,” Computer Networks,

vol. 50, no. 13, pp. 2127–2159, May 2006.

[136] B. A. Fette, Cognitive Radio Technology. Amsterdam: Newnes/Elsevier, 2006.

[137] S. Haykin, “Cognitive Radio: Brain-Empowered Wireless Communications,” IEEE

Journal on Selected Areas in Communications, vol. 23, no. 2, pp. 201–220, Feb.

2005.

[138] F. K. Jondral, “Software-defined radio – basics and evolution to cognitive radio,”

EURASIP Journal on Wireless Communications and Networking, vol. 2005, no. 3,

pp. 275–283, Apr. 2005.

[139] C. Tran, R. P. Lu, A. D. Ramirez, R. C. Adams, T. O. Jones, C. B. Phillips, and S. Thai,

“Dynamic spectrum access: Architectures and implications,” in Military Communications

Conference, 2008. MILCOM 2008., pp. 1–7. IEEE, Nov. 2008.

[140] T. Yücek and H. Arslan, “A survey of spectrum sensing algorithms for cognitive radio

applications,” IEEE Communications Surveys and Tutorials, vol. 11, no. 1, Jan. 2009.

[141] A. Gatherer, “Market and technology drivers for cellular infrastructure modem development,”

Wireless Personal Communications, vol. 38, no. 1, pp. 43–53, June 2006.

[142] S. Jennis and P. Burns, “What does mainstream telecom need to embrace ’true

SDR’?” in SDR’08 Technical Conference and Product Exposition, Oct. 2008.

[143] Huawei. Huawei Industry First with 3G/2G Software Defined Radio (SDR) Single

RAN Product Launch. (Sep. 2008). [Online]. Available: http://www.huawei.com/

news/view.do?id=5587&cid=42

[144] ZTE. ZTE Launches First SDR Base Station at GSMA Mobile World Congress

Barcelona 2008. (Feb. 2008). [Online]. Available: http://wwwen.zte.com.cn

[145] A. Kaul, “Software Defined Radio - The Transition from Defense to Commercial

Markets,” Nov. 2007, Key note at the 2007 Software Defined Radio Technical Conference

and Product Exposition (presentation slides).

[146] NXP Semiconductors. Samsung, NXP and T3G showcase world’s first TD-

SCDMA HSDPA/GSM multi-mode mobile phone. (Oct. 2008). [Online]. Available:

http://www.nxp.com

[147] J. Martin and F. Clermidy, “Mobile digital baseband: From configurable to evolutive

platforms,” in 16th IST Mobile and Wireless Communications Summit, 2007., pp. 1–5.

IEEE, July 2007.

[148] J. Mitola III, “Cognitive Radio An Integrated Agent Architecture for Software Defined

Radio,” Doctoral dissertation, Royal Institute of Technology (KTH), SE-164 40

Kista, Sweden, May 2000.

141


REFERENCES

[149] I. F. Akyildiz, L. Won-Yeol, M. C. Vuran, and M. Shantidev, “A survey on spectrum

management in cognitive radio networks,” IEEE Communications Magazine, vol. 46,

no. 4, pp. 40–48, Apr. 2008.

[150] C. Cordeiro, K. Challapali, D. Birru, and S. Shankar, “IEEE 802.22: An Introduction

to the First Wireless Standard based on Cognitive Radios,” Journal of Communications,

vol. 1, no. 1, pp. 38–47, Apr. 2006.

[151] E. L. Org, R. J. Cyr, R. Cook, and D. Donovan, “Software defined radio - programmable

transceivers for femtocells,” in SDR’08 Technical Conference and Product

Exposition, Oct. 2008.

142


Paper B

On Workload in an SCA-based System, with Varying Component and Data Packet

Sizes

Tore Ulversøy and Jon Olavsson Neset

Presented at

NATO Research and Technology Organization (RTO) Information Systems Technology

Panel (IST) Symposium on Military Communications With a Special Focus on Tactical

Communications for Network Centric Operations

Prague, Czech Republic

April 21-22, 2008

Published online at:

http://www.rta.nato.int/Pubs/RDP.asp?RDP=RTO-MP-IST-083

Summary of this candidate’s contribution to this paper:

This candidate has contributed with the major parts (> 90%) of the article text, as well as the computer

measurements. Further, the underlying basis for and the development of the models, and the

graphs. Also, parts of the computer programming (other parts by the co-author, J.E. Neset)

143


Abstract

An SCA-based SDR-application has components which exchange data through

ports. This enables scalability through the possibility of deployment on multiple processors,

makes code reuse easier in other SDR applications, and makes it easier to

port applications. There is wide freedom as regards the granularity of the component

structure, e.g. few components with a large amount of processing in each component

or a higher number of components with less processing in each component. While a

fine structure further enhances scalability, reuse and portability, it has side-effects in

the form of increased workload overheads. Here the effects of varying this granularity

on the workload of the total system are examined. The analysis is done for a varying

number of components, while splitting the processing functional work such that the total

functional processing work remains the same. Also, the effects of varying the size

of the data packets between the components are studied. The analysis is done both

by model calculations and experimental measurements. The analysis is important for

efficient utilization of the processing elements in an SDR system.

145


1. Introduction

1 Introduction

The Software Communications Architecture (SCA) is the dominant architecture for Software

Defined Radio (SDR) in the military domain. SCA defines a run-time environment

for applications and allows applications to be built as compositions of components. SCA

also defines a distributed architecture, allowing the application components to be deployed

on several processing elements. The communication between the components is enabled

through port interfaces, and uses CORBA as the underlying communications middleware

between the various parts of the application running on the various processors. On processing

elements where CORBA is not available, SCA prescribes the use of CORBA-adapters,

of which several solutions exist.

The splitting of SDR applications into components, with their defined input- and output

interfaces, and their defined requirements to their runtime environment, enables scalable

systems in that the various components may be deployed on additional processors as needed.

Components also make reuse of parts of an application easier, e.g. if two standards have the

same common interleaver, it makes sense to define this interleaver as one component that

can be used in the waveform applications for both standards. Generally, an application

composition with many small components increases the probability of being able to reuse

the components in other waveform applications.

The component approach, however, has side effects in the form of processor workload

overhead, where we here define the processor workload implied by a task or a group of tasks

as the fraction of available processor cycles occupied over a time period. The component

approach adds CORBA overhead through the processing of the CORBA invocations and

format conversions in the system. Also, the approach increases the number of separate processes

and threads, which increases context switching in the cases where several components

are deployed on the same processor.

In the following we will consider such a case where the application granularity is so fine

that several components are deployed on the same processor. We examine this by using a

scenario where we have one CORBA-capable General-Purpose Processor (GPP) which runs

the same functional application work, but implemented as a variable number of components.

As part of the experiment, we also vary the workload of components, the size of the data

packets transferred between the components, and the data rate in the system.

2 Analysis Approach

Our aim in this work is to quantify and understand the effects of application granularity of

SCA-based applications, when the granularity is such that several components will need to

be deployed on the same CORBA-capable GPP processor. We will be strictly interested in

granularity effects on workload, i.e. we will not consider other aspects that may also be

affected, such as latency, throughput, and statistical variation of latency and throughput.

In analysing processor workload or processor system performance in general, a wide

range of analysis approaches are available, as illustrated in Figure B.1. Simple analytical

system models may provide good insight and be easy to understand, however for most practical

model sizes such models will only provide a course representation of the system, and

will not capture the system behaviour in fine detail. At the other end of the scale, workload

measurements on the actual system will provide good accuracy in determining the actual

147


B. ON WORKLOAD IN AN SCA-BASED SYSTEM, WITH VARYING COMPONENT AND

DATA PACKET SIZES

workload, however the level of detail in the total system may result in a less clear picture

as to what are the dominant causes for the measured system performance. By analysing

and structuring the measured information however, we may also get an understanding of the

underlying dominant mechanisms.

An approach that makes it easier to capture more details and hence makes it easier to

provide a higher accuracy than an analytical model is e.g. a Petri-Net model [1]. This type

of model is however also more complex to understand and interpret. An even further model

refinement will be a simulation model, which is even more complex to build and understand,

but which has the potential of capturing finer details. If we did not have a system to perform

measurements on, a further alternative would be to build a testbed that captured the specific

characteristics we were interested in.

We have chosen in this work to use a combination of empirical analysis and simple

analytical models. Empirical analysis in the form of workload measurements on sample

applications provide quantification of granularity effects with good accuracy, and also allow

us to do detailed analysis where appropriate. The simple analytical models intend to help

provide better insight into the effects we are observing.

Increasing

clarity /

simplicity

Analytical

model

Concurrency model,

e.g. Petri-net

Simulation

model

Measurements

on testbed

Measurements on

actual system

Increasing

accuracy

FIGURE B.1: Illustration of system performance analysis methods

3 Workload Assessment Through Empirical Analysis

3.1 General

We have chosen to perform the empirical analysis using the OSSIE Core Framework (CF)

from VirginiaTech [2], which uses the omniORB [3]. Since this is an open source CF, it

gives us the advantage of having the full source code available for our analysis. The system

runs on Linux on an x86 hardware architecture. A Linux system is convenient due to both

good availability of public documentation and due to a rich variety of publicly available

analysis tools. Specifics of software and hardware are listed in Table B.1.

We have further chosen to do the empirical analysis on the basis of two groups of waveform

applications. The first group is based on a sample waveform application with a low

computational load (relative to the capacity of the processor), the transmitter (TX) part of

Stanag 4285. This waveform is being used in a study in the RTO-IST-080 RTG-038 on

SDR, and the base code has been provided by Telefunken Racoms. For purpose of the study

here, the waveform processing is configured into applications with a variable number of

components, as described in section 3.2.

148


3. Workload Assessment Through Empirical Analysis

TABLE B.1: HW and SW used in the experiments

Experiment 3.2 Experiment 3.3

OS Linux 2.6.9-34.EL Linux 2.6.9-34.EL

OSSIE revision 0.6.0 0.6.2

Processor Pentium M 1.86GHz Pentium M 1.86GHz

RAM 1,5 GByte 1,5 GByte

Cache specifics

L1i=first level instruction cache

L1d=first level data cache

L2=second level cache

L1i=32kB

L1i=32kB

L2=2MB

L1i=32kB

L1d=32kB

L2=2MB

The second group of applications is a synthetic workload, also configured into applications

with a variable number of components, and as described more specifically in section

3.3. A synthetic workload allows us to more easily vary the computational load of the components,

and also the size of the transmitted packages between components, which will help

us understand the system behaviour.

The system workload, and how the workload is distributed among various processes and

modules in the system, may be measured in a variety of ways, ranging from instrumentation

(time measurements in the code) to various profiling and monitoring tools. Here we have

chosen to use the tools ’OProfile’ [4] and ’SYSSTAT sar’ [5]. OProfile is a statistical profiler

that samples the complete system including the kernel, shared libraries, and executables.

This makes it possible to assess the workload distribution over the total system, including

e.g. what portion of the CPU time was spent in the various modules of the applications and

what portion of it was used with CORBA. SYSSTAT sar is a performance monitoring tool

that works through collecting operating system information at intervals specified by the user.

SYSSTAT sar will be used to monitor the CPU utilization during execution of the different

sample applications.

3.2 Empirical Workload Assessment Using a Sample Waveform

Application With Low Computational Complexity: Stanag 4285

TX

In this experiment, we compare a 2-component, 7- and 11-component application implementation

of the Stanag 4285 TX waveform, as illustrated in Figure B.2, in terms of CPU

workload. Each of the implementations contain the same functional processing code, such

that the amount of processing work that is carried out is the same in all three cases. Additionally

we compare these application configurations against a non-SCA-based version, this

version also doing the same processing work.

For regulating the number of data packets processed per second (PR) in the application

chain, the data sink contains a packet rate regulator in the form of SW instructions that

perform a blocking read of a number of samples from an audio card. PR can then be set as

the number of samples read relative to the audio card sample rate.

The CPU workload is quantified using SYSSTAT sar (sar -u 40 5), and quantified as the

CPU % in the user application space, and as the sum of the CPU % in the user application

space and in the system space.

149


On Workload in an SCA-Based System with Varying Component and Data Packet Sizes

implementations contain the same functional processing code, such that the amount of processing work

that is carried out is the same in all three cases. Additionally we compare these application configurations

B. ON WORKLOAD

against a non-SCA-based

IN AN SCA-BASED

version, this version

SYSTEM,

also doing

WITH

the

VARYING

same processing

COMPONENT

work.

AND

DATA PACKET SIZES

2 comp.

Stanag

4285 TX

Data

Sink

7 comp.

Data

Source

FEC

Encoder

Interleaver

Symbol

Mapper &

Scrambler

Symbol to

I/Q & TX

Filter

Float to

fixed

converter

Data

Sink

11

comp.

Data

Source

FEC

Encoder

Interleaver

Symbol

Mapper &

Scrambler

Symbol to

I/Q & TX

Filter

Float to

fixed

converter

Forwarder Forwarder Forwarder Forwarder

Data

Sink

FIGURE B.2: Stanag 4285 TX implemented in three different application configurations: As two

SCA components (top), as 7 SCA components (middle) and as 11 components (bottom). The data

sink contains a packet rate regulator.

Figure 2: Stanag 4285 TX implemented in three different application configurations: As two

SCA components (top), as 7 SCA components (middle) and as 11 components (bottom). The

data sink contains a packet rate regulator.

For regulating the number of data packets processed per second (PR) in the application chain, the data sink

Figure contains B.3 shows a packet the rate user regulator CPU in % the results form of from SW instructions the experiment, that perform when a blocking running read the of application

samples at 2395from symbols an audio / second card. PR (which can then corresponds be set as the number to 9,36 of data samples packets read relative per second), to the audio card

a number of

and then sample at 25600 rate. and 64000 symbols second. 2400 symbols/sec is the rate of the actual

waveform, the reason for additionally measuring at the two higher speeds is to get more

The CPU workload is quantified using SYSSTAT sar (sar –u 40 5), and quantified as the CPU % in the

significant user CPU application % readings. space, and as the sum of the CPU % in the user application space and in the system space.

Figure

303 shows the user CPU % results from the experiment, when running the application at 2395

symbols / second (which corresponds to 9,36 data packets per second), and then at 25600 and 64000

symbols second. 2400 symbols/sec is the rate of the actual waveform, the reason for additionally

25

measuring at the two higher speeds is to get more significant CPU% readings.

User CPU %

20

15

10

Non-SCA

Single component+sink

6-components+sink

10-components + sink

5

0

2395 25600 64000

Symbol rate

FIGURE B.3: User applications CPU workload in %, measured by SYSSTAT sar, for Stanag 4285

TX, for four different application configurations, each having the same processing functionality. The

processing symbol rates are also varied, by varying the number of packets processed per second (the

original Stanag 4285TX symbol rate is 2400 symb/sec).

17 - 4 RTO IST-083 - Czech Republic - 21/04/2008 - 22/04/2008

We see from Figure B.3 that the user CPU workload is increased more than 2,5 times,

going from the non-SCA basecode to the 11-components implementation, and roughly 2,5

times from a two-component implementation to an 11-component one. When considering

user+system CPU%, Figure B.4, we see that this difference is even larger. Hence we may

conclude, for this processing code, that the CPU workload increases significantly as the

processing code is split into more components.

150


3. Workload Assessment Through Empirical Analysis

45

40

User + System CPU %

35

30

25

20

15

10

5

Non-SCA

Single component+sink

6-components+sink

10-components + sink

0

2395 25600 64000

Symbol rate

FIGURE B.4: Same as previous figure, but showing the User+System CPU workload in %.

3.3 Empirical Workload Assessment Using a Synthetic Load

In this experiment we compare a 2-component (W2), 3-component (W3), 5-component

(W5) and 11-component (W11) application, see Figure B.5. The computational load in each

case consists of 9 FIR-filter loads, each of B · N multiply-and-addition operations (plus the

additional necessary load and store operations). Here B is the packet (block) size in number

of floats (1 float=4 bytes), and N is the number of filter taps.

In this setup, we may easily vary B and the total functional load (≈ const · 9 · B · N ),

which we term W L F UNC . In the same way as in 3.2, we vary PR by using the audio card as

a packet rate regulator. For W2 the regulator is in the FTOT component, and for W3...W11

in the SRC component.

Due to subtle details as to how the functional code runs on the processor, the proportionality

of W L F UNC to 9 · B · N is not exact for all B and N. In order for this not to confuse

our interpretation of measurements, we include reference measurements of W L F UNC where

appropriate, defining W L F UNC as the measured workload when running the functional code

as a standalone c-program.

As an initial experiment we want to verify that we see CPU workload differences between

the various configurations also of the synthetic load. Since we have the liberty of

changing the load in the components, we do this for two different loads, N*B=20000 and

100000, keeping the packet size B at 2000 floats. The results are provided in Figure B.6.

For N*B=20000, we observe a CPU (U+S) workload ratio between the standalone program

and the 11-component SCA-based configuration of 1,94, hence we clearly observe workload

overhead effects. When N*B is increased 5 times, we still observe differences between the

configurations, but the same (U+S) workload ratio is now decreased to 1,17. Hence, under

these conditions, we see a clear workload ratio dependency on N*B.

Next we want to investigate the effects on workload from varying the packet size, while

keeping W L F UNC approximately constant, Figure B.7. Since the component functional

workload, as mentioned previously, is not ideally proportional to N*B, we have selected

combinations of N and B points that result in W L F UNC = 10 ± 0.3% as a reference. We

see again that dividing the processing into more components leads to increased processor

151


Due to subtle details as to how the functional code runs on the processor, the proportionality of WL FUNC to

9 ⋅ B ⋅ N is not exact for all B and N. In order for this not to confuse our interpretation of measurements,

we include reference measurements of WL FUNC where appropriate, defining WL FUNC as the measured

workload when running the functional code as a standalone c-program.

B. ON WORKLOAD IN AN SCA-BASED SYSTEM, WITH VARYING COMPONENT AND

DATA PACKET SIZES

W2:

FTOT

SNK

W3:

SRC F1TO9 SNK

W5:

SRC F123 F456 F789 SNK

W11:

SRC F1 F2 F3 F4

F5 F6 F7 F8 F9

SNK

FIGURE B.5: The synthetic load on the system, in four different application configurations. In

all configurations, Figure the functional 5: The synthetic load isload that on of the 9 FIR-filters, system, in each four with different B x application N multiplications configurations. and In all

additions. configurations, the functional load is that of 9 FIR-filters, each with B x N multiplications and

additions.

workload. As an We initial also experiment see a clearwe processor want to workload verify that dependency we see CPU onworkload B, with differences the workload between the various

generally configurations increasing with also increasing of the synthetic B. load. Since we have the liberty of changing the load in the

components, we do this for two different loads, N*B=20000 and 100000, keeping the packet size B at

2000 floats. The results are provided in figure 6. For N*B=20000, we observe a CPU (U+S) workload

ratio between the standalone program and the 11-component SCA-based configuration of 1,94, hence we

4 Workload clearly observe Assessment workload overhead through effects. When Low-Complexity

N*B is increased 5 times, we still observe differences

between the configurations, but the same (U+S) workload ratio is now decreased to 1,17. Hence, under

Analytical

these conditions,

Models

we see a clear workload ratio dependency on N*B.

4.1 A Simple Lower Bound Model

Model

This model is an optimistic one, and will serve as a lower bound on the workload in the SCAbased

system. The model will tell us how much of the additional workload on the processor,

due to increased granularity, can be explained by accounting for the added workload due to

the CORBA-based communication, i.e. the CORBA client, data format conversions, ORB

activities, data transport, and CORBA servant.

In this model we will assume no loss due to context switches. We will assume that all

the functional code in the components, and all other activities on the processor, execute in

sequence without any performance losses due to there being several processes competing

for the resources, and we will assume that the processor goes into idle state when it has

processed one data package through the chain of components. Specifically we here assume

that separating the functional code into more components does not lead to an increased

amount of cache misses and hence does not lead to a lower average memory fetch/store

speed. In the model, we have chosen to separate out the data conversion to and from the

’FloatSequence’ type that is used for the port data packet communication, as we found that

to be important workload contributors.

Under these conditions we may write the CPU workload in % for our synthetic applica-

152

17 - 6 RTO IST-083 - Czech Republic - 21/04/2008 - 22/04/2008


4. Workload Assessment through Low-Complexity Analytical Models

CPU Workload versus Configuration

BLSZ=2000, PR=40

50

45

40

Ratio, user: 1,10

Ratio, user+syst: 1,17

35

CPU WL [%]

30

25

20

15

Ratio, user: 1,67

Ratio, user+syst: 1,94

System

User

10

5

0

FUNC

N=10

W2

N=10

W3

N=10

W5

N=10

W11

N=10

FUNC

N=50

W2

N=50

W3

N=50

W5

N=50

W11

N=50

Configuration

FIGURE B.6: User and system CPU workload in %, measured by SYSSTAT sar, for the four synthetic

waveform application configurations, and compared to the standalone c-implementation, FUNC. The

packet size is 2000 floats, and N=10 for the 5 leftmost measurements, N=50 for the other 5 measurements.

PR=40.

tion with M components (including SRC and SNK) as

W L i% = 100·P R ·[9t CL+t SRC +t SNK +(M−1)t packet +(M−1)t T S +(M−2)t T F ]

P CR

(B.1)

where t CL is the number of processor cycles to process one unit component load CL, t SRC

is the number of processor cycles to process the functionality in the SRC component, t SNK

is the number of processor cycles to process the functionality in the SNK component, t T S is

the number of processor cycles to convert array of floats data to its FloatSequence representation,

t T F is the number of processor cycles to convert the FloatSequence representation

to array of floats data, t packet is the number of processor cycles to transfer one data packet

between components, and P CR is the cycle rate of the processor. (The (M−2) is due to

there not being any conversion from the FloatSequence type in the SNK component in our

specific case.)

Estimation of Parameters

Allthough it is possible to calculate estimates of the above parameters based on the source

code of our system and processor specifics, it is a lengthy exercise and it is also difficult to

get good accuracy. For our purpose of describing the workload overheads in the system, it

suffices to instead measure the specific parameters in (1), as follows:

• 9·t CL is measured, using OProfile, as a function of N and B, by running the functional

code as a standalone c-program

• t SRC ,t SNK ,t T S ,t T F and t packet : Measured using both SYSSTAT Sar and OProfile, with

a test application consisting of merely one source and one sink component, see Figure

B.8.

153


B. ON WORKLOAD IN AN SCA-BASED SYSTEM, WITH VARYING COMPONENT AND

DATA PACKET SIZES

CPU Workload versus Packet Size

35

CPU WL [%]

30

25

20

15

10

5

FUNC U

FUNC U+S

W3 U

W3 U+S

W5 U

W5 U+S

W11 U

W11 U+S

W2 U

W2 U+S

0

0 10000 20000 30000 40000 50000 60000 70000

Packet Size

FIGURE B.7: User(U) and User+System (U+S) CPU workload in %, measured by SYSSTAT sar,

as a function of packet size in # of floats, for the four synthetic waveform application configurations

and with the standalone C-program as a reference (FUNC). FUNC has been adjusted, by selecting

combinations of B and N, to approximately 10% workload (see FUNC graph).

– t SRC ,t SNK : Measured using OProfile, with no conversion to/from FloatSequence

taking place in the source- and sink-components, as a function of B,

and excluding all ORB-related processor cycles

– t T S ,t T F : Measured with OProfile and sar, with a large number of conversions to

FloatSequence taking place in the source component and/or and a large number

of conversions from FloatSequence taking place in the sink component

– t packet : Measured with OProfile as the sum of the processor cycles judged to be

ORB-related. Estimated with SAR with no conversion to/from FloatSequence

and assuming t SRC and t SNK ≈ 0.

CORBATEST:

CORBATSTSRC

CORBATSTSINK

FIGURE B.8: Test application for measurements of parameters in equation (B.1)

Comparison with Measured WL

Figure B.9 shows a comparison of measured WL (user %) for the W11 configuration

as a function of B (same data as in Figure B.7), and W L i% , when using the OProfile

t SRC ,t SNK ,t T S ,t T F and t packet estimates from table 2, and setting M=11 in the model. As

expected W L i% underestimates the real workload, in particular for high B’s, and in particular

as we have not accounted for any context switching. Notably though, W L i% predicts the

major part of the observed WL overhead, particularly for the low B region.

154


4. Workload Assessment through Low-Complexity Analytical Models

TABLE B.2: The parameters in model (B.1) measured with OProfile and SYSSTAT sar, using the

test application of Figure B.8.

User

User

System

System

Estimate

based on

’sar’ measurements:

Estimate

based on

OProfile

measurements:

Estimate

based on

’sar’ measurements:

Estimate

based on

OProfile

measurements:

t SRC t SNK t T S (B) t T F (B) t packet (B)

Assumed 0 Assumed 0 14, 5·B 10, 6·B 79200 + 5, 5·B

650 200 14, 6·B 10, 6·B 80700 + 5, 5·B

Assumed 0 Assumed 0 ≈ 0 ≈ 0 160000 + 23, 6·B

≈ 0 ≈ 0 ≈ 0 ≈ 0 150000 + 23, 0·B

4.2 A Simple Model Including Context Switching

Model

A context switch refers to the switching of the CPUs execution from one process or thread

to another.

Each SCA component will run as a process of its own. The switching of which process

is the active executing one, is determined by the scheduling algorithm of the OS, in our case

by the Linux 2.6.9 scheduler.

When a context switch occurs, this has a workload cost for the processor. We refer to

the direct cost of the context switch as the cost directly associated with almost every context

switch, as the saving and restoring of processor registers, the execution of the OS scheduler,

reloading of TLB entries and the flushing of the processor pipeline [6]. The indirect cost

is related to the cache sharing between processes [6], which may cause parts of the cache

having to be stored onto lower-level cache or memory, and new values to be read into the

cache, i.e. a ’rewarming-up’ of the caches to the process or thread being executed.

The costs due to context switching is an added workload to that of equation (1), hence

we write the workload including context switching W L CS% as

W L CS% = W L i% + CSR · (t CSD + t CSI )

P CR

· 100 (B.2)

where CSR is the rate of context switches, t CSD is the direct cost in number of processor

cycles due to a context switch, whereas t CSI is the indirect cost, in processor cycles, of the

context switch.

Estimation of Parameters

155


B. ON WORKLOAD IN AN SCA-BASED SYSTEM, WITH VARYING COMPONENT AND

DATA PACKET SIZES

Comparison of W11 Measured Data and Simple Lower Bound Model

24

22

20

WL User [%]

18

16

14

12

10

0 1 2 3 4 5 6

B

x 10 4

FIGURE B.9: Comparison of measured (upper graph) WL (user %) for W11 , and W L i% (lower

graph), when using the OProfile t SRC ,t SNK ,t T S ,t T F and t packet estimates from table 2 when calculating

W L i% as a function of the packet size. FUNC has been adjusted to approximately 10% for all

B.

As our interest here is mainly that of verifying that B.2 provides useful explanations of our

problem, we will here only use course estimates of the parameters in B.2:

156

• CSR is a function of both the scheduling algorithm and the processor load. For the

comparison with the W11 measured data in Figure B.9, we here measure CSR directly,

using ’vmstat’ in Linux. We use a constant CSR of 1300 as a course average of

observed CSR’s for the various B’s in Figure B.9.

• t CSD is assumed to be a constant, and can be measured through methods described in

the literature. In [6], t CSD is measured to be 3,8µsec for a dual 2.0GHz Intel Xeon

system. We will use 5µsec or 9300 cycles as a course estimate for our system.

• t CSI is dependent on the L1 and L2 caches of the processor. In a simplified view, if

the sum of the data and instruction code areas that we are addressing are beyond the

size of the L2 cache, the system will increasingly need to write data buffers to memory

and read data buffers and instruction code from memory, at context switches. If this

is the case, we assume that we will need to store our changed data of the past process

to memory, and then read in the data and instructions of the next process that is to

execute. We make a course estimate t CSI for the W11 case as a fraction c multiplied

by the cycle count for writing one component data buffer of size B floats to memory,

reading two buffers of size B from memory and reading 10kBytes of instructions, for

the case that the sum of the buffers and code of all 11 components exceed the L2 size.


5. Conclusions

The blue line in Figure B.10 is W L CS% , for W11 and the same general conditions as in

figure B.9, and when accounting only for CSR and t CSD . We see that this provides a better

agreement with the measured data for W11 than using only the simple lower bound model,

in particular for the left part of the curve (low B).

When using a t CSI based on c=1 (full buffers written to and read from memory at each

context switch) we vastly overestimate the processor workload. When setting c=0.02 we

get the dashed black curve in Figure B.10, which, given our many assumptions, should be

regarded as an example of CS model output only. Interestingly we see however that the

exhaustion of the L2 explains the deviation of the measured data from that of the Simple

Lower Bound model for high B’s.

Comparison of W11 Measured Data and CS Model with Approximate Parameters

24

22

20

WL User [%]

18

16

14

12

10

0 1 2 3 4 5 6

B

x 10 4

FIGURE B.10: Comparison of measured (red diamond graph) WL (user %) for W11, and W L CS%

accounting for direct cost only (blue solid line), with same conditions as in Figure B.10. The black

dashed graph is an example of a W L CS% graph including both direct and indirect CS cost under

certain assumptions, see specific conditions in the text.

5 Conclusions

We have used empirical analysis and simple analytical models to understand the effects of

component granularity in an SCA-based system, when the granularity is such that several

components are deployed on the same CORBA-capable processor. For the empirical analysis,

we have used the OSSIE CF from VirginiaTech, and omniORB. We have used variable

number of components implementations of a real TX waveform processing, the Stanag

4285TX, and we have also used a synthetic waveform where we have been able to vary both

the component workloads and the data packet sizes.

157


REFERENCES

When executing the same total functional processing work, but with a varying number of

SCA-based components, we observe that the processor workload increases as the number of

components increases, and increasingly so for decreasing total functional processing work

as well as for increasing data packet sizes. Hence the scalability and reusability benefits that

result from implementing the SDR-application with a high number of components, must

be balanced against the processing efficiency loss that occurs when having to run several

components on the same processor.

We have proposed two simple models that explain the major effects of the processor

workload overheads in an SCA-based system, and that with proper determination of their

parameters may be used to predict actual workload in a system.

6 Acknowledgements

The first author would like to thank Torleiv Maseng at FFI (the Norwegian Defence Research

Establishment) for his encouragement and for valuable inputs and advice, and would also

like to thank Sarvpreet Singh at FGAN for his inputs on how to design the packet rate

regulator.

References

[1] P. J. Fortier and H. E. Michel, Computer Systems Performance Evaluation and Prediction.