november 2010 volume 1 number 2 - Advances in Electronics and ...
november 2010 volume 1 number 2 - Advances in Electronics and ...
november 2010 volume 1 number 2 - Advances in Electronics and ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
NOVEMBER <strong>2010</strong> VOLUME 1 NUMBER 2
Journal Editorial Board<br />
KRZYSZTOF WESOŁOWSKI, Editor-<strong>in</strong>-Chief<br />
Poznan University of Technology<br />
Piotrowo 3A, 60-965 Poznań, Pol<strong>and</strong><br />
krzysztof.wesolowski@et.put.poznan.pl<br />
WOJCIECH BANDURSKI<br />
Poznan University of Technology<br />
ANNA DOMAŃSKA<br />
Poznan University of Technology<br />
MACIEJ STASIAK<br />
Poznan University of Technology<br />
Advisory Board<br />
FLAVIO CANAVERO<br />
Politecnico di Tor<strong>in</strong>o<br />
Italy<br />
LAJOS HANZO<br />
University of Southampton<br />
UK<br />
MACIEJ OGORZAŁEK<br />
AGH Technical University<br />
Jagiellonian University<br />
Cracow, Pol<strong>and</strong><br />
Cover design Barbara Wesołowska<br />
ANNA PAWLACZYK, Secretary<br />
Poznan University of Technology<br />
Piotrowo 3A, 60-965 Poznań, Pol<strong>and</strong><br />
anna.pawlaczyk@et.put.poznan.pl<br />
HANNA BOGUCKA<br />
Poznan University of Technology<br />
MAREK DOMAŃSKI<br />
Poznan University of Technology<br />
RYSZARD STASIŃSKI<br />
Poznan University of Technology<br />
TADEUSZ CZACHÓRSKI<br />
Polish Academy of Science<br />
Institute of Theretical <strong>and</strong> Applied<br />
Informatics<br />
Gliwice, Pol<strong>and</strong><br />
MICHAEL LOGOTHETIS<br />
University of Patras<br />
Greece<br />
JOHN G. PROAKIS<br />
University of California<br />
San Diego, USA<br />
c○ Copyright by POZNAN UNIVERSITY OF TECHNOLOGY, Poznań, Pol<strong>and</strong>, <strong>2010</strong><br />
Edition based on ready-to-pr<strong>in</strong>t materials submitted by authors<br />
Materials published without further edit<strong>in</strong>g at the responsibility of the authors<br />
ISBN 978-83-7143-899-8<br />
ISSN 2081-8580<br />
PUBLISHING HOUSE OF POZNAN UNIVERSITY OF TECHNOLOGY<br />
60-965 Poznań, pl. M. Skłodowskiej-Curie 2<br />
tel. +48 (61) 6653516, fax +48 (61) 6653583<br />
e-mail: office_ed@put.poznan.pl<br />
www.ed.put.poznan.pl<br />
ADRIAN LANGOWSKI, Technical Editor<br />
Poznan University of Technology<br />
Piotrowo 3A, 60-965 Poznań, Pol<strong>and</strong><br />
adrian.langowski@et.put.poznan.pl<br />
ANDRZEJ DOBROGOWSKI<br />
Poznan University of Technology<br />
WOJCIECH KABACIŃSKI<br />
Poznan University of Technology<br />
PAWEŁ SZULAKIEWICZ<br />
Poznan University of Technology<br />
PIERRE DUHAMEL<br />
CNRS - Supélec<br />
France<br />
JÓZEF MODELSKI<br />
Warsaw University of Technology<br />
Pol<strong>and</strong><br />
RALF SCHÄFER<br />
Fraunhofer He<strong>in</strong>rich-Hertz-Institut<br />
Berl<strong>in</strong>, Germany<br />
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS is a peer-reviewed journal published at Poznań University of Technology, Faculty<br />
of <strong>Electronics</strong> <strong>and</strong> Telecommunications. It publishes scientific papers address<strong>in</strong>g crucial issues <strong>in</strong> the area of contemporary electronics <strong>and</strong><br />
telecommunications. Detailed <strong>in</strong>formation about the journal can be found at: www.advances.et.put.poznan.pl.
NOVEMBER <strong>2010</strong> VOLUME 1 NUMBER 2<br />
Radio Communication Series:<br />
Poznań Telecommunications Workshop<br />
Issue Editor: Paweł Szulakiewicz<br />
Note from the Issue Editor<br />
Paweł Szulakiewicz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1<br />
Wireless Systems <strong>and</strong> Networks<br />
Multipurpose Radio for Railways.Construction <strong>and</strong> Applications<br />
J. Kasperek, A. Nikoniuk, <strong>and</strong> P. Rajda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />
Simulation Study of the IEEE 802.15.4 St<strong>and</strong>ard Low Rate Wireless Personal Area Networks<br />
D. Ko´scielnik <strong>and</strong> J. St˛epień . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7<br />
Diversity <strong>and</strong> Multiplex<strong>in</strong>g Techniques<br />
M. Krasicki . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12<br />
Spectral Analysis of Boosted Space-Time Diversity Scheme<br />
M. Krasicki . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17<br />
Krylov Subspace Methods <strong>in</strong> Application to WCDMA Network Optimization<br />
R. Zdunek <strong>and</strong> M. Nawrocki . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22<br />
Networks<br />
Stream<strong>in</strong>g Video over TFRC with L<strong>in</strong>ear Throughput Equation<br />
A. Chodorek <strong>and</strong> R. R. Chodorek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26<br />
Simulation model for evaluation of packet sequence changed order of stream <strong>in</strong> DiffServ network<br />
M. Czarkowski <strong>and</strong> S. Kaczmarek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Packet dispatch<strong>in</strong>g schemes support<strong>in</strong>g uniform <strong>and</strong> nonuniform traffic distribution patterns <strong>in</strong> msm clos-network<br />
switches<br />
J. Kleban . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35<br />
Time <strong>and</strong> Synchronization<br />
Methods of Real-Time Calculation of Allan Deviation <strong>and</strong> Time Deviation<br />
A. Dobrogowski <strong>and</strong> M. Kasznia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42<br />
Application of Vernier Interpolation for Digital Time Error Measurement<br />
K. Lange <strong>and</strong> M. Kasznia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47<br />
Communication Theory<br />
Improv<strong>in</strong>g Statistical Properties of Number Sequences Generated by Multiplicative Congruential Pseudor<strong>and</strong>om<br />
Generator<br />
M. Jessa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51<br />
New Tailbit<strong>in</strong>g Convolutional Codes over R<strong>in</strong>gs<br />
P. Remle<strong>in</strong> <strong>and</strong> D. Szłapka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55<br />
Fiber Optics<br />
Model<strong>in</strong>g Step Index Fiber to Soliton Propagation<br />
T. Kaczmarek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59<br />
Are Carrier Transport Effects Important for Chirp Model<strong>in</strong>g of Quantum-Well Lasers?<br />
P. Krehlik . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63<br />
Precise Measurements of Highly Attenuated Optical Eye Diagrams<br />
P. Krehlik, Ł. ´Sliwczyński, <strong>and</strong> G. Sikorski . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67<br />
Bit Error Rate Tester for 10 Gb/s Fibre Optic L<strong>in</strong>k<br />
Ł. ´Sliwczyński <strong>and</strong> P. Krehlik . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 1<br />
Note from the Issue Editor<br />
The second issue of <strong>Advances</strong> <strong>in</strong> <strong>Electronics</strong> <strong>and</strong> Telecommunications<br />
conta<strong>in</strong>s sixteen selected papers presented at the last two editions<br />
of Poznań Telecommunications Workshop: PWT 2007 <strong>and</strong> 2008<br />
(www.pwt.et.put.poznan.pl). The conference held annually <strong>in</strong> Poznań at<br />
the beg<strong>in</strong>n<strong>in</strong>g of December is devoted to topics concern<strong>in</strong>g research <strong>and</strong><br />
education <strong>in</strong> telecommunications, electronics, <strong>and</strong> related fields. These most<br />
important areas of Information <strong>and</strong> Communication Technologies focus the<br />
attention of the workshop participants, who are ma<strong>in</strong>ly young researchers<br />
<strong>and</strong> PhD students from Polish universities of technology.<br />
The PWT workshops <strong>in</strong> Poznań have become a forum for develop<strong>in</strong>g a<br />
wide range of professional relationships. Both the presentations of research<br />
results <strong>and</strong> the discussions that follow provide the young authors with<br />
valuable opportunities to <strong>in</strong>teract with more experienced scientists, <strong>in</strong>dustry<br />
professionals <strong>and</strong> <strong>in</strong>novators <strong>in</strong> the fields of their particular <strong>in</strong>terests.<br />
We hope that the selection of PWT papers <strong>in</strong>cluded <strong>in</strong> this issue of<br />
<strong>Advances</strong> proves to be <strong>in</strong>terest<strong>in</strong>g to the readers.<br />
The presented papers are divided <strong>in</strong>to the follow<strong>in</strong>g five groups: Wireless Systems <strong>and</strong> Networks, Networks,<br />
Time <strong>and</strong> Synchronization, Communication Theory, Fiber Optics.<br />
We would like the <strong>Advances</strong> journal <strong>and</strong> the Poznań Telecommunications Workshop to bridge the gap between<br />
research <strong>and</strong> development, <strong>and</strong> eng<strong>in</strong>eer<strong>in</strong>g <strong>and</strong> implementation. Our readers will judge how far we are from that<br />
goal.<br />
Paweł Szulakiewicz<br />
Issue Editor
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 3<br />
Multipurpose Radio for Railways. Construction <strong>and</strong><br />
Applications<br />
Abstract—This paper provides <strong>in</strong>formation on the construction<br />
<strong>and</strong> presents experience from the “Koliber” project: a modern<br />
multipurpose radio system for railways. The radio equipment is<br />
produced by Radionika Ltd., <strong>and</strong> was designed <strong>in</strong> cooperation<br />
with Department of <strong>Electronics</strong>, AGH University of Science <strong>and</strong><br />
Technology. Discussed here are system architecture,technical <strong>and</strong><br />
functional parameters, <strong>and</strong> <strong>in</strong>novative radio system applications<br />
possible thanks to its <strong>in</strong>novatory construction.<br />
Index Terms—VHF railway radio, GSM-R<br />
I. INTRODUCTION<br />
Jerzy Kasperek, Andrzej Nikoniuk, <strong>and</strong> Paweł Rajda<br />
“<br />
KOLIBER” is a modern solution for radio communication,<br />
designed exclusively for railway needs. The device<br />
works as a mobile set <strong>in</strong> double-cab<strong>in</strong>locomotivesof all types<br />
<strong>and</strong> <strong>in</strong> any other rail vehicles. The stationary version of the<br />
radio is <strong>in</strong>tended to work as a base station, operated by the<br />
railway dispatcher.<br />
The device provides radio connections of all types <strong>in</strong> radio<br />
networksoperatedby railway companies,us<strong>in</strong>g VHF 150MHz<br />
b<strong>and</strong>. The device provides a specific signal<strong>in</strong>g used <strong>in</strong> Polish<br />
railways: tone selected calls (Zew1, Zew3) <strong>and</strong> emergency<br />
tra<strong>in</strong> stop protocol [1], [2]. Among the m<strong>and</strong>atory functions<br />
presented above, the solution offers more advanced functions<br />
available <strong>in</strong> contemporary radio communication. In particular,<br />
the device enables a range of functions <strong>in</strong>clud<strong>in</strong>g selective<br />
call signal<strong>in</strong>g (SelCall), CTCSS/DCS encod<strong>in</strong>g <strong>and</strong> decod<strong>in</strong>g,<br />
modem data transmission, <strong>and</strong> GPS navigation.<br />
Furthermore, the architecture <strong>and</strong> technology of the equipment<br />
allow also us<strong>in</strong>g the device <strong>in</strong> other communication<br />
network st<strong>and</strong>ards (<strong>in</strong>clud<strong>in</strong>g GSM <strong>and</strong> GSM-R). Besides the<br />
obvious economic benefits (s<strong>in</strong>gle device support<strong>in</strong>g multiple<br />
communication systems), this solution significantly simplifies<br />
the operationof radio for railway vehicle drivers<strong>and</strong> dispatchers.<br />
“Koliber” is a solution that not only serves the needs<br />
of current users of the railway network but also ensures the<br />
operationofequipmentaftermodernizationofthe network<strong>and</strong><br />
dur<strong>in</strong>g switch<strong>in</strong>g to a new digital communication st<strong>and</strong>ard.<br />
The device is fully compatible with mount<strong>in</strong>g <strong>and</strong> connectors<br />
currently used <strong>in</strong> vehicles <strong>and</strong> dispatcher desks. The dimensions<strong>and</strong>solutionsofthedeviceweredesignedtoenablequick<br />
assembly<strong>and</strong>setupwithuseoftheexist<strong>in</strong>gwir<strong>in</strong>g<strong>and</strong>fixtures.<br />
II. RADIO SET ARCHITECTURE<br />
Fig. 1 presents the architecture of the radio set version<br />
designed for the double-cab<strong>in</strong> locomotives. Both cab<strong>in</strong>s are<br />
Fig. 1. “Koliber” radio system architecture.<br />
equipped with a manipulator (DMI – Driver Mach<strong>in</strong>e Interface).<br />
Each DMI is connected with an <strong>in</strong>telligent switch<br />
module which commutes signals to the radio module. The<br />
switchmodulemayoptionallybeequippedwithaGSMeng<strong>in</strong>e<br />
to carry on voice communication through the mobile phone<br />
network of any operator <strong>and</strong>/or to transmit GPRS messages,<br />
<strong>in</strong>clud<strong>in</strong>g the status, geographical coord<strong>in</strong>ates, <strong>and</strong> parameters<br />
of the locomotive (e.g. the consumption of fuel <strong>in</strong> combustion<br />
locomotives).<br />
The entire set is powered through a universal DC/DC<br />
converter,work<strong>in</strong>gwith<strong>in</strong>awiderangeofvoltage(15...212V).<br />
S<strong>in</strong>gle-cab<strong>in</strong> locomotive sets have only one DMI mounted,<br />
while stationary sets feature an AC/DC power supply mounted<br />
<strong>in</strong>stead of the DC/DC converter. Moreover, the open architecture<br />
of the device enables <strong>in</strong>tegration of any ready-to-use<br />
GSM-R external modules [3]. To date, successful <strong>in</strong>tegration<br />
with certified PortBox Ultralight GSM-R module of HFWK<br />
(formerly Kapsch) was performed.<br />
III. DRIVER-MACHINE INTERFACE MODULE<br />
The DMI (Driver-Mach<strong>in</strong>e Interface) module performs the<br />
role of the user’s <strong>in</strong>terface radio. Its ma<strong>in</strong> operationalelements<br />
<strong>in</strong>clude:<br />
• high resolution graphic LCD display with backlight,<br />
• contextually illum<strong>in</strong>ated numeric <strong>and</strong> functional keypad,<br />
• “RadioStop” button be<strong>in</strong>g a part of the emergency tra<strong>in</strong><br />
stop system,<br />
• set of signal<strong>in</strong>g LEDs,<br />
• microphone with the PTT (Push To Talk) key,<br />
• speaker,<br />
• 1-Wire <strong>in</strong>terface for identification/authentication.<br />
The block diagram of the DMI module is presented <strong>in</strong><br />
Fig. 2. It is a typical microcontroller application based on<br />
a 8-bit Atmel RISC ATmega128 device. The DMI features a
4 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 2. Driver Mach<strong>in</strong>e Interface block diagram.<br />
large <strong>and</strong> clear graphic LCD display unit with resolution of<br />
240×64 pixels to present the current state of the whole radio<br />
set. The display presents also contextual description of the<br />
keyboard functions. The mean<strong>in</strong>g of particular keys depends<br />
on the menu selected, <strong>and</strong> contextual illum<strong>in</strong>ation facilitates<br />
their operation further. 1-Wire contact devices are used for<br />
access authorization <strong>and</strong> radio operator log-<strong>in</strong> & log-out.<br />
Communication with other modules of the set is performed<br />
via RS422 bus, while the radio voice is sent as analogue.<br />
IV. SWITCH MODULE<br />
The primary task of the switch module is commutation of<br />
signals between the radio module <strong>and</strong> the active DMI <strong>in</strong> one<br />
of the two locomotive cab<strong>in</strong>s. This module was designed <strong>and</strong><br />
developed as a natural replacement for the mechanical switch<br />
used before <strong>in</strong> most locomotives <strong>in</strong> Pol<strong>and</strong> [4]. The switch<br />
module can optionally be equipped with the GSM Motorola<br />
G24 eng<strong>in</strong>e. This solution enables concurrent usage of the<br />
audio <strong>and</strong> data GSM services parallel to st<strong>and</strong>ard work <strong>in</strong> the<br />
VHF b<strong>and</strong>.This allows us<strong>in</strong>g emergencycalls as well as SMS.<br />
What is more, once a GPS module has been <strong>in</strong>stalled, it<br />
is also possible to transfer tra<strong>in</strong> location data via the GPRS<br />
data l<strong>in</strong>k. The GPRS network l<strong>in</strong>k is a convenient medium<br />
for transmission of all k<strong>in</strong>ds of status messages between the<br />
driver <strong>and</strong> the stationary rail service. The switch module<br />
architecture is presented <strong>in</strong> Fig. 3. It uses an ATmega128<br />
microcontrollerasthema<strong>in</strong>processor.Duetothelarge<strong>number</strong><br />
of serial port controlled modules, a quadruple UART is used.<br />
This module can also be used <strong>in</strong> st<strong>and</strong> alone mode, e.g.<br />
as an <strong>in</strong>telligent GPRS modem for localization systems <strong>and</strong><br />
for various data acquisition solutions. To enable its operation<br />
even after the locomotive’s on-board supply failure (or while<br />
locomotive is be<strong>in</strong>g moved), the module is equipped with a<br />
power management system with a high capacity battery cell.<br />
V. RADIO MODULE<br />
The radio module conta<strong>in</strong>s the ma<strong>in</strong> execution unit for the<br />
whole set. The Tait TM8100 VHF transceiver module is used<br />
as the RF eng<strong>in</strong>e, <strong>and</strong> all the other functions – <strong>in</strong>clud<strong>in</strong>g audio<br />
signal<strong>in</strong>g, data transmission <strong>and</strong> voice record<strong>in</strong>g – are carried<br />
out by a dedicated unit control module.<br />
Fig. 3. Switch module block diagram.<br />
Fig. 4. Radio system architecture.<br />
Below listed are the parameters of the radio module:<br />
• 134-174 MHz frequency VHF b<strong>and</strong>,<br />
• 256 radio channels,<br />
• scann<strong>in</strong>g,<br />
• programmable channels frequency, RF power, <strong>and</strong> channel<br />
spac<strong>in</strong>g,<br />
• generation <strong>and</strong> detection of the emergency “Radiostop”<br />
signal,<br />
• generation <strong>and</strong> detection of sub-audio CTCSS/DCS signals<br />
<strong>and</strong> selective call audio signal<strong>in</strong>g,<br />
• 1200/2400bps modem data transmission,<br />
• call party ID generation <strong>and</strong> detection,<br />
• radio voice <strong>and</strong> system event record<strong>in</strong>g with optional<br />
external record<strong>in</strong>g channel,<br />
• GPS <strong>in</strong>ternal module option, external DCF module port,<br />
• RTC clock.<br />
The Fig. 4 presents the block diagram of the radio module<br />
architecture. The TM8100 VHF transceiver is controlled by<br />
a serial port with the Tait company proprietary comm<strong>and</strong><br />
protocol. All RF parameters are controlled by respective<br />
appropriate software comm<strong>and</strong>s. A dedicated audio processor
KASPEREK et al.: MULTIPURPOSE RADIO FOR RAILWAYS.CONSTRUCTION AND APPLICATIONS 5<br />
Fig. 5. “Koliber” GPS System architecture.<br />
Fig. 6. “Qguar Qpilot” localization w<strong>in</strong>dow.<br />
CMX7041 chip from CML is used for all audio <strong>and</strong> sub-audio<br />
signal<strong>in</strong>g,<strong>and</strong>also formodemtransmission.InPol<strong>and</strong>,thecall<br />
party ID signals are transmitted as modem messages.<br />
The radio module is controlled by the same type of microcontroller(ATmega128fromAtmel).However,duetoasignificant<br />
need of hardwareresources, the rema<strong>in</strong><strong>in</strong>gmodule architecture<br />
is implemented <strong>in</strong> 200k gates FPGA Spartan3 device<br />
from Xil<strong>in</strong>x. The ma<strong>in</strong> subsystem implemented <strong>in</strong> FPGA is<br />
the Secure Digital flash memorycard host controller.SD cards<br />
are used as the archive repository for voice <strong>and</strong> event records.<br />
To enable quick archive content read<strong>in</strong>g without remov<strong>in</strong>g the<br />
SD card, an SD controller was designed to work <strong>in</strong> a highspeed<br />
parallel (4-bit data bus) mode with troughput exceed<strong>in</strong>g<br />
10MB/s. In addition, the FPGA implements an <strong>in</strong>terface to a<br />
USB 2.0 controller, two UARTs (one for communication with<br />
the TM8100 VHF transceiver <strong>and</strong> the other for the service),<br />
two CVSD codec drivers (one for record<strong>in</strong>g audio from the<br />
radio set; i.e. VHF GSM calls, the other for an optional<br />
externalvoicerecorder),externaldatamemory<strong>in</strong>terfaceforthe<br />
microcontroller, <strong>and</strong> the authentication subsystem based on a<br />
hardware implementationof Blowfish cryptographicalgorithm<br />
with external “1-Wire” ID device.<br />
The module uses a small backup battery for the real-time<br />
clock device. Time synchronization is provided by the GPS<br />
eng<strong>in</strong>e, which – <strong>in</strong> the case of desktop solutions – may be<br />
replaced with an external DCF77 receiver.<br />
Fig. 7. Real-time locomotive cockpit visualization.<br />
Fig. 8. Architecture of DSR radio dispatcher system.<br />
VI. SYSTEM FIRMWARE<br />
Microcontrollers software was written <strong>in</strong> C language <strong>in</strong> the<br />
IAR AVR environment, <strong>and</strong> the FPGA project was created<br />
with VHDL.<br />
The <strong>in</strong>telligent switch module with GPRS option uses UIP<br />
TCP/IP freeware stack [5]. The web server <strong>and</strong> client ensure<br />
HTTP support for the “post” <strong>and</strong> “get” comm<strong>and</strong>s. When<br />
the external monitor<strong>in</strong>g device is connected, one can use<br />
the proprietary protocol to query the module for numerous<br />
parameters of the locomotive <strong>and</strong> localization. There is also<br />
an option to remotely change any EEPROM configuration<br />
memory content, e.g. the APN name <strong>and</strong> other GPRS network<br />
connection parameters.
6 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 9. GUI of DSR radio dispatcher system.<br />
Fig. 10. GIS RSSI data report.<br />
The AVR bootloader feature may be used to change any<br />
module microcontroller program memory <strong>and</strong>/or the FPGA<br />
configuration memory content, which facilitates firmware upgrades.<br />
All radio parameters can be set up us<strong>in</strong>g a dedicated<br />
software connected to the DMI module service RS232 port.<br />
VII. APPLICATIONS AND EXPERIENCES<br />
Based on the referred solution, some <strong>in</strong>terest<strong>in</strong>g applications<br />
of the radio set have been implemented. Their <strong>number</strong><br />
<strong>in</strong>cludes:<br />
• tra<strong>in</strong> localization <strong>and</strong> locomotive parameters monitor<strong>in</strong>g<br />
system,<br />
• DSR dispatcher system – remotely controlled VHF base<br />
station sets for railway ma<strong>in</strong> tracks,<br />
• G.sHDSL modem for the radio remote controll,<br />
• GIS RSSI measurement system for railway tracks.<br />
Presented below is a selection of screens <strong>and</strong> diagrams of the<br />
aplications mentioned.<br />
Fig. 5 presents “Qguar Qpilot” fleet management system<br />
architecture from Quantum Software S.A. The “Koliber” radio<br />
set sends localization data from GPS via the GPRS l<strong>in</strong>k to the<br />
company’s APN GSM <strong>in</strong>frastracture. Fig. 6 presents a sample<br />
GUI w<strong>in</strong>dow from the application.<br />
Fig. 7 presents real time visualization of the locomotive<br />
parameters from the “Koliber” switch module, connected to<br />
the CL400 module of the locomotive monitor<strong>in</strong>g unit (manufactured<br />
by ZEPWN).<br />
Fig. 9 presents an architecture of the DSR dispatcher radio<br />
system,whichconsistsofseveralradiobasestationscontrolled<br />
by dispatchers from the Local Control Center. Each base<br />
station <strong>in</strong>cludes up to 4 radios, along with a service DMI<br />
module <strong>and</strong> a control unit. Base stations are connected with<br />
Local Control Center by SDH based E1 l<strong>in</strong>ks, form<strong>in</strong>g a star<br />
structure.<br />
Fig. 10 presents example results of the radio signal strength<br />
measurement system done with the “Koliber” set on one of<br />
the ma<strong>in</strong> Polish rail tracks.<br />
VIII. CONCLUSION<br />
A few years of us<strong>in</strong>g the “Koliber” system <strong>in</strong> railway<br />
radio networks allow the conclusion that the design based<br />
on a simple 8-bit microcontroller, equipped with few external<br />
devices for dedicated functions, is fully justified. The design<br />
was verified by approved<strong>in</strong>dustrial bodies, positively tested <strong>in</strong><br />
the real consumer world, <strong>and</strong> opened many new application<br />
fields.<br />
REFERENCES<br />
[1] “R-12 <strong>in</strong>strukcja o u˙zytkowaniu urz˛ adzeń radioł˛ aczno´sci poci˛ agowej na<br />
pkp,” Biuletyn PKP, zał˛ acznik do nr 25 z dn. 18.12.1992, poz 102, (<strong>in</strong><br />
Polish).<br />
[2] E36 Instrukcja o organizacji i u˙zytkowaniu sieci urzadzeń ˛ radiołaczno´sci ˛<br />
w przedsi˛ebiorstwie państwowym PKP, (<strong>in</strong> Polish).<br />
[3] “GSM-R Procurement Guide,” [onl<strong>in</strong>e], Feb. 2007, www.uic.asso.fr.<br />
[4] R. Markowski, “Stan obecny radioł˛ aczno´sci na pkp – problemy i wyzwania,”<br />
<strong>in</strong> Proc. of Radiołaczno´sć ˛ w kolejnictwie wczoraj - dzi´s - jutro,<br />
Telekomunikacja Kolejowa Warszawa, Sep. 2003, (<strong>in</strong> Polish).<br />
[5] A. Dunkels, “Full TCP/IP for 8-Bit Architectures,” <strong>in</strong> Proc. of 1st<br />
International Conference on Mobile Applications, Systems <strong>and</strong> Services,<br />
MOBISYS, San Francisco, May 2003.<br />
Jerzy Kasperek, Paweł J. Rajda (kasperek@agh.edu.pl, pjrajda@agh.edu.pl)<br />
– Department of <strong>Electronics</strong>, AGH University of Science <strong>and</strong> Technology,<br />
30-059 Kraków, Al. Mickiewicza 30. Interest areas: digital design, hardware<br />
description languages, programmable logic <strong>and</strong> microcontroller applications,<br />
hardware accelerated signal process<strong>in</strong>g, custom comput<strong>in</strong>g mach<strong>in</strong>es.<br />
Andrzej Nikoniuk (<strong>and</strong>rzej.nikoniuk@radionika.com) – Radionika sp. z o.o.,<br />
30-003 Kraków, ul. Lubelska 14-18, Interest areas: railway radiocommunication<br />
systems design, bus<strong>in</strong>ess development manag<strong>in</strong>g.
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 7<br />
Simulation Study of the IEEE 802.15.4 St<strong>and</strong>ard<br />
Low Rate Wireless Personal Area Networks<br />
Abstract—This article presents a description of the simulation<br />
study of the low rate wireless personal area networks, def<strong>in</strong>ed<br />
by the IEEE 802.15.4 st<strong>and</strong>ard. The obta<strong>in</strong>ed results make<br />
it available to evaluate the effective transmission rate of a<br />
transmission channel,theresistance tothephenomenonof hidden<br />
station as well as the sensibility to the problem of exposed node.<br />
Index Terms—Exposed station, hidden station, low rate wireless<br />
area network<br />
I. INTRODUCTION<br />
THE IEEE 802.15.4 st<strong>and</strong>ard was created <strong>in</strong> 2003, <strong>and</strong> its<br />
current form results from the modifications <strong>in</strong>troduced<br />
three years later. The specification def<strong>in</strong>es the physical layer<br />
(PHY), the medium access control sublayer (MAC), as well<br />
as the pr<strong>in</strong>ciple of their <strong>in</strong>teraction with the higher layers.<br />
The LR-WPAN are characterized by very low energy consumption,<br />
simplicity of their structure mak<strong>in</strong>g it possible to<br />
implementthe transmissionprotocolon8-bit microcontrollers,<br />
as well as low costs of receiv<strong>in</strong>g <strong>and</strong> transmitt<strong>in</strong>g equipment.<br />
LR-WPAN aredesignedto beused<strong>in</strong> different<strong>in</strong>dustrial,agricultural<br />
<strong>and</strong> alarm systems, build<strong>in</strong>g automatics, monitor<strong>in</strong>g,<br />
<strong>in</strong>teractive toys <strong>and</strong> <strong>in</strong> particular <strong>in</strong> wireless sensor networks<br />
(WSN).<br />
The bit rate of the IEEE 802.15.4 network can be equal to:<br />
20 kb/s, 40 kb/s, 100 kb/s or 250 kb/s. The nodes realize the<br />
transmission <strong>in</strong> a discont<strong>in</strong>uous way, try<strong>in</strong>g to rema<strong>in</strong> for the<br />
longest possible time <strong>in</strong> <strong>in</strong>active mode – this make it possible<br />
to achievelow energyconsumption.The radiated poweris less<br />
than 1 mW, <strong>and</strong> the transmission range, characteristic for the<br />
personal operat<strong>in</strong>g space class solutions (POS), equals 10 m.<br />
The IEEE 802.15.4 st<strong>and</strong>ard offers a high capacity of the<br />
system <strong>and</strong> a very fast identification of equipment appear<strong>in</strong>g<br />
<strong>in</strong> its range. The <strong>number</strong> of operat<strong>in</strong>g nodes can equal 216 or<br />
264 , dependent on the length of addresses, whereas <strong>in</strong> general<br />
the time of registration of a new node does not exceed 30 ms.<br />
Moreover, a precious advantage is the automatic modification<br />
of connections with mov<strong>in</strong>g equipment.<br />
The IEEE 802.15.4 st<strong>and</strong>ard offers two ways of transmission:<br />
<strong>in</strong> non-synchronized (non-beacon) <strong>and</strong> <strong>in</strong> synchronized<br />
(bacon enabled) mode. The first one def<strong>in</strong>es only a<br />
contention access, us<strong>in</strong>g a simple mechanism permitt<strong>in</strong>g to<br />
identify the channel state <strong>and</strong> avoid collisions – unslotted-<br />
CSMA/CA (carriersense,multipleaccesswithcollisionavoidance).<br />
In the second method a less developed, slotted contention<br />
protocol has been implemented – slotted-CSMA/CA,<br />
as well as a no-collision access mechanism.<br />
Dariusz Ko´scielnik <strong>and</strong> Jacek St˛epień<br />
II. SIMULATION TESTS OF THE CONTENTION PROTOCOL<br />
The ma<strong>in</strong> objective of the tests of the contention protocol<br />
implemented <strong>in</strong> the IEEE 802.15.4 network was to def<strong>in</strong>e its<br />
efficiency <strong>and</strong> resistance to the appearance of hidden stations<br />
or exposed stations <strong>in</strong> the system, named also blocked nodes.<br />
The simulation was realized us<strong>in</strong>g a NetSim package created<br />
<strong>in</strong> the Department of <strong>Electronics</strong>, AGH University of Science<br />
<strong>and</strong> Technology. The NetSim software has been written <strong>in</strong><br />
C++language.Thepackageusesanevent-plann<strong>in</strong>gtechnology<br />
(event queue). Its mechanisms permit to correctly render<br />
the reciprocal time <strong>in</strong>terrelations exist<strong>in</strong>g between several<br />
simultaneous processes. The importance of simulated time as<br />
well as the <strong>number</strong> of stages of the tested processes can be<br />
dynamically adapted to the follow<strong>in</strong>g factors: the character of<br />
the observed events, the momentary importance of the offered<br />
traffic, the size of the tested system as well as the required<br />
precision of obta<strong>in</strong>ed results.<br />
In the further part of this work we have presented the<br />
results of tests relat<strong>in</strong>g to the evaluation of the efficiency<br />
of CSMA/CA protocol implemented <strong>in</strong> non-synchronized <strong>and</strong><br />
synchronized LR-WPAN network. In all the studied cases the<br />
assumptions are as follows: transmission rate of 250 kb/s, the<br />
DATAframestransmitdatafieldswithmaximalpermittedsize,<br />
the node emitters are equipped with buffers with a capacity<br />
of 50 packets <strong>and</strong> every successful transaction ends with an<br />
ACK frame. Moreover, we have admitted a two-ray ground<br />
propagation model, mean<strong>in</strong>g that the nodes located with<strong>in</strong><br />
the emitter range correctly receive its transmission with a<br />
probability equal to 1. The other stations do not hear the<br />
transmission – their probability of packet reception equals 0.<br />
In the simulation model, we did not take <strong>in</strong>to consideration<br />
the possible impact of any external <strong>in</strong>terference that might<br />
decreasethe efficiencyof the transmission.Therefore,the only<br />
possible cause of unsuccessful transfer can be a collision.<br />
A. Effective transmission rate of the transmission channel<br />
The effective transmission rate of the transmission channel<br />
<strong>in</strong>dicates a maximum<strong>number</strong> of user’s data transmitted with<strong>in</strong><br />
a time unit [1]. Usually, the value of this parameter is largely<br />
different from the used transmission rate, because of the<br />
overhead <strong>in</strong>troduced by the second <strong>and</strong> first layers as well<br />
as because of the <strong>in</strong>activity periods related to the duration of<br />
transmission delay times <strong>and</strong> the test<strong>in</strong>g of channel occupation<br />
dur<strong>in</strong>g the contention.<br />
For the identification of effective transmission rate of the<br />
system, we have used a model conta<strong>in</strong><strong>in</strong>g two nodes, one<br />
of them work<strong>in</strong>g as coord<strong>in</strong>ator. The transmission is realized
8 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
only <strong>in</strong> one direction – towards the coord<strong>in</strong>ator.Therefore, the<br />
network is free of collisions <strong>and</strong> the <strong>in</strong>tensity of the operated<br />
traffic is the maximal possible.<br />
The results obta<strong>in</strong>ed for both network operation modes<br />
(non-synchronized <strong>and</strong> synchronized) are summarized <strong>in</strong><br />
Fig. 1. The effective transmission rate <strong>in</strong> the non-synchronized<br />
mode equals to 116 kb/s, correspond<strong>in</strong>g to the utilization of<br />
46 % of the channel operation time. The rema<strong>in</strong><strong>in</strong>g transmission<br />
rate of the system is absorbed by the transmission<br />
overhead <strong>and</strong> by the dead periods, related to the r<strong>and</strong>om<br />
delay of the moment start<strong>in</strong>g transmission. The effective<br />
transmission rate of the synchronized network is even worse<br />
<strong>and</strong> equals about 98 kb/s, correspond<strong>in</strong>g to 39 % of the<br />
assumed transmission rate. The supplementary b<strong>and</strong> losses<br />
result from the necessity of the periodical transmission of<br />
BEACON frame, the <strong>in</strong>creas<strong>in</strong>g of the channel occupation<br />
test, the <strong>in</strong>creas<strong>in</strong>g of the contention w<strong>in</strong>dow size <strong>and</strong> the<br />
non-utilization of the last fragment of the superframe, which<br />
rema<strong>in</strong>s empty because the transmitt<strong>in</strong>g node cannot manage<br />
to fit the entire transaction <strong>in</strong> it. The average length of this<br />
section corresponds to the half of the transaction time.<br />
The Fig. 1.b presents the relation between the coefficient of<br />
delivered packets <strong>and</strong> the <strong>in</strong>tensity of the offered traffic. The<br />
losses of frames appear only dur<strong>in</strong>g the overload<strong>in</strong>g of the<br />
system. The superiority of the traffic offered over the traffic<br />
operated leads to the overfill<strong>in</strong>g of the emitter’s queue <strong>and</strong> the<br />
result<strong>in</strong>g refusal of a certa<strong>in</strong> part of the requests.<br />
Thesame modelofthesystem, loadedwith a trafficdirected<br />
<strong>in</strong>asymmetricalwaytobothnodes,makesitpossibletodef<strong>in</strong>e<br />
the <strong>in</strong>fluence of the bidirectionaltransmission for the available<br />
transmission rate of the network. The obta<strong>in</strong>ed results are<br />
summarized <strong>in</strong> Fig. 2. Their values are not significantly worse,<br />
evenif it couldseemthatthe nodesshould<strong>in</strong>itiateacontention<br />
concern<strong>in</strong>g the access to the common channel, lead<strong>in</strong>g to<br />
collisions. In the LR-WPAN, the transactions realized <strong>in</strong><br />
both directions are <strong>in</strong>itiated by a s<strong>in</strong>gle slave station, so any<br />
contention is excluded. The decrease <strong>in</strong> the transmission rate<br />
of the transmission channel results from a worse efficiency<br />
of transmission directed towards the slave node. Any such<br />
transaction must start with the transmission of REQUEST <strong>and</strong><br />
ACK frames [1], <strong>in</strong>creas<strong>in</strong>g its duration.<br />
The coefficient of delivered packets, def<strong>in</strong>ed for the discussed<br />
configuration, has slightly changed because of the<br />
decrease <strong>in</strong> the transmission rate of the network (Fig. 2.b).<br />
The form of both curves rema<strong>in</strong>s identical, confirm<strong>in</strong>g a total<br />
operation of the requests directed toward a system free of<br />
overload<strong>in</strong>g.<br />
B. Influence of a hidden station on the transmission rate of<br />
the system<br />
The collisions caused by hidden stations are much more<br />
troublesome for the system than those result<strong>in</strong>g from the<br />
contention for the access to the radio channel. A long time<br />
of emission of a s<strong>in</strong>gle frame significantly <strong>in</strong>creases the<br />
probability of generat<strong>in</strong>g a new request directed to the hidden<br />
station <strong>in</strong> this period [2]. Its immediate realization will disturb<br />
the transactionbe<strong>in</strong>g already<strong>in</strong> progresswith the distant node.<br />
Fig. 1. Unidirectional transmission <strong>in</strong> a system consist<strong>in</strong>g of two nodes: a)<br />
<strong>in</strong>tensity of the operated traffic, b) coefficient of delivered packets<br />
Study<strong>in</strong>g the <strong>in</strong>fluence of the presence of a hidden station<br />
on the operation of the LR-WPAN network, we have used<br />
the model presented <strong>in</strong> Fig. 3. A centrally placed coord<strong>in</strong>ator<br />
works with two slave nodes, located out of their reciprocal<br />
range. The entire offered traffic is evenly divided between<br />
slave stations, which direct their transfers exclusively to the<br />
coord<strong>in</strong>ator.<br />
The results of simulation tests, summarized <strong>in</strong> Fig. 4,<br />
<strong>in</strong>dicate a radical decrease <strong>in</strong> the transmission rate of the<br />
system – for both transmission modes it equals only 23 % of<br />
the effective channel transmission rate. Moreover, the network<br />
works with the efficiency close to maximal only <strong>in</strong> certa<strong>in</strong>,<br />
relatively narrow <strong>in</strong>terval of the <strong>in</strong>tensity of the offered traffic.<br />
A further <strong>in</strong>crease <strong>in</strong> the <strong>number</strong> of requests results <strong>in</strong> an<br />
important worsen<strong>in</strong>g of the quality of their servic<strong>in</strong>g <strong>and</strong><br />
<strong>in</strong> system overload<strong>in</strong>g. The shape of obta<strong>in</strong>ed characteristics<br />
correspondsto the panic curve,def<strong>in</strong><strong>in</strong>g the operationof many<br />
systems with collision access.<br />
The reason of the decrease <strong>in</strong> the network transmission rate<br />
– when the <strong>in</strong>tensity of the offered traffic exceeds of a given<br />
threshold value – is the <strong>in</strong>crease <strong>in</strong> the channel occupation<br />
time, favorable to the appearance of collisions with the hidden<br />
stations. The retransmissions activated by both nodes <strong>in</strong>crease<br />
<strong>in</strong> an artificial way the <strong>in</strong>tensity of requests directed towards<br />
the system, lead<strong>in</strong>g to its overload<strong>in</strong>g. It is worth mention<strong>in</strong>g
KO´SCIELNIK AND STEPIEŃ: ˛ SIMULATION STUDY OF THE IEEE 802.15.4 STANDARD LOW RATE WIRELESS PERSONAL AREA NETWORKS 9<br />
Fig. 2. Bidirectional transmission <strong>in</strong> a system consist<strong>in</strong>g of two nodes: a)<br />
<strong>in</strong>tensity of the operated traffic, b) coefficient of delivered packets<br />
Fig. 3. Model of a system conta<strong>in</strong><strong>in</strong>g hidden stations<br />
that <strong>in</strong> congestion conditions the transmission rate of a nonsynchronized<br />
network decreases to zero, whereas a synchronized<br />
system always guarantees a certa<strong>in</strong> m<strong>in</strong>imal level of<br />
servic<strong>in</strong>g the transmission requests. Such an advantage is a<br />
side effect of the algorithm realized by the node of the LR-<br />
WPAN network, verify<strong>in</strong>g before the start of each transaction<br />
if its duration does not exceed the limits of the f<strong>in</strong>ish<strong>in</strong>g<br />
superframe. Thanks to that, the hidden station rarely disturbs<br />
the last transmission that can fit <strong>in</strong>to the superframe.<br />
The def<strong>in</strong>ed characteristics of the coefficient of delivered<br />
packets (Fig. 4.b) <strong>in</strong>dicate that the loss of frames appears<br />
even with a very little <strong>in</strong>tensity of the offered traffic. The<br />
reason is the cancellation of further retransmissions of these<br />
packets, not delivered with a pre-def<strong>in</strong>ed admissible <strong>number</strong><br />
of attempts. As the <strong>in</strong>tensity of the requests <strong>in</strong>creases, this<br />
phenomenon appears more <strong>and</strong> more often. In an overloaded<br />
system, the queues of s<strong>in</strong>gle emitters become overfilled <strong>and</strong> a<br />
Fig. 4. Unidirectional transmission <strong>in</strong> a system conta<strong>in</strong><strong>in</strong>g hidden stations:<br />
a) <strong>in</strong>tensity of the operated traffic, b) coefficient of delivered pa<br />
more significant part of the offered traffic is refused.<br />
The objective of successive series of tests consisted <strong>in</strong><br />
verify<strong>in</strong>g the <strong>in</strong>fluence of the hidden station on the node<br />
located <strong>in</strong> the range of its signal. In the system presented<br />
<strong>in</strong> Fig. 3 this function is assumed by the coord<strong>in</strong>ator. We<br />
should rem<strong>in</strong>d that the transactions of the coord<strong>in</strong>ator are<br />
<strong>in</strong>itialized by other nodes of the cluster, strongly <strong>in</strong>fluenced<br />
by the presence of the hidden station. Based on this, we can<br />
presume that the hidden station will also disturb the servic<strong>in</strong>g<br />
of requests directed towards the coord<strong>in</strong>ator.<br />
The diagrams presented <strong>in</strong> Fig. 5 have been obta<strong>in</strong>ed us<strong>in</strong>g<br />
the model given <strong>in</strong> Fig. 3, <strong>in</strong> which the offered traffic has<br />
been evenly divided between all the nodes. Contrary to the<br />
assumptions, the presence of the hidden station has only a<br />
limited <strong>in</strong>fluence for on transactions realized by the coord<strong>in</strong>ator.<br />
Moreover, the <strong>in</strong>tensity of traffic realized by this station is<br />
not suddenly decreased when the threshold value is exceeded,<br />
as it was the case for the other nodes.<br />
The differencesexist<strong>in</strong>g <strong>in</strong> the way of servic<strong>in</strong>g the transactions<br />
realized <strong>in</strong> each direction are connected with the length<br />
of <strong>in</strong>itiat<strong>in</strong>g frames. A transaction directed to the coord<strong>in</strong>ator<br />
startswithalongDATApacket,whereasthetransfer<strong>in</strong>another<br />
directionis<strong>in</strong>itiatedwithamuchshorterREQUESTframe[3].<br />
Therefore, <strong>in</strong> the second case the probability of a collision<br />
caused by the hidden station is much lower. Moreover, if a
10 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 5. Bidirectional transmission <strong>in</strong> a system conta<strong>in</strong><strong>in</strong>g hidden stations: a)<br />
<strong>in</strong>tensity of the operated traffic, b) coefficient of delivered packets<br />
Fig. 6. System with exposed stations<br />
collision appears, its duration will also be shorter, reduc<strong>in</strong>g its<br />
<strong>in</strong>fluence on the channel transmission rate. The frames ACK<br />
<strong>and</strong> DATA <strong>in</strong>itiated by the coord<strong>in</strong>ator are received by all<br />
the nodes of the cluster, so the hidden stations have not any<br />
<strong>in</strong>fluence on further part of the transaction. The transmission<br />
directedtotheslave nodeis similarto atransactionconcern<strong>in</strong>g<br />
the reservation of channels with RTS <strong>and</strong> CTS frames, used <strong>in</strong><br />
IEEE 802.11 st<strong>and</strong>ard, <strong>and</strong> protect<strong>in</strong>g WLAN network aga<strong>in</strong>st<br />
problems created by the hidden stations.<br />
Irrespective of the status of the system, when the threshold<br />
value of the <strong>in</strong>tensity of offered traffic is exceeded, due to<br />
the transmission realized by the coord<strong>in</strong>ator, the coefficient of<br />
delivered packets does not decrease to zero, as it was <strong>in</strong> the<br />
previous case (Fig. 5.b). Its value gradually decreases because<br />
the overfill<strong>in</strong>g of the buffer <strong>in</strong> the coord<strong>in</strong>ator’s emitter results<br />
<strong>in</strong> the refusal of an <strong>in</strong>creas<strong>in</strong>g <strong>number</strong> of requests.<br />
Fig. 7. Unidirectional transmission <strong>in</strong> a system conta<strong>in</strong><strong>in</strong>g exposed stations<br />
C. Effect of the exposed station<br />
Study<strong>in</strong>gtheeffectsoftheexposedstation,wehaveusedthe<br />
modelpresented<strong>in</strong> Fig. 6.The<strong>in</strong>tensity ofthe offeredtraffic is<br />
evenly divided between nodes N1 <strong>and</strong> N3. The characteristics<br />
obta<strong>in</strong>ed <strong>in</strong> these conditions are summarized <strong>in</strong> Fig. 7.<br />
The obta<strong>in</strong>ed characteristics, as it concerns their shape <strong>and</strong><br />
values, are very similar to those observed for the system<br />
consist<strong>in</strong>g of two nodes<strong>and</strong> realiz<strong>in</strong>g the transmission towards<br />
the coord<strong>in</strong>ator(see Fig. 1). The total transmissionrate of both<br />
clusters is slightly higher than the effective transmission rate<br />
of a s<strong>in</strong>gle channel. The coefficients of delivered packets are<br />
also slightly higher, thanks to a double capacity of the buffers<br />
of both nodes. Therefore, the presence of exposed stations<br />
permits only a half of transmission resources of each cluster<br />
to be used.<br />
III. CONCLUSION<br />
The ma<strong>in</strong> objective of the authors of the IEEE 802.15.4<br />
st<strong>and</strong>ardwastocreateasystemthatcouldconta<strong>in</strong>anenormous<br />
<strong>number</strong> of nodes (even 2 64 ) <strong>and</strong> at the same time us<strong>in</strong>g a<br />
transmission protocol very simple to implement, guarantee<strong>in</strong>g<br />
m<strong>in</strong>imal energy consumption. The fulfill<strong>in</strong>g of all the abovementioned<br />
assumptionsprovesto be very difficult <strong>and</strong> – as the<br />
realized studies have shown – leads to an important decrease<br />
<strong>in</strong> the available transmission rate of the transmission channel.<br />
Important problems result also from the presence of a hidden<br />
station <strong>and</strong> exposed station.<br />
REFERENCES<br />
[1] A. Kouba, M. Alves, <strong>and</strong> Tovar, “A comprehensive simulation study of<br />
slotted CSMA/CA for IEEE 802.15.4 wireless sensor networks,” [onl<strong>in</strong>e],<br />
IPPHURRAY Research Group, Polytechnic Institute of Porto, http://<br />
www.iis.s<strong>in</strong>ica.edu.tw/cclljj/publication/2006/06_WCNC_802.15.4.pdf.<br />
[2] T. Sun, C. L<strong>in</strong>g-Jyh, H. Chih-Chieh, G. Yang, <strong>and</strong> M. Gerla, “Measur<strong>in</strong>g<br />
effective capacity of IEEE 802.15.4 beaconless mode,” <strong>in</strong> IEEE Wireless<br />
Communications <strong>and</strong> Network<strong>in</strong>g Conference, WCNC 2006, Las Vegas,<br />
Apr. 2006, pp. 493–498.<br />
[3] A. Herms, G. Lukas, <strong>and</strong> S. Ivanov, “Realism <strong>in</strong> design <strong>and</strong> evaluation<br />
of wireless rout<strong>in</strong>g protocols,” <strong>in</strong> Proceed<strong>in</strong>gs of First <strong>in</strong>ternational<br />
Workshop on Mobile Services <strong>and</strong> Personalized Environments MSPE‘06,<br />
2006.
KO´SCIELNIK AND STEPIEŃ: ˛ SIMULATION STUDY OF THE IEEE 802.15.4 STANDARD LOW RATE WIRELESS PERSONAL AREA NETWORKS 11<br />
Dariusz Ko´scielnik graduated <strong>in</strong> <strong>Electronics</strong> Eng<strong>in</strong>eer<strong>in</strong>g (1990) <strong>and</strong> <strong>in</strong><br />
Telecommunication (1993) from AGH – University of Science <strong>and</strong> Technology<br />
<strong>in</strong> Cracow, Pol<strong>and</strong>. He received his Ph.D degree <strong>in</strong> <strong>Electronics</strong><br />
Eng<strong>in</strong>eer<strong>in</strong>g (2000) from AGH – University of Science <strong>and</strong> Technology.<br />
Currently he is an Assistant Professor at the Institute of <strong>Electronics</strong> of<br />
AGH. His ma<strong>in</strong> research <strong>in</strong>terests have been <strong>in</strong> <strong>in</strong>ter-processor networks <strong>and</strong><br />
transmission protocols for control systems with spread <strong>in</strong>telligence. He is the<br />
author of books: Logical <strong>and</strong> Hardware Structure of ISDN (WPT, Cracow,<br />
1994), ISDN – Integrated Services Digital Network (WKiŁ, Warsaw, 1996)<br />
<strong>and</strong> Nitron Microcontrollers – Motorola M68HC08 (WKiŁ, Warsaw, 2005).<br />
Jacek St˛epień graduated <strong>in</strong> <strong>Electronics</strong> Eng<strong>in</strong>eer<strong>in</strong>g (1992) from AGH –<br />
University of Science <strong>and</strong> Technology <strong>in</strong> Cracow, Pol<strong>and</strong>. He received his<br />
Ph.D degree <strong>in</strong> <strong>Electronics</strong> Eng<strong>in</strong>eer<strong>in</strong>g (2001) from AGH – University of<br />
Science <strong>and</strong> Technology. Currently, he is an Assistant Professor at the Institute<br />
of <strong>Electronics</strong> of AGH. His research is focused on wired <strong>and</strong> wireless sensor<br />
networks <strong>and</strong> transmission protocols.
12 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Diversity <strong>and</strong> Multiplex<strong>in</strong>g Techniques<br />
of 802.11n WLAN<br />
Abstract—This paper is devoted to analyze an improvement<br />
<strong>in</strong> the performance of WLAN (Wireless Local Area Network)<br />
systems <strong>in</strong>troduced by space <strong>and</strong> space-time diversity, as well<br />
as spatial multiplex<strong>in</strong>g. These MIMO (Multiple-Input Multiple-<br />
Output) techniques are approved <strong>in</strong> the latest 802.11n specification.<br />
In order to perform the experiment, a Matlab application<br />
that simulates WLAN physical layer has been developed.<br />
Index Terms—Signal process<strong>in</strong>g, MIMO systems, diversity<br />
schemes, cod<strong>in</strong>g, modulation.<br />
I. INTRODUCTION<br />
COMMON WLAN st<strong>and</strong>ards def<strong>in</strong>ed by IEEE operate <strong>in</strong><br />
the ISM (Industrial, Scientific, Medical) b<strong>and</strong>s, i.e. 2.4<br />
GHz <strong>and</strong> 5.2 GHz. OFDM (Orthogonal Frequency Division<br />
Multiplex<strong>in</strong>g) is applied to overcome <strong>in</strong>tersignal <strong>in</strong>terference<br />
(ISI). The transmission runs <strong>in</strong> a frame mode. Numerous<br />
Modulation <strong>and</strong> Cod<strong>in</strong>g Schemes (MCS) are provided, which<br />
are switched by the transmitter adaptively, accord<strong>in</strong>g to the<br />
channel condition.<br />
The new specification of WLAN systems [1] has <strong>in</strong>troduced<br />
many techniques to improve data rate <strong>in</strong> the physical layer.<br />
Apart from modification of the OFDM symbol (52 subcarriers<br />
dedicated for data transmission <strong>in</strong>stead of 48 <strong>in</strong> 802.11a/g,<br />
shorter guard <strong>in</strong>terval), two groups of methods can be dist<strong>in</strong>guished:<br />
with backward signal<strong>in</strong>g <strong>and</strong> without it. The first<br />
group comprises beamform<strong>in</strong>g, i.e. based on knowledge of<br />
the channel state, the transmitter forms the signals <strong>in</strong> such a<br />
way that their performance at the receiver’s <strong>in</strong>put is optimized.<br />
These methods are not considered <strong>in</strong> the paper, which focuses<br />
on the space <strong>and</strong> space-time diversity techniques, <strong>in</strong>stead.<br />
Spatial multiplex<strong>in</strong>g is also addressed.<br />
Some results of multi-antenna OFDM systems preformance<br />
have been delivered <strong>in</strong> a few articles, e.g. [2], [3]. They can be<br />
treated as a reference to the present work to verify the accuracy<br />
of the simulation Matlab code developed by the author.<br />
The article is organized as follows: Section 2 reviews space<br />
<strong>and</strong> space-time diversity techniques, while Section 3 refers to<br />
spatial multiplex<strong>in</strong>g. The simulation results are presented <strong>in</strong><br />
Section 4. F<strong>in</strong>ally, Section 5 concludes the work.<br />
II. SPACE AND SPACE-TIME DIVERSITY SCHEMES<br />
The aim of space <strong>and</strong> space-time diversity is to improve<br />
radio l<strong>in</strong>k quality, by means of MIMO technology. In the first<br />
M. Krasicki is with the Faculty of <strong>Electronics</strong> <strong>and</strong> Telecommunications,<br />
Pozna University of Technology, Poznan, Pol<strong>and</strong> (phone: +48 61 665 39 36;<br />
fax: +48 61 665 38 23; e-mail: mkrasic@et.put.poznan.pl).<br />
This work was supported by the Polish M<strong>in</strong>istry of Science <strong>and</strong> Higher<br />
Education under Grant PBZ-MNiSW-02/II/2007.<br />
Maciej Krasicki<br />
Fig. 1. Transmitter <strong>and</strong> receiver of system exploit<strong>in</strong>g space (space-time)<br />
diversity<br />
place, the systems with only receive diversity will be considered.<br />
Afterwards, a smart idea of Space-Time Block Cod<strong>in</strong>g<br />
(STBC) [4], which is proposed by 802.11n specification, will<br />
be exam<strong>in</strong>ed. A general model of the transmitter <strong>and</strong> the<br />
receiver of a system employ<strong>in</strong>g space (space-time) diversity<br />
is shown <strong>in</strong> Fig. 1. At the transmitter, adjacent data bits are<br />
encoded by a convolutional encoder. Consecutive codewords<br />
are distributed among adjacent subcarriers accord<strong>in</strong>g to the<br />
block <strong>in</strong>terleav<strong>in</strong>g rule, after which they are mapped onto<br />
signals Ck(p), where k is the <strong>number</strong> of subcarrier <strong>and</strong> p<br />
denotes the <strong>number</strong> of OFDM symbol.<br />
The STBC encoder (if implemented) takes the consecutive<br />
signals Ck(p) <strong>and</strong> Ck(p + 1), occupy<strong>in</strong>g a given subcarrier k,<br />
which fall to the p-th <strong>and</strong> the (p + 1)-th OFDM symbols, <strong>and</strong><br />
creates their modified copies. All the signals are transmitted<br />
accord<strong>in</strong>g to the orthogonal Alamouti scheme [4], i.e. the<br />
first antenna transmits Ck1(p) = Ck(p) <strong>and</strong> Ck1(p + 1) =<br />
−C∗ k (p + 1) on the p-th <strong>and</strong> the (p + 1)-th OFDM symbol,<br />
respectively. Simultaneously, the second antenna transmits<br />
Ck2(p) = Ck(p + 1) <strong>and</strong> Ck2(p + 1) = C∗ k (p). The signals to<br />
be transmitted via the second antenna are cyclically rotated,<br />
accord<strong>in</strong>g to 802.11n specification, but this operation does not<br />
result <strong>in</strong> further diversity ga<strong>in</strong>.<br />
If space-time diversity is not implemented, STBC block is<br />
“transparent”, i.e. Ck1(p) = Ck(p), Ck1(p + 1) = Ck(p + 1),<br />
etc. In this case only one stream is transmitted.<br />
Next, OFDM is performed by means of Inverse Fast Fourier<br />
Transformation (IFFT). F<strong>in</strong>ally, Cyclic Prefix is added to<br />
avoid <strong>in</strong>ter-signal <strong>in</strong>terference. In a real system Digital/Analog<br />
conversion <strong>and</strong> carrier modulation should be done before<br />
the signals are transmitted. These steps can be omitted <strong>in</strong><br />
simulations s<strong>in</strong>ce the transmission <strong>in</strong> a baseb<strong>and</strong> channel is<br />
considered.<br />
At the receiver, after Cyclic Prefix removal (CPR) <strong>and</strong>
MACIEJ KRASICKI: DIVERSITY AND MULTIPLEXING TECHNIQUES OF 802.11N WLAN 13<br />
OFDM demodulation (FFT algorithm), each subchannel <strong>in</strong> the<br />
frequency doma<strong>in</strong> is ideally estimated, i.e. the frequency responses<br />
Hknm of the subchannel between the mth transmit <strong>and</strong><br />
the n-th receive antenna at the k-th subcarrier are calculated<br />
for all m, n, k. If the frequency response does not vary while<br />
a data frame is transmitted, the time <strong>in</strong>dex p can be omitted.<br />
The signal received from the nth antenna at the k-th subcarrier<br />
<strong>in</strong> the p-th OFDM symbol is<br />
Rkn (p) = �<br />
HknmCkm (p) + ηkn (p) , (1)<br />
m<br />
where Ckm(p) is a signal transmitted from the m-th antenna<br />
at the kth subcarrier <strong>in</strong> the p-th OFDM symbol, ηnk is a<br />
component represent<strong>in</strong>g additive noise. The diversity comb<strong>in</strong>er<br />
computes estimates � Ck (p) of the transmitted signals, <strong>in</strong> a<br />
way depend<strong>in</strong>g on the employed diversity scheme. It delivers<br />
estimates � Hk of the effective channel frequency response to the<br />
Maximum Likelihood detector, which makes decisions about<br />
the transmitted codewords. F<strong>in</strong>ally, the de<strong>in</strong>terleaved bits are<br />
passed to the Viterbi decoder.<br />
A. Receive Diversity<br />
The follow<strong>in</strong>g diversity algorithms are to be exam<strong>in</strong>ed:<br />
Antenna Selection, Subcarrier Selection, Equal Ga<strong>in</strong><br />
Comb<strong>in</strong><strong>in</strong>g (EGC) <strong>and</strong> Maximal Ratio Comb<strong>in</strong><strong>in</strong>g (MRC).<br />
S<strong>in</strong>ce only one transmit <strong>and</strong> two receive antennas are<br />
used, let us denote Hn = [H1n1 . . . H64n1] T , Rn(p) =<br />
[R1n(p) . . . R64n(p)] T , � �<br />
C(p) = �C1(p) . . . � �T C64(p) , <strong>and</strong> f<strong>in</strong>ally<br />
� �<br />
H = �H1 . . . � �T H64 .<br />
1) Antenna Selection: The diversity comb<strong>in</strong>er chooses a<br />
signal with higher average power from the signals received<br />
by adjacent antennas. Thus � C(p) = R1(p) <strong>and</strong> � H = H1<br />
if �<br />
k |Hk11| 2 > �<br />
k |Hk21| 2 . Otherwise, � C(p) = R2(p) <strong>and</strong><br />
�H = H2. It is noticeable that the comparison of average power<br />
is executed only once per frame due to the assumption of<br />
channel stationarity.<br />
2) Subcarrier Selection: The choice of antenna is made<br />
separately for each subcarrier k, depend<strong>in</strong>g on the magnitude<br />
response. That is � Ck(p) = Rk1(p) <strong>and</strong> � Hk = Hk11 if |Hk11| ><br />
|Hk21|. Otherwise � Ck(p) = Rk2(p) <strong>and</strong> � Hk = Hk21.<br />
3) Equal Ga<strong>in</strong> Comb<strong>in</strong><strong>in</strong>g (EGC): The signals from both<br />
receive antennas are exploited, i.e. they are added after the<br />
compensation of phase offsets:<br />
�Ck(p) = Rk1(p)e −j arg(Hk11) + Rk2(p)e −j arg(Hk21) .<br />
Consequently � Hk = |Hk11|+|Hk21|. The same operation runs<br />
for each subcarrier.<br />
4) Maximal Ratio Comb<strong>in</strong><strong>in</strong>g (MRC): This technique is<br />
very similar to EGC. The only modification is that the signals<br />
from both antennas are weighted accord<strong>in</strong>g to their power.<br />
Hence, the estimated transmitted signals are computed as<br />
�Ck(p) = Rk1(p)H ∗ k11 + Rk2(p)H ∗ k21 , while the estimates<br />
of the effective channel response can be written as � Hk =<br />
|Hk11| 2 + |Hk21| 2 .<br />
Fig. 2. Transmitter <strong>and</strong> receiver of spatially multiplexed system<br />
B. Space-Time Block Codes<br />
In case of space-time cod<strong>in</strong>g, the diversity comb<strong>in</strong>er computes<br />
the estimates of transmitted signals aga<strong>in</strong>. It is done<br />
accord<strong>in</strong>g to the follow<strong>in</strong>g rout<strong>in</strong>e. The signals received by<br />
adjacent antennas <strong>in</strong> consecutive timeslots p, <strong>and</strong> p + 1 can be<br />
written as:<br />
Rk1(p) = Hk11Ck(p) + Hk12Ck(p + 1)e −jθ<br />
+ηk1(p)<br />
Rk1(p + 1) = −Hk11C ∗ k (p + 1) + Hk12C ∗ k (p)e−jθ<br />
+ηk1(p + 1)<br />
Rk2(p) = Hk21Ck(p) + Hk22Ck(p + 1)e −jθ<br />
+ηk2(p)<br />
Rk2(p + 1) = −Hk21C ∗ k (p + 1) + Hk22C ∗ k (p)e−jθ<br />
+ηk2(p + 1)<br />
The factor denoted by e−jθ represents the phase rotation, required<br />
by 802.11n specification, which has to be compensated<br />
at the receiver. The author of this paper proposes to modify<br />
the orig<strong>in</strong>al rout<strong>in</strong>e of diversity comb<strong>in</strong>er [4] to mitigate the<br />
effect of cyclic rotation, <strong>in</strong>troduced by the transmitter:<br />
�Ck(p) = H∗ k11Rk1(p) �<br />
+ Hk12 Rk1(p + 1)ejθ�∗ +H∗ k21Rk2(p) �<br />
+ Hk22 Rk2(p + 1)ejθ�∗ (3)<br />
�Ck(p + 1) = H ∗ k12 Rk1(p)e jθ − Hk11 (Rk1(p + 1)) ∗<br />
+H ∗ k22 Rk2(p)e jθ − Hk21 (Rk1(p + 1)) ∗ .<br />
It can be proved that each of these comb<strong>in</strong>ed signals relates<br />
to a s<strong>in</strong>gle transmitted signal. In case of the 2 × 1 STBC<br />
system, the components associated with signals received from<br />
the second antenna should be omitted <strong>in</strong> (3).<br />
III. SPATIAL MULTIPLEXING<br />
Spatial multiplex<strong>in</strong>g offers higher data rate than any of<br />
diversity techniques analyzed above. The transmitter <strong>and</strong> receiver<br />
structures are shown <strong>in</strong> Fig. 2. Consecutive bits outgo<strong>in</strong>g<br />
from the encoder are distributed among different space streams<br />
<strong>and</strong> are subject to constellation mapp<strong>in</strong>g, cyclic shift <strong>and</strong> IFFT.<br />
As two <strong>in</strong>dependent signals are transmitted simultaneously<br />
through different antennas, they <strong>in</strong>terfere with one another at<br />
the <strong>in</strong>put of the receiver. To overcome this disadvantage, a<br />
simple Zero Forc<strong>in</strong>g comb<strong>in</strong>er is employed, which evaluates<br />
the estimates of signals Ck(p) = [Ck1(p) . . . Ckm(p)] T , transmitted<br />
from antennas 1 . . . m at the k-th subcarrier. Let us<br />
(2)
14 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 3. Average power delay profile<br />
denote Rk(p) = [Rk1(p) . . . Rkn(p)] T <strong>and</strong><br />
⎡<br />
⎢<br />
Hk = ⎣<br />
Hk11<br />
.<br />
. . .<br />
. ..<br />
Hk1m<br />
.<br />
⎤<br />
⎥<br />
⎦ .<br />
Hkn1 . . . Hknm<br />
It is noticeable that Rk(p) = HkCk(p) + ηk(p). To recover<br />
the transmitted signals, Rk(p) is multiplied by the <strong>in</strong>verse<br />
channel matrix H −1<br />
k . Note that <strong>in</strong> case of spatial multiplex<strong>in</strong>g<br />
there is no need to balance the cyclic shifts, which can be<br />
h<strong>and</strong>led as if they were <strong>in</strong>troduced by the channel. After ZF<br />
comb<strong>in</strong><strong>in</strong>g, the signals are demapped <strong>and</strong> de<strong>in</strong>terleaved, as for<br />
diversity techniques, but separately <strong>in</strong> different space streams.<br />
F<strong>in</strong>ally, demultiplexed bits undergo convolutional decod<strong>in</strong>g.<br />
A. Simulation setup<br />
IV. SIMULATION RESULTS<br />
Tim<strong>in</strong>g-related properties are <strong>in</strong>herited from 802.11n specification.<br />
Transmission runs <strong>in</strong> the 20 MHz b<strong>and</strong>width mode,<br />
52 subcarriers are dedicated for data transmission, 4 of them<br />
are assigned to pilot signals. The convolutional encoder characterized<br />
by [171 133]OCT generator polynomials is employed<br />
(resultant data rate is 1/2). Two modulation schemes are<br />
considered: QPSK <strong>and</strong> 16-QAM. An average total power is<br />
1 W. It is <strong>in</strong>dependent of the <strong>number</strong> of transmit antennas, for<br />
a fair comparison.<br />
A subchannel between each transmit <strong>and</strong> each receive<br />
antenna is simulated accord<strong>in</strong>g to the 11-tap exponential model<br />
(see e.g. [5]) with the root-mean-square delay spread τrms of<br />
92.435 ns. The average power delay profile of the assumed<br />
subchannel is shown <strong>in</strong> Fig. 3. R<strong>and</strong>omly generated fad<strong>in</strong>g<br />
coefficients are normalized to achieve unitary average signal<br />
power at the <strong>in</strong>put of each receive antenna. The assumed<br />
subchannel model is similar to ETSI B [6] <strong>in</strong> terms of the<br />
rms delay spread but much easier to simulate.<br />
The Doppler effect, a result of evolv<strong>in</strong>g channel state, has<br />
been neglected. To justify this approach, let us assume the<br />
term<strong>in</strong>al speed v = 3 km/h <strong>and</strong> the carrier frequency fc = 2.45<br />
GHz. Then, the maximum Doppler shift is fDmax = vfc/c ≈<br />
6.8 Hz (c is the speed of light). In the auto-regressive channel<br />
model (see e.g. [7]), the time-doma<strong>in</strong> channel response of the<br />
j-th tap of the subchannel at discrete time t + iTs is<br />
gj(t + iTs) = αigj(t) + wj(t + iTs) (4)<br />
where αi = E � gj(t)g ∗ j (t + iTs) � = J0(2πfD maxiTs), E (•)<br />
denotes the expected value, J0(•) is the zeroth-order Bessel<br />
function of the first k<strong>in</strong>d, wj(t + iTs) is an <strong>in</strong>dependent complex<br />
Gaussian r<strong>and</strong>om variable with zero mean <strong>and</strong> variance<br />
σ 2 w = 1 − α 2 i . Ts is the sample time. As the worst case, 4096<br />
<strong>in</strong>formation bytes per frame are to be transmitted <strong>in</strong> mode 1<br />
(BPSK) without spatial multiplex<strong>in</strong>g. The resultant <strong>number</strong> of<br />
the OFDM symbols is 1261, that gives 100880 samples <strong>in</strong><br />
time doma<strong>in</strong> (<strong>in</strong>clud<strong>in</strong>g the cyclic prefix). The autocorrelation<br />
value of tap responses fall<strong>in</strong>g to a frame decl<strong>in</strong>es only from<br />
1 to 0.988. It proves that the Doppler effect can be neglected.<br />
Assum<strong>in</strong>g that each frame is transmitted <strong>in</strong> different channel<br />
condition due to r<strong>and</strong>om channel access, fad<strong>in</strong>g coefficients<br />
can be generated <strong>in</strong>dependently for each frame.<br />
B. Results<br />
First, let us consider S<strong>in</strong>gle-Input S<strong>in</strong>gle-Output systems<br />
(MCS ∈ {1, 3}). The BER curves for 16-QAM <strong>and</strong> QPSK<br />
are presented <strong>in</strong> Fig. 4.a <strong>and</strong> Fig. 5.a, respectively, with th<strong>in</strong><br />
solid l<strong>in</strong>es. The analyzed curves are asymptotically parallel<br />
s<strong>in</strong>ce both systems have the same <strong>number</strong> of antennas. The<br />
higher modulation order, i.e. the <strong>number</strong> of bits mapped onto<br />
one constellation po<strong>in</strong>t, the worse BER performance. But it<br />
does not mean that 16-QAM is worse than QPSK <strong>in</strong> any case.<br />
To make the comparison fair, higher data rate of the former<br />
should be taken <strong>in</strong>to account. Moreover, any erroneously<br />
decoded bit is the cause of frame retransmission. Therefore,<br />
T hroughput = R(1 − FER), where R denotes the data rate<br />
<strong>and</strong> FER is the Frame Error Rate, is a more accurate measure<br />
of the l<strong>in</strong>k quality. Charts display<strong>in</strong>g the throughput are shown<br />
<strong>in</strong> Fig. 4.b <strong>and</strong> Fig. 5.b, respectively. The notation of particular<br />
curves is the same as before. It turns out that the 16-QAM<br />
system outperforms the QPSK one for SNRs > 19 dB, giv<strong>in</strong>g<br />
higher throughput.<br />
The receive diversity schemes reviewed <strong>in</strong> Section 2 have<br />
been exam<strong>in</strong>ed for 16-QAM <strong>and</strong> QPSK. It is noticeable that<br />
Antenna Selection is rather an <strong>in</strong>ferior technique, while the<br />
others significantly improve data l<strong>in</strong>k quality (higher slope of<br />
BER curve, diversity ga<strong>in</strong> of about 10 dB around the BER of<br />
10 −6 ). The difference <strong>in</strong> BER between particular algorithms is<br />
negligible, but only EGC <strong>and</strong> MRC are comparable with each<br />
other <strong>in</strong> the throughput, so there is a suggestion to employ<br />
Equal Ga<strong>in</strong> Comb<strong>in</strong><strong>in</strong>g, due to its easier implementation.<br />
For comparison, the 2 × 1 system with Space-Time Block<br />
Code has been analyzed. The BER <strong>and</strong> throughput curves are<br />
shifted right by about 3 dB <strong>in</strong> comparison with EGC. It is<br />
justified by the fact that the total transmitted power is normalized.<br />
In consequence, the power per receive antenna is still the<br />
same, <strong>and</strong> hence the systems with multiplied receive antennas<br />
perform better. Therefore, receive diversity techniques are<br />
more advantageous than Space-Time Block Cod<strong>in</strong>g, the more<br />
so as they are easier to implement. Nevertheless, space-time
MACIEJ KRASICKI: DIVERSITY AND MULTIPLEXING TECHNIQUES OF 802.11N WLAN 15<br />
Fig. 4. BER vs. SNR (a) <strong>and</strong> Throughput vs. SNR (b) for 16-QAM<br />
modulation<br />
codes are still useful to build a system with diversity only at<br />
one (Access Po<strong>in</strong>t’s) side.<br />
The performance of the 2 × 2 STBC 16-QAM system has<br />
been exam<strong>in</strong>ed, too. It appears to be much better than any 1×2<br />
or 2 × 1 system s<strong>in</strong>ce the signals are transmitted through 4<br />
<strong>in</strong>dependent subchannels (additive noise varies from one time<br />
sample to another). SNR ga<strong>in</strong> of about 15 dB around the BER<br />
of 10 −6 is observed <strong>in</strong> comparison with the SISO system.<br />
F<strong>in</strong>ally, the advantages of spatial multiplex<strong>in</strong>g have been<br />
analyzed. The BER <strong>and</strong> throughput curves of 2×2 <strong>and</strong> 4×4 16-<br />
QAM (MCS ∈ {9, 27}) as well as 2 × 2 QPSK (MCS = 11)<br />
systems are shown <strong>in</strong> Fig. 4 <strong>and</strong> Fig. 5, respectively. As it<br />
can be noticed, the multiplexed systems offer the same BER<br />
performance as 1x1 ones, asymptotically. Nevertheless, at low<br />
SNRs the signal detection is destroyed by the additive noise<br />
ga<strong>in</strong>ed by the ZF comb<strong>in</strong>er. In the region of high SNRs, the<br />
throughput is higher than for the 1 × 1 system, proportionally<br />
to the <strong>number</strong> of space streams om both sides of the system.<br />
V. CONCLUSIONS<br />
In this paper some transmit <strong>and</strong> receive diversity algorithms,<br />
approved by 802.11n specification, have been analyzed. These<br />
MIMO techniques have appeared to be powerful tools to enhance<br />
data rate regardless of the channel state. Thanks to 2×1<br />
Space-Time Block Codes, the system with antennas doubled<br />
only on the Access Po<strong>in</strong>t’s side can improve the l<strong>in</strong>k quality <strong>in</strong><br />
Fig. 5. BER vs. SNR (a) <strong>and</strong> Throughput vs. SNR (b) for QPSK modulation<br />
both directions. Spatial multiplex<strong>in</strong>g enhances the throughput<br />
but it fails <strong>in</strong> case of poor channel condition, which is caused<br />
by the ZF operation. To overcome this disadvantage, other<br />
algorithms, such as M<strong>in</strong>imum Mean Square Error (MMSE)<br />
[8] <strong>and</strong> Successive Interference Cancellation (e.g. [9]), should<br />
be exam<strong>in</strong>ed <strong>in</strong> the future.<br />
The conclusions the author arrived at agree with earlier<br />
works related to MIMO-OFDM schemes. The simulation<br />
Matlab code passed the validation test <strong>and</strong>, therefore, it can<br />
be used <strong>in</strong> further research.<br />
REFERENCES<br />
[1] 802.11n-2009 IEEE St<strong>and</strong>ard for Information Technology-Part 11: Wireless<br />
LAN Medium Access Control (MAC) <strong>and</strong> Physical Layer (PHY)<br />
Specifications Amendment: Enhancements for Higher Throughput.<br />
[2] J.D. Moreira et al., “Diversity techniques for OFDM based WLAN<br />
systems,” <strong>in</strong> Proc. of IEEE Int. Symposium on Personal, Indoor <strong>and</strong><br />
Mobile Radio Commnications (PIMRC), Lisbon, 2002.<br />
[3] L. Jee-Hye, B. Myung-Sun, <strong>and</strong> S. Hyoung-Kyu, “Efficient MIMO<br />
Receiv<strong>in</strong>g Technique <strong>in</strong> IEEE 802.11n System for Enhanced Services,”<br />
IEEE Trans. Consum. Electron., vol. 53, no. 2, May 2007.<br />
[4] S. Alamouti, “A simple transmit diversity technique for wireless communications,”<br />
IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp. 1452–1458,<br />
Oct. 1998.<br />
[5] Y. Sun, A. Nix, <strong>and</strong> J. McGeehan, “HIPERLAN performance analysis<br />
with dual antenna diversity <strong>and</strong> decision feedback equalization,” <strong>in</strong> Proc.<br />
of Vehicular Technology Conference, vol. 3, 1996.<br />
[6] BRAN TS 101 475 v1.2.2 BRAN; HIPERLAN Type 2; Physical (PHY)<br />
layer.<br />
[7] F. C. Zheng <strong>and</strong> A. G. Burr, “Signal detection for non-orthogonal spacetime<br />
block cod<strong>in</strong>g over time-selective fad<strong>in</strong>g channels,” IEEE Commun.<br />
Lett., vol. 8, no. 8, Aug. 2004.
16 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
[8] H. Gao, P. J. Smith, <strong>and</strong> M. Clark, “Theoretical reliability of MMSE l<strong>in</strong>ear<br />
diversity comb<strong>in</strong><strong>in</strong>g <strong>in</strong> rayleigh-fad<strong>in</strong>g additive <strong>in</strong>terference channels,”<br />
IEEE Trans. Commun., vol. 46, no. 5, pp. 666–672, May 1998.<br />
[9] L. Yang, M. Chen, S. Cheng, <strong>and</strong> H. Wang, “Comb<strong>in</strong>ed maximum likelihood<br />
<strong>and</strong> ordered successive <strong>in</strong>terference cancellation grouped detection<br />
algorithm for multistream mimo,” <strong>in</strong> Proc. of 8th IEEE Int. Symposium on<br />
Spread Spectrum Techniques <strong>and</strong> Applications, Aug. 2004, pp. 250–254.<br />
Maciej Krasicki received the M.S. degree <strong>in</strong> <strong>Electronics</strong> <strong>and</strong> Telecommunications<br />
from Poznan University of Technology, Pol<strong>and</strong>, <strong>in</strong> 2006. S<strong>in</strong>ce then he<br />
has been work<strong>in</strong>g towards the Ph.D. degree. His dissertation work concerns<br />
a new (‘boosted’) space-time diversity scheme, designed to support iterative<br />
decod<strong>in</strong>g at the receiver of WLAN systems. His Ph.D. defense took place <strong>in</strong><br />
<strong>2010</strong>.<br />
From 2009 he has been with the Faculty of <strong>Electronics</strong> <strong>and</strong> Telecommunications,<br />
Poznan University of Technology, as a Research Assistant. His<br />
research <strong>in</strong>terests <strong>in</strong>clude multi-antenna transmission, space-time cod<strong>in</strong>g <strong>and</strong><br />
iterative signal process<strong>in</strong>g. He has published several papers <strong>in</strong> journals (e.g.<br />
<strong>Electronics</strong> Letters) <strong>and</strong> conference proceed<strong>in</strong>gs.
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 17<br />
Spectral Analysis of Boosted<br />
Space-Time Diversity Scheme<br />
Abstract—In this paper the asymptotic performance of a<br />
new <strong>in</strong>tuitive space-time diversity scheme is analyzed. So called<br />
boosted scheme is compatible with today’s WLAN specifications<br />
with regard to convolutional cod<strong>in</strong>g <strong>and</strong> bit labell<strong>in</strong>g,<br />
<strong>and</strong> m<strong>in</strong>imizes the <strong>number</strong> of decod<strong>in</strong>g iterations, required to<br />
obta<strong>in</strong> a reasonable Bit Error Rate. Good properties of the<br />
proposed scheme are proved by high asymptotic cod<strong>in</strong>g ga<strong>in</strong><br />
<strong>and</strong> advantageous distance spectrum. A simulation experiment<br />
is run to <strong>in</strong>vestigate the system performance <strong>in</strong> terms of poor<br />
channel state. The boosted scheme is compared with its ancestor<br />
– Bit-Interleaved Space-Time Coded Modulation with Iterative<br />
Decod<strong>in</strong>g (BI-STCM-ID).<br />
Index Terms—Multiple-<strong>in</strong>put multiple-output channels, bit<strong>in</strong>terleaved<br />
space-time coded modulation, Alamouti scheme, constellation<br />
label<strong>in</strong>g, block fad<strong>in</strong>g, distance spectrum, cod<strong>in</strong>g ga<strong>in</strong>.<br />
I. INTRODUCTION<br />
WIRELESS Local Area Networks have recently become<br />
a very popular Internet access technique. Almost each<br />
notebook is equipped with an 802.11 card. Expectations<br />
for WLAN throughput are still grow<strong>in</strong>g. The new 802.11n<br />
[1] specification provides some promis<strong>in</strong>g techniques such<br />
as multi-antenna transmission with space-time block cod<strong>in</strong>g<br />
(STBC). The key issue is to make use of higher Multiple-Input<br />
Multiple-Output channel capacity. Reviewed <strong>in</strong> Section 2 BI-<br />
STCM-ID [2], that exploits iterative process<strong>in</strong>g at the receiver,<br />
seems to be an excellent solution. Unfortunately, the 802.11n<br />
specification accepts only Gray constellation labell<strong>in</strong>g, which<br />
makes iterative process<strong>in</strong>g worthless [3]. On the opposite,<br />
there are some constellation labell<strong>in</strong>gs, optimized for the<br />
lowest Bit Error Rate (BER) <strong>in</strong> case of error-free feedback.<br />
The author is an advocate of a new <strong>in</strong>tuitive approach to<br />
an overall mapp<strong>in</strong>g (constellation labell<strong>in</strong>g <strong>and</strong> space-time<br />
cod<strong>in</strong>g) described <strong>in</strong> Section 3. The proposed boosted spacetime<br />
diversity scheme m<strong>in</strong>imizes the <strong>number</strong> of passes while<br />
reasonable BER is kept. Theoretical analysis <strong>and</strong> simulation<br />
results are presented <strong>in</strong> Section 4. Section 5 of this paper is<br />
designed to conclude the work.<br />
A. System model<br />
II. BI-STCM-ID OVERVIEW<br />
A BI-STCM-ID system is shown <strong>in</strong> Fig. 1. In the first<br />
<strong>in</strong>stance, <strong>in</strong>formation bits are encoded by a convolutional<br />
encoder of rate RC = 1/kc. Next, K <strong>in</strong>terleaved encoded<br />
bits [v 1 t . . . v K t ] choose a vector [x 1 t . . . x q<br />
t ] of constellation<br />
M. Krasicki is with the Faculty of <strong>Electronics</strong> <strong>and</strong> Telecommunications,<br />
Poznan University of Technology, Poznan, Pol<strong>and</strong> (phone: +48 61 665 39 36;<br />
fax: +48 61 665 38 23; e-mail: mkrasic@et.put.poznan.pl).<br />
Maciej Krasicki<br />
Fig. 1. Transmitter (a) <strong>and</strong> receiver (b) of BI-STCM-ID<br />
po<strong>in</strong>ts, each of them accord<strong>in</strong>g to labell<strong>in</strong>g rule ω. Next, q<br />
constellation po<strong>in</strong>ts form the space-time (ST) symbol Xt ∈ ℵ.<br />
The ST symbol consists of modified constellation po<strong>in</strong>ts<br />
transmitted by Nt antennas with<strong>in</strong> L time slots. As it can<br />
be noticed, each ST symbol is unequivocally associated with<br />
K encoded bits, so overall mapp<strong>in</strong>g rule ϖ : {0, 1} K → ℵ<br />
can be def<strong>in</strong>ed. In case of orthogonal 2 × 2 Alamouti scheme,<br />
q = 2, L = 2, <strong>and</strong><br />
�<br />
x<br />
Xt =<br />
1 t x2 �<br />
� � t<br />
2 ∗ � �<br />
1 ∗ . (1)<br />
−xt xt The signals received by Nr antennas with<strong>in</strong> L time slots are<br />
expressed by<br />
Yt = XtHt + Wt<br />
Matrix Ht describes the channel, i.e. [hi,j]t is a temporal ga<strong>in</strong><br />
of the path between ith transmit- <strong>and</strong> jth receive antenna. Wt<br />
represents the Gaussian noise.<br />
Space-time demapper evaluates its output log-likelihood ratios<br />
(LLRs) [4] λ � v k t ; O � accord<strong>in</strong>g to a priori LLRs λ � v k t ; I �<br />
<strong>and</strong> the <strong>in</strong>formation received from the channel. SISO decoder<br />
[5] <strong>in</strong>creases LLR’s reliability accord<strong>in</strong>g to max-log-MAP<br />
rout<strong>in</strong>e.<br />
B. BI-STCM-ID Asymptotic Performance<br />
When ideal <strong>in</strong>terleav<strong>in</strong>g is assumed, the union bound of bit<br />
error probability is given by [3]:<br />
Pb ≤ 1<br />
kc<br />
d=df<br />
(2)<br />
∞�<br />
WI(d)f (d, ϖ, ℵ), (3)<br />
where df is the free distance of the convolutional code,<br />
<strong>and</strong> WI(d) denotes the total <strong>in</strong>put weight of error events at
18 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Hamm<strong>in</strong>g distance d. F<strong>in</strong>ally, f (d, ϖ, ℵ) is the pairwise error<br />
probability (PEP). Its loos<strong>in</strong>g Chernoff bound [3] is<br />
⎡<br />
f (d, ϖ, ℵ)≤⎣<br />
1<br />
K2K ⎤<br />
K� 1� � �<br />
m<strong>in</strong>Φ∆(X,Z)(s)<br />
⎦, (4)<br />
s<br />
k=1 b=0 X∈ℵk b Z∈ℵk ¯b where Z is a “neighbor” of X, the label of which has opposite<br />
kth bit ( ¯ b <strong>in</strong>stead of b). Φ ∆(X,Z)(s) is the Laplace transform<br />
of probability density function<br />
∆(X, Z) = �Y − ZH� 2 − �Y − XH� 2 . (5)<br />
Follow<strong>in</strong>g [4], it can be written that<br />
�<br />
r�<br />
�−Nr (1 + λi/4N0) , (6)<br />
m<strong>in</strong><br />
s Φ ∆(X,Z)(s) =<br />
i=1<br />
where λi are the nonzero eigenvalues of matrix<br />
A = (X − Z) H (X − Z),<br />
hav<strong>in</strong>g rank r. Tak<strong>in</strong>g only the nearest neighbor � Z ∈ ℵ k<br />
b of<br />
X <strong>in</strong> (4), one arrives at so-called expurgated PEP [3]:<br />
⎡<br />
fex (d, ϖ, ℵ) ≤ ⎣ 1<br />
K2K ⎤<br />
K� 1� �<br />
m<strong>in</strong> Φ<br />
s ∆(X, Z) � (s) ⎦ (7)<br />
If N0 → 0,<br />
fex (d, ϖ, ℵ) ∼<br />
where<br />
�Ω 2 ⎡<br />
(ℵ, ϖ, Nr)= ⎣ 1<br />
K2K k=1 b=0 X∈ℵk b<br />
K�<br />
�<br />
1�<br />
4<br />
�Ω 2 /N0<br />
k=1b=0<br />
X∈ℵk b<br />
� �rNrd<br />
�<br />
� �r�<br />
�λi<br />
i=1<br />
, (8)<br />
�−Nr<br />
⎤<br />
⎦<br />
1<br />
�rNr<br />
can be <strong>in</strong>terpreted as an asymptotic cod<strong>in</strong>g ga<strong>in</strong> associated<br />
with both space-time cod<strong>in</strong>g <strong>and</strong> constellation label<strong>in</strong>g. In the<br />
above statements � λi <strong>and</strong> �r are the nonzero eigenvalues <strong>and</strong><br />
the rank of matrix � A = (X − � Z) H (X − � Z), respectively. (The<br />
expurgated PEP is accurate only for Gray-labelled schemes.<br />
In such case, there is exactly one nearest neighbor � Z. For<br />
other labell<strong>in</strong>gs (7) is an overoptimistic approximation. [3])<br />
Note that (8) is valid only for mapp<strong>in</strong>g rules ω with the same<br />
�r value for each (X, � Z) pair. It has been checked that such<br />
condition is satisfied by the BI-STCM-ID with the Alamouti<br />
space-time code, considered <strong>in</strong> this paper.<br />
Hav<strong>in</strong>g taken only the first term (for d = df ) <strong>in</strong> (3) <strong>and</strong><br />
assumed that energy per <strong>in</strong>formation bit Eb = 1/R, where R<br />
is the overall <strong>in</strong>formation rate, the BER for BI-STCM system<br />
(after the first pass or without iterative process<strong>in</strong>g) is bounded<br />
on the logarithmic scale by [4]<br />
log 10 � Pb ≈ − �rNrdf<br />
10<br />
��<br />
R� Ω 2�<br />
+ (Eb/N0) dB<br />
dB<br />
(9)<br />
�<br />
+ const.<br />
(10)<br />
Note that the slope of the asymptotic bound is associated<br />
with the rank of � A. So only if all � A matrixes (for each<br />
(X, � Z) pair) are full-ranked, full diversity ga<strong>in</strong> can be reached.<br />
Additionally, the comparison of different mapp<strong>in</strong>g rules can be<br />
Fig. 2. The boosted space-time diversity scheme:<br />
fair only if the convolutional code of the same free distance df<br />
is employed. It is worth mention<strong>in</strong>g that the asymptotic cod<strong>in</strong>g<br />
ga<strong>in</strong> � Ω 2 of a mapp<strong>in</strong>g rule <strong>in</strong>fluences the horizontal offset of<br />
the bound (the higher cod<strong>in</strong>g ga<strong>in</strong>, the better position of the<br />
asymptotic bound).<br />
If the iterative decod<strong>in</strong>g runs, one can assume the error-free<br />
feedback, i.e. all bits are assumed to be perfectly known at<br />
the demapper, except the one for which the LLR is currently<br />
be<strong>in</strong>g evaluated. In such case, BER is asymptotically bounded<br />
by<br />
log 10 ˜ Pb ≈ − ˜rNrdf<br />
10<br />
��<br />
R˜ Ω 2�<br />
�<br />
+ (Eb/N0) dB +const, (11)<br />
dB<br />
where ˜ Ω 2 (ℵ, ϖ, Nr) is similar to � Ω 2 (ℵ, ϖ, Nr) from (9), but<br />
�λi <strong>and</strong> �r must be replaced with ˜ λi <strong>and</strong> ˜r, that are respectively<br />
the nonzero eigenvalues <strong>and</strong> the rank of<br />
à = (X − ˜ Z) H (X − ˜ Z).<br />
The bit labels of signals X <strong>and</strong> ˜ Z differ only on the kth bit<br />
position. Note that <strong>in</strong> the considered case there is exactly one<br />
˜Z symbol for each X.<br />
An accurate way to characterize labell<strong>in</strong>g of Bit-Interleaved<br />
Coded Modulation with Iterative Decod<strong>in</strong>g (an ancestor of BI-<br />
STCM-ID) is the Euclidean distance spectrum [6]. The idea<br />
is briefly depicted below. For each constellation po<strong>in</strong>t x <strong>and</strong><br />
each k-th position of its bit label, all neighbor<strong>in</strong>g po<strong>in</strong>ts z with<br />
the opposite k-th bit are found on the constellation. Distance<br />
spectrum D is just a histogram of all |x − z| 2 entries. Such<br />
spectrum is proper to judge the asymptotic performance of the<br />
system without iterative process<strong>in</strong>g. In the error-free feedback<br />
case, which can be approached after many iterations, Def<br />
spectrum of |x − ˜z| 2 distances should be evaluated, <strong>in</strong>stead.<br />
The <strong>in</strong>terpretation of distance spectra is as follows. The<br />
lower frequency of short distances <strong>in</strong> D, the better asymptotic<br />
system performance after the first iteration. Similarly, low<br />
frequency of short distances <strong>in</strong> Def suggests good asymptotic<br />
system performance <strong>in</strong> case of error-free feedback. Note that<br />
the spectrum analysis is useful to compare different mapp<strong>in</strong>g<br />
rules, <strong>and</strong> does not cover the impact of the employed convolutional<br />
code on overall system performance.<br />
Let us extend the idea of distance spectrum for any spacetime<br />
diversity scheme. If an orthogonal space-time code is<br />
used, the issue of the overall mapp<strong>in</strong>g rule ϖ optimization is<br />
reduced to search for optimal constellation labell<strong>in</strong>g ω. To f<strong>in</strong>d
MACIEJ KRASICKI: SPECTRAL ANALYSIS OF BOOSTED SPACE-TIME DIVERSITY SCHEME 19<br />
this statement true, see Theorem 1 <strong>in</strong> [4]. As a more general<br />
approach, the author proposes to associate the spectrum D<br />
with ( � r<br />
i=1 λi) values. In the same manner Def should<br />
consist of ( � ˜r<br />
i=1 ˜ λi) values. The correspondence between the<br />
mean<strong>in</strong>g of the Euclidean distance for 2-dimensional space<br />
<strong>and</strong> the mean<strong>in</strong>g of the product of eigenvalues for matrices<br />
makes this approach justified. The idea of distance spectrum<br />
is utilized <strong>in</strong> Section 4 to exam<strong>in</strong>e the performance of the<br />
proposed space-time diversity scheme.<br />
III. BOOSTED SPACE-TIME DIVERSITY SCHEME<br />
FOR WLAN SYSTEMS<br />
The most common approach to BI-STCM-ID is to use<br />
an orthogonal STBC <strong>and</strong> f<strong>in</strong>d a constellation labell<strong>in</strong>g ω to<br />
maximize cod<strong>in</strong>g ga<strong>in</strong> ˜ Ω 2 . In the region of BI-STCM-ID<br />
potential applications, like WLAN systems, decod<strong>in</strong>g time is<br />
the key issue. Unfortunately, any optimized labell<strong>in</strong>g makes<br />
the convergence of iterative process slower [2]. The solution<br />
would be a new overall mapp<strong>in</strong>g rule ϖ, thanks to which a<br />
dem<strong>and</strong>ed BER can be achieved after only a few iterations.<br />
The compatibility with the IEEE WLAN specifications would<br />
be highly appreciated.<br />
The author proposed <strong>in</strong> [7] an <strong>in</strong>tuitive space-time diversity<br />
scheme for WLAN systems, which is shown <strong>in</strong> Fig. 2. The<br />
convolutional encoder is taken from 802.11a/g/n specifications<br />
([171 133]OCT ). The idea is to take advantage of both Gray<br />
<strong>and</strong> “optimal” [4] labell<strong>in</strong>gs of 16-QAM. There are two<br />
signal streams at the transmitter. The first one, with the Gray<br />
mapper, is expected to provide good performance after the<br />
first decod<strong>in</strong>g pass. The second one is responsible for high<br />
asymptotic cod<strong>in</strong>g ga<strong>in</strong>. (The author has proved <strong>in</strong> [8] that<br />
à matrices are full-ranked for each (X, ˜ Z) pair, i.e. ˜r = 2.<br />
Therefore, eq. (11) rema<strong>in</strong>s valid.)<br />
The shaded blocks <strong>in</strong> Fig. 2 are used optionally, <strong>and</strong> should<br />
be turned off when other devices <strong>in</strong> a network run <strong>in</strong> legacy<br />
mode. The block denoted by Π is a symbol <strong>in</strong>terleaver of<br />
unitary depth. The resultant space-time codeword is<br />
Xt =<br />
� x 1 t (Gray) x 2 t (Gray)<br />
x 2 t (opt.) x 1 t (opt.)<br />
�<br />
. (12)<br />
The block diagram of the receiver is the same as for conventional<br />
BI-STCM-ID. Unfortunately, the boosted scheme suffers<br />
from non-orthogonality. In consequence, the rout<strong>in</strong>e of spacetime<br />
demapper is more complex.<br />
IV. EVALUATION OF BOOSTED SPACE-TIME<br />
DIVERSITY SCHEME<br />
Let us analyze the distance spectra D <strong>and</strong> Def of the<br />
boosted system <strong>and</strong> BI-STCM-ID – the latter with both “optimal”<br />
<strong>and</strong> Gray labell<strong>in</strong>gs. For convenience, spectrum entries<br />
can be scaled by the shortest possible distance d0, as <strong>in</strong> [6]<br />
for BICM-ID. (In that paper the entries were written as multiplicities<br />
of the m<strong>in</strong>imum squared Euclidean distance |x − z| 2<br />
between a constellation po<strong>in</strong>t x <strong>and</strong> its nearest neighbor z).<br />
For better legibility, entries d/d0 of the spectra will be<br />
treated as values of r<strong>and</strong>om variable D, whose cumulative<br />
distribution function P r(D < d/d0) will be plotted <strong>in</strong>stead<br />
Fig. 3. Distance spectra D<br />
Fig. 4. Distance spectra Def<br />
of the orig<strong>in</strong>al spectrum. Note that the abscissa will be scaled<br />
logarithmically.<br />
Fig. 3 represents the distance spectra D of Gray- <strong>and</strong><br />
“optimally”-labelled BI-STCM-ID <strong>and</strong> the boosted space-time<br />
diversity scheme. As it can be noticed, the spectrum of the<br />
boosted scheme is the worst one, i.e. the highest frequency of<br />
the lowest possible entry occurs. Moreover, the CDF of the<br />
proposed scheme grows much faster than for both BI-STCM-<br />
ID systems, considered <strong>in</strong> this paper. So it can be concluded<br />
that the spectrum of the boosted scheme conta<strong>in</strong>s many lowvalued<br />
entries. The best mapp<strong>in</strong>g rule <strong>in</strong> this competition is the<br />
Gray-labelled BI-STCM-ID with its slowly <strong>in</strong>creas<strong>in</strong>g CDF.<br />
It is not a surpris<strong>in</strong>g conclusion s<strong>in</strong>ce the Gray labell<strong>in</strong>g is<br />
recognized as the best solution for non-iteratively BICM-like<br />
systems [3].<br />
Note that poor asymptotic performance of the boosted<br />
scheme does not result <strong>in</strong> slow convergence of iterative<br />
process. The last can be exam<strong>in</strong>ed by means of EXtr<strong>in</strong>sic<br />
Information Transfer (EXIT) chart [9]. The author of this paper<br />
showed <strong>in</strong> [8] that convergence of the boosted scheme is very
20 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
TABLE I<br />
ASYMPTOTIC CODING GAIN FOR DIFFERENT MAPPING RULES<br />
OVERALL MAPPING RULE ϖ Asymptotic cod<strong>in</strong>g ga<strong>in</strong> ˜ Ω 2<br />
Gray-labelled BI-STCM-ID 0.4298<br />
Optimally-labelled BI-STCM-ID 2.3414<br />
Boosted space-time diversity scheme 1.0896<br />
Fig. 5. Bit Error Rate vs. Eb/N0<br />
fast, i.e. the improvement <strong>in</strong> the system performance from one<br />
iteration to another is significant.<br />
The boosted scheme <strong>in</strong>volves iterative decod<strong>in</strong>g. Therefore,<br />
its advantages should occur <strong>in</strong> the error-free feedback case.<br />
The distance spectra Def of all considered systems are plotted<br />
In Fig. 4 . As it was said above, to obta<strong>in</strong> good asymptotic<br />
performance, the frequency of short distances <strong>in</strong> the spectrum<br />
should be m<strong>in</strong>imized. The shortest distance occurr<strong>in</strong>g <strong>in</strong> the<br />
spectrum should be maximized as well. In light of these<br />
assumptions, the Gray-labelled BI-STCM-ID is the worst<br />
one (most of the spectrum entries have the lowest possible<br />
value d/d0 = 1). Therefore, such system is <strong>in</strong>appropriate for<br />
iterative decod<strong>in</strong>g.<br />
The boosted scheme is far better than the Gray-labelled BI-<br />
STCM-ID, as the shortest possible distance d/d0 = 1 does<br />
not occur at all. Instead, the most common entry d/d0 = 5<br />
accounts for as much as 3/8 of the total, <strong>and</strong> the highest d/d0<br />
(one <strong>in</strong> every eight entries) equals 90.<br />
The “optimally”-labelled BI-STCM-ID w<strong>in</strong>s the competition<br />
for the best mapp<strong>in</strong>g rule <strong>in</strong> the error-free feedback case.<br />
The lowest distance (every other entry) is d/d0 = 25, <strong>and</strong> the<br />
highest (one <strong>in</strong> every four) equals 169.<br />
A “compact” quality measure of a mapp<strong>in</strong>g rule is the<br />
asymptotic cod<strong>in</strong>g ga<strong>in</strong> ˜ Ω 2 . Its values for all the considered<br />
systems are listed <strong>in</strong> Table I. The results confirm the analysis<br />
of distance spectrum, i.e. the Gray-labelled BI-STCM-ID is<br />
the worst system under the condition of error-free feedback,<br />
the “optimally”-labelled BI-STCM-ID is the best one, <strong>and</strong> the<br />
boosted scheme is <strong>in</strong> the middle.<br />
The mapp<strong>in</strong>g rule is not the only one that <strong>in</strong>fluences the<br />
whole system performance. To work properly, the system<br />
requires a good match between the mapp<strong>in</strong>g rule <strong>and</strong> the<br />
convolutional code. The author of this paper showed <strong>in</strong> [7]<br />
that the “optimally”-labelled BI-STCM-ID, <strong>in</strong> contrary to the<br />
boosted scheme, cannot cooperate with the [171 133]OCT code<br />
at low Eb/N0 values (i.e. the decod<strong>in</strong>g trajectory on the EXIT<br />
chart is p<strong>in</strong>ched off ). In fact, “optimally”-labelled BI-STCM-<br />
ID has only been considered <strong>in</strong> the literature <strong>in</strong> comb<strong>in</strong>ation<br />
with convolutional codes of short free distance to avoid the<br />
p<strong>in</strong>ch-off effect.<br />
Thanks to the fact that the boosted scheme is well matched<br />
to the [171 133]OCT code, two goals are achieved: compatibility<br />
with the 802.11n specification, <strong>and</strong> the asymptotic<br />
bound steeper than for “optimally”-labelled BI-STCM-ID. In<br />
consequence, the latter performs worse asymptotically, <strong>in</strong> spite<br />
of higher asymptotic cod<strong>in</strong>g ga<strong>in</strong>.<br />
Till now it has been shown that the boosted space-time<br />
diversity scheme performs better asymptotically than the Graylabelled<br />
BI-STCM-ID. To compare the performance at low<br />
Eb/N0, Monte Carlo simulation was conducted. Each frame<br />
consisted of 10 000 bits. The convolutional code <strong>in</strong>herited<br />
from WLAN specifications <strong>and</strong> flat Rayleigh fad<strong>in</strong>g MIMO<br />
channel were assumed. For statistical reliability, 5 × 10 8 data<br />
bits were transmitted for each Eb/N0 value. The simulation<br />
results are shown <strong>in</strong> TABLE I. It can be observed that the<br />
decrement <strong>in</strong> Bit Error Rate from one iteration to another<br />
is <strong>in</strong>significant for Gray-labelled BI-STCM-ID, which makes<br />
the iterative process<strong>in</strong>g worthless. The first-pass performance<br />
of the boosted scheme is worse than for BI-STCM-ID. This<br />
fact results from disadvantageous D spectrum of the boostedscheme.<br />
Nevertheless, the iterative process converges fast <strong>and</strong><br />
a reasonable BER can be reached after only a few iterations.<br />
V. CONCLUSION<br />
A boosted scheme deriv<strong>in</strong>g advantages from both Gray<br />
<strong>and</strong> “optimal” constellation labell<strong>in</strong>gs has been analyzed. The<br />
proposed scheme outperforms the Gray-labelled BI-STCM-ID<br />
for any Eb/N0 value. The “optimally”-labelled BI-STCM-ID<br />
has been excepted from the comparison due to a mismatch<br />
between optimal labell<strong>in</strong>g <strong>and</strong> [171 133]OCT convolutional<br />
code.<br />
As orthogonality of the space-time code <strong>in</strong> the proposed<br />
scheme has been lost, signal detection is more complex <strong>and</strong><br />
further research on its simplification is necessary.<br />
REFERENCES<br />
[1] 802.11n-2009 IEEE St<strong>and</strong>ard for Information Technology-Part 11: Wireless<br />
LAN Medium Access Control (MAC) <strong>and</strong> Physical Layer (PHY)<br />
Specifications Amendment: Enhancements for Higher Throughput.<br />
[2] Y. Huang <strong>and</strong> J. Ritcey, “Tight ber bounds for iteratively decoded bit<strong>in</strong>terleaved<br />
space-time coded modulation,” IEEE Commun. Lett., vol. 8,<br />
Mar. 2004.<br />
[3] G. Caire <strong>and</strong> G. Taricco, “Bit-<strong>in</strong>terleaved coded modulation,” IEEE Trans.<br />
Inf. Theory, vol. 44, May 1998.<br />
[4] Y. Huang <strong>and</strong> J. Ritcey, “Optimal constellation label<strong>in</strong>g for iteratively<br />
decoded bit-<strong>in</strong>terleaved space-time coded modulation,” IEEE Trans. Inf.<br />
Theory, vol. 51, no. 5, May 2005.<br />
[5] S. Benedetto, D. Divsalar, G. Montorsi, <strong>and</strong> F. Pollara, “A soft-<strong>in</strong>put softoutput<br />
app module for iterative decod<strong>in</strong>g of con-catenated codes,” IEEE<br />
Commun. Lett., vol. 1, Jan. 1997.<br />
[6] F. Schreckenbach <strong>and</strong> P. Henkel, “Analysis <strong>and</strong> design of mapp<strong>in</strong>gs for<br />
iterative decod<strong>in</strong>g of BICM,” <strong>in</strong> Proc. of URSI Symposium, Poznan, 2005.<br />
[7] M. Krasicki <strong>and</strong> P. Szulakiewicz, “A new space-time diversity scheme for<br />
WLAN systems,” <strong>in</strong> Proc. of 19-th IEEE Personal, Indoor <strong>and</strong> Mobile<br />
Radio Communications Conference, Cannes, 2008.<br />
[8] ——, “Boosted space-time diversity scheme for wireless communications,”<br />
IET Electron. Lett., vol. 45, no. 16, pp. 843–844, Jul. 2009.
MACIEJ KRASICKI: SPECTRAL ANALYSIS OF BOOSTED SPACE-TIME DIVERSITY SCHEME 21<br />
[9] S. ten Br<strong>in</strong>k, “Convergence of iterative decod<strong>in</strong>g,” IET Electron. Lett.,<br />
vol. 25, no. 10, pp. 806–808, May 1999.<br />
Maciej Krasicki received the M.S. degree <strong>in</strong> <strong>Electronics</strong> <strong>and</strong> Telecommunications<br />
from Poznan University of Technology, Pol<strong>and</strong>, <strong>in</strong> 2006. S<strong>in</strong>ce then he<br />
has been work<strong>in</strong>g towards the Ph.D. degree. His dissertation work concerns<br />
a new (“boosted”) space-time diversity scheme, designed to support iterative<br />
decod<strong>in</strong>g at the receiver of WLAN systems. His Ph.D. defense took place <strong>in</strong><br />
<strong>2010</strong>.<br />
From 2009 he has been with the Faculty of <strong>Electronics</strong> <strong>and</strong> Telecommunications,<br />
Poznan University of Technology, as a Research Assistant. His<br />
research <strong>in</strong>terests <strong>in</strong>clude multi-antenna transmission, space-time cod<strong>in</strong>g <strong>and</strong><br />
iterative signal process<strong>in</strong>g. He has published several papers <strong>in</strong> journals (e.g.<br />
<strong>Electronics</strong> Letters) <strong>and</strong> conference proceed<strong>in</strong>gs.
22 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Krylov Subspace Methods <strong>in</strong> Application to<br />
WCDMA Network Optimization<br />
Abstract—Krylov subspace methods, which <strong>in</strong>clude, e.g. CG,<br />
CGS, Bi-CG, QMR or GMRES, are commonly applied as l<strong>in</strong>ear<br />
solvers for sparse large-scale l<strong>in</strong>ear least squares problems. In the<br />
paper, we discuss the usefulness of such methods to the optimization<br />
of WCDMA networks. We compare the selected methods<br />
with respect to their convergence properties <strong>and</strong> computational<br />
complexity, us<strong>in</strong>gatypical upl<strong>in</strong>kmodel for a WCDMA network.<br />
The comparison shows that GMRES is the most suitable method<br />
for our task.<br />
Index Terms—Krylov subspace methods, WCDMA network<br />
optimization, l<strong>in</strong>ear solvers, CG, GMRES<br />
Rafal Zdunek <strong>and</strong> Maciej Nawrocki<br />
I. INTRODUCTION<br />
OUR considerations are restricted to WCDMA network<br />
optimization at the stage of layout design. In this approach,the<br />
variablesof the cost functionare usually expressed<br />
<strong>in</strong> terms of transmitted powers that depend on the parameters<br />
to be optimized. The parameters basically concern base stations,<br />
i.e. their <strong>number</strong>, locations, antenna azimuth <strong>and</strong> tilt as<br />
well as pilot channel powers. The details on this are given <strong>in</strong><br />
[1]. Exclud<strong>in</strong>g very simplified models, the transmitted powers<br />
usually cannot be presented as analytical functions of the<br />
desired parameters.This implies the use of numerical methods<br />
for the computations of transmitted powers. Comput<strong>in</strong>g these<br />
powersis the most computationally<strong>in</strong>tensive task <strong>in</strong> an overall<br />
optimization problem, so f<strong>in</strong>d<strong>in</strong>g a proper (fast) method seems<br />
to be crucial. Assum<strong>in</strong>g the target Signal-to-Interference(SIR)<br />
valuesfor each l<strong>in</strong>k betweenaBase Station (BS) <strong>and</strong> a Mobile<br />
Station (MS), the transmitted powers can be computed from<br />
the system of l<strong>in</strong>ear equations:<br />
Ap = b, (1)<br />
where A =∈ IR K×K is the system matrix of coefficients that<br />
depend on the l<strong>in</strong>k ga<strong>in</strong>s, othogonality factors (for downl<strong>in</strong>k)<br />
<strong>and</strong> target SIR values, p ∈ IR K is the vector of unknown<br />
transmitted powers, b ∈ IR K is the noise vector. The aim<br />
is to f<strong>in</strong>d a possible best estimation of the vector p at the<br />
least computational cost. It should be noted that the task is<br />
very challeng<strong>in</strong>g s<strong>in</strong>ce the system (1) can be very large (even<br />
after apply<strong>in</strong>g the dimension reduction technique [2], [3]) <strong>and</strong><br />
such estimations must be repeated many times to provide<br />
many Monte Carlo (MC) samples used <strong>in</strong> static simulators<br />
for network plann<strong>in</strong>g <strong>and</strong> optimization [4], [5]. The system<br />
(1) has rather good numerical properties (square, consistent,<br />
R. Zdunek is with Institute of Telecommunications, Tele<strong>in</strong>formatics, <strong>and</strong><br />
Acoustics, Wroclaw University of Technology, 50–370 Wroclaw, Pol<strong>and</strong>, email:<br />
rafal.zdunek@pwr.wroc.pl<br />
M. Nawrocki is with OPTYME Consult<strong>in</strong>g, Wroclaw, Pol<strong>and</strong>, email:<br />
maciej.nawrocki@optyme.pl<br />
well-conditioned), <strong>and</strong> hence many l<strong>in</strong>ear solvers can be used<br />
<strong>in</strong> our application. Nevertheless, not all the methods have the<br />
same convergence properties <strong>and</strong> computational complexity,<br />
thus there is a need to study the usefulness of these methods<br />
to our task. The problem has been already tackled for <strong>in</strong><br />
our previous works [1], [6], [7], where we compared the<br />
Gaussian elim<strong>in</strong>ation <strong>and</strong> some iterative methods such as<br />
Jacobi, Gauss-Seidel, Successive Over-Relaxation (SOR), <strong>and</strong><br />
Conjugate Gradient Square (CGS). Some numerical results<br />
from [6], [7] will be rem<strong>in</strong>ded here. F<strong>in</strong>ally, we concluded<br />
<strong>in</strong> [6] that the Gauss-Seidel <strong>and</strong> CGS gave the best results.<br />
S<strong>in</strong>ce the CGS belongsto a class of Krylovsubspacemethods,<br />
we decided to cont<strong>in</strong>ue our tests with respect to the Krylov<br />
subspace methods which we shortly present <strong>in</strong> Section II. The<br />
comparison results are presented <strong>in</strong> Section III, <strong>and</strong> f<strong>in</strong>ally<br />
some conclud<strong>in</strong>g remarks are given <strong>in</strong> Section IV.<br />
II. KRYLOV SUBSPACE METHODS<br />
Krylov subspace methods are widely applied to solve largescale<br />
l<strong>in</strong>ear systems aris<strong>in</strong>g <strong>in</strong> many areas of science, especially<br />
to solve discretized Partial Differential Equations (PDE)<br />
[8], [9]. Due to their low computational cost, the methods can<br />
be also useful <strong>in</strong> the optimization of WCDMA networks. A<br />
short survey of the Krylov subspace methods that are used <strong>in</strong><br />
our experiments is given below.<br />
• CGLS<br />
Thefirst versionofthe ConjugateGradients(CG) method<br />
was proposed by Hestenes <strong>and</strong> Stiefel [10], <strong>and</strong> it is<br />
commonly used for solv<strong>in</strong>g symmetric l<strong>in</strong>ear systems. It<br />
iteratively m<strong>in</strong>imizesthe gradientof a quadraticobjective<br />
function with gradient updates derived from orthogonal<br />
directions. S<strong>in</strong>ce <strong>in</strong> our application the symmetry condition<br />
is not met, the CG method is applied to the normal<br />
equations. In the literature, such method is known as<br />
CGLS <strong>and</strong> it may be found <strong>in</strong> many implementations.<br />
We used the Hansen’s implementation [11].<br />
• CGS<br />
The Conjugate Gradient Square (CGS) method was <strong>in</strong>vented<br />
by Sonneveld [12] <strong>and</strong> it <strong>in</strong>volvesthe CG scheme.<br />
In contrary to the CG, it can be used to non-symmetric<br />
systems. Moreover, it is not sensitive to so-called the<br />
serious breakdown that may occur <strong>in</strong> the CG.<br />
• BiCG<br />
The Bi-Conjugate Gradient (BiCG) method, proposed<br />
by Fletcher [13], belongs to a group of bi-orthogonal<br />
methods <strong>and</strong> extends the st<strong>and</strong>ard CG method to nonsymmetric,<br />
large <strong>and</strong> sparse systems of l<strong>in</strong>ear equations.<br />
Hence, it may be suitable for our application.
ZDUNEK AND NAWROCKI: KRYLOV SUBSPACE METHODS IN APPLICATION TO WCDMA NETWORK OPTIMIZATION 23<br />
TABLE I<br />
COMPUTATIONAL COST OF ONE ITERATIVE STEP FOR THE ANALYZED<br />
METHODS.THE SUBSCRIPTS m, d, a, s DENOTE ELEMENTARY<br />
MULTIPLICATIVE, DIVISION, ADDITION, AND SUBTRACTION OPERATIONS.<br />
THE SUBSCRIPT f STANDS FOR A FUNCTION EVALUATION (SQUARE<br />
ROOTING OR POWERING).<br />
Method Computational cost of one iteration<br />
CGLS (5K2 + 6K) m/d + (5K2 + 7K) a/s<br />
CGS (2K2 + 10K) m/d + (2K2 + 12K) a/s<br />
BiCG (4K2 + 9K) m/d + (5K2 + 10K) a/s<br />
BiCGSTAB (6K 2 + 12K) m/d + (6K 2 + 14K) a/s<br />
QMR (3K 2 + 14K) m/d + (3K 2 + 14K) a/s + (2K + 2)f<br />
GMRES depends on many factors (sparsity)<br />
• BiCGSTAB<br />
The BiConjugate Gradients Stabilized (BiCGSTAB)<br />
method was developed by Van der Vorst [8], [9]. The<br />
BiCGSTAB differs from the CGS only with the way<br />
of comput<strong>in</strong>g a residual vector. It is reported <strong>in</strong> [8]<br />
that the BiCGSTAB has better convergence properties<br />
due to local m<strong>in</strong>imization of successive updates for<br />
the residual vector. The curve of the l2 norm of the<br />
residual vector is smoother <strong>and</strong> steeper than for the<br />
CGS. Unfortunately, some perturbations <strong>in</strong> convergence<br />
or even a serious breakdown of an iterative process may<br />
occasionally happend, especially if the system matrix has<br />
complex eigenvalues.<br />
• QMR<br />
The Quasi-M<strong>in</strong>imal Residual (QMR) method that was<br />
designed by Freund <strong>and</strong> Nachtigal [14] uses the similar<br />
assumptions as the BiCG but considerable difference<br />
exists <strong>in</strong> the residual smooth<strong>in</strong>g technique. Its highest<br />
advantage is a numerical stability, i.e. it avoids the case<br />
of serious breakdown. There are many implementations<br />
of the QMR [9], [14]. In the experiments we used the<br />
implementations given <strong>in</strong> MATLAB 7.0.<br />
• GMRES<br />
The GMRES method was proposed by Saad <strong>and</strong> Schultz<br />
[15] for solv<strong>in</strong>g l<strong>in</strong>ear least squares problems with nonsymmetric<br />
matrices without a necessity of creat<strong>in</strong>g the<br />
normal equations. In the experiments we used the MAT-<br />
LAB implementation where the Gram-Schmidt orthogonalization<br />
is obta<strong>in</strong>ed with the Givens rotations.<br />
The roughly estimated computational costs of all the algorithms<br />
used <strong>in</strong> our experiments are given <strong>in</strong> Table. 1. The<br />
computational cost for the GMRES is not easy to estimate<br />
because it depends on the system matrix used. For a sparse<br />
matrix, the cost is considerably lower than for a dense matrix<br />
because the related <strong>number</strong> of the <strong>in</strong>volved Givens rotations<br />
is much smaller. In our application, the system matrix may be<br />
very sparse if a large network is analyzed (without us<strong>in</strong>g the<br />
dimension reduction technique [2], [3]).<br />
III. NUMERICAL RESULTS<br />
The experiments demonstrat<strong>in</strong>g the efficiency of the analyzed<br />
methods are performed for a r<strong>and</strong>omly selected MC<br />
snapshot <strong>in</strong> upl<strong>in</strong>k transmission with both omnidirectional<br />
y−axis [km]<br />
20<br />
18<br />
16<br />
14<br />
12<br />
10<br />
8<br />
6<br />
4<br />
2<br />
Distribution of BSs <strong>and</strong> MSs<br />
BS<br />
MS<br />
0<br />
0 5 10 15<br />
x−axis [km]<br />
20 25 30<br />
(a)<br />
(b)<br />
Fig. 1. (a) Layout of BSs <strong>and</strong> MSs; (b) The <strong>number</strong>s of users assigned to<br />
each cell.<br />
antennas <strong>and</strong> Smart Antennas (SA). Typically, we assume<br />
1000 users r<strong>and</strong>omly distributed <strong>in</strong> 104 cells with a mixture<br />
of uniform <strong>and</strong> skrew-Gaussian distributions. Hence, we have<br />
A ∈ IR 1000×1000 , K = 1000 <strong>and</strong> M = 104. In our approach,<br />
weassumethattheanalyzednetworkisnotoverloaded.Forthe<br />
overloaded case, some values of the target SIR vector should<br />
be decreased, which can be done with many techniques, e.g.<br />
with the one described <strong>in</strong> [3]. The layout of BSs <strong>and</strong> MSs is<br />
presented <strong>in</strong> Fig. 1(a). The geometry of the tested area <strong>and</strong><br />
the <strong>number</strong> of the users <strong>in</strong> each cell are shown <strong>in</strong> Fig. 1(b).<br />
Half of the users work with a voice service (Rb = 12.2kbps),<br />
<strong>and</strong> the other half with a data service (Rb = 64kbps).<br />
For thissnapshot<strong>and</strong>traditionalantennas(omnidirectional):<br />
maxi{|λi(A)|} = 2.1 × 10 −7 <strong>and</strong> m<strong>in</strong>i{|λi(A)|} = 8.7 ×<br />
10 −13 , <strong>and</strong> for the SA: maxi{|λi(A (SA))|} = 2.1 × 10 −6 <strong>and</strong><br />
m<strong>in</strong>i{|λi(A (SA))|} = 9.2 × 10 −12 . Hence, the convergence<br />
of the Krylov subspace method is def<strong>in</strong>itely guaranteed [8],<br />
[9], [12]–[15]. All the iterative algorithms are run until the<br />
stopp<strong>in</strong>g criterion e k = ||p k − p k−1 ||∞ ≥ ǫ is met, where<br />
for arbitrary u: ||u||∞ = maxi{ui}, <strong>and</strong> ǫ is a small positive<br />
<strong>number</strong>.We assume that the solution should be computedwith<br />
the accuracy up to the fifth significant digit, thus ǫ = 16 −6 .<br />
The plots of e k versus iterations are illustrated <strong>in</strong> Fig. 2(a)<br />
<strong>and</strong> Fig. 2(b) for the cases of traditional antennas <strong>and</strong> SAs,<br />
respectively.
24 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
(a)<br />
(b)<br />
Fig. 2. History of error e k versus iterations for: (a) traditional antennas, (b)<br />
SA.<br />
The dashed horizontal l<strong>in</strong>es <strong>in</strong> Fig. 2 mark the error level<br />
of 10 −6 at which the iterative process is stopped. It follows<br />
from Fig. 2(a) that this level or lower is reached by the CGS,<br />
CGLS, BiCG, BiCGSTAB, QMR <strong>and</strong> GMRES after runn<strong>in</strong>g<br />
7, 10, 9, 6, 9 <strong>and</strong> 9 iterations, respectively. For SAs (see<br />
Fig. 2(b)), this level is reached with<strong>in</strong> 3, 5, 5, 3, 5 <strong>and</strong><br />
5 iterations for respective methods. In [6] the Richardson,<br />
Jacobi, Gauss-Seidel, SOR methods stopped at the same error<br />
level after perform<strong>in</strong>g 36, 50, 29, 14 iterations for traditional<br />
antennas, <strong>and</strong> 15, 4, 3, 5 iterations for SAs, respectively. All<br />
the discussed methodshavebeenappliedto the preconditioned<br />
versionofthesystem(1),wheretheright-h<strong>and</strong>precondition<strong>in</strong>g<br />
was applied as <strong>in</strong> [1], [6], [7].<br />
To simplifythecomparisonanalysis,letusdropthenotation<br />
of the k<strong>in</strong>d of arithmetic operations. First, let us consider<br />
traditional antennas. Thus, it follows from Table I that the<br />
computational cost of perform<strong>in</strong>g 7 iterations with the CGS is<br />
about 28K 2 + 154K arithmetic operations. For the CGLS,<br />
BiCG, BiCGSTAB <strong>and</strong> QMR we have: 100K 2 + 130K,<br />
91K 2 +171K, 72K 2 +156K <strong>and</strong> 54K 2 +252K+18K,respectively,<br />
where additional 18K <strong>in</strong> QMR means the cost related<br />
to the function evaluation, which may be quite expensive but<br />
dependent on the software <strong>and</strong> hardware used. To rem<strong>in</strong>d, we<br />
got 72K 2 + 108K, 100K 2 + 150K, 87K 2 , <strong>and</strong> 44K 2 + 16K<br />
for the preconditionedRichardson,Jacobi’s, Gauss-Seidel,<strong>and</strong><br />
SOR methods. A similar analysis for the case of SAs gives<br />
the follow<strong>in</strong>g rough estimations of the costs: 32K 2 + 46K,<br />
8K 2 + 12K, 9K 2 , 16K 2 + 6K, 14K 2 + 67K, 50K 2 + 65K,<br />
45K 2 + 95K, 36K 2 + 78K, <strong>and</strong> 30K 2 + (140 + 10)K for<br />
the correspond<strong>in</strong>gmethods: Richardson, Jacobi, Gauss-Seidel,<br />
SOR, CGS, CGLS, BiCG, BiCGSTAB, <strong>and</strong> QMR. Because<br />
the estimation of the cost for GMRES is not so easy, we<br />
compare this method only with respect to the elapsed time<br />
of perform<strong>in</strong>g 10 iterations <strong>in</strong> the same computational <strong>and</strong><br />
hardware environment (MATLAB 7.0).<br />
The elapsed time [<strong>in</strong> seconds] measured <strong>in</strong> MATLAB is<br />
given<strong>in</strong> TableIIwherewe comparethe methodsappliedto the<br />
problemsof different scales. The first two columnsrefer to the<br />
small-scale problem that occured after apply<strong>in</strong>g the dimension<br />
reduction technique ( [2], [3]) to the snapshot described above<br />
(M = 104, K = 1000). Thus, our system matrix is reduced<br />
to the size 104 by 104. S<strong>in</strong>ce <strong>in</strong> real applications much bigger<br />
problems must be resolved, we analyze a bigger case – the<br />
snapshot with 300 cells <strong>and</strong> 3000 users – without us<strong>in</strong>g the<br />
dimension reduction technique but with the above-mentioned<br />
precondition<strong>in</strong>g. The elapsed times are given <strong>in</strong> the last two<br />
columns.Notethatthemeasuredtimeisexemplary<strong>and</strong><strong>in</strong>each<br />
snapshot it may be slightly different due to the difference <strong>in</strong><br />
properties of the system matrix.<br />
IV. CONCLUSIONS<br />
Compar<strong>in</strong>g the estimations of the computational costs, we<br />
canconcludethattheGauss-Seidelmethodisthe mostpromis<strong>in</strong>g,<br />
especially for the SA case. For the traditional antennas,<br />
the CGS is the fastest, <strong>and</strong> then the SOR.<br />
However, with reference to Table II, we can conclude that<br />
for large-scale problems the GMRES is the fastest algorithm.<br />
Thus, for the analysis of a large network (with many BSs), the<br />
GMRES should be used <strong>in</strong> a static simulator. For small-scale<br />
problems, especially for a small <strong>number</strong> of BSs, the Gauss-<br />
Seidel <strong>and</strong> CGS are optimal.<br />
ACKNOWLEDGMENTS<br />
This work was supported with the Grant No. N517 010<br />
32/1675 from Polish State Committee for Scientific Research.<br />
REFERENCES<br />
[1] M. J. Nawrocki, M. Dohler, <strong>and</strong> A. H. Aghvami, Eds., Underst<strong>and</strong><strong>in</strong>g<br />
UMTS Radio Network Modell<strong>in</strong>g, Plann<strong>in</strong>g <strong>and</strong> Automated Optimisation:<br />
Theory <strong>and</strong> Practice. John Wiley <strong>and</strong> Sons, 2006.<br />
[2] L. Mendo <strong>and</strong> J. M. Hern<strong>and</strong>o, “On dimension reduction for the power<br />
control,” IEEE Trans. On Communications, vol. 49, no. 2, pp. 243–248,<br />
2001.<br />
[3] R. Zdunek <strong>and</strong> M. J. Nawrocki, “Improved model<strong>in</strong>g of highly loaded<br />
UMTS network with nonnegative constra<strong>in</strong>ts,” <strong>in</strong> IEEE 17th International<br />
Symposium on Personal, Indoor <strong>and</strong> Mobile Radio Communications<br />
(PIMRC 2006), Hels<strong>in</strong>ki, F<strong>in</strong>l<strong>and</strong>, September 2006.<br />
[4] J. Laiho, A. Wacker, <strong>and</strong> T. Novosad, Radio Network Plann<strong>in</strong>g <strong>and</strong><br />
Optimization for UMTS. Chichester: John Wiley <strong>and</strong> Sons, 2002.<br />
[5] A.Wacker, J.Laiho-Steffens, K.Sipila, <strong>and</strong> M.Jasberg,“Static simulator<br />
for study<strong>in</strong>g WCDMA radio network plann<strong>in</strong>g issues,” <strong>in</strong> Proc. IEEE<br />
Vehicular Technology Conference, Houston, Texas, USA, May 1999, pp.<br />
2436–2440.
ZDUNEK AND NAWROCKI: KRYLOV SUBSPACE METHODS IN APPLICATION TO WCDMA NETWORK OPTIMIZATION 25<br />
TABLE II<br />
ELAPSED TIME [IN SECONDS] OF PERFORMING10 ITERATIONS WITH DIFFERENT ALGORITHMS AND FOR DIFFERENT SIZE OF THE ANALYZED NETWORK<br />
EQUIPPED WITH TRADITIONAL(T) AND INTELLIGENT(SMART) ANTENNAS.<br />
Problem/Method K = 1000 K = 1000 K = 3000 K = 3000<br />
M = 104 M = 104 M = 300 M = 300<br />
A ∈ IR 104×104 A ∈ IR 1000×1000 A ∈ IR 3000×3000 A ∈ IR 3000×3000<br />
(T) (SMART) (T) (SMART)<br />
Richardson 0.04 0.13 1.056 1.101<br />
Jacobi 0.01 0.13 1.072 1.081<br />
Gauss-Seidel 0.01 0.231 1.952 1.923<br />
SOR 0.02 0.311 2.943 2.824<br />
CGLS 0.03 0.12 1.121 0.991<br />
BiCG 0.06 0.211 1.562 1.523<br />
BiCGSTAB 0.088 0.41 2.053 2.403<br />
CGS 0.011 0.257 1.572 1.701<br />
QMR 0.091 0.241 1.592 1.643<br />
GMRES 0.10 0.15 0.691 0.771<br />
[6] R. Zdunek, M. J. Nawrocki, M. Dohler, <strong>and</strong> A. H. Aghvami, “Application<br />
of l<strong>in</strong>ear solvers to UMTS network optimization without <strong>and</strong> with<br />
smart antennas,” <strong>in</strong> IEEE 16th International Symposium on Personal,<br />
Indoor <strong>and</strong> Mobile Radio Communications (PIMRC 2005), vol. 4,<br />
Berl<strong>in</strong>, Germany, September 11–14 2005, pp. 2322–2326.<br />
[7] R. Zdunek <strong>and</strong> M. J. Nawrocki, “On l<strong>in</strong>ear solvers <strong>in</strong> applications<br />
to WCDMA network optimization,” <strong>in</strong> Proc. National Conference on<br />
Radio-communication, Radio <strong>and</strong> Television (KKRRiT),Krakow, Pol<strong>and</strong>,<br />
June 15–17 2005, pp. 77–80.<br />
[8] G. H. Golub <strong>and</strong> H. A. V. der Vorst, “Closer to the solution: Iterative<br />
l<strong>in</strong>ear solvers,” <strong>in</strong> The State of the Art <strong>in</strong> Numerical Analysis, I. Duff<br />
<strong>and</strong> G. Watson, Eds. Clarendon Press, Oxford, 1997, pp. 63–92.<br />
[9] Y. Saad <strong>and</strong> H. A. V. der Vorst, “Iterative solution of l<strong>in</strong>ear systems <strong>in</strong><br />
the 20-th century,” Journal of Computational <strong>and</strong> Applied Mathematics,<br />
vol. 123, no. 1–2, pp. 1–33, 2000.<br />
[10] M. R. Hestenes <strong>and</strong> E. Stiefel, “Method of conjugate gradients for<br />
solv<strong>in</strong>g l<strong>in</strong>ear systems,” J. Res. Nat. Bur. St<strong>and</strong>ards, vol. 49, pp. 409–<br />
436, 1952.<br />
[11] P. C. Hansen, “Regularization tools version 4.0 for matlab 7.3,” Numerical<br />
Algorithms, vol. 46, pp. 189–194, 2007.<br />
[12] P. Sonneveld, “CGS: A fast lanczos-type solver for nonsymmetric l<strong>in</strong>ear<br />
systems,” SIAM J. Sci. Statist. Comput., vol. 10, pp. 36–52, 1989.<br />
[13] R. Fletcher, “Conjugate gradient methods for <strong>in</strong>def<strong>in</strong>ite systems,” <strong>in</strong><br />
Numerical Analysis, ser. Lecture Notes Math., G. Watson, Ed. Berl<strong>in</strong>-<br />
Heidelberd-New York: Spr<strong>in</strong>ger-Verlag, 1976, vol. 506, pp. 73–89.<br />
[14] R. W. Freund <strong>and</strong> N. M. Nachtigal, “QMR: a quasi-m<strong>in</strong>imal residual<br />
method for non-hermitian l<strong>in</strong>ear systems,” Numer. Math., vol. 60, pp.<br />
315–339, 1991.<br />
[15] Y. Saad <strong>and</strong> M. H. Schultz, “GMRES: a generalized m<strong>in</strong>imal residual<br />
algorithm for solv<strong>in</strong>g nonsymmetic l<strong>in</strong>ear systems,” SIAM J. Sci. Statist.<br />
Comput., vol. 7, pp. 856–869, 1986.<br />
Rafal Zdunek received the M.Sc. <strong>and</strong> Ph.D. degrees <strong>in</strong> telecommunications<br />
from Wroclaw University of Technology, Pol<strong>and</strong>, <strong>in</strong> 1997 <strong>and</strong> 2002, respectively.<br />
S<strong>in</strong>ce 2002, he has been a Lecturer <strong>in</strong> Institute of Telecommunications,<br />
Tele<strong>in</strong>formatics <strong>and</strong> Acoustics, Wroclaw University of Technology, Pol<strong>and</strong>.<br />
In 2004, he was a Visit<strong>in</strong>g Associate Professor <strong>in</strong> the Institute of Statistical<br />
Mathematics, Tokyo, Japan. From 2005 to 2007, he worked as Research<br />
Scientist <strong>in</strong> Bra<strong>in</strong> Science Institute, RIKEN, Saitama, Japan. His area of<br />
<strong>in</strong>terest <strong>in</strong>cludes numerical methods <strong>and</strong> <strong>in</strong>verse problems <strong>in</strong> application<br />
to WCDMA network optimization, nonnegative matrix factorization, bl<strong>in</strong>d<br />
source separation, <strong>and</strong> tomographic image reconstruction. He has published<br />
over 60 journal <strong>and</strong> conference papers. He is a co-author of the monograph<br />
Nonnegative Matrix <strong>and</strong> Tensor Factorizations: Applications to Exploratory<br />
Multi-way Data Analysis <strong>and</strong> Bl<strong>in</strong>d Source Separation, published by John<br />
Wiley <strong>and</strong> Sons <strong>in</strong> 2009.<br />
Maciej Nawrocki received his M.Sc. <strong>and</strong> Ph.D. degrees <strong>in</strong> telecommunications<br />
from Wroclaw University of Technology, Pol<strong>and</strong>, <strong>in</strong> 1997 <strong>and</strong><br />
2002, respectively. S<strong>in</strong>ce then he was an Assistant Professor at the Wroclaw<br />
University of Technoogy, Research Fellow at the K<strong>in</strong>gs College London, UK<br />
(Centre for Telecommunications Research) <strong>and</strong> Visit<strong>in</strong>g Lecturer at University<br />
of Wroclaw. He also created ICT Research Centre with<strong>in</strong> Wroclaw Research<br />
Centre EIT+, be<strong>in</strong>g its first Director of Research. Now, he is with OPTYME<br />
Consult<strong>in</strong>g, concentrat<strong>in</strong>g on complex mobile network solutions. His areas<br />
of <strong>in</strong>terest <strong>in</strong>clude automated optimization, auto-tun<strong>in</strong>g, SON, measurement<br />
oriented optimization <strong>and</strong> propagation for mobile networks. He is the author<br />
of a <strong>number</strong> of publications, <strong>in</strong>clud<strong>in</strong>g a book about UMTS optimisation<br />
published by John Wiley <strong>and</strong> Sons.
26 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Stream<strong>in</strong>g Video over TFRC with L<strong>in</strong>ear<br />
Throughput Equation<br />
Agnieszka Chodorek <strong>and</strong> Robert R. Chodorek<br />
Abstract—The TCP-Friendly Rate Control (TFRC) protocol<br />
manifests strong equality towards compet<strong>in</strong>g TCP or TCPfriendly<br />
flows. Although the RFC 3448 suggests that TFRC is<br />
suitable for multimedia, this equality is a great disadvantage <strong>in</strong><br />
the case of transmitt<strong>in</strong>g multimedia over the TFRC.<br />
The TFRC emulates TCP-like congestion control us<strong>in</strong>g the<br />
TCP throughput equation. In the paper, we substitute the TCP<br />
throughput equation recommended for the TFRC with a l<strong>in</strong>ear<br />
throughput equation. Simulation results show that the proposed<br />
solution is more suitable for multimedia than the equation<br />
proposed <strong>in</strong> RFC 3448. Experiments were carried out us<strong>in</strong>g an<br />
event-driven ns-2 simulator, developed <strong>in</strong> U. C. Berkeley.<br />
Index Terms—congestion control, multimedia, TCP-friendly<br />
protocol<br />
I. INTRODUCTION<br />
THE phenomenon of the collapse of TCP transmissions<br />
which compete for b<strong>and</strong>width with multimedia over<br />
RTP/UDP or UDP, was the reason for the design of so-called<br />
TCP-friendly transport protocols. One of the best known, <strong>and</strong><br />
the first st<strong>and</strong>ardized TCP-friendly protocol was the TCP-<br />
Friendly Rate Control (TFRC) [1], [2]. This multipurposeprotocol<br />
was designed to carry different k<strong>in</strong>ds of data, <strong>in</strong>clud<strong>in</strong>g<br />
real-time multimedia.<br />
TCP-friendly transport protocols implement TCP-like congestion<br />
control<strong>and</strong> behaveunder congestionlike TCP. Among<br />
others, they equally share the throughput of bottleneck l<strong>in</strong>ks<br />
with TCP flows or other TCP-friendly flows. This feature is<br />
a great advantage <strong>in</strong> the case of bulk data transfer because<br />
it allows for the achievement of Quality of Service (QoS)<br />
appropriate for each transmission. In the case of real-time<br />
multimedia transmission, we can see the opposite tendency. If<br />
flow equalityis contraryto real-time requirements,we observe<br />
degradation of the QoS of the multimedia transmission. The<br />
deeper the conflict between equality <strong>and</strong> real-time becomes,<br />
the larger degradation can be observed [3], [4].<br />
The TFRC emulates TCP-like congestion control us<strong>in</strong>g<br />
the TCP throughput equation. The equation is used for the<br />
estimationof<strong>in</strong>stantaneousthroughputofTFRCundercongestion.<br />
In the paper, we substitute the TCP throughput equation<br />
recommended for the TFRC a l<strong>in</strong>ear function of packet error<br />
rate. The aim of such substitution is to develop a transport<br />
A. Chodorek is with the Department of Telecommunications, Photonics<br />
<strong>and</strong> Nanomaterials Kielce University of Technology, Kielce, Pol<strong>and</strong> (e-mail:<br />
a.chodorek@tu.kielce.pl).<br />
R. R. Chodorek is with the Department of Telecommunications The<br />
AGH University of Science <strong>and</strong> Technology, Kraków, Pol<strong>and</strong> (e-mail:<br />
chodorek@kt.agh.edu.pl).<br />
Manuscript received on July 29, <strong>2010</strong>. This work is supported by the Polish<br />
Government under Grant No. N517 012 32/2108 (years 2007¬2009).<br />
protocol which is more suitable for multimedia than TFRC<br />
<strong>and</strong> more TCP-friendly than RTP.<br />
The paper is organized as follows. Section 2 briefly describes<br />
the TFRC protocol. Section 3 proposes a l<strong>in</strong>ear functionwhichwillbeusedasathroughputequationfortheTFRC.<br />
Section 4 describessimulationexperiments.Section 5 presents<br />
the simulation results of TFRC <strong>and</strong> TCP transmissions <strong>in</strong><br />
shared l<strong>in</strong>k. Section 6 summarizes our experiences.<br />
II. THE TFRC PROTOCOL<br />
The TFRC protocol represents the modern approach to<br />
transport layer protocols, which treats the protocols as a set of<br />
build<strong>in</strong>g blocks – <strong>in</strong>dependent components from which transport<br />
protocols are assembled [5]. The TFRC is a congestion<br />
control build<strong>in</strong>g block designed to be reasonably fair when<br />
compet<strong>in</strong>g for b<strong>and</strong>width with TCP flows. As other control<br />
systems, the TFRC consists of:<br />
• a controller which makes decisions about the value of the<br />
controlled quantity,<br />
• a control device which adjusts the controlled quantity to<br />
the value given by the controller.<br />
In the case of TFRC, the controller (the congestion control<br />
mechanism) evaluates the output throughput of flow us<strong>in</strong>g<br />
the so-called TCP throughput equation, which is, <strong>in</strong> fact, an<br />
analytical model of the TCP behaviour under congestion. The<br />
equation describes TCP throughput as a function of packet<br />
errorrate. TheTFRC uses Padhye’smodelofTCP throughput,<br />
described <strong>in</strong> [6], [7]. Accord<strong>in</strong>g to this model, the throughput<br />
of the TCP protocol (<strong>and</strong>, <strong>in</strong> result, the TFRC throughput) is<br />
equal to:<br />
T (P ER) =<br />
MSS √ C<br />
RT T 2<br />
3 P ER+12P ER√ 3<br />
8 P ER(1+32P ER2 ) .<br />
where P ER denotes the packet error rate, T is a TCP<br />
throughput, <strong>and</strong> C is the scale coefficient.<br />
The output throughput of TFRC is adjusted to the value<br />
given by the controller us<strong>in</strong>g the rate control mechanism. This<br />
mechanism modulates the TFRC send<strong>in</strong>g rate <strong>in</strong> packets per<br />
second.<br />
The authors of RFC 3448 recommend that the TFRC is<br />
suitableforapplicationssuch astelephonyorstream<strong>in</strong>gmedia.<br />
They suggest also that the TFRC could be used <strong>in</strong> a transport<br />
protocol such as Real-time Transport Protocol (RTP) [8],<br />
which is commonly used as a transport protocol for audio<br />
<strong>and</strong> video transmission.<br />
(1)
CHODOREK AND CHODOREK: STREAMING VIDEO OVER TFRC WITH LINEAR THROUGHPUT EQUATION 27<br />
Fig. 1. Throughput of the RTP as a function of Packet Error Rate (PER).<br />
III. LINEAR THROUGHPUT EQUATION<br />
A typical feature of TCP-friendly protocols is an equality<br />
of compet<strong>in</strong>g TCP flows. This equality means that sometimes<br />
the TFRC is not able to meet the real-time requirements of<br />
multimedia transmission. It means that TCP is too aggressive<br />
(whencomparedwithTFRC)toallowtheTFRCtomanagethe<br />
real-time transmission of multimedia. As a result, the TFRC<br />
is not able to preserve QoS for multimedia traffic.<br />
A protocol which is aggressive enough to force real-time<br />
transmission <strong>in</strong> the presence of TCP flows is the RTP –<br />
the transport protocol <strong>in</strong>tended for the real-time multimedia<br />
transmission. However, the RTP protocol is not designed for<br />
TCP-friendl<strong>in</strong>ess <strong>and</strong> some researchers have reported that it<br />
can cause the degradation of TCP connections <strong>in</strong> a shared<br />
l<strong>in</strong>k.<br />
Our proposition is a comb<strong>in</strong>ation of two features: TCPfriendl<strong>in</strong>ess<br />
of the TFRC <strong>and</strong> good QoS of real-time multimedia<br />
transmission, presented by the RTP. We want to achieve<br />
this goal by apply<strong>in</strong>g elements of the real-time behavior of the<br />
RTP to the TFRC. As a result, the new TFRC should be more<br />
aggressive than the st<strong>and</strong>ard one <strong>and</strong> still able to co-operate<br />
with the TCP <strong>in</strong> a shared l<strong>in</strong>k.<br />
Because the RTP implements neither congestion control,<br />
flow control, nor error control, theo traffic offered will be<br />
reduced only by packet losses (Fig. 1). As a result, <strong>in</strong> a<br />
networkthat iswell-dimensionedformultimediathe analytical<br />
model of RTP throughputshould depend only on the target bit<br />
rate of the carried multimedia stream <strong>and</strong> the packet errorrate.<br />
Thus, the RTP throughput equation should be as follows:<br />
T (P ER) = Sp<br />
t0<br />
− Sl<br />
. (2)<br />
t0<br />
where P ER denotes the packet error rate, T is the RTP<br />
throughput, Sp is the amount of <strong>in</strong>formation (<strong>in</strong> bits) sent <strong>in</strong><br />
RTP packets(both<strong>in</strong>headers<strong>and</strong>payloads)dur<strong>in</strong>gthetime t0,<br />
Sl is an amount of <strong>in</strong>formation carried <strong>in</strong> RTP packets which<br />
were lost or damaged dur<strong>in</strong>g the time t0, t0 is the observation<br />
time.<br />
Note that the above analytical model of RTP throughput<br />
describes both the transmission of stream<strong>in</strong>g media over<br />
RTP/UDP <strong>and</strong> the transmission of stream<strong>in</strong>g media over UDP.<br />
In the paper, we propose to substitute the TCP throughput<br />
equation (1) used by the TFRC with the l<strong>in</strong>ear throughput<br />
equation:<br />
T (P ER) = T BR (1 − P ER) (3)<br />
where T BR is the target bit rate of multimedia stream.<br />
Because<br />
Sp<br />
= T BR (4)<br />
<strong>and</strong><br />
t0<br />
Sl<br />
Sp<br />
= P ER (5)<br />
The l<strong>in</strong>ear throughput equation describes, <strong>in</strong> fact, the<br />
RTP/UDP throughput as a function of packet error rate.<br />
Because the proposed equation is based on the RTP model,<br />
we believe it is aggressive enough to preserve the real-time<br />
character of transmitted steam<strong>in</strong>g media. However, it does not<br />
mean that TFRC will behave under congestion like the RTP if<br />
the l<strong>in</strong>ear throughput equation is used. The RTP protocol does<br />
not implement congestion control. It is not able to change the<br />
transmission rate due to congestion.<br />
TheTFRCwiththel<strong>in</strong>earthroughputequationwillstillhave<br />
congestion control, although the usage of this equation causes<br />
congestion control to be a “light” version. The send<strong>in</strong>g rate is<br />
reduced only by packets which are lost due to congestion. It<br />
means that TFRC can not aggressively avoid congestion but<br />
it does not allow the congestion to grow.<br />
IV. SIMULATION EXPERIMENTS<br />
Simulation experiments were carried out us<strong>in</strong>g s<strong>in</strong>glebottleneck<br />
topology (Fig. 2.). Senders S are connected to<br />
router R1 via 100 Mb/s l<strong>in</strong>ks with 1 µs propagation delay.<br />
The same l<strong>in</strong>ks are used to connect receivers R <strong>and</strong> router R2.<br />
Routers are connected via 4 Mb/s bottleneck l<strong>in</strong>k with 10 ms<br />
propagation delay.<br />
Constant Bit Rate (CBR) video stream is transmitted between<br />
S <strong>and</strong> R end-systems <strong>and</strong> the target bit rate of the<br />
stream is equal to B. Because we assume that the network is<br />
well-dimensioned for multimedia, 0 Mb/s ≤ B ≤ 4 Mb/s.<br />
Real-time CBR transmission is carried out us<strong>in</strong>g the TFRC<br />
<strong>and</strong> modified TFRC with l<strong>in</strong>ear throughput equation. For the<br />
sake of comparison, RTP/UDP protocols also are used. FTP<br />
over TCP transmissions are carried out between the pair of<br />
nodes S<br />
T CP<br />
i<br />
<strong>and</strong> R<br />
T CP<br />
i<br />
, i = 1,...,N. All transport protocols<br />
used <strong>in</strong> experimentshave the same size of data packets – 1000<br />
B (960 B of data + 40 B overheads).<br />
Dur<strong>in</strong>g the experiments we <strong>in</strong>vestigated achieved the<br />
throughput (both for multimedia <strong>and</strong> bulk data transfer).<br />
Experiments were carried out us<strong>in</strong>g Berkeley’s ns-2 simulator<br />
[8].<br />
V. SIMULATION RESULTS<br />
Inthe first experimentwe changedthe <strong>number</strong>ofcompet<strong>in</strong>g<br />
TCP flows N from 0 to 10. The target bit rate of CBR<br />
transmission was set to 1 Mb/s (1/4 of throughput of the<br />
bottleneck l<strong>in</strong>k). Results are shown <strong>in</strong> Fig. 3.<br />
Simulation results show that CBR video transmissions will<br />
preserve their real-time character if a modified TFRC with a<br />
l<strong>in</strong>ear equation is used <strong>in</strong> the transport layer. Stream<strong>in</strong>g video
28 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 2. Topology of simulated network.<br />
Fig. 3. Throughput of CBR the transmission as a function of N.<br />
over classic TFRC (with TCP throughput equation) causes<br />
strong degradation of a CBR connection <strong>in</strong> the case of larger<br />
values of N.<br />
The usage of the proposed solution <strong>in</strong>stead of classic<br />
TFRC allows one to achieve throughput of the CBR stream<br />
comparableto thethroughputof CBR overRTP. Moreover,the<br />
parameters of TCP transmissions are approximately the same<br />
as those observed when classic TFRC is used. It means that<br />
the l<strong>in</strong>ear equation avoids the collapse of the TCP connections<br />
<strong>and</strong> allows the TCP to utilize available b<strong>and</strong>width (b<strong>and</strong>width<br />
of the bottleneck l<strong>in</strong>k reduced by target bit rate of multimedia<br />
stream).<br />
In the second experiment we changed the throughput of the<br />
CBR transmission B from 0.5 Mb/s to 4 Mb/s (throughput of<br />
the bottleneckl<strong>in</strong>k). The <strong>number</strong> of compet<strong>in</strong>gTCP flows was<br />
set to 1. Results of experiments are shown <strong>in</strong> Fig. 4.<br />
As we can see <strong>in</strong> Fig. 4, TFRC with the l<strong>in</strong>ear equation<br />
allows one to transmit real-time multimedia even if the target<br />
bit rate of the CBR stream is close to the throughput of<br />
bottleneckl<strong>in</strong>k.BothRTP<strong>and</strong>classicTFRCwereabletocarry<br />
out real-time transmission up to about a half of the throughput<br />
of the bottleneck l<strong>in</strong>k (at least <strong>in</strong> this experiment). In the case<br />
of both modified TFRC <strong>and</strong> classic TFRC, concurrent TCP<br />
streams were able to utilize all rema<strong>in</strong><strong>in</strong>g b<strong>and</strong>width of the<br />
bottleneck l<strong>in</strong>k.<br />
Fig. 4. Throughput of CBR transmission as a function of B.<br />
VI. CONCLUSION<br />
Although the authors of TFRC suggest that the protocol<br />
is suitable for multimedia transmission, it is not aggressive<br />
enough to meet the QoS requirements of carried stream<strong>in</strong>g<br />
media when it competes for b<strong>and</strong>width with the TCP. In the<br />
paper we propose to substitute the orig<strong>in</strong>al TFRC throughput<br />
equation with a l<strong>in</strong>ear throughput equation. This substitution<br />
makes the TFRC more aggressive, which allows the protocol<br />
to preserve the real-time character of the transmitted flow<br />
no worse than the RTP or the UDP protocol. Moreover, <strong>in</strong><br />
situations when the usage of the RTP causes the collapse of<br />
TCP transmission (or, at least, worsen<strong>in</strong>gof the QoS of one or<br />
more TCP flows), the proposed solution is “friendly” enough<br />
for compet<strong>in</strong>g TCP flows to equally share the rema<strong>in</strong><strong>in</strong>g<br />
b<strong>and</strong>width. Such results allow us to believe that the proposed<br />
l<strong>in</strong>ear equation is more suitable for multimedia transmission<br />
than the equation orig<strong>in</strong>ally <strong>in</strong>cluded <strong>in</strong> the RFC 3448.<br />
REFERENCES<br />
[1] M. H<strong>and</strong>ley, S. F. J. Padhye, <strong>and</strong> J. Widmer, TCP Friendly Rate Control<br />
(TFRC): Protocol Specification, IETF RFC 3448, Jan. 2003.<br />
[2] J. P. S. Floyd, M. H<strong>and</strong>ley <strong>and</strong> J. Widmer, TCP Friendly Rate Control<br />
(TFRC): Protocol Specification, IETF RFC 5348, Sep. 2008.<br />
[3] A. Chodorek <strong>and</strong> R. R. Chodorek, “Applicability of TCP-friendly protocols<br />
for real-time multimedia transmission,” <strong>in</strong> Proc. XII Poznan Telecommunications<br />
Workshop (PWT), Poznan, 2007.<br />
[4] A. Chodorek, “Stream<strong>in</strong>g video with TFRC - simulation approach,” <strong>in</strong><br />
Proc. of SympoTIC’04, Oct. 2004.<br />
[5] A. Chodorek, R. R. Chodorek, <strong>and</strong> A. R. Pach, Dystrybucja danych w<br />
sieci Internet. Warszawa: WKŁ, 2007.<br />
[6] J. Padhye, “Model-based approach to TCP-friendly congestion control,”<br />
Ph.D. dissertation, Department of Computer Science, University of Massachusetts<br />
at Amherst, 2000.<br />
[7] J.Padhye, V. Firoiu, D.Towsley, <strong>and</strong> J.Kurose, “Model<strong>in</strong>g TCPThroughput:<br />
A Simple Model <strong>and</strong> its Empirical Validation,” <strong>in</strong> Proc. Proceed<strong>in</strong>gs<br />
of ACM SIGCOMM, 1998.<br />
[8] H. Schulzr<strong>in</strong>ne, S. Casner, R. Frederick, <strong>and</strong> V. Jacobson, RTP: A<br />
Transport Protocol for Real-Time Applications, IETF RFC 3550, Jul.<br />
2003.<br />
Agnieszka Chodorek received her M.Sc. degree <strong>in</strong> electrical eng<strong>in</strong>eer<strong>in</strong>g<br />
from the Kielce University of Technology <strong>in</strong> Kielce, Pol<strong>and</strong>, <strong>in</strong> 1991, <strong>and</strong><br />
her Ph.D. degree <strong>in</strong> telecommunications from the AGH University of Science<br />
<strong>and</strong> Technology <strong>in</strong> Krakow, Pol<strong>and</strong>, <strong>in</strong> 2001. She is an assistant professor<br />
at the Department of Telecommunications, Photonics <strong>and</strong> Nanomaterials,<br />
Kielce University of Technology <strong>in</strong> Kielce, Pol<strong>and</strong>. She is currently lectur<strong>in</strong>g
CHODOREK AND CHODOREK: STREAMING VIDEO OVER TFRC WITH LINEAR THROUGHPUT EQUATION 29<br />
on Satellite <strong>and</strong> Mobile Communications, Computer Networks, Multimedia<br />
Technology, <strong>and</strong> Internet Multimedia Services. Her research <strong>in</strong>terests lie <strong>in</strong> the<br />
area of telecommunication networks, with emphasis on Internet technology<br />
<strong>and</strong> multimedia transmission. She has authored many publications <strong>in</strong> these<br />
areas, <strong>in</strong>clud<strong>in</strong>g two books.<br />
Robert R. Chodorek received his M.Sc. degree <strong>in</strong> electrical eng<strong>in</strong>eer<strong>in</strong>g<br />
from the Kielce University of Technology <strong>in</strong> Kielce, Pol<strong>and</strong>, <strong>in</strong> 1990, <strong>and</strong><br />
his Ph.D. degree <strong>in</strong> computer sciences from the AGH University of Science<br />
<strong>and</strong> Technology <strong>in</strong> Krakow, Pol<strong>and</strong>, <strong>in</strong> 1996. He is currently an assistant<br />
professor at the Department of Telecommunications, AGH University of<br />
Science <strong>and</strong> Technology <strong>in</strong> Krakow, Pol<strong>and</strong>. His current areas of research<br />
<strong>in</strong>clude performance evaluation of telecommunication networks, <strong>in</strong> particular<br />
broadb<strong>and</strong> communications, IP multicast<strong>in</strong>g <strong>and</strong> multimedia communications.<br />
He is author or co-author of over 80 research papers <strong>and</strong> two books.
30 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Simulation model for evaluation of packet sequence<br />
changed order of stream <strong>in</strong> DiffServ network<br />
M. Czarkowski <strong>and</strong> S. Kaczmarek<br />
Abstract—Current packet networks use a large variety of<br />
mechanisms which should support QoS (Quality of Service). One<br />
of those mechanisms is rout<strong>in</strong>g (calculat<strong>in</strong>g connection paths for<br />
<strong>in</strong>com<strong>in</strong>g service requests). The most effective mechanism <strong>in</strong> QoS<br />
context is dynamic rout<strong>in</strong>g, based on the current network state<br />
described by the offered traffic matrix <strong>and</strong> l<strong>in</strong>k states. After<br />
switch<strong>in</strong>g between calculated available paths, connection path<br />
changes may cause received packets to change order with<strong>in</strong> a<br />
s<strong>in</strong>gle stream. This paper <strong>in</strong>cludes the problem def<strong>in</strong>ition <strong>and</strong> the<br />
analysis of all additional effects. A comb<strong>in</strong>ed simulation/analytic<br />
model was proposed <strong>in</strong> order to answer whether the <strong>number</strong> of<br />
changed-orderpackets issignificant<strong>and</strong>ifitshouldbeconsidered<br />
when calculat<strong>in</strong>g the end-to-end delay balance <strong>in</strong> analytical models<br />
for packet networks withdifferentiatedservices. Furthermore,<br />
the proposed model gave the answer on how often calculated<br />
paths may be switched to avoid the network be<strong>in</strong>gout of tune.<br />
Index Terms—IP, QoS, DiffServ, QoS rout<strong>in</strong>g<br />
I. INTRODUCTION<br />
CURRENT telecommunications networks are based on a<br />
largevarietyoftechnologies.Manyofthosenetworksare<br />
packet based networks with focus on networks which use IP<br />
protocol (so called IP networks). If they are applied <strong>in</strong> a local<br />
scope (IP network connect<strong>in</strong>g just neighbor devices), they<br />
work accord<strong>in</strong>g to the provided design <strong>and</strong> they do not cause<br />
any additional problems with configuration <strong>and</strong> ma<strong>in</strong>tenance;<br />
however, when they are used <strong>in</strong> a global scope (IP network<br />
as a core network), they are the source of many problems<br />
<strong>and</strong> unexpected network behavior. Those problems are mostly<br />
comb<strong>in</strong>ed with servic<strong>in</strong>g requested QoS <strong>and</strong>, simultaneously,<br />
optimal network resources utilization. It is due to very strong<br />
dynamic traffic changes from multiple traffic sources. Those<br />
sources vary <strong>in</strong> their traffic characteristics. That is why any<br />
mechanism used should be resistant to such strong traffic<br />
dynamics. Unfortunately,current network control mechanisms<br />
provided for IP networks fail to solve this problem [1],<br />
[2]. One of network control mechanisms is connection path<br />
calculation process – rout<strong>in</strong>g. The important condition which<br />
should provide effective rout<strong>in</strong>g <strong>in</strong> these terms is to calculate<br />
paths to support requested QoS for differentiated services.<br />
Effective path calculation means also avoid<strong>in</strong>g network congestion<br />
states <strong>and</strong> optimization of available resources. Current<br />
rout<strong>in</strong>g mechanisms do not meet those requirements [3], [4].<br />
The key element to solve this problem is to use dynamic<br />
M. Czarkowski is with the Gdansk University of Technology, Faculty<br />
<strong>Electronics</strong>, Telecommunications <strong>and</strong> Informatics, Gdansk, Pol<strong>and</strong> (e-mail:<br />
czarka@eti.pg.gda.pl).<br />
S. Kaczmarek is with the Gdansk University of Technology, Faculty<br />
<strong>Electronics</strong>, Telecommunications <strong>and</strong> Informatics, Gdansk, Pol<strong>and</strong> (e-mail:<br />
kasyl@eti.pg.gda.pl).<br />
This work was supported <strong>in</strong> part by the Polish National Centre for Research<br />
<strong>and</strong> Development under the project PBZ MNiSW – 02/II/2007.<br />
rout<strong>in</strong>g – the process of path calculation which follows the<br />
network changes <strong>and</strong> path selection decision, based solely<br />
on the current network state. In addition, the <strong>in</strong>troduction of<br />
dynamic rout<strong>in</strong>g causes some consequences. One of them are<br />
<strong>in</strong>com<strong>in</strong>g packets order changes with<strong>in</strong> a s<strong>in</strong>gle stream, which<br />
is due to the switch<strong>in</strong>g of available paths. Change of packets<br />
order is caused by switch<strong>in</strong>g from a path with longer delay<br />
<strong>in</strong>to a path with shorter delay. The packet delay is directly<br />
comb<strong>in</strong>edwiththe<strong>number</strong>oftransitnodes<strong>and</strong>trafficcurrently<br />
located <strong>in</strong> the network. Unfortunately, there is no scientific<br />
literature which considers the problem <strong>and</strong> no research results<br />
on the subject of reordered packets. Most authors deal<strong>in</strong>g<br />
with dynamic rout<strong>in</strong>g mechanisms assume <strong>in</strong> their works that<br />
packet reorder<strong>in</strong>gdur<strong>in</strong>g path switch<strong>in</strong>g is not significant. The<br />
authors who noticed the problem of packet reorder<strong>in</strong>g made<br />
<strong>in</strong>itial assumption that reorder<strong>in</strong>g will be solved by upper<br />
layers <strong>and</strong> they just shift the responsibility. Other analyzed<br />
papers <strong>in</strong>cluded the assumption that packet reorder<strong>in</strong>g due to<br />
path switch<strong>in</strong>g will not be considered because it is not an<br />
important issue. It seems to be a wrong assumption. In this<br />
paper we give the answer to the question whether the packet<br />
sequencechangedorderisasignificanteffectfromthe po<strong>in</strong>tof<br />
view of dynamic rout<strong>in</strong>g. The rest of the paper is organizedas<br />
specified below. Section II describes the problem <strong>in</strong> general<br />
<strong>in</strong> terms of generated traffic relations <strong>and</strong> available system<br />
resources. Section III is a short description of the proposed<br />
simulation model used for problem evaluation <strong>and</strong> extended<br />
experiments. Section IV conta<strong>in</strong>s the research results <strong>and</strong> the<br />
analysis of those results. Some <strong>in</strong>vestigated relations are also<br />
identified. The f<strong>in</strong>al section V provides a short summary with<br />
focus on further work directions.<br />
II. PROBLEM DEFINITION AND DECOMPOSITION<br />
Some basic assumptions were made for further <strong>in</strong>vestigations.<br />
The analyzed network supports prioritized services.<br />
Packetscome<strong>in</strong>to/comeoutofthenetworkviaedgenodes.All<br />
core nodes support transit nodes functionality. Additionally,<br />
the service <strong>in</strong> the node is based on the non-preemptivepriority<br />
model. The considered problem is illustrated <strong>in</strong> Fig. 1.<br />
Packets come <strong>in</strong>to the network <strong>in</strong>to edge node A <strong>and</strong> are<br />
transferredviacorenodeCtoedgenodeB.Thefirstcalculated<br />
path1 from node A to node B is A→C→B. All packets<br />
with dest<strong>in</strong>ation address B are transported us<strong>in</strong>g this path.<br />
After sudden traffic changes on path1, congestion state has<br />
been detected <strong>and</strong> the entire path had to be calculated aga<strong>in</strong><br />
(dynamicrout<strong>in</strong>g).Letusassumethatthenewcalculatedpath2<br />
fromAtoBis:A→D→E→C→B.Packetssentbeforethepath<br />
recalculation, which were be<strong>in</strong>g transported via path1 (<strong>and</strong>
CZARKOWSKI AND KACZMAREK: SIMULATION MODEL FOR EVALUATION OF PACKET SEQUENCE CHANGED ORDER OF STREAM IN DIFFSERV NETWORK31<br />
Fig. 1. Basic problem visualization.<br />
have not reached the out-node yet), were not discarded <strong>and</strong><br />
are processed <strong>in</strong> the network.<br />
In Fig. 1 this one situation refers to packets with <strong>number</strong>s<br />
1 <strong>and</strong> 2. Packets with <strong>number</strong>s 3 <strong>and</strong> 4 were sent through<br />
the new path2. After some time, the connection paths were<br />
transformed aga<strong>in</strong> <strong>in</strong>to path1 A→C→B (packet 5) <strong>and</strong> aga<strong>in</strong><br />
<strong>in</strong>to path2 A→D→E→C→B (packet 6). Let us assume that<br />
each l<strong>in</strong>k <strong>in</strong> Fig. 1 <strong>in</strong>troduces the same propagation time (the<br />
same medium<strong>and</strong> the same length foreach correspond<strong>in</strong>gl<strong>in</strong>k<br />
on the path). All l<strong>in</strong>ks are one direction symmetric l<strong>in</strong>ks with<br />
the same b<strong>and</strong>width. Moreover, each core node <strong>in</strong>troduces the<br />
same wait<strong>in</strong>g time (for service <strong>in</strong> the queue). Both paths from<br />
A to B differ only <strong>in</strong> the transit nodes <strong>number</strong>. Packets sent<br />
via path2 will be received later than they would be received<br />
from path1. This will cause switched packets order <strong>in</strong> node B<br />
(packet 5 will be received by node B before packets 3 <strong>and</strong> 4).<br />
The proposed model does not simulate delays on the path (the<br />
behavior of service systems). Therefore, an analytical part has<br />
been <strong>in</strong>troduced for delays calculation (buffer<strong>in</strong>g delay, send<br />
delay <strong>and</strong> propagation delay). The end-to-end delay time may<br />
be described us<strong>in</strong>g the follow<strong>in</strong>g equations when we assume<br />
PQ systems <strong>in</strong> nodes [5]:<br />
E(tend−to−end) = k · (E(twait) + E(tsend) + tprop) (1)<br />
E(twait) =<br />
R�<br />
i=1<br />
�<br />
2 1 − i−1 �<br />
ρj<br />
j=1<br />
where: R(= 3) – <strong>number</strong> of classes<br />
ρj – offered traffic for class j<br />
λi – packets <strong>in</strong>tensity for class i<br />
m (2)<br />
i<br />
– second moment for class i<br />
λim (2)<br />
i<br />
� �<br />
1 − i�<br />
ρj<br />
j=1<br />
� (2)<br />
k – <strong>number</strong> of core node (=1 for a shorter path <strong>and</strong><br />
=3 for a longer path)<br />
E(tsend) = E(Li)<br />
Cl<br />
where: Li – length of the packet for class i<br />
Cl – l<strong>in</strong>k b<strong>and</strong>width <strong>in</strong> a given direction<br />
tprop = αmdu−v<br />
where: αm – delay factor for medium type m<br />
du−v – length between nodes u <strong>and</strong> v<br />
Threebasictypesoftime (wait<strong>in</strong>gtime,sendtime <strong>and</strong>propagation<br />
time) may <strong>in</strong>fluence the problem under consideration.<br />
The end user connectedto the edge node may generate several<br />
traffic classes (e.g. stream<strong>in</strong>g, elastic, best effort). The time<br />
distribution between packets is assumed to be exponential.<br />
Packets generated from each user are transmitted through a<br />
common l<strong>in</strong>k to the <strong>in</strong> edge node. In the edge node rout<strong>in</strong>g<br />
a decision is made (path selection) <strong>and</strong> packets are forwarded<br />
to the path chosen from the two available paths. If they reach<br />
the out edge node, they are marked off from the aggregated<br />
DiffServ stream <strong>and</strong> forwarded to the dest<strong>in</strong>ation end user.<br />
III. SIMULATION MODEL<br />
Based on the above delays model of events, a simulation<br />
model was proposed, i.e. a comb<strong>in</strong>ation of simulation <strong>and</strong><br />
analytical delay rules. A scheme of the proposedmodel is presented<br />
<strong>in</strong> Fig. 2 <strong>and</strong> demonstrated<strong>in</strong> omnet++simulation tools<br />
[6]. The <strong>in</strong>put <strong>in</strong> the model are traffic sources limited to three<br />
traffic classes: stream<strong>in</strong>g services sensitive to delay <strong>and</strong> jitter<br />
– classified to EF; elastic services sensitive to loss probability<br />
– classified to AF; other services not sensitive to any factor<br />
– classified to BE. AF has been limited only to a s<strong>in</strong>gle class<br />
(3)<br />
(4)
32 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 2. Screenshot from omnet++ simulation model.<br />
just to identify the problem. Those three service classes are<br />
generated by any user connected to the network. Each traffic<br />
class is def<strong>in</strong>ed by priority <strong>and</strong> packet <strong>in</strong>tensity. Inter-arrival<br />
time between <strong>in</strong>com<strong>in</strong>g packets is calculated on the basis of<br />
packet <strong>in</strong>tensity with<strong>in</strong> the class. Stream<strong>in</strong>g services use short<br />
packetswith160byteslength,elastic servicesusepacketswith<br />
500 bytes length, other services – packets with 1,500 bytes<br />
length. Before simulation is run, traffic classes proportions<br />
are calculated. Users send their packets (User traffic) to the<br />
edge node which actually correspondsto the aggregat<strong>in</strong>gnode<br />
(aggregator block connected to In-node block <strong>in</strong> Fig. 2).<br />
Connection paths are calculated <strong>in</strong> the edge node because we<br />
have source rout<strong>in</strong>g <strong>and</strong> packets are transmitted through the<br />
service system (<strong>in</strong> the edge node each path has its own service<br />
system). The rema<strong>in</strong><strong>in</strong>g connection path (Path simulation) is<br />
calculated <strong>in</strong> block devices (D), which <strong>in</strong> fact are a cha<strong>in</strong> of<br />
service systems present <strong>in</strong> the path.<br />
Those devices simulate each type of delay, i.e. send delay,<br />
buffer<strong>in</strong>g delay <strong>and</strong> propagation delay, over the connection<br />
path. All global data used <strong>in</strong> the simulation are stored <strong>in</strong> the<br />
board object which is not l<strong>in</strong>ked to any block<strong>in</strong> the simulation<br />
model.<br />
Given connection paths have vary<strong>in</strong>g delay values. Packets<br />
switched order is detected <strong>in</strong> the declassifier block (Outnode)<br />
<strong>and</strong> statistics are collected separately for each traffic<br />
source. Packets are deleted <strong>in</strong> the s<strong>in</strong>k block (leave). The <strong>in</strong>put<br />
parameters of simulation: the <strong>number</strong> of transit nodes present<br />
<strong>in</strong> the path, nodes distance, b<strong>and</strong>width between nodes, l<strong>in</strong>k<br />
load, packets <strong>in</strong>terarrival time (given as exponential distribution),<br />
time values between successive rout<strong>in</strong>g table changes<br />
(paths recalculation). The follow<strong>in</strong>g functional blocks have<br />
been def<strong>in</strong>ed:<br />
A. User traffic<br />
• EF_i – stream<strong>in</strong>g class traffic generator for user i<br />
• AF_i – elastic class traffic generator for user i<br />
• BE_i – best effort traffic generator for user i<br />
• User_i – aggregator of all available traffic classes<br />
B. Background traffic<br />
• EF_back_i – background traffic generator for stream<strong>in</strong>g<br />
class for user i<br />
• AF_back_i–backgroundtrafficgeneratorforelastic class<br />
for user i<br />
• BE_back_i - background traffic generator for best effort<br />
class for user i<br />
• Background_i – aggregator of all available traffic classes<br />
for background traffic
CZARKOWSKI AND KACZMAREK: SIMULATION MODEL FOR EVALUATION OF PACKET SEQUENCE CHANGED ORDER OF STREAM IN DIFFSERV NETWORK33<br />
C. Aggregator – switches the traffic onto the proper path<br />
D. In node<br />
• Classifier – separates aggregated traffic <strong>in</strong>to separated<br />
class queues<br />
• Delay_i – receiver process<strong>in</strong>g delay (<strong>in</strong> this research set<br />
to zero)<br />
• Qserver – PQ queue model<br />
E. Path simulation<br />
• delaySend_j – simulates send<strong>in</strong>g delay dependent on l<strong>in</strong>k<br />
speed <strong>and</strong> packet length for path j<br />
• delayBuff_j–simulatesbuffer<strong>in</strong>gdelaydependentonnon<br />
preemptive service model of path j<br />
• dealyProp_j – simulates propagation delay of path j<br />
F. Out node<br />
• declassifier – splits packets received <strong>in</strong> aggregatedstream<br />
<strong>in</strong>to sub-streams <strong>and</strong> collects required statistics<br />
• leave – s<strong>in</strong>k for created packets<br />
G. Board – global storage of simulation parameters <strong>and</strong><br />
common data<br />
IV. RESULTS ANALYSIS<br />
A set of simulation results with confidence level of 0.95<br />
have been collected for various configurations across many<br />
possibilities. The follow<strong>in</strong>g charts represent some selected<br />
results. The figures outl<strong>in</strong>e the situation when background<br />
trafficis80%<strong>and</strong>therest(20%ofthetraffic)isbe<strong>in</strong>gswitched<br />
between paths. The background traffic has been <strong>in</strong>troduced so<br />
that two service systems (for path 1 <strong>and</strong> path 2) are work<strong>in</strong>g<strong>in</strong><br />
parallel while the paths are switched. Each of the charts shows<br />
different time values between rout<strong>in</strong>g tables recalculation.<br />
The first one is when a rout<strong>in</strong>g table is updated every 5<br />
seconds (Fig. 3), the second one when the table is updated<br />
every 20 seconds (Fig. 4), <strong>and</strong> f<strong>in</strong>ally every 40 seconds<br />
(Fig. 5). The results have been grouped <strong>in</strong> three parts: the<br />
first part (marked with EF on the x axis) collects EF class, the<br />
second one (marked with AF) collects AF class <strong>and</strong> the last<br />
one (marked with BE) collects BE class. The presented values<br />
are the ratio between switched packets with<strong>in</strong> a s<strong>in</strong>gle stream<br />
of class i to all packets sent for this stream class i. For all of<br />
the charts n<strong>in</strong>e simulation series are presented.<br />
Each series differs as far as proportions of traffic share for<br />
EF, AF <strong>and</strong> BE classes are concerned. Classes’ shares are<br />
listed <strong>in</strong> TABLE I.<br />
All charts show that for EF class a lower ratio of switched<br />
packets to all packets is when EF class has more shares with<strong>in</strong><br />
the overall traffic. It can be expla<strong>in</strong>ed with the highest EF<br />
priority of all traffic classes <strong>and</strong> the fact that EF are short<br />
(160 bytes) – more share, will cause more <strong>in</strong>tensity of EF,<br />
<strong>and</strong> less <strong>in</strong>tensity with<strong>in</strong> longer packets (AF <strong>and</strong> BE), so the<br />
residual time due to non-preemptive priorities, will not affect<br />
EF as strongly. No unexpected effect has been observed also<br />
for BE traffic class. The ratio of BE switched order packets<br />
was high for low BE share <strong>and</strong> high for EF <strong>and</strong> AF shares <strong>in</strong><br />
TABLE I<br />
CLASSES PROPORTIONS FOR EACH SIMULATION SERIES<br />
Series EF [%] AF [%] BE[%]<br />
1 10 10 80<br />
2 10 45 45<br />
3 10 70 20<br />
4 20 10 70<br />
5 20 40 40<br />
6 20 60 20<br />
7 30 10 60<br />
8 30 35 35<br />
9 30 50 20<br />
Fig. 3. Results chart for 20% traffic switched every 5 seconds.<br />
the overall traffic. It may be expla<strong>in</strong>ed by the mean<strong>in</strong>g of BE<br />
priority (the weakest) as well as by the low <strong>in</strong>tensity of BE.<br />
EF <strong>and</strong> AF have much higher <strong>in</strong>tensity than BE.<br />
A peculiar effect was observed for AF class <strong>in</strong> the case of<br />
some classes proportions. When EF class had the share above<br />
40% <strong>and</strong> the rema<strong>in</strong><strong>in</strong>g traffic (60%) was divided between<br />
AF <strong>and</strong> BE, AF had much higher switched sequence changed<br />
orderpacketsratiothanusual.AlthoughAFsharewasgrow<strong>in</strong>g<br />
(with<strong>in</strong> 60% of traffic for AF <strong>and</strong> BE), the ratio did not<br />
fall (though it should due to the priority higher than BE).<br />
It may be partially expla<strong>in</strong>ed with the residual time of BE;<br />
but when BE share falls, the residual time shall not <strong>in</strong>fluence<br />
the AF class so strongly. The nature of the observed relations<br />
shows that they are <strong>in</strong>fluenced by many other factors which<br />
require further extended experiments. Only then will it be<br />
possible to identify all the relations <strong>and</strong> f<strong>in</strong>d the explanation<br />
of <strong>in</strong>vestigated effect. The current research stage allows us to<br />
confirmthattheproblem<strong>in</strong>vestigated<strong>in</strong>thisworkissignificant<br />
<strong>in</strong> terms of dynamically controlled rout<strong>in</strong>g.<br />
V. SUMMARY<br />
Dynamic rout<strong>in</strong>g may <strong>in</strong>troduce many additional problems.<br />
Some of them seem to be simple <strong>and</strong> their explanation should<br />
be obvious (they are already analyzed <strong>and</strong> solved). Unfortunately,<br />
sometimes they cause unexpected system behavior<br />
<strong>and</strong> <strong>in</strong>troduce additional effects that have not been solved yet.<br />
Such effect is packet sequence changed order with<strong>in</strong> a s<strong>in</strong>gle<br />
stream caused by changes <strong>in</strong> the path transit node <strong>number</strong>
34 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 4. Results chart for 20% traffic switched every 20 seconds.<br />
Fig. 5. Results chart for 20% traffic switched every 40 seconds.<br />
(different delays on different paths). Further considerations<br />
gave several <strong>in</strong>terest<strong>in</strong>g answers on the mean<strong>in</strong>g of dynamic<br />
rout<strong>in</strong>g mechanisms. The proposed simulation model made it<br />
possible to answer some questions <strong>and</strong> to shed light on the<br />
scope of other problems. Us<strong>in</strong>g some proportions between<br />
classes <strong>in</strong> differentiated services doma<strong>in</strong> packets reorder<strong>in</strong>g<br />
caused by path switch<strong>in</strong>g should be marked <strong>in</strong> end-to-end<br />
balance. It may not be skipped <strong>and</strong> omitted <strong>in</strong> the system<br />
analysis. The AF switched sequence changed order packets to<br />
all AF sendpacketsratiomaynotbeexpla<strong>in</strong>edbyapply<strong>in</strong>gthe<br />
known analytical equations (for the non-preemptive priority<br />
system). The ratio value is significant for flexible services <strong>and</strong><br />
should be taken <strong>in</strong>to consideration. Furthermore, an important<br />
conclusion for EF traffic was found. The stream<strong>in</strong>g services<br />
have lower switched sequence changed order than all EF sent<br />
packets ratio when EF share <strong>in</strong> the overall traffic amount<br />
is 20–40 %. Some additional remarks were also found for<br />
different time values between rout<strong>in</strong>g table recalculations. It<br />
turned out that the optimal time between rout<strong>in</strong>g table updates<br />
(<strong>in</strong> short term changes – seconds) was 35–40 seconds <strong>in</strong>terval.<br />
This statement is based on simulation results but will not be<br />
discussed <strong>in</strong> this paper due to space limitation. For rout<strong>in</strong>g<br />
table switch<strong>in</strong>g time a local m<strong>in</strong>imum of the 35–40 seconds<br />
was observed. For all analyzed situations residual time is<br />
important when packet length differs between given traffic<br />
classes (EF – 160 bytes, AF – 500 bytes, BE – 1,500 bytes).<br />
Further <strong>in</strong>vestigationswill be aimed at f<strong>in</strong>d<strong>in</strong>gthe relationsfor<br />
AF traffic <strong>and</strong> expla<strong>in</strong><strong>in</strong>g the issue us<strong>in</strong>g the newly developed<br />
analytical equations.<br />
REFERENCES<br />
[1] S. Chen <strong>and</strong> K. Nahrstedt, “An Overview – of – Service Rout<strong>in</strong>g for the<br />
Next Generation High – Speed Networks: Problems <strong>and</strong> Solutions,” IEEE<br />
Network Magaz<strong>in</strong>e, vol. 12, no. 6, pp. 64–79, Dec. 1998.<br />
[2] G. Feng, K. Makki, N. Piss<strong>in</strong>ou, <strong>and</strong> C. Douligeris, “Heuristic <strong>and</strong> Exact<br />
Algorithms for QoS Rout<strong>in</strong>g with Multiple Constra<strong>in</strong>ts,” IEICE Trans.<br />
Commun., no. 12, pp. 2838–2850, Dec. 2002.<br />
[3] J. T. Moy, OSPF Anatomy of an Internet Rout<strong>in</strong>g Protocol, 2001.<br />
[4] ——, OSPF Complete Implementation, 2001.<br />
[5] J. N. Daigle, Queu<strong>in</strong>g Theory with Applications to Packet Telecommunication,<br />
2005.<br />
[6] [onl<strong>in</strong>e], http://www.omnetpp.org.<br />
M. Czarkowski received the M.Sc. degree <strong>in</strong> telecommunication systems<br />
from Gdansk University of Technology (GUT), Gdansk, Pol<strong>and</strong>, <strong>in</strong> July<br />
2004. He is currently pursu<strong>in</strong>g for the Ph.D. degree <strong>in</strong> Telecommunication<br />
Networks <strong>and</strong> Systems, GUT. His Ph.D. work focuses ma<strong>in</strong>ly on dynamic<br />
rout<strong>in</strong>g algorithms with Quality of Service (QoS.)<br />
S. Kaczmarek received the M.Sc./B.Sc. <strong>in</strong> electronics eng<strong>in</strong>eer<strong>in</strong>g, Ph.D<br />
<strong>and</strong> D.Sc <strong>in</strong> switch<strong>in</strong>g <strong>and</strong> teletraffic science from Gdansk University of<br />
Technology, Gdansk, Pol<strong>and</strong>, <strong>in</strong> 1972, 1981 <strong>and</strong> 1994, respectively. His<br />
research <strong>in</strong>terests <strong>in</strong>clude: IP QoS <strong>and</strong> GMPLS networks, switch<strong>in</strong>g, rout<strong>in</strong>g,<br />
teletraffic <strong>and</strong> quality of service. He has published more than 190 papers.<br />
Now he is the Head of Tele<strong>in</strong>formation Networks Department.
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 35<br />
Packet dispatch<strong>in</strong>g schemes support<strong>in</strong>g uniform <strong>and</strong><br />
nonuniform traffic distribution patterns <strong>in</strong> MSM<br />
Clos-network switches<br />
Abstract—In this paper new packet dispatch<strong>in</strong>g schemes for<br />
efficient support of the uniform as well as the nonuniform traffic<br />
distribution patterns <strong>in</strong> Memory-Space-Memory (MSM) Closnetworkswitchesare<br />
presented.Threesuchschemes, calledStatic<br />
Dispatch<strong>in</strong>g-First Choice (SD-FC), Static Dispatch<strong>in</strong>g-Optimal<br />
Choice (SD-OC) <strong>and</strong> Input Module (IM)-Output Module (OM)<br />
Match<strong>in</strong>g (IOM), are proposed <strong>and</strong> evaluated. The algorithms<br />
are able to unload the overloaded <strong>in</strong>put buffers employ<strong>in</strong>g a<br />
central arbiter. This effect is a desirable feature especially for<br />
effective support of the nonuniform traffic distribution patterns.<br />
We show via simulation that the proposed schemes deliver very<br />
good performance <strong>in</strong> terms of throughput, cell delay, <strong>and</strong> <strong>in</strong>put<br />
buffers size under different traffic distribution patterns. The<br />
results obta<strong>in</strong>ed for the proposed algorithms are compared with<br />
the results obta<strong>in</strong>ed for selected request-grant-accept iterative<br />
packet dispatch<strong>in</strong>g schemes.<br />
Index Terms—Clos-network, packet schedul<strong>in</strong>g, packet switch<strong>in</strong>g,<br />
virtual output queu<strong>in</strong>g.<br />
Janusz Kleban<br />
I. INTRODUCTION<br />
THE switch<strong>in</strong>g fabric <strong>in</strong> high-performance packet switch<strong>in</strong>g<br />
nodes may be built as a s<strong>in</strong>gle stage-switch (e.g.<br />
crossbar) or a multi-stage switch, such as the Clos switch<strong>in</strong>g<br />
fabric. The switch<strong>in</strong>g process <strong>in</strong> a multi-stage switch<strong>in</strong>g fabric<br />
consists of two activities, namely <strong>in</strong>put-output match<strong>in</strong>g <strong>and</strong><br />
route assignment between the first <strong>and</strong> last stages. These two<br />
phases can be processed separately or simultaneously. S<strong>in</strong>ce<br />
the high-speed switch<strong>in</strong>g fabrics support fixed-length packets<br />
called cells, packets of variable size must be segmented <strong>in</strong>to<br />
cells at switch <strong>in</strong>put ports, <strong>and</strong> cells must be reassembled <strong>in</strong>to<br />
packets at switch output ports [1].<br />
While cells are be<strong>in</strong>g routed <strong>in</strong> a switch<strong>in</strong>g fabric, it is very<br />
likely that more than one cell is dest<strong>in</strong>ed for the same output<br />
port or for a physical l<strong>in</strong>k <strong>in</strong>side the multi-stage switch<strong>in</strong>g<br />
fabric. Cells that have lost contention must be either discarded<br />
or buffered. Buffers may be placed at <strong>in</strong>puts, outputs, <strong>in</strong>puts<br />
<strong>and</strong> outputs,<strong>and</strong>/or with<strong>in</strong> the switch<strong>in</strong>g fabric [2]. The virtual<br />
output queu<strong>in</strong>g (VOQ) is widely implemented as a good<br />
solution for <strong>in</strong>put queued (IQ) switches, to avoid the Head-<br />
Of-L<strong>in</strong>e (HOL) block<strong>in</strong>g problem encountered <strong>in</strong> the <strong>in</strong>putbuffered<br />
switches. In VOQ switches every <strong>in</strong>put provides<br />
a s<strong>in</strong>gle <strong>and</strong> separate FIFO for each output. Such a FIFO<br />
is called a Virtual Output Queue. When a new cell arrives<br />
at the <strong>in</strong>put port, it is stored <strong>in</strong> the dest<strong>in</strong>ed queue <strong>and</strong><br />
waits for transmission through a switch<strong>in</strong>g fabric. To solve<br />
<strong>in</strong>ternal block<strong>in</strong>g<strong>and</strong> output port contentionproblems<strong>in</strong> VOQ<br />
switches, fast arbitration schemes are needed. The arbitration<br />
scheme decides which items of <strong>in</strong>formation should be passed<br />
from <strong>in</strong>puts to arbiters, <strong>and</strong> – based on that decision – how<br />
each arbiter picks one cell from among all <strong>in</strong>put cells dest<strong>in</strong>ed<br />
for the output. Algorithmswhich can assign the route between<br />
<strong>in</strong>put<strong>and</strong>outputmodulesare usuallycalledpacketdispatch<strong>in</strong>g<br />
schemes. Considerable work has been done on schedul<strong>in</strong>g<br />
algorithms for VOQ switches. Most of them achieve 100%<br />
throughputunder uniform traffic, but the throughputis usually<br />
reduced under nonuniform traffic [1], [3]–[14]. A switch can<br />
achieve 100% throughputunder uniform or nonuniformtraffic<br />
if the switch is stable, as was def<strong>in</strong>ed <strong>in</strong> [15]. In general, a<br />
switch is stable for a particular arrival process if the expected<br />
length of the <strong>in</strong>put queues does not grow without limits.<br />
Multiple-stage Clos-network switches are a potential solution<br />
to overcome the limited scalability of s<strong>in</strong>gle-stage<br />
switches, <strong>in</strong> terms of the <strong>number</strong> of I/O chip p<strong>in</strong>s <strong>and</strong> the<br />
<strong>number</strong> of switch<strong>in</strong>g elements. Different dispatch<strong>in</strong>g schemes<br />
forthethree-stageClos-networkswitcheswereproposed<strong>in</strong>the<br />
literature [4]–[6], [9]–[14]. The basic idea of these algorithms<br />
is to use the effect of desynchronizationof arbitration po<strong>in</strong>ters<br />
<strong>and</strong> a common request-grant-accept h<strong>and</strong>shak<strong>in</strong>g scheme. All<br />
high speed switch<strong>in</strong>g fabrics implemented by the manufacturers<br />
of switches/routersare now based on SERDES technology.<br />
The signals pass<strong>in</strong>g through these serial l<strong>in</strong>ks are with<strong>in</strong> the<br />
range of several hundred nanoseconds. It is very difficult to<br />
implement the algorithms with multiple-phase iterations <strong>in</strong> a<br />
three-stage environment with currently available technologies,<br />
because of time constra<strong>in</strong>ts (one slot time <strong>in</strong> a 10 Gbps<br />
switch<strong>in</strong>g fabric lasts around 50 ns).<br />
In this paper SD-FC, SD-OC, <strong>and</strong> IOM packet dispatch<strong>in</strong>g<br />
schemes are presented. These algorithms give better performance<br />
results than other dispatch<strong>in</strong>g schemes proposed<br />
for the MSM Clos switch<strong>in</strong>g fabric, <strong>and</strong> can achieve 100%<br />
throughput for both the uniform <strong>and</strong> the nonuniform traffic<br />
distribution patterns. The rema<strong>in</strong>der of this paper is organized<br />
as follows. Section II <strong>in</strong>troduces some background knowledge<br />
concern<strong>in</strong>g the MSM Clos switch<strong>in</strong>g fabric that we refer to<br />
throughoutthis paper. Section III presentsthe SD-FC, SD-OC,<br />
<strong>and</strong> IOM packet dispatch<strong>in</strong>g schemes. Section IV is devoted<br />
to performance evaluation of the proposed algorithms. The<br />
comparisonof cell delay betweenproposedalgorithms<strong>and</strong> the<br />
selected multiple-phaseiterativepacketdispatch<strong>in</strong>gschemesis<br />
also shown. We conclude this paper <strong>in</strong> Section V.
36 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 1. The MSM Clos switch<strong>in</strong>g fabric architecture.<br />
II. MSM CLOS SWITCHING NETWORK<br />
Clos-networks are well known <strong>and</strong> widely analyzed <strong>in</strong> the<br />
literature [16]. The three-stage Clos-network architecture is<br />
denoted by C(m, n, k), where parameters m, n, <strong>and</strong> k entirely<br />
determ<strong>in</strong>e the structure of the network. We def<strong>in</strong>e the MSM<br />
Clos switch<strong>in</strong>g fabric based on the term<strong>in</strong>ology used <strong>in</strong> [4]<br />
(see Fig. 1).<br />
IntheMSMClosswitch<strong>in</strong>gfabricarchitecturethefirst stage<br />
consists of k IMs, <strong>and</strong> each of them has an n × m dimension<br />
<strong>and</strong> nk V OQ(i, j, h) to elim<strong>in</strong>ateHead-Of-L<strong>in</strong>eblock<strong>in</strong>g.The<br />
second stage consists of m bufferless CMs, <strong>and</strong> each of them<br />
has a k × k dimension. The third stage consists of k OMs of<br />
capacity m × n, where each OP (j, h) has an output buffer.<br />
Each output buffer can receive at most m cells from m CMs,<br />
so a memory speedup is required here.<br />
Generally speak<strong>in</strong>g, <strong>in</strong> the MSM Clos switch<strong>in</strong>g fabric<br />
architecture each V OQ(i, j, h) located <strong>in</strong> IM(i) stores cells<br />
go<strong>in</strong>g from IM(i) to the OP (j, h) at OM(j). In one cell<br />
time slot VOQ can receive at most n cells from n <strong>in</strong>put ports<br />
<strong>and</strong> send one cell to any CMs. A memory speedup of n is<br />
required here because the rate of memory work has to be n<br />
times higherthan the l<strong>in</strong>e rate. Each IM(i) has m outputl<strong>in</strong>ks<br />
LI(i, r) connected to each CM(r), respectively. A CM(r)<br />
has k output l<strong>in</strong>ks LC(r, j) which are connected to each<br />
OM(j), respectively.<br />
In simulation experiments we consider the Clos switch<strong>in</strong>g<br />
fabric without any expansion, denoted by C(n, n, n), so <strong>in</strong><br />
the description of the packet dispatch<strong>in</strong>g schemes <strong>in</strong> Section<br />
III, parameters k <strong>and</strong> m are not used. We also use Virtual<br />
Output Module Queues (VOMQs), <strong>in</strong>stead of VOQs. In this<br />
case, an <strong>in</strong>put buffer <strong>in</strong> each IM is divided <strong>in</strong>to k parallel<br />
queues, each of them stor<strong>in</strong>g cells dest<strong>in</strong>ed for different OMs.<br />
It is possible to arrange buffers <strong>in</strong> such way because OMs<br />
are nonblock<strong>in</strong>g. Memory speedup of n is necessary here.<br />
There are fewer queues <strong>in</strong> each IM but they are longer than<br />
VOQs. Each V OMQ(i, j) stores cells go<strong>in</strong>g from IM(i) to<br />
the OM(j).<br />
III. PACKET DISPATCHING SCHEMES<br />
Under the nonuniform traffic distribution patterns, selected<br />
VOQs store more cells than others. Because of that, it is<br />
Fig. 2. Static connection patterns <strong>in</strong> CMs, C(3, 3, 3).<br />
necessary to implement a special mechanism for a packet<br />
dispatch<strong>in</strong>g scheme, which is able to send up to n cells from<br />
IM(i) to OM(j) <strong>in</strong> the same time slot, <strong>in</strong> order to unload<br />
overloaded buffers. Three dispatch<strong>in</strong>g schemes presented <strong>in</strong><br />
this paper have such possibility.<br />
Theproposedpacketdispatch<strong>in</strong>gschemesperformmatch<strong>in</strong>g<br />
between each IM <strong>and</strong> OM, tak<strong>in</strong>g <strong>in</strong>to account the <strong>number</strong><br />
of cells wait<strong>in</strong>g <strong>in</strong> VOMQs. Each VOMQ has its own<br />
counter P V (i, j) which shows the <strong>number</strong> of cells dest<strong>in</strong>ed<br />
for OM(j). The value of P V (i, j) is <strong>in</strong>creased by 1 when<br />
a new cell is written <strong>in</strong>to memory, <strong>and</strong> decreased by 1 when<br />
the cell is sent out to OM(j). The algorithms use the central<br />
arbiter to <strong>in</strong>dicate the matched pairs of IM(i) − OM(j) but<br />
the set of data sent to the arbiter by each scheme is different.<br />
Therefore, the architecture <strong>and</strong> functionality of each arbiter<br />
is also different. After match<strong>in</strong>g phase, <strong>in</strong> the next time slot<br />
IM(i) is allowed to send up to n cells to the selected OM(j).<br />
In the SD-OC <strong>and</strong> SD-FC schemes the central arbiter<br />
matches IM(i) <strong>and</strong> OM(j) only if the <strong>number</strong> of cells<br />
buffered <strong>in</strong> V OMQ(i, j) is at least equal to n. Under the<br />
nonuniform traffic distribution patterns this happens very often,contraryto<br />
theuniformtrafficdistribution.Intheproposed<br />
packet dispatch<strong>in</strong>g schemes, each VOMQ has to wait until at<br />
least n cellsare stored beforebe<strong>in</strong>gallowedto makearequest.<br />
To reduce latency <strong>and</strong> avoid starvation, a very simple packet<br />
dispatch<strong>in</strong>g rout<strong>in</strong>e, called Static Dispatch<strong>in</strong>g (SD), is also<br />
used. Underthis algorithm,connect<strong>in</strong>gpaths<strong>in</strong> the MSM Clos<br />
switch<strong>in</strong>g fabric are set up accord<strong>in</strong>g to connection patterns<br />
which are static but different <strong>in</strong> each CM (see Fig. 2). These<br />
fixed connection paths between IMs <strong>and</strong> OMs elim<strong>in</strong>ate the<br />
h<strong>and</strong>shak<strong>in</strong>g process with the second stage, <strong>and</strong> no <strong>in</strong>ternal<br />
conflicts<strong>in</strong> the switch<strong>in</strong>gfabricwill occur.Also, noarbitration<br />
process is necessary. Cells dest<strong>in</strong>ed for the same OM, but<br />
located <strong>in</strong> different IMs, will be sent through different CMs.<br />
In detail, the SD algorithm works as follows:<br />
Step 1: Accord<strong>in</strong>g to the connection pattern of IM(i),<br />
match all output l<strong>in</strong>ks LI(i, r) with cells from VOMQs.<br />
Step 2: Send the matched cells <strong>in</strong> the next time slot. If there<br />
is any unmatched output l<strong>in</strong>k, it rema<strong>in</strong>s idle.<br />
The SD-OC <strong>and</strong> SD-FC schemes are very similar but the<br />
central arbiter which matches the IMs <strong>and</strong> OMs works <strong>in</strong>
JANUSZ KLEBAN: PACKET DISPATCHING SCHEMES SUPPORTING UNIFORM AND NONUNIFORM TRAFFIC DISTRIBUTION PATTERNS 37<br />
a different way. In both algorithms the P V (i, j) counter<br />
which reaches the value equal to or greater than n sends the<br />
<strong>in</strong>formation about an overloaded buffer to the central arbiter.<br />
In the central arbiterthere is a b<strong>in</strong>arymatrix of VOMQ buffers<br />
load. If the value of matrix element x[i, j] = 1, it means that<br />
IM(i) has at least n cells that should be sent to OM(j).<br />
In the SD-OC scheme the ma<strong>in</strong> task of the central arbiter is<br />
to f<strong>in</strong>d an optimal set of 1s <strong>in</strong> the matrix. The best case is n<br />
1s but it is possible to choose only one 1 from column i <strong>and</strong><br />
row j. If there is no such set of 1s, the arbiter tries to f<strong>in</strong>d a<br />
set of n − 1 1s which fulfill the same conditions, <strong>and</strong> so on.<br />
The round-rob<strong>in</strong> rout<strong>in</strong>e is used for the start<strong>in</strong>g po<strong>in</strong>t of the<br />
search<strong>in</strong>g process. Otherwise, the MSM Clos switch<strong>in</strong>g fabric<br />
works under the SD scheme.<br />
The ma<strong>in</strong> difference between the SD-OC <strong>and</strong> SD-FC lies<br />
<strong>in</strong> the operation of the central arbiter. In the SD-FC scheme<br />
the central arbiter does not look for an optimal set of 1s but<br />
tries to match IM(i) with OM(j), choos<strong>in</strong>g the first 1 found<br />
<strong>in</strong> column i <strong>and</strong> row j. No optimization process for select<strong>in</strong>g<br />
the IM-OM pairs is employed. In detail, the SD-OC algorithm<br />
works as follows:<br />
Step 1: If the value of the P V (i, j) counter is equal to or<br />
greater than n, send a request to the central arbiter.<br />
Step 2: If the central arbiter receives the request from<br />
IM(i), it sets the value of the buffer load matrix element<br />
x[i, j] to 1 (the values of i <strong>and</strong> j come from the counter<br />
P V (i, j)).<br />
Step 3: After receiv<strong>in</strong>g all requests, the central arbiter tries<br />
to f<strong>in</strong>d an optimal set of 1s which allows the greatest <strong>number</strong><br />
of cells to be sent from IMs to OMs. The central arbiter has to<br />
go through all rows of the buffer load matrix to f<strong>in</strong>d a set of n<br />
1s represent<strong>in</strong>g IM(i)−OM(j) match<strong>in</strong>g. If it is not possible<br />
to f<strong>in</strong>d a set of n 1s, it attempts to f<strong>in</strong>d a set of (n − −1) 1s,<br />
<strong>and</strong> so on.<br />
Step 4: In the next time slot send n cells from IMs to the<br />
matched OMs. Decrease the value of P V (i, j) by n. For the<br />
IM-OM pairs not matched by the central arbiter, use the SD<br />
scheme <strong>and</strong> decrease the value of P V counters by 1.<br />
The steps <strong>in</strong> the SD-FC scheme are very similar to the steps<br />
<strong>in</strong> the SD-OC scheme but the optimizationprocess <strong>in</strong> the third<br />
step is not carried out. The central arbiter chooses the first 1<br />
which fulfills the requirements <strong>in</strong> each row. The row searched<br />
asthefirstoneisselectedaccord<strong>in</strong>gtotheround-rob<strong>in</strong>rout<strong>in</strong>e.<br />
The IOM packet dispatch<strong>in</strong>g scheme also employs the<br />
central arbiter to make a match between each IM <strong>and</strong> OM.<br />
The cells are sent only between IM-OM pairs matched by<br />
the arbiter. The SD scheme is not used. In detail, the IOM<br />
algorithm works as follows:<br />
Step 1 (each IM): Sort the values of P V (i, j) <strong>in</strong> descend<strong>in</strong>g<br />
order. Send a request to the central arbiter, conta<strong>in</strong><strong>in</strong>g a list<br />
of the OMs identifiers. The identifier of OM(j) for which<br />
V OMQ(i, j) stores the greatest <strong>number</strong> of cells should be<br />
placed on the list as the first one, <strong>and</strong> the identifier of OM(s)<br />
for which V OMQ(i, s) stores the smallest <strong>number</strong> of cells<br />
should be placed on the list as the last one.<br />
Step 2 (central arbiter): The central arbiter analyzes the<br />
request receivedfrom IM(i) <strong>and</strong> checks whetherit is possible<br />
to match this IM with OM(j) whose identifier was sent as the<br />
first one on the list <strong>in</strong> the request. If match<strong>in</strong>g is not possible<br />
because the OM(j) was matched with other IM, the arbiter<br />
selects the next OM on the list. The round-rob<strong>in</strong> arbitration is<br />
employed for the selection of IM(i) for which the request is<br />
analyzed as the first one.<br />
Step 3 (central arbiter): The central arbiter sends confirmation<br />
to each IM with the identifier of OM(t) to which the IM<br />
is allowed to send cells.<br />
Step 4 (each IM): Match all output l<strong>in</strong>ks LI(i, r) with cells<br />
from V OMQ(i, t). If there are less than n cells to be sent to<br />
OM(t), some output l<strong>in</strong>ks rema<strong>in</strong> unmatched.<br />
Step 5 (each IM): Decrease the value of P V (i, t) by the<br />
<strong>number</strong> of cells to be sent to OM(t).<br />
Step 6 (each IM): In the next time slot send the cells<br />
from the matched V OMQ(i, t) to the OM(t) selected by the<br />
central arbiter.<br />
A. Packet arrival models<br />
IV. SIMULATION EXPERIMENTS<br />
Two packet arrival models are considered <strong>in</strong> simulation<br />
experiments: the Bernoulli arrival model <strong>and</strong> the bursty traffic<br />
model.Inthe Bernoulliarrivalmodel,cellsarriveat each <strong>in</strong>put<br />
<strong>in</strong> a slot-by-slot manner. Under the Bernoulli arrival process,<br />
the probability that there is a cell arriv<strong>in</strong>g <strong>in</strong> each time slot is<br />
identical to <strong>and</strong> <strong>in</strong>dependent of all other slots. The probability<br />
that a cell may arrive <strong>in</strong> a time slot is denoted by p <strong>and</strong><br />
is referred to as the load of the <strong>in</strong>put. In the bursty traffic<br />
model, each <strong>in</strong>put alternates between active <strong>and</strong> idle periods.<br />
Dur<strong>in</strong>gactive periods,cells dest<strong>in</strong>edfor the same outputarrive<br />
cont<strong>in</strong>uously <strong>in</strong> consecutive time slots. The average burst<br />
(active period) length is set to 10 cells.<br />
B. Traffic distribution models<br />
We consider several traffic distribution models which determ<strong>in</strong>e<br />
the probability that a cell which arrives at an <strong>in</strong>put will<br />
be directed to a certa<strong>in</strong> output. The considered traffic models<br />
are:<br />
Uniform traffic. This type of traffic is the most commonly<br />
used traffic profile. In uniformly distributed traffic, the probability<br />
pij that a packet from <strong>in</strong>put i will be directed to output<br />
j is uniformly distributed through all outputs, that is:<br />
pij = p/N ∀i, j. (1)<br />
Trans-diagonal traffic. In this traffic model some outputs<br />
have a higher probability of be<strong>in</strong>g selected, <strong>and</strong> respective<br />
probability pij was calculated accord<strong>in</strong>g to the follow<strong>in</strong>g<br />
equation:<br />
pij =<br />
� p<br />
2<br />
p<br />
2(N−1)<br />
for i = j<br />
for i �= j.<br />
Bi-diagonal traffic. This type of traffic is very similar to<br />
the trans-diagonal traffic but packets are directed to one of<br />
two outputs, <strong>and</strong> respective probability pij was calculated<br />
accord<strong>in</strong>g to the follow<strong>in</strong>g equation:<br />
pij =<br />
⎧<br />
⎨<br />
⎩<br />
2<br />
3p for i = j<br />
p<br />
3 for j = (i + 1) mod N<br />
0 otherwise.<br />
(2)<br />
(3)
38 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 3. Average cell delay, uniform traffic.<br />
Chang’s traffic. This model is def<strong>in</strong>ed as:<br />
�<br />
0 for i = j<br />
pij =<br />
otherwise.<br />
p<br />
N−1<br />
C. Results of simulation experiments<br />
The experiments have been carried out for the MSM Clos<br />
switch<strong>in</strong>g fabric of size 64 × 64 – C(8, 8, 8), <strong>and</strong> for a wide<br />
range of traffic loads per <strong>in</strong>put port: from p = 0.05 to<br />
p = 1, with a step of 0.05. The 95% confidence <strong>in</strong>tervals that<br />
have been calculated after t-student distribution for ten series<br />
with 50000 cycles (after the start<strong>in</strong>g phase compris<strong>in</strong>g 15000<br />
cycles, whichenablesthe stable state ofthe switch<strong>in</strong>g fabricto<br />
be reached)are at least oneorderlower than the mean valueof<br />
the simulationresults,<strong>and</strong>,therefore,theyarenotshown<strong>in</strong> the<br />
figures.Wehaveevaluatedtwoperformancemeasures:average<br />
cell delay <strong>in</strong> time slots <strong>and</strong> maximum VOMQs size (we have<br />
<strong>in</strong>vestigatedtheworst case).Thesize ofthebuffersat the<strong>in</strong>put<br />
<strong>and</strong> output side of switch<strong>in</strong>g fabric is not limited, so cells are<br />
not discarded.However,theyencounterdelay<strong>in</strong>stead. Because<br />
oftheunlimitedsizeofbuffers,nomechanismcontroll<strong>in</strong>gflow<br />
control between the IMs <strong>and</strong> OMs (to avoid buffer overflows)<br />
is implemented. The results of the simulation are shown <strong>in</strong><br />
the charts (Fig. 3, Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8, Fig. 9,<br />
Fig. 10, Fig. 11, Fig. 12). Fig. 3, Fig. 5, Fig. 7, Fig. 9, <strong>and</strong><br />
Fig. 11 show the average cell delay <strong>in</strong> time slots obta<strong>in</strong>ed for<br />
the uniform, trans-diagonal, bi-diagonal, Chang’s, <strong>and</strong> bursty<br />
traffic patterns, whereas Fig. 4, Fig. 6, Fig. 8, Fig. 10 <strong>and</strong><br />
Fig.12showthemaximumVOMQsize <strong>in</strong>the<strong>number</strong>ofcells.<br />
Fig. 11 <strong>and</strong> Fig. 12 show the results for the bursty traffic with<br />
average burst size b = 10 (10 is the <strong>number</strong> of cells).<br />
We can see that the MSM Clos switch<strong>in</strong>g fabric with all<br />
the schemes proposed has 100% throughput for all k<strong>in</strong>ds<br />
of <strong>in</strong>vestigated traffic distribution patterns <strong>and</strong> for the bursty<br />
traffic. The average cell delay is less than 10 for a wide range<br />
of <strong>in</strong>putloads, regardlessof the traffic distributionpattern.It is<br />
a very <strong>in</strong>terest<strong>in</strong>g result especially for the trans-diagonal <strong>and</strong><br />
the bi-diagonal traffic patterns. Both traffic patterns are highly<br />
dem<strong>and</strong><strong>in</strong>g <strong>and</strong> many packet dispatch<strong>in</strong>g schemes proposed<br />
<strong>in</strong> the literature cannot provide 100% throughput for the<br />
<strong>in</strong>vestigatedswitch<strong>in</strong>gfabric.Fortheburstytraffic,theaverage<br />
cell delay becomes very similar to a l<strong>in</strong>ear function of <strong>in</strong>put<br />
load with the maximum value less than 150. We can see<br />
(4)<br />
Fig. 4. The maximum VOMQ size, uniform traffic.<br />
Fig. 5. Average cell delay, trans-diagonal traffic.<br />
that the very complicated arbitration rout<strong>in</strong>e used <strong>in</strong> the SD-<br />
OC scheme does not improve the performance of MSM Clos<br />
switch<strong>in</strong>g fabric. In some cases the results are even worse<br />
than for the IOM scheme (the trans-diagonal traffic with very<br />
high <strong>in</strong>put load <strong>and</strong> bursty traffic). Generally, the IOM scheme<br />
gives higher latency than the SD schemes, especially for low<br />
to medium <strong>in</strong>put load. This is due to match<strong>in</strong>g IM(i) to that<br />
OM(j) to which it is possible to send the greatest <strong>number</strong> of<br />
cells. As a consequence, it is less probable that IM-OM pairs<br />
will be matched to serve one, two, or three cells per cycle.<br />
The size of VOMQ <strong>in</strong> the MSM Clos switch<strong>in</strong>g network<br />
depends on the traffic distribution pattern. For all proposed<br />
packet distribution schemes <strong>and</strong> uniform <strong>and</strong> Chang’s traffic<br />
themaximumsize ofVOMQislessthan140cells.Thismeans<br />
that <strong>in</strong> the worst case the average <strong>number</strong> of cell wait<strong>in</strong>g for<br />
transmission to a particular output was not bigger than 16. For<br />
the trans-diagonal traffic <strong>and</strong> the IOM scheme the maximum<br />
size of VOMQ is less than 200 but for SD-OC <strong>and</strong> SD-FC the<br />
sizes are greater <strong>and</strong> reach 700 <strong>and</strong> 3000, respectively. For the<br />
bi-diagonal traffic the smallest size of VOMQ was obta<strong>in</strong>ed<br />
for the SD-OC scheme for which it was less than 290. For<br />
the bursty traffic the maximal size of VOMQ reaches 750 for<br />
SD-FC, 500 for SD-OC, <strong>and</strong> 350 for the IOM scheme.<br />
D. Comparison of cell delay between proposed schemes <strong>and</strong><br />
selected multiple-phase packet dispatch<strong>in</strong>g algorithms<br />
The primary multiple-phase dispatch<strong>in</strong>g algorithms for the<br />
buffered Clos-network switches were proposed <strong>in</strong> [4]. The<br />
basic ideaofthese algorithmsisto use theeffectofdesynchronization<br />
of arbitration po<strong>in</strong>ters <strong>in</strong> the Clos-network switch <strong>and</strong>
JANUSZ KLEBAN: PACKET DISPATCHING SCHEMES SUPPORTING UNIFORM AND NONUNIFORM TRAFFIC DISTRIBUTION PATTERNS 39<br />
Fig. 6. The maximum VOMQ size, trans-diagonal traffic.<br />
Fig. 7. Average cell delay, bi-diagonal traffic.<br />
the common request-grant-accept h<strong>and</strong>shak<strong>in</strong>g scheme. The<br />
well known algorithm with multiple-phase iterations is the<br />
CRRD (Concurrent Round-Rob<strong>in</strong> Dispatch<strong>in</strong>g). Other algorithmslike<br />
the CMSD (ConcurrentMaster-Slave Round-Rob<strong>in</strong><br />
Dispatch<strong>in</strong>g)[4],SRRD(StaticRound-Rob<strong>in</strong>Dispatch<strong>in</strong>g)[6],<br />
<strong>and</strong>,asproposedbyus<strong>in</strong>[11],CRRD-OG(ConcurrentRound-<br />
Rob<strong>in</strong> Dispatch<strong>in</strong>gwith Open Grants) use the ma<strong>in</strong> idea of the<br />
CRRD scheme <strong>and</strong> try to improvethe results by implement<strong>in</strong>g<br />
different mechanisms.<br />
Fig. 13, Fig. 14, Fig. 15 show the comparison between<br />
average cell delays obta<strong>in</strong>ed for the CRRD, CMSD, SRRD,<br />
<strong>and</strong> CRRD-OG schemes with four iterations (more than n/2<br />
iterations do not change the performance of all <strong>in</strong>vestigated<br />
iterative schemes significantly) <strong>and</strong> average cell delay obta<strong>in</strong>ed<br />
for the schemes proposed <strong>in</strong> this paper. The simulation<br />
experiments were carried out for all k<strong>in</strong>ds of <strong>in</strong>vestigated<br />
traffic distribution patterns, but only results for the uniform,<br />
trans-diagonal, <strong>and</strong> bi-diagonal traffic patterns are shown. The<br />
conditions of computer simulation experiments were the same<br />
for all <strong>in</strong>vestigated schemes.<br />
For the uniform traffic distribution pattern all schemes can<br />
achieve 100% throughput. The best results can be obta<strong>in</strong>ed<br />
by us<strong>in</strong>g the CRRD-OG scheme, but the results are almost<br />
the same as for SD schemes. For highly dem<strong>and</strong><strong>in</strong>g traffic<br />
distribution patterns like the trans-diagonal <strong>and</strong> bi-diagonal<br />
ones, only SD-FC, SD-OC, <strong>and</strong> IOM schemes can provide<br />
100% throughput for the MSM Clos switch<strong>in</strong>g fabric. The<br />
<strong>in</strong>vestigated request-grant-accept packet dispatch<strong>in</strong>g schemes<br />
are not able to provide such high efficiency. The best results<br />
Fig. 8. The maximum VOMQ size, bi-diagonal traffic.<br />
Fig. 9. Average cell delay, Chang’s traffic.<br />
Fig. 10. The maximum VOMQ size, Chang’s traffic.<br />
Fig. 11. Average cell delay, bursty traffic.<br />
from among multiple-phase algorithms have been obta<strong>in</strong>ed<br />
for the CRRD-OG scheme. These are respectively: under<br />
the trans-diagonal traffic pattern: 85% throughput for four<br />
iterations (Fig. 14), <strong>and</strong> under the bi-diagonal traffic pattern,<br />
95% (Fig. 15).
40 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 12. The maximum VOMQ size, bursty traffic.<br />
Fig. 13. Average cell delay for selected request-grant-accept algorithms (four<br />
iterations) <strong>and</strong> the proposed schemes, uniform traffic.<br />
Fig. 14. Average cell delay for selected request-grant-accept algorithms (four<br />
iterations) <strong>and</strong> the proposed schemes, trans-diagonal traffic.<br />
The <strong>in</strong>vestigated request-grant-accept packet dispatch<strong>in</strong>g<br />
schemes are based on the effect of desynchronization of<br />
arbitration po<strong>in</strong>ters <strong>in</strong> the Clos-network switch. We have<br />
made an attempt to improve the desynchronization method<br />
for the CRRD-OG scheme to ensure 100% throughput for the<br />
nonuniform traffic distribution patterns. Additional po<strong>in</strong>ters<br />
<strong>and</strong> arbiters for open grants were added to the MSM Clos<br />
switch<strong>in</strong>g fabric but the scheme was not able to provide 100%<br />
throughput for the nonuniform traffic distribution patterns.<br />
To the best of our knowledge, it is not possible to achieve<br />
very good desynchronization of po<strong>in</strong>ters us<strong>in</strong>g the methods<br />
implemented <strong>in</strong> the iterative packet dispatch<strong>in</strong>g schemes. In<br />
our op<strong>in</strong>ion, the decisions of distributed arbiters have to be<br />
Fig. 15. Average cell delay for selected request-grant-accept algorithms (four<br />
iterations) <strong>and</strong> the proposed schemes, bi-diagonal traffic.<br />
supportedbythecentralarbiterbutthe implementationofsuch<br />
solutions <strong>in</strong> the real equipment will be very complex. Therefore<br />
the algorithms, which are able to unload the overloaded<br />
<strong>in</strong>put buffers like SD-FC <strong>and</strong> IOM should be implemented.<br />
V. CONCLUSION<br />
We have proposed the SD-FC, SD-OC, <strong>and</strong> IOM packet<br />
dispatch<strong>in</strong>g schemes for the MSM Clos switch<strong>in</strong>g fabric. The<br />
algorithmsemploy the central arbiter to match IMs with OMs.<br />
In SD-FC <strong>and</strong> IOM schemes the arbiter performs relatively<br />
simple functions. Simulation experiments have shown that the<br />
proposed schemes are very promis<strong>in</strong>g <strong>and</strong> give very good<br />
resultsfor boththe uniform<strong>and</strong> nonuniformtraffic distribution<br />
patterns. The algorithms can manage all <strong>in</strong>vestigated traffic<br />
patterns very effectively, provid<strong>in</strong>g 100% throughput. This is<br />
a highlydesirablepropertyofthe packetdispatch<strong>in</strong>galgorithm<br />
for the switch<strong>in</strong>g fabric of the next generation packet node.<br />
A hardware implementation of the central arbiters required by<br />
the proposed schemes will be subject to further research.<br />
REFERENCES<br />
[1] J. Chao <strong>and</strong> B. Liu, High Performance Switches <strong>and</strong> Routers. New<br />
Jersey: Wiley, Hoboken, 2007.<br />
[2] K. Yoshigoe <strong>and</strong> K. J. Christensen, “An evolution to crossbar switches<br />
with virtual ouptut queu<strong>in</strong>g <strong>and</strong> buffered cross po<strong>in</strong>ts,” IEEE Network,<br />
vol. 17, no. 5, pp. 48–56, 2003.<br />
[3] E. Oki, R. Rojas-Cessa, <strong>and</strong> H. J. Chao, “A pipel<strong>in</strong>e-based approach for<br />
maximal-sized match<strong>in</strong>g schedul<strong>in</strong>g <strong>in</strong> <strong>in</strong>put-buffered switches,” IEEE<br />
Commun. Lett., vol. 5, no. 6, pp. 263–265, 2001.<br />
[4] E. Oki, Z. J<strong>in</strong>g, R. Rojas-Cessa, <strong>and</strong> H. J. Chao, “Concurrent<br />
round-rob<strong>in</strong>-based dispatch<strong>in</strong>g schemes for Clos-network switches,”<br />
IEEE/ACM Trans. on Network<strong>in</strong>g, vol. 10, no. 6, pp. 830–844, 2002.<br />
[5] R. Rojas-Cessa <strong>and</strong> H.J.Chao, “Maximum weight match<strong>in</strong>g dispatch<strong>in</strong>g<br />
scheme <strong>in</strong> buffered Clos-network packet switches,” <strong>in</strong> Proc. of IEEE<br />
International Conference on Communications, ICC 2004, Paris, France,<br />
2004, pp. 830–844.<br />
[6] K. Pun <strong>and</strong> M. Hamdi, “Dispatch<strong>in</strong>g schemes for Clos-network<br />
switches,” Computer Networks, no. 44, pp. 667–679, 2004.<br />
[7] Y. Jiang <strong>and</strong> M. Hamdi, “A fully desynchronized round-rob<strong>in</strong> match<strong>in</strong>g<br />
scheduler for a VOQ packet switch architecture,” <strong>in</strong> Proc. of IEEE High<br />
Performance Switch<strong>in</strong>g <strong>and</strong> Rout<strong>in</strong>g, HPSR 2001, May 2001, pp. 407–<br />
411.<br />
[8] J. Y. Hui <strong>and</strong> E. Arthurs, “A broadb<strong>and</strong> packet switch for <strong>in</strong>tegrated<br />
transport,” IEEE J. Sel. Areas Commun., vol. 5, no. 8, pp. 1264–1273,<br />
Oct. 1987.<br />
[9] C. B. L<strong>in</strong> <strong>and</strong> R. Rojas-Cessa, “Frame occupancy-based dispatch<strong>in</strong>g<br />
schemes for buffered three-stage Clos-network switches,” <strong>in</strong> Proc. of<br />
13th IEEE International Conference on Networks 2005, 2005.
JANUSZ KLEBAN: PACKET DISPATCHING SCHEMES SUPPORTING UNIFORM AND NONUNIFORM TRAFFIC DISTRIBUTION PATTERNS 41<br />
[10] R. Rojas-Cessa <strong>and</strong> C. B. L<strong>in</strong>, “Scalable two-stage Clos-network switch<br />
<strong>and</strong> module-first match<strong>in</strong>g,” <strong>in</strong> Proc. of High Performance Switch<strong>in</strong>g<br />
<strong>and</strong> Rout<strong>in</strong>g, HPSR 2006, 2006, pp. 303–308.<br />
[11] J. Kleban <strong>and</strong> A. Wieczorek, “CRRD-OG – a packet dispatch<strong>in</strong>g algorithm<br />
with open grants for three-stage buffered Clos-network switches,”<br />
<strong>in</strong> Proc. of High Performance Switch<strong>in</strong>g <strong>and</strong> Rout<strong>in</strong>g, HPSR2006, 2006,<br />
pp. 315–320.<br />
[12] J. Kleban, M.Sobieraj, <strong>and</strong> S. W˛eclewski, “The modified MSM Clos<br />
switch<strong>in</strong>g fabric with efficient packet dispatch<strong>in</strong>g scheme,” <strong>in</strong> Proc. of<br />
IEEEHigh Performance Switch<strong>in</strong>g <strong>and</strong> Rout<strong>in</strong>g, HPSR2007, New York,<br />
May 2007.<br />
[13] J. Kleban <strong>and</strong> H. Santos, “Packet dispatch<strong>in</strong>g algorithms with the<br />
static connection patterns scheme for three-stage buffered Clos-network<br />
switches,” <strong>in</strong> Proc. of IEEE International Conference on Communica-<br />
tions, ICC-2007, Glasgow, UK, Jun. 2007.<br />
[14] J. Kleban <strong>and</strong> M. Sobieraj, “Delayed response of central arbiter <strong>in</strong> threestage<br />
bufferless Clos-network switches,” <strong>in</strong> Proc. of 5th Polish-German<br />
Teletraffic Symposium, PGTS 2008, Berl<strong>in</strong>, Oct. 2008, pp. 51–60.<br />
Janusz Kleban Faculty of <strong>Electronics</strong> <strong>and</strong> Telecommunications, Chair of<br />
Communication <strong>and</strong> Computer Networks, Poznan University of Technology,<br />
ul. Polanka 3, 60-965 Poznan, e-mail: janusz.kleban@et.put.poznan.pl.<br />
The ma<strong>in</strong> area of <strong>in</strong>terest of the author covers packet dispatch<strong>in</strong>g <strong>and</strong><br />
schedul<strong>in</strong>g algorithms for both electronic <strong>and</strong> optical switch<strong>in</strong>g fabrics.
42 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Methods of Real-time Calculation of Allan<br />
Deviation <strong>and</strong> Time Deviation<br />
Andrzej Dobrogowski <strong>and</strong> Michał Kasznia<br />
Abstract—The methods enabl<strong>in</strong>g real-time calculation of two<br />
commonly used parameters of tim<strong>in</strong>g signals – Allan deviation<br />
(ADEV) <strong>and</strong> time deviation (TDEV) – are presented <strong>in</strong> this<br />
paper. The idea of real-time computation of both parameters<br />
is described. The results of experimental tests of the methods<br />
enabl<strong>in</strong>g separate as well as jo<strong>in</strong>t real-time ADEV <strong>and</strong> TDEV<br />
computation are presented <strong>and</strong> discussed.<br />
Index Terms—tim<strong>in</strong>g signal, time error, Allan deviation, time<br />
deviation<br />
I. INTRODUCTION<br />
THE Allan deviation ADEV <strong>and</strong> time deviation TDEV<br />
allow the type of phase noise affect<strong>in</strong>g the tim<strong>in</strong>g signal<br />
to be recognized. The parameters are commonly used for<br />
evaluation of signals generated by atomic clocks as well as<br />
for describ<strong>in</strong>g the quality of synchronization signal <strong>in</strong> the<br />
telecommunication networks [1]–[3]. The evaluation of the<br />
synchronizationsignal is commonlyatwo-stage process.First,<br />
the sequence of time error samples between the analyzed<br />
signal <strong>and</strong>somereferencehasto bemeasuredat somenetwork<br />
<strong>in</strong>terface. When the measurement is completed, the calculation<br />
of the parameter’s estimate us<strong>in</strong>g time error samples is<br />
performed. Such procedure causes an obvious delay <strong>in</strong> the<br />
evaluation process.<br />
This paper describes the real-time methods of ADEV <strong>and</strong><br />
TDEV computation, which enable the reduction of the evaluation<br />
time. These methods allow the estimates of ADEV<br />
<strong>and</strong> TDEV (which characterizes of more complex estimator’s<br />
formula) to be computed <strong>in</strong> the real time, dur<strong>in</strong>g the measurementprocess,simultaneouslyforaset<br />
ofobservation<strong>in</strong>tervals.<br />
Additionally,thecomputationprocesscanbeperformedjo<strong>in</strong>tly<br />
for both parameters.<br />
In order to calculate the ADEV <strong>and</strong> TDEV estimate simultaneously<br />
for several observation <strong>in</strong>tervals <strong>in</strong> the real<br />
time, all necessary operations should be performed <strong>in</strong> the<br />
time period between two sampl<strong>in</strong>g <strong>in</strong>stants, i.e. dur<strong>in</strong>g the<br />
sampl<strong>in</strong>g <strong>in</strong>terval τ0. The ability of perform<strong>in</strong>g the realtime<br />
assessment depends on several conditions: computation<br />
ability of the measurement equipment, sampl<strong>in</strong>g <strong>in</strong>terval <strong>and</strong><br />
<strong>number</strong> of the observation <strong>in</strong>tervals considered. The methods<br />
described <strong>in</strong> the paper are developed for a measur<strong>in</strong>g system<br />
A. Dobrogowski is with Chair of Telecommunication Systems <strong>and</strong> Optoelectronics,<br />
Poznan University of Technology, ul. Polanka 3, 60-965 Poznań,<br />
Pol<strong>and</strong> (e-mail: dobrog@et.put.poznan.pl).<br />
M. Kasznia is with Chair of Telecommunication Systems <strong>and</strong> Optoelectronics,<br />
Poznan University of Technology, ul. Polanka 3, 60-965 Poznań, Pol<strong>and</strong><br />
(e-mail: mkasznia@et.put.poznan.pl).<br />
This work was supported by the M<strong>in</strong>istry of Science <strong>and</strong> Higher Education<br />
<strong>in</strong> the frame of the project <strong>number</strong> N N517 1645 33 <strong>in</strong> the years 2007-<strong>2010</strong>.<br />
where the time error counter <strong>and</strong> the computer controll<strong>in</strong>g the<br />
measurement are two separate units. Therefore, the computer<br />
may be changed depend<strong>in</strong>g on the comput<strong>in</strong>g requirements.<br />
The results of experimental tests of the methods proposed<br />
for different conditions are presented <strong>in</strong> the paper. The calculations<br />
were performed for the time error sequences taken<br />
with sampl<strong>in</strong>g <strong>in</strong>terval τ0 = 1/30 s, which is often used<br />
<strong>in</strong> the telecommunication applications. Different <strong>number</strong>s <strong>and</strong><br />
lengths of observation <strong>in</strong>tervals simultaneously analyzed were<br />
considered.<br />
II. ALLAN DEVIATION AND TIME DEVIATION<br />
The computations of the Allan deviation <strong>and</strong> time deviation<br />
estimates are based on the averag<strong>in</strong>g of second differences of<br />
the phase process x(t) of the analyzed tim<strong>in</strong>g signal. We can<br />
assume for the telecommunication applications, <strong>in</strong> the case<br />
of negligible <strong>in</strong>fluence of frequency drift, that ADEV <strong>and</strong><br />
TDEVareestimatedbasedonthetimeerrorfunctionmeasured<br />
between the analyzed tim<strong>in</strong>g signal <strong>and</strong> the reference one [1]–<br />
[3].<br />
The formulae for the estimators of Allan deviation ADEV<br />
<strong>and</strong> the time deviation TDEV take the form:<br />
A ˆ �<br />
�<br />
�<br />
DEV (t)= � 1<br />
2n2τ 2 0 (N−2n)<br />
N−2n �<br />
(xi+2n − 2xi+n + xi) 2 (1)<br />
i=1<br />
T ˆ �<br />
�<br />
�<br />
� 1<br />
DEV(t)= �<br />
6n2 ⎡<br />
� �<br />
(N−3n+1)<br />
N−3n+1 j+n−1<br />
⎣ (xi+2n−2xi+n+xi)<br />
j=1<br />
(2)<br />
where {xi} is a sequence of N samples of time error function<br />
x(t) taken with <strong>in</strong>terval τ0; τ = nτ0 is an observation<strong>in</strong>terval.<br />
In order to simplify the computation process, the formula<br />
of the TDEV estimator (2) can be changed [4], [5]. After<br />
conversion the formula takes the form:<br />
i=j<br />
T ˆ �<br />
�<br />
�<br />
DEV (nτ0) = �1 1 1<br />
6 N − 3n + 1 n2 where<br />
N−3n+1 �<br />
j=1<br />
⎤<br />
⎦<br />
2<br />
S2 j (n), (3)<br />
Sj(n) = Sj−1(n)−xj−1+3xj+n−1−3xj+2n−1+xj+3n−1 (4)<br />
<strong>and</strong><br />
S1(n) =<br />
n�<br />
(xi+2n − 2xi+n + xi) (5)<br />
i=1<br />
When comput<strong>in</strong>g <strong>in</strong> the real time, we do not have access<br />
to the time error samples <strong>in</strong>dexed by i + n or i + 2n for
DOBROGOWSKI AND KASZNIA: METHODS OF REAL-TIME CALCULATION OF ALLAN DEVIATION AND TIME DEVIATION 43<br />
a time <strong>in</strong>stant described by <strong>in</strong>dex i (the currently measured<br />
sample) because these samples have not been measured yet.<br />
We have access to the sample currently measured (for the<br />
current sampl<strong>in</strong>g <strong>in</strong>stant i) <strong>and</strong> the samples measured earlier<br />
(with <strong>in</strong>dexes smaller than i) <strong>and</strong> stored <strong>in</strong> the equipment<br />
memory. Therefore, the <strong>in</strong>dexes <strong>in</strong> formulae for ADEV <strong>and</strong><br />
TDEV estimators must be changed <strong>in</strong> the case of real-time<br />
calculation. The rearrangement of <strong>in</strong>dexes for both estimators<br />
was performed<strong>in</strong>[6].As a result,we haveobta<strong>in</strong>edthe ADEV<br />
estimator formula for a current <strong>in</strong>stant i <strong>in</strong> the form depend<strong>in</strong>g<br />
on the sum of squares of second differences computed for the<br />
<strong>in</strong>stant i − 1:<br />
A ˆ � �<br />
DEVi(nτ0)= K(i, nτ0) Ai−1(n)+(xi−2xi−n+xi−2n) 2�<br />
(6)<br />
were K(i, nτ0) = 1/(2n 2 τ 2 0 (i − 2n)) <strong>and</strong> Ai(n) is the sum<br />
of squares of second differences of time error samples:<br />
Ai(n) =<br />
i�<br />
j=2n+1<br />
(xj − 2xj−n + xj−2n) 2 , i > 2n (7)<br />
The rearrangement of <strong>in</strong>dexes of the time deviation estimator<br />
is more complex than <strong>in</strong> the case of Allan deviation [6].<br />
After chang<strong>in</strong>g the <strong>in</strong>dexesof the simplified formula (3-5), we<br />
have obta<strong>in</strong>ed:<br />
T ˆ �<br />
1 1 1<br />
DEVi(nτ0) =<br />
6 i − 3n + 1 n2 Sov,i(n) (8)<br />
where Sov,i(n) is the overall sum updated for each sample i,<br />
given <strong>in</strong> the form:<br />
where<br />
Sov,i(n) = Sov,i−1(n) + S 2 i (n) (9)<br />
Si(n) = Si−1(n)−xi−3n+3xi+2n−3xi+n+xi, i > 3n (10)<br />
<strong>and</strong><br />
S3n(n) =<br />
3n�<br />
j=2n+1<br />
(xj − 2xj−n + xj−2n), j > 2n (11)<br />
F<strong>in</strong>ally, the operations of the real-time TDEV computation<br />
for i-th sampl<strong>in</strong>g <strong>in</strong>terval are performedus<strong>in</strong>g the formula [6]:<br />
T ˆ � �<br />
DEVi(nτ0)= L(i, n) Sov,i−1(n)+(Si−1(n)+∆i(n)) 2�<br />
where L(i, n) = 1/ � 6n 2 (i − 3n + 1) � <strong>and</strong>:<br />
∆i(n) = xi − 3xi+n + 3xi+2n − xi−3n<br />
(12)<br />
(13)<br />
As a result of the rearrangement of the parameters formulae,<br />
<strong>in</strong> order to compute both parameters, ADEV <strong>and</strong> TDEV, for<br />
a current sampl<strong>in</strong>g <strong>in</strong>stant i <strong>and</strong> given observation <strong>in</strong>terval<br />
τ = nτ0, we need the values of appropriate sum Ai−1(n),<br />
Sov,i−1(n), <strong>and</strong> Si−1(n), currently measured sample xi <strong>and</strong><br />
the samples xi−n, xi−2n, <strong>and</strong> xi−3n previously measured <strong>and</strong><br />
stored <strong>in</strong> the equipment memory.<br />
III. REAL-TIME COMPUTATION<br />
The formulae of ADEV <strong>and</strong> TDEV estimators <strong>in</strong> the forms<br />
given by (6) <strong>and</strong> (12) allow us to perform the calculation <strong>in</strong><br />
the real time, dur<strong>in</strong>g the measurement of time error samples.<br />
A general procedureof the real-time quasi-parallel ADEV <strong>and</strong><br />
TDEV computation for a series of observation <strong>in</strong>tervals is as<br />
follows [6]:<br />
1) Measure a new time error sample <strong>and</strong> store it <strong>in</strong> a data<br />
file.<br />
2) Compute the appropriated differences (for ADEV <strong>and</strong><br />
TDEV) for a given n (observation <strong>in</strong>terval τ = nτ0)<br />
us<strong>in</strong>g the current sample, <strong>and</strong> the samples measured n,<br />
2n or 3n sampl<strong>in</strong>g <strong>in</strong>tervals earlier.<br />
3) Update the sum for TDEV <strong>and</strong> sum of squares for<br />
ADEV, <strong>and</strong> compute the square for TDEV.<br />
4) Compute current averages <strong>and</strong> their square roots.<br />
5) Execute Steps 2-4 for successive larger observation<br />
<strong>in</strong>tervals (larger n).<br />
6) Return to Step 1 (measure a new sample).<br />
7) When the measurement is f<strong>in</strong>ished, the values of the parameter<br />
estimate for the observation<strong>in</strong>tervalsconsidered<br />
are known.<br />
Steps 2-5 can be executed when a sufficient <strong>number</strong> of time<br />
error samples were measured, i.e. 2n + 1 samples for a given<br />
n. We can compute the first value of ADEV estimate when<br />
the sample no. 2n + 1 has been measured. However, for the<br />
TDEV the computation of the <strong>in</strong>ternal sum Si(n), given by<br />
(11), only just starts. The first value of TDEV estimate we can<br />
compute when the sample no. 3n + 1 has been measured.<br />
An example of the real-time ADEV computation for the<br />
three observation <strong>in</strong>tervals – 3τ0, 5τ0, 7τ0 – is presented <strong>in</strong><br />
Fig. 1. Fifteen samples have been measured until now. Three<br />
w<strong>in</strong>dows,relatedwiththe observation<strong>in</strong>tervalsconsidered,are<br />
active. These w<strong>in</strong>dows – the operators of second difference –<br />
<strong>in</strong>dicate adequate samples engaged for calculat<strong>in</strong>g a proper<br />
second difference, e.g. the w<strong>in</strong>dow related with n=3 <strong>in</strong>dicates<br />
the samples no. 15, 12, <strong>and</strong> 9.<br />
The computation of TDEV for the first observation <strong>in</strong>terval<br />
τ = nτ0 beg<strong>in</strong>s when the first 2n + 1 samples are measured<br />
– for this <strong>in</strong>stant the first item of <strong>in</strong>ternal sum S3n(n) can be<br />
computed.Thesum S3n(n)isupdateduntilthesample<strong>number</strong><br />
3n is measured. Start<strong>in</strong>g from this <strong>in</strong>stant, the sum Si(n) is<br />
updated us<strong>in</strong>g the samples <strong>number</strong> i − 3n, i − 2n, i − n <strong>and</strong><br />
i, accord<strong>in</strong>g to (10), <strong>and</strong> the overall sum Sov,i(n) is updated<br />
accord<strong>in</strong>g to (9). When the updat<strong>in</strong>g for a given n is f<strong>in</strong>ished,<br />
the conditions for successive (greater) observation <strong>in</strong>tervals<br />
are checked, <strong>and</strong> necessary operations for the <strong>in</strong>tervals are<br />
performed.<br />
An exampleof the real-timeTDEV computationfor the two<br />
observation <strong>in</strong>tervals – 3τ0 <strong>and</strong> 5τ0 – is presented <strong>in</strong> Fig. 2.<br />
The stage of the process after measurement of the sample<br />
<strong>number</strong> 16 is presented. The overall sum Sov,i(3) is updated<br />
us<strong>in</strong>g the samples <strong>number</strong> 10, 13, <strong>and</strong> 16. The <strong>in</strong>ternal sum<br />
S1(3) was computed at the early stages of the process <strong>and</strong> its<br />
operator (second difference operator) is not active now. The<br />
<strong>in</strong>ternalsum S1(5) wascomputed<strong>and</strong>theoverallsum Sov,i(5)<br />
is updated us<strong>in</strong>g the samples <strong>number</strong> 1, 6, 11, <strong>and</strong> 16.
44 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 1. Real-time ADEV calculation for observation <strong>in</strong>tervals observation<br />
<strong>in</strong>tervals 3τ0, 5τ0, 7τ0 sample <strong>number</strong> 15 is measured.<br />
On-l<strong>in</strong>ecomputationofTDEVismorecomplexthanADEV<br />
computation, especially for the early stages of the process<br />
when the <strong>in</strong>ternal <strong>and</strong> overall sums are computed <strong>and</strong> the<br />
computations for some greater observation <strong>in</strong>tervals are not<br />
active yet (the conditions of beg<strong>in</strong>n<strong>in</strong>g the computations must<br />
be checked for each step). In general, <strong>in</strong> the case of real-time<br />
ADEV computation, three samples are <strong>in</strong>volved for a given n:<br />
onesample currentlymeasured,<strong>and</strong>two samplesfromthe past<br />
– measured n <strong>and</strong> 2n sampl<strong>in</strong>g <strong>in</strong>tervals earlier; <strong>in</strong> the case<br />
of real-time TDEV computation, four samples are <strong>in</strong>volved<br />
(besides these three samples, also the sample measured 3n<br />
sampl<strong>in</strong>g <strong>in</strong>tervals earlier) except for the early stages, when<br />
the <strong>in</strong>ternal sum is updated.<br />
Because the same samples are used for updat<strong>in</strong>g the sums<br />
<strong>in</strong> the real-timecalculation processesof ADEV <strong>and</strong>TDEV, we<br />
could compute both parameters jo<strong>in</strong>tly. The samples needed<br />
for computation of both parameters <strong>in</strong> the current <strong>in</strong>stant can<br />
be read out from the equipment memory at once, us<strong>in</strong>g one<br />
procedure <strong>in</strong>volv<strong>in</strong>g three samples (<strong>in</strong>dexed by i, i − n, <strong>and</strong><br />
i − 2n) at the early stages of the measurement process <strong>and</strong><br />
four samples (additionally the sample <strong>in</strong>dexed by i − 3n) at<br />
the late stages. Therefore, the <strong>in</strong>fluence of the most critical<br />
issue – access to the measured data – on the calculation time<br />
with<strong>in</strong> one sampl<strong>in</strong>g <strong>in</strong>terval can be reduced [7].<br />
An example of the real-time computationof Allan deviation<br />
<strong>and</strong> time deviation for s<strong>in</strong>gle observation <strong>in</strong>terval 3τ0 performedjo<strong>in</strong>tlyispresented<strong>in</strong>Fig.3<strong>and</strong>Fig.4.Theearlystage<br />
ofthe processispresented<strong>in</strong>Fig. 3.Thistime seventimeerror<br />
samples have been measured until now <strong>and</strong> the ADEV sum<br />
operator <strong>and</strong> TDEV <strong>in</strong>ternal sum operator are active, start<strong>in</strong>g<br />
from this <strong>in</strong>stant. The operator of the overall sum Sov(3) is<br />
still not active. Fig. 4 presents the stage of the process after<br />
the sample no. 10 has been measured. The ADEV operator is<br />
active <strong>and</strong> its sum of squares was updated us<strong>in</strong>g samples no.<br />
10, 7, <strong>and</strong> 4. The TDEV <strong>in</strong>ternal operator is not active now<br />
– the sum S(3) is computed now <strong>and</strong> from this <strong>in</strong>stant the<br />
overall sum operator (<strong>in</strong>dicat<strong>in</strong>g four samples) is active – the<br />
first item of the sum Sov(3) can be computed.<br />
Fig. 2. Real-time TDEV calculation for observation <strong>in</strong>tervals observation<br />
<strong>in</strong>tervals 3τ0 <strong>and</strong> 5τ0, sample <strong>number</strong> 16 is measured.<br />
Fig. 3. Jo<strong>in</strong>t real-time ADEV <strong>and</strong> TDEV calculation for observation <strong>in</strong>terval<br />
3τ0, sample <strong>number</strong> 7 is measured.<br />
IV. RESULTS OF COMPUTATION EXPERIMENT<br />
The methodsof separate as well as jo<strong>in</strong>t real-time computation<br />
of ADEV <strong>and</strong> TDEV described above were tested <strong>in</strong> the<br />
calculation experiments. The results of the experimental tests<br />
were presented <strong>in</strong> [6], [7]. The calculations were performed<br />
off-l<strong>in</strong>e but the onl<strong>in</strong>e work was imitated. The data sequence<br />
used <strong>in</strong> the experiment conta<strong>in</strong>s time error samples taken with<br />
the sampl<strong>in</strong>g <strong>in</strong>terval τ0 = 1/30 s, represent<strong>in</strong>g white phase<br />
noise.<br />
The calculations were performed for variable <strong>number</strong>s of<br />
observation <strong>in</strong>tervals, arranged <strong>in</strong> the logarithmic scale <strong>in</strong><br />
a range between 0.1 s <strong>and</strong> 1000 s. The start<strong>in</strong>g (smallest)<br />
observation <strong>in</strong>terval was τm<strong>in</strong> = 0.1 s (n = 3). The longest<br />
observation <strong>in</strong>terval was changed from 1 s till 1000 s. The<br />
calculations were performed for 5, 10, <strong>and</strong> 20 observation
DOBROGOWSKI AND KASZNIA: METHODS OF REAL-TIME CALCULATION OF ALLAN DEVIATION AND TIME DEVIATION 45<br />
Fig. 4. Jo<strong>in</strong>t real-time ADEV <strong>and</strong> TDEV calculation for observation <strong>in</strong>terval<br />
3τ0, sample <strong>number</strong> 10 is measured.<br />
Range of <strong>in</strong>tervals [s]<br />
TABLE I<br />
TIME OF ADEV CALCULATION<br />
Number of <strong>in</strong>tervals per decade<br />
5 10 20<br />
t-max [s] t-max [s] t-max [s]<br />
0.1-1 0.00012 0.00025 0.0005<br />
0.1-10 0.00024 0.00050 0.0010<br />
0.1-100 0.00034 0.00078 0.0015<br />
0.1-1000 0.00055 0.00110 0.0020<br />
<strong>in</strong>tervals per decade for each range.<br />
The maximum time used for calculation with<strong>in</strong> one sampl<strong>in</strong>g<br />
<strong>in</strong>terval was the observed quantity. We have assumed<br />
that this time cannot exceed the length of sampl<strong>in</strong>g <strong>in</strong>terval<br />
τ0 = 1/30 s = 0.0333. . . s. Personal computer with Intel<br />
Pentium IV 3.0 GHz microprocessor was used <strong>in</strong> the experimental<br />
tests.<br />
The time of ADEV computation is presented <strong>in</strong> TABLE I<br />
<strong>and</strong> the time of TDEV computation is presented <strong>in</strong> TABLE II<br />
[6]. The time of jo<strong>in</strong>t ADEV <strong>and</strong> TDEV computation is<br />
presented <strong>in</strong> TABLE III [7].<br />
The results presented were satisfactory for all cases considered.<br />
Even the most time-consum<strong>in</strong>g case – simultaneous<br />
computation for 81 observation <strong>in</strong>tervals (the range of τ from<br />
0.1 s till 1000 s <strong>and</strong> 20 observation <strong>in</strong>tervals for decade)<br />
– brought good result. The maximum time of operations<br />
performed for one sampl<strong>in</strong>g <strong>in</strong>terval does not exceed the<br />
sampl<strong>in</strong>g <strong>in</strong>terval 1/30 s. Compar<strong>in</strong>g the time of jo<strong>in</strong>t TDEV<br />
<strong>and</strong> ADEV computation with the time of TDEV computation,<br />
wecanseethatadditionaloperationsofADEVcomputationdo<br />
not <strong>in</strong>fluence the maximum time observed for one sampl<strong>in</strong>g<br />
<strong>in</strong>terval. The comparison of average time of operations performed<br />
with<strong>in</strong> one sampl<strong>in</strong>g <strong>in</strong>terval for TDEV computation<br />
<strong>and</strong> jo<strong>in</strong>t TDEV <strong>and</strong> ADEV computation presented <strong>in</strong> [7]<br />
confirmsthe expectationthatan additionaloperationof ADEV<br />
computation does not burden the whole process of real-time<br />
computation.<br />
Range of <strong>in</strong>tervals [s]<br />
TABLE II<br />
TIME OF TDEV CALCULATION<br />
Number of <strong>in</strong>tervals per decade<br />
5 10 20<br />
t-max [s] t-max [s] t-max [s]<br />
0.1-1 0.00018 0.00030 0.00060<br />
0.1-10 0.00030 0.00060 0.00120<br />
0.1-100 0.00050 0.00090 0.00180<br />
0.1-1000 0.00070 0.00130 0.00260<br />
TABLE III<br />
TIME OF TDEV AND ADEV JOINT COMPUTATION<br />
Range of <strong>in</strong>tervals [s]<br />
Number of <strong>in</strong>tervals per decade<br />
5 10 20<br />
t-max [s] t-max [s] t-max [s]<br />
0.1-1 0.00018 0.00032 0.00060<br />
0.1-10 0.00030 0.00060 0.00120<br />
0.1-100 0.00050 0.00090 0.00180<br />
0.1-1000 0.00070 0.00130 0.00260<br />
The computation complexity does not depend on the length<br />
of observation <strong>in</strong>terval; the <strong>number</strong> of observation <strong>in</strong>tervals<br />
considered is the only limit<strong>in</strong>g factor. Therefore, hav<strong>in</strong>g limited<br />
computational capacities, we can choose wider range<br />
of observation <strong>in</strong>tervals or greater <strong>number</strong> of observation<br />
<strong>in</strong>tervals for one decade (resolution of the computation results<br />
on the scale of observation <strong>in</strong>tervals). Small <strong>number</strong> of observation<br />
<strong>in</strong>tervals per decade (5 or 10) is sufficient for prompt<br />
analysis of tim<strong>in</strong>g signal, especially when perform<strong>in</strong>g <strong>in</strong> the<br />
real-time. More precise evaluation with the use of greater<br />
resolution (greater <strong>number</strong> of observation <strong>in</strong>tervals) could be<br />
performed off-l<strong>in</strong>e.<br />
V. CONCLUSIONS<br />
The results of the experimental tests have proved the<br />
ability of the real-time computation of Allan deviation <strong>and</strong><br />
time deviation as well as the real-time computation of both<br />
parameters performed jo<strong>in</strong>tly. The computation can be performed<br />
simultaneously for numerous series <strong>and</strong> wide range<br />
of observation <strong>in</strong>tervals (up to 81 simultaneously analyzed<br />
observation<strong>in</strong>tervalswere tested).Rather shortmaximumtime<br />
spent for computation with<strong>in</strong> one sampl<strong>in</strong>g <strong>in</strong>terval allows us<br />
to consider jo<strong>in</strong>t computation of another additional parameter<br />
based on the averag<strong>in</strong>g of second or third difference of time<br />
error.<br />
REFERENCES<br />
[1] ETSI EN 300 462, “Generic requirements for synchronization networks,”<br />
Tech. Rep., 1998.<br />
[2] ITU-T Rec. G.810, “Considerations on tim<strong>in</strong>g <strong>and</strong> synchronization issues,”<br />
Tech. Rep., 1996.<br />
[3] ANSI T1.101-1999, “Synchronization <strong>in</strong>terface st<strong>and</strong>ard,” Tech. Rep.<br />
[4] S. Bregni, Synchronization of Digital Telecommunications Networks. J.<br />
Wiley & Sons, 2002.<br />
[5] M. Kasznia, “Some approach to computation of ADEV, TDEV <strong>and</strong><br />
MTIE,” <strong>in</strong> Proc. of the 11th European Frequency <strong>and</strong> Time Forum,<br />
Neuchatel, Mar. 1997, pp. 544–548.
46 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
[6] A. Dobrogowski <strong>and</strong> M. Kasznia, “Real-time assessment of Allan deviation<br />
<strong>and</strong> time deviation,” <strong>in</strong> Proc. of the 2007 IEEE International<br />
Frequency Control Symposium Jo<strong>in</strong>tly with the 21st European Frequency<br />
<strong>and</strong> Time Forum, Geneva, May 2007, pp. 887–882.<br />
[7] ——, “Jo<strong>in</strong>t real-time assessment of Allan deviation <strong>and</strong> time deviation,”<br />
<strong>in</strong> Proc. of the 22nd European Frequency <strong>and</strong> Time Forum, Toulouse,<br />
France, Apr. 2008.<br />
Andrzej Dobrogowski was born <strong>in</strong> Poznan, Pol<strong>and</strong>, <strong>in</strong> 1938. He received his<br />
M.Sc. degree <strong>in</strong> electrical eng<strong>in</strong>eer<strong>in</strong>g from Poznan University of Technology<br />
<strong>in</strong> 1962, Ph.D. degree <strong>in</strong> telecommunications from Warsaw University of<br />
Technology <strong>in</strong> 1971 <strong>and</strong> Doctor habilitus degree from Poznan University of<br />
Technology <strong>in</strong> 1984. Heconcentrated his research <strong>in</strong>terests on synchronization<br />
<strong>in</strong> telecommunication networks <strong>and</strong> systems, optical networks, <strong>and</strong> estimation<br />
of signals’ parameters. He has been a manager of several projects carried out<br />
for Polish Telecom, deal<strong>in</strong>g with network <strong>and</strong> system synchronization. In his<br />
research group several unique measurement <strong>and</strong> timesource devices havebeen<br />
constructed mostly for Polish Telecom. He currently holds the position of Full<br />
Professor at the Chair of Telecommunication Systems <strong>and</strong> Optoelectronics,<br />
PUT.<br />
Michal Kasznia was born <strong>in</strong> Poznan, Pol<strong>and</strong>, <strong>in</strong> 1971. He received his<br />
M.Sc. degree <strong>in</strong> electronics <strong>and</strong> telecommunications <strong>in</strong> 1994 <strong>and</strong> Ph.D. degree<br />
<strong>in</strong> telecommunications <strong>in</strong> 2002 from Poznan University of Technology. His<br />
research concentrates on synchronization <strong>in</strong> telecommunication networks <strong>and</strong><br />
systems, especially on tim<strong>in</strong>g <strong>and</strong> carrier recovery us<strong>in</strong>g DSP technology, <strong>and</strong><br />
analysis of the quality of synchronization signals. He is currently an Assistant<br />
Professor at the Chair of Telecommunication Systems <strong>and</strong> Optoelectronics,<br />
PUT
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 47<br />
Application of Vernier Interpolation for Digital<br />
Time Error Measurement<br />
Krzysztof Lange <strong>and</strong> Michał Kasznia<br />
Abstract—The paper discusses potential applications of the<br />
time vernier pr<strong>in</strong>ciple, based on the so-called Vernier <strong>in</strong>terpolation.<br />
It presents the application of this method to precise time<br />
<strong>in</strong>terval measurement <strong>and</strong> to the results of construction work.<br />
Index Terms—phase detector, time error<br />
I. INTRODUCTION<br />
TIMING (synchronization) signals <strong>in</strong> telecommunication<br />
networksare affectedby a variety of distortion processes,<br />
which lower their quality. One of such processes is longterm<br />
r<strong>and</strong>om phase variation (w<strong>and</strong>er), characterized by a b<strong>and</strong>width<br />
below 10 Hz. The basic measure for estimat<strong>in</strong>g the<br />
quality of tim<strong>in</strong>g signal is time error TE, be<strong>in</strong>g the difference<br />
of phases of the <strong>in</strong>vestigated signal <strong>and</strong> the reference signal,<br />
expressed <strong>in</strong> time units. The precise measurement of TE is of<br />
key significance for appropriate estimation of the quality of<br />
the tim<strong>in</strong>g signal under test.<br />
II. TIME ERROR MEASUREMENT<br />
A typical (st<strong>and</strong>ard) technical implementation of TE measurement<br />
is the use of a circuit of digital phase detector; its<br />
general diagram is shown <strong>in</strong> Fig. 1. There are two signals, A<br />
<strong>and</strong> B, appended to the detector <strong>in</strong>puts; their phase difference<br />
is the subject of the measurement.The signal com<strong>in</strong>g out from<br />
thephasedetectorisaperiodicpattern<strong>in</strong>whichthedurationof<br />
high state ∆t is equal to the phase difference between signals<br />
A <strong>and</strong> B. Precise measurement of the phase difference is then<br />
reduced to the accurate measurement of time <strong>in</strong>terval ∆t.<br />
The measurement of time <strong>in</strong>terval ∆t accord<strong>in</strong>g to the idea<br />
shown <strong>in</strong> Fig. 1 consists <strong>in</strong> fill<strong>in</strong>g this <strong>in</strong>terval with pulses<br />
from a referencegeneratorwith frequency fw, which performs<br />
a gate circuit.<br />
Input signals A <strong>and</strong> B are <strong>in</strong>troduced on the <strong>in</strong>put circuits,<br />
which – except for the st<strong>and</strong>ardization of the form of these<br />
signals – often divide their frequency, reduc<strong>in</strong>g it to kilohertz<br />
values. This operation is favorable because the extension of<br />
duration of the exam<strong>in</strong>ed <strong>in</strong>terval ∆t is proportional to the<br />
division ratio , which enables the measurement range to be<br />
<strong>in</strong>creased already at this stage of measurement. Unfortunately,<br />
K.Langeis with Chair of Telecommunication Systems <strong>and</strong> Optoelectronics,<br />
Poznań University of Technology, ul. Polanka 3, 60-965 Poznañ, Pol<strong>and</strong> (email:<br />
lange@et.put.poznan.pl).<br />
M. Kasznia is with Chair of Telecommunication Systems <strong>and</strong> Optoelectronics,<br />
Poznañ University of Technology, ul. Polanka 3, 60-965 Poznań, Pol<strong>and</strong><br />
(e-mail: mkasznia@et.put.poznan.pl).<br />
This work was supported by the M<strong>in</strong>istry of Science <strong>and</strong> Higher Education<br />
<strong>in</strong> the frame of the project <strong>number</strong> <strong>number</strong> N N517 1545 33 <strong>in</strong> the years<br />
2007-<strong>2010</strong><br />
Fig. 1. diagram of a st<strong>and</strong>ard phase detector.<br />
apply<strong>in</strong>g large values of the division ratio of <strong>in</strong>put signals<br />
makes the period T proportionally <strong>in</strong>creased, which extends<br />
the time between particular time <strong>in</strong>tervals ∆t, <strong>and</strong>, consequently,<br />
it significantly limits the measurement dynamics. The<br />
counter, however, determ<strong>in</strong>es the <strong>number</strong> of pulses pass<strong>in</strong>g<br />
through a gate <strong>in</strong> time ∆t. The counted <strong>number</strong> N is certa<strong>in</strong>ly<br />
also proportional to reference frequency fw, <strong>and</strong> the st<strong>and</strong>ard<br />
generator period determ<strong>in</strong>es the phase detector resolution,<br />
equal to 1/fw. To obta<strong>in</strong> high precision <strong>in</strong> measur<strong>in</strong>g the<br />
phase difference of signals A <strong>and</strong> B, a high value of reference<br />
generator frequency is required. This requirement encounters<br />
two troublesome barriers. The underly<strong>in</strong>g cause of the first<br />
barrier is the st<strong>and</strong>ard frequency generator itself. The higher<br />
its frequency, the higher should be the multiplication factor<br />
of source signal, which is usually generated by the quartz<br />
oscillator. A high multiplication factor <strong>in</strong>creases the phase<br />
noise of this signal, which may cause errors greater than the<br />
error of <strong>in</strong>sufficient resolution. The other barrier results from<br />
the technologyof count<strong>in</strong>gpulsesbythecounter.Toobta<strong>in</strong>the<br />
expected high resolution, it is necessary to use digital counters<br />
with capacities of several dozen bits, <strong>in</strong> which first stages<br />
must operate correctly with gigahertz frequencies. It leads to<br />
emitt<strong>in</strong>g huge amounts of heat <strong>in</strong> them <strong>and</strong> significantly raises<br />
the costs of this solution.<br />
III. TIME VERNIER METHOD<br />
The Vernier <strong>in</strong>terpolation is commonly applied <strong>in</strong> the form<br />
of vernier (i.e. nonius) [1], [2] (<strong>in</strong> honor of a XV-century<br />
mathematician) to precisely measure lengths <strong>in</strong> two devices:<br />
micrometer screw <strong>and</strong> slide caliper. In each of those devices,<br />
depend<strong>in</strong>g on the length of the applied base, it is possible to<br />
<strong>in</strong>crease the measurement resolution from 10 to 100 or more
48 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
times. This method is adopted to the digital measurement of<br />
time <strong>in</strong>terval by the implementation of a time vernier device<br />
with two or three generators. Both circuits operate <strong>in</strong> similar<br />
way as slide caliper. They have two scales: the ma<strong>in</strong> scale<br />
<strong>and</strong> vernier scale. These scales have different densities, i.e.<br />
– for time <strong>in</strong>terval measurement – different periods of their<br />
generators. Respective tim<strong>in</strong>g diagrams for methods with 3<br />
<strong>and</strong> with 2 generators are shown <strong>in</strong> Fig. 2 <strong>and</strong> Fig. 3 [1].<br />
The vernier circuit with 3 generators needs a precise reference<br />
generator with period T0 <strong>and</strong> two quick-start auxiliary<br />
generators T1 <strong>and</strong> T2 with periods equal to one another but<br />
different from the period of generator T0. The analysis of the<br />
diagram <strong>in</strong> Fig. 2 allows us to determ<strong>in</strong>e a relation between<br />
the periodsof particular generators<strong>and</strong> the <strong>number</strong> of counted<br />
pulses <strong>in</strong> the method with 3 generators, which is shown <strong>in</strong> (1).<br />
∆t = T1 + T3 − T2 �<br />
= n1T0 1 + 1<br />
�<br />
+ n0T0 − n2T0<br />
n<br />
�<br />
�<br />
∆t = T0 n0 + (n1 − n2) 1 + 1<br />
��<br />
n<br />
�<br />
1 + 1<br />
�<br />
n<br />
Generator T1 starts at the <strong>in</strong>stance of the beg<strong>in</strong>n<strong>in</strong>g of<br />
exam<strong>in</strong>ed <strong>in</strong>terval ∆t. Generator T2 starts at the <strong>in</strong>stance of<br />
the end of this<strong>in</strong>terval.Valuesnwith appropriate<strong>in</strong>dexdenote<br />
the <strong>number</strong> of pulses counted by counters between the time<br />
co<strong>in</strong>cidences of pulses from generators T0 <strong>and</strong> T1 as well as<br />
T0 <strong>and</strong> T2. Without the vernier circuit <strong>in</strong>terval ∆t would be<br />
determ<strong>in</strong>ed accord<strong>in</strong>g to the formula:<br />
∆t = T0n0<br />
where n0 is the <strong>number</strong> of pulses counted by the counter.<br />
Expression (2) is – as we can easily notice – a fragment<br />
of equation (1). A measure of advantage result<strong>in</strong>g from the<br />
applicationoftheverniermethodisadditionaltermofequation<br />
(1). This is shown <strong>in</strong> the follow<strong>in</strong>g expression:<br />
∆t ′ = T0<br />
�<br />
(n1 − n2) (<br />
n + 1<br />
n )<br />
where n1 <strong>and</strong> n2 are the <strong>number</strong>s of pulses counted by<br />
respective counters, <strong>and</strong> n is a coefficient between periods<br />
T0, T1, T2, which is shown <strong>in</strong> formulae:<br />
� �<br />
1<br />
T1 = T2 = T0 + 1 (4)<br />
n<br />
n =<br />
T0<br />
T1 − T0<br />
From equation (5) it results that T1 = T2 > T0, which means<br />
that the frequency of auxiliary generators must be lower than<br />
the frequency of st<strong>and</strong>ard generator.<br />
The resolution of method with 3 generators is a result of<br />
the period of st<strong>and</strong>ard generator <strong>and</strong> coefficient n, which is<br />
shown <strong>in</strong> dependence:<br />
τ = T0<br />
(6)<br />
n<br />
A solution of the vernier circuit with 3 generators was proposed<br />
by Hewlett Packard <strong>in</strong> 1980 <strong>in</strong> a frequency counter.<br />
This method is effective because it makes it possible to obta<strong>in</strong><br />
�<br />
(1)<br />
(2)<br />
(3)<br />
(5)<br />
a resolution at 20 ps level; its practical realization, however,<br />
is troublesome. Construction difficulties result from a need to<br />
structuretwo generatorswithsimultaneousquickstart,without<br />
delay <strong>in</strong> the trigger pulse, with frequencies equal to one<br />
another <strong>and</strong> fixed frequency relation with the third generator.<br />
A certa<strong>in</strong> simplification is the solution with two generators.<br />
In order to preserve the vernier idea, the generators are <strong>in</strong><br />
mutual frequency relation, which is expressed by a fractional<br />
<strong>number</strong>, similarly as that of equations (4, 5). The elim<strong>in</strong>ation<br />
of one generatorfrom this solution makes the circuit operation<br />
easier because it is easier to design two generators with<br />
mutuallyfixedfrequencydifferencethanthreesuchgenerators.<br />
We should remember at the same time that we cannot use the<br />
quartz resonator when construct<strong>in</strong>g such generator due to its<br />
very high quality factor (with values of order 105 – 106). That<br />
is the reasonforaconsiderabletime delay at the <strong>in</strong>stanceof its<br />
start (of millisecond order) – it does not fulfill the assumption<br />
of rapid start [1].<br />
The operat<strong>in</strong>g pr<strong>in</strong>ciple of the vernier method with 2 generators<br />
is shown <strong>in</strong> a tim<strong>in</strong>g diagram <strong>in</strong> Fig. 3 [3].<br />
The operation of the measurement system of the exam<strong>in</strong>ed<br />
time <strong>in</strong>terval Tx beg<strong>in</strong>s at the <strong>in</strong>stance of appear<strong>in</strong>g of the<br />
edgethat triggersthe beg<strong>in</strong>n<strong>in</strong>gofthe exam<strong>in</strong>ed<strong>in</strong>terval.Then<br />
the generator with time T1 starts. After the time duration of<br />
exam<strong>in</strong>ed <strong>in</strong>terval ends, the other generator with the duration<br />
T2 is triggered. Both generators produce their signals so long<br />
as a time co<strong>in</strong>cidence occurs between the pulses of these<br />
generators. Up to this moment, each generator will produce<br />
the <strong>number</strong>s of pulses, respectively n1 <strong>and</strong> n2. This leads to<br />
the follow<strong>in</strong>g relation:<br />
∆t = n1 · T1 − n2 · T2 = (n1 − n2)T1 + n2τ (7)<br />
where τ = T1 − T2 is the difference of the generator<br />
periods, which at the same time expresses the resolution of<br />
the method. The fundamental difficulty <strong>in</strong> the realization of<br />
the method is a problem similar to that of the vernier with 3<br />
generators – it is difficult to construct a quick-start generator<br />
with good stability parameters. As mentioned above, quartz<br />
oscillators cannot be used for that purpose due to their long<br />
start, <strong>and</strong> the key<strong>in</strong>g of these generators causes a r<strong>and</strong>om<br />
error related with asynchronism between the generator trigger<br />
pulse <strong>and</strong> the generator period. Apart from the difficulties<br />
with manufactur<strong>in</strong>g the quick-start generator, there are other<br />
difficultieswith practical implementationof this method.They<br />
concern different problems, <strong>and</strong> an attempt to m<strong>in</strong>imize their<br />
effects unfortunately limits the possibilities to achieve good<br />
measurement parameters.<br />
These limitations result from the very idea of the vernier<br />
circuit. We can easily notice that the co<strong>in</strong>cidence of signals<br />
from two generators will never appear when their output<br />
frequencies are equal (to one another). A natural relation appears,therefore,betweenthetimeofachiev<strong>in</strong>gtheco<strong>in</strong>cidence<br />
<strong>and</strong> the measurement resolution, which is determ<strong>in</strong>ed by the<br />
difference of the periods of considered generators, accord<strong>in</strong>g<br />
to equation (7). Also the frequency of quick-start generators<br />
<strong>in</strong>fluences the co<strong>in</strong>cidence time, which is f<strong>in</strong>ally shown <strong>in</strong> an
LANGE AND KASZNIA: APPLICATION OF VERNIER INTERPOLATION FOR DIGITAL TIME ERROR MEASUREMENT 49<br />
Fig. 2. Tim<strong>in</strong>g diagram of vernier with 2 generators.<br />
TABLE I<br />
RELATION BETWEEB COINCIDENCE TIME AND MEASUREMENT<br />
RESOLUTION<br />
Resolution τ [ps] 1 5 10 50 100 500<br />
Co<strong>in</strong>cidence time tk [µs] 400 80 40 8 4 0,8<br />
<strong>in</strong>tuitive dependence:<br />
tk = T1T2<br />
≈<br />
T1 − T2<br />
1<br />
f 2τ + ∆t (8)<br />
where tk is the time of achiev<strong>in</strong>g co<strong>in</strong>cidence, f is an average<br />
frequency of generators, ∆t is the duration of exam<strong>in</strong>ed time<br />
<strong>in</strong>terval, <strong>and</strong> τ is the measurement resolution.<br />
The dependence express<strong>in</strong>g this relation – without tak<strong>in</strong>g<br />
<strong>in</strong>to account the impact of ∆t as well as for an assumed f<br />
average value of frequency of the vernier generators 50 MHz<br />
<strong>and</strong> a few possible sett<strong>in</strong>gs of resolution – is presented <strong>in</strong><br />
TABLE I. From the relations it results that <strong>in</strong> order to achieve<br />
a low resolution τ of the measurement of time <strong>in</strong>terval ∆t, the<br />
vernier circuit requires a process<strong>in</strong>g time that can significantly<br />
fulfill the <strong>in</strong>equality:<br />
tk > ∆t (9)<br />
Conclusionof<strong>in</strong>equality(9)limitsthemeasurementdynamics,<br />
which implies that each considered <strong>in</strong>terval must appear at the<br />
<strong>in</strong>put of the vernier circuit <strong>in</strong> time <strong>in</strong>terval greater than time<br />
tk [4].<br />
Technological limitations are result of the potential of implement<strong>in</strong>g<br />
the quick-start generators. As already mentioned,<br />
it is impossible to <strong>in</strong>troduce <strong>in</strong>to a generator a resonance<br />
system with large quality factor which will ensure a frequency<br />
<strong>in</strong>stability sufficient dur<strong>in</strong>g the measurement.<br />
In this situation a possible solution is e.g. to use a generator<br />
with delay l<strong>in</strong>e, <strong>in</strong> which the vibration period of this generator<br />
will be a function of delay time. Unfortunately, the performance<br />
of generator of that type is not stable enough <strong>and</strong> its<br />
output frequency depends, among others, on: temperature, the<br />
repeatability of applied circuits, supply voltage or the stability<br />
of the delay l<strong>in</strong>e itself. Summariz<strong>in</strong>g, technological problems<br />
can be, to some degree,reducedto “<strong>in</strong>fect<strong>in</strong>g”adigital system<br />
with a quasianalog unit with all consequences of such move.<br />
Fig. 3. Schematic diagram of generator.<br />
IV. DESCRIPTION OF CIRCUIT CONSTRUCTION<br />
In the experiment carried out the real signal from the<br />
real phase detector was replaced with a precise generator<br />
of time <strong>in</strong>terval. It is unimportant from the functional po<strong>in</strong>t<br />
of view; such replacement, however, makes it easier to set<br />
different time <strong>in</strong>tervals, which was proved to be suitable <strong>in</strong><br />
the analysis of errors of manufactured vernier circuits. The<br />
researchcarriedoutontheverniercircuitconcernsthesolution<br />
with 2 generators. The most important circuit <strong>in</strong> this case is<br />
the quick-start generator. The authors decided to <strong>in</strong>troduce<br />
programmable delay circuits to the generator. Thanks to the<br />
controlofdelayvalues,attemptsto optimizethissolutionwere<br />
possible. A schematic diagram of the generator is given <strong>in</strong><br />
Fig. 4 [4].<br />
The generator consists of two flip-flops 74AC74, delay<br />
l<strong>in</strong>e Dallas DS1020-25 [5], <strong>and</strong> gates 74AC00, 74AC86 <strong>and</strong><br />
74AC04.AXORgateisfedwiththeexam<strong>in</strong>edsignal.Because<br />
there are two such generators <strong>in</strong> the vernier circuit, the task<br />
of the gate is to compensate delays related to the start of the<br />
generators, as well as to determ<strong>in</strong>e whether a generator should<br />
be triggered by the lead<strong>in</strong>g (ris<strong>in</strong>g) or trail<strong>in</strong>g (fall<strong>in</strong>g) edge of<br />
the <strong>in</strong>put signal. The operation of the circuit utilizes the idea<br />
of positive feedback with the delay l<strong>in</strong>e.<br />
The construction of the l<strong>in</strong>e requires that the pulse duration<br />
to be delayed is longer than the delay time. This is the reason<br />
for <strong>in</strong>troduc<strong>in</strong>g a negator <strong>in</strong>to the reset circuit of flip-flop U1.<br />
The digital l<strong>in</strong>e DS1020-25 [4] applied <strong>in</strong> the circuit has the<br />
delay programmed with 8-bit word with step 150 ps. Only<br />
4 lower bits are used, however, because a delay longer than<br />
12,4 ns is not required. This value results from summ<strong>in</strong>g up<br />
16x150 ps <strong>and</strong> 10 ns. 10 ns is the m<strong>in</strong>imum value of the delay<br />
that can be achieved <strong>in</strong> the digital l<strong>in</strong>e DS1020-15.<br />
Theperiodofsuch generatoris a sum ofdelaysofparticular<br />
elements form<strong>in</strong>g the generator. After the digital word is<br />
provided at the output, its edge is propagated through flipflop<br />
U1A with time τU1. Then, the edge is delayed by the<br />
digital delay l<strong>in</strong>e DS1020 with controllable delay τDS, passes<br />
aga<strong>in</strong> onto the flip-flop – this time U2A with delay τU2 – <strong>and</strong><br />
is sent at the generator output. To pulses generated after the<br />
start pulse we should add also the propagation time τNAND<br />
from the output to gate NAND to flip-flop U1A. Therefore,<br />
the period of generated measurement signal equals:<br />
T = τU1 + τU2 + τNAND + τDS<br />
(10)<br />
In formula (10) the first three factors are almost constant.
50 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 4. Vernier block diagram.<br />
They depend exclusively on supply voltage <strong>and</strong> temperature.<br />
Simultaneous control over the period length <strong>and</strong> frequency is<br />
realized by the last factor.<br />
The rema<strong>in</strong><strong>in</strong>g part of the system is a technical realization<br />
of expression (7). It is presented <strong>in</strong> a block diagram <strong>in</strong> Fig. 5.<br />
Theexam<strong>in</strong>edmeasurementsignalissentatthe<strong>in</strong>putcircuit<br />
which generated signals START <strong>and</strong> STOP at its <strong>in</strong>put. The<br />
time unit between these signals is directly proportional to the<br />
exam<strong>in</strong>ed <strong>in</strong>terval.<br />
From equation (10) we can calculate that the shortest<br />
possible period that can be achieved is ca 18,5 ns, which<br />
gives a frequency approximately equal to 54 MHz. When<br />
the co<strong>in</strong>cidence of pulses from both generators appears, the<br />
detection system generates a co<strong>in</strong>cidence signal ST stopp<strong>in</strong>g<br />
the work of both generators. Dur<strong>in</strong>g the whole operation the<br />
systems of counters count pulses from both generators n1 <strong>and</strong><br />
n2. After read<strong>in</strong>g, a microprocessor makes the calculations of<br />
the exam<strong>in</strong>ed <strong>in</strong>terval ∆t, resets the counters, <strong>and</strong> then grants<br />
another measurement. The measurement result is displayed on<br />
the monitor of a computer cooperat<strong>in</strong>g with the microprocessor.<br />
Thanks to the application of two generators which conta<strong>in</strong><br />
<strong>in</strong>dependentdelay systemsDS1020-15,it ispossible to choose<br />
an appropriate value of τ, dependent only on the difference<br />
of words programm<strong>in</strong>g the delay systems. Assum<strong>in</strong>g a too<br />
small difference, e.g. 50 ps, causes an <strong>in</strong>stability of generators<br />
because the <strong>in</strong>stability comprises that difference.<br />
Assum<strong>in</strong>g a too big value of τ places the application of<br />
this methodunderthe questionmark becausethe improvement<br />
of resolution is very slight. In the manufactured model the<br />
value of τ 200 ps was assumed, which is an equivalent to the<br />
frequency of reference generator with value 5 GHz.<br />
V. SUMMARY<br />
The develop<strong>in</strong>g of the vernier circuit enabled the estimation<br />
of its performance. The fundamental objective has<br />
been achieved, i.e. the test<strong>in</strong>g of the vernier method <strong>and</strong> its<br />
optimization <strong>in</strong> the frame of the technology applied. The tests<br />
have answered the follow<strong>in</strong>g questions: which elements of the<br />
slotted l<strong>in</strong>e are responsible for process<strong>in</strong>g errors, <strong>and</strong> which<br />
po<strong>in</strong>ts of the system corrections should be <strong>in</strong>troduced. The<br />
ma<strong>in</strong> task at present is to reduce the manufactured device<br />
to FPGA technology, which will elim<strong>in</strong>ate the trouble of<br />
generators of the vernier itself <strong>and</strong> improve their parameters;<br />
a decrease <strong>in</strong> resolution is especially desired. The purpose of<br />
further effort is to achieve a resolution level of 20 ps.<br />
REFERENCES<br />
[1] S. Bregni, Synchronization of Digital Telecommunications Networks. J.<br />
Wiley&Sons, 2002.<br />
[2] [onl<strong>in</strong>e], pl.wikipedia.org/wiki/Noniusz.<br />
[3] J. Kalisz, R. Pe´lka, <strong>and</strong> R. Szplet, “Design problems <strong>in</strong> precise metrology<br />
of time units,” <strong>in</strong> Proc.of MWK conference, Rynia, 2001, pp. 117–165.<br />
[4] J. J˛erzejewski, “The application of the idea vernier to precise measurement<br />
of time <strong>in</strong>terval,” Master’s thesis, Poznan University of Technology,<br />
2008, supervisor Krzysztof Lange.<br />
[5] Catalog note of Dallas company.<br />
Krzysztof Lange was born <strong>in</strong> Poznan, Pol<strong>and</strong>, <strong>in</strong> 1945. He received his<br />
M.Sc. degree <strong>in</strong> 1969 <strong>and</strong> Ph.D. degree <strong>in</strong> 1978 from Poznan University<br />
of Technology. His research concentrates on synchronization <strong>in</strong> telecommunication<br />
networks <strong>and</strong> systems, digital circuit application <strong>and</strong> time <strong>and</strong><br />
frequency metrology. He is currently an Assistant Professor at the Chair of<br />
Telecommunication Systems <strong>and</strong> Optoelectronics, PUT<br />
Michal Kasznia was born <strong>in</strong> Poznan, Pol<strong>and</strong>, <strong>in</strong> 1971. He received his<br />
M.Sc. degree <strong>in</strong> electronics <strong>and</strong> telecommunications <strong>in</strong> 1994 <strong>and</strong> Ph.D. degree<br />
<strong>in</strong> telecommunications <strong>in</strong> 2002 from Poznan University of Technology. His<br />
research concentrates on synchronization <strong>in</strong> telecommunication networks <strong>and</strong><br />
systems, especially on tim<strong>in</strong>g <strong>and</strong> carrier recovery us<strong>in</strong>g DSP technology, <strong>and</strong><br />
analysis of the quality of synchronization signals. He is currently an Assistant<br />
Professor at the Chair of Telecommunication Systems <strong>and</strong> Optoelectronics,<br />
PUT
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 51<br />
Improv<strong>in</strong>g Statistical Properties of Number<br />
Sequences Generated by Multiplicative Congruential<br />
Pseudor<strong>and</strong>om Generator<br />
Abstract—A new method of improv<strong>in</strong>g the properties of <strong>number</strong><br />
sequences produced by a multiplicative congruential pseudor<strong>and</strong>om<br />
generator (MCPG) was proposed. The characteristic<br />
feature of the method is the simultaneous usage of <strong>number</strong>s<br />
generated by the sawtooth chaotic map, realized <strong>in</strong> a f<strong>in</strong>itestate<br />
mach<strong>in</strong>e, <strong>and</strong> symbols produced by the same map. The<br />
period of generated sequences can be significantly longer than<br />
theperiodof sequencesproducedbyamultiplicativecongruential<br />
pseudor<strong>and</strong>om generator realized <strong>in</strong> the same mach<strong>in</strong>e. It is<br />
shown that sequences obta<strong>in</strong>ed with the use of the proposed<br />
method pass all statistical tests from thest<strong>and</strong>ard NISTstatistical<br />
test suite v.1.8.<br />
Index Terms—pseudor<strong>and</strong>om generators, shuffl<strong>in</strong>g, comb<strong>in</strong>ed<br />
generators, sequences of symbols, statistical properties<br />
Mieczysław Jessa<br />
I. INTRODUCTION<br />
PSEUDORANDOM <strong>number</strong> sequences are used <strong>in</strong> many<br />
fieldsof science.Everyprogramm<strong>in</strong>glanguageprovidesa<br />
pseudor<strong>and</strong>om <strong>number</strong> generator that produces a sequence of<br />
nonnegative <strong>in</strong>tegers {p0, p1, ...} with <strong>in</strong>teger upper bound b,<br />
<strong>and</strong>thenuses {x0 = p0/b, x1 = p1/b, ...} asanapproximation<br />
of an <strong>in</strong>dependent <strong>and</strong> identically distributed (i.i.d.) sequence<br />
from unit <strong>in</strong>terval I = (0, 1). In almost all programm<strong>in</strong>g languages,<br />
<strong>number</strong>s {p0, p1, ...} are generated by a multiplicative<br />
congruential pseudor<strong>and</strong>om generator (MCPG) of the form<br />
pn = (apn−1) mod b n = 1, 2, .... (1)<br />
The properties of generated sequences depend strongly on the<br />
choice of two parameters: a multiplier a <strong>and</strong> a modulus b. To<br />
obta<strong>in</strong> maximal-length sequences (m-sequences), modulus b<br />
hastobe aprime<strong>number</strong><strong>and</strong>multiplier ahastobe aprimitive<br />
element modulo b [1]–[3]. Because the value for b is usually<br />
determ<strong>in</strong>ed by the <strong>number</strong> of bits used to encode <strong>number</strong>s,<br />
the statistical properties of generated sequences depend on the<br />
choice of the multiplier. In general, the choice of a “good” a<br />
is not simple <strong>and</strong> the <strong>number</strong>of multipliersgenerat<strong>in</strong>g<strong>number</strong><br />
sequences with good statistical properties is quite small [1],<br />
[2].<br />
In this paper, we propose a new method of improv<strong>in</strong>g<br />
properties of m-sequences produced by generator (1). The<br />
method exploits a sequence of symbols produced by the sawtooth<br />
chaotic map, implemented <strong>in</strong> computer <strong>in</strong> the modular<br />
arithmetic. The sequence is used to shuffle the output stream<br />
of MCPG. The same stream is shuffled <strong>in</strong> different ways,<br />
M. Jessa is with the Poznan University of Technology, Faculty of <strong>Electronics</strong><br />
<strong>and</strong> Telecommunications (e-mail: mjessa@et.put.poznan.pl).<br />
produc<strong>in</strong>g different sequences. The obta<strong>in</strong>ed sequences are<br />
comb<strong>in</strong>ed <strong>in</strong>to a s<strong>in</strong>gle sequence which forms the output<br />
stream. The generation of successive <strong>number</strong>s is slightly<br />
slower but we obta<strong>in</strong> additional control parameters (degrees<br />
of freedom) which can be used for improv<strong>in</strong>g the statistical<br />
properties of generated sequences, <strong>in</strong>clud<strong>in</strong>g the possibility<br />
of <strong>in</strong>creas<strong>in</strong>g the period of the sequences. The statistical<br />
properties of output streams are verified with the use of the<br />
st<strong>and</strong>ard NIST statistical test suite v.1.8 [4].<br />
This paper is organized as follows. Section II describes the<br />
method <strong>and</strong> the period of generated sequences. The results<br />
of the statistical tests from the st<strong>and</strong>ard NIST statistical test<br />
suitev.1.8,appliedtosequencesproducedbytheMCPG<strong>and</strong>to<br />
sequences produced by the proposed generator, are presented<br />
<strong>in</strong> Section III. Conclusions are drawn <strong>in</strong> Section IV.<br />
II. THE METHOD<br />
One of the characteristic features of many pseudor<strong>and</strong>om<br />
<strong>number</strong> generators is that <strong>number</strong>s obta<strong>in</strong>ed <strong>in</strong> the iterative<br />
procedure are simultaneously the output of the generator.<br />
MacLaren <strong>and</strong> Marsaglia suggested that the output stream of<br />
l<strong>in</strong>ear congruential pseudor<strong>and</strong>om <strong>number</strong> generator should<br />
be shuffled by us<strong>in</strong>g another, perhaps simpler, generator to<br />
obta<strong>in</strong> sequences with better statistical properties [2], [3].<br />
The first generator produces sequences which fill a table <strong>and</strong><br />
the second one is used to read off elements from this table.<br />
Because a s<strong>in</strong>gle pseudor<strong>and</strong>om<strong>number</strong>generatorcan be used<br />
to generate <strong>in</strong>dependent pseudor<strong>and</strong>om <strong>number</strong>s, it can also<br />
be used to shuffle itself [2], [3]. This method, us<strong>in</strong>g only one<br />
generator, was applied by Gebhardt to improve the statistical<br />
properties of <strong>number</strong> sequences produced by the Fibonacci<br />
generator[5].In1976Bays<strong>and</strong>Durhamproposeda methodof<br />
us<strong>in</strong>g a s<strong>in</strong>gle generatorto shuffle <strong>number</strong>sequencesproduced<br />
by the MCPG, known as RANDU [6]. Although shuffl<strong>in</strong>g can<br />
improvethe statistical propertiesof sequencesproducedby the<br />
MCPG, it is <strong>in</strong>sufficient to ensure that all statistical tests from<br />
the st<strong>and</strong>ard NIST statistical test suite v.1.8 could be passed<br />
for many a. Another approach uses comb<strong>in</strong>ed generators. In<br />
such type of generator the output streams of two or more<br />
generators (called source generators) are comb<strong>in</strong>ed, usually<br />
with the use of modulo 2 operation, <strong>in</strong>to a s<strong>in</strong>gle stream. The<br />
output sequence of the comb<strong>in</strong>ed generator has significantly<br />
longer period <strong>and</strong> better statistical properties than the output<br />
sequences of the source generators. Examples of comb<strong>in</strong>ed<br />
generators can be found, e.g., <strong>in</strong> [1], [3]. To achieve a positive
52 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
result of all tests from the NIST test suite, we must use<br />
many source generators, which is numerically <strong>in</strong>efficient. In<br />
this section, we <strong>in</strong>troduce a new method for generat<strong>in</strong>g many<br />
source streams by a s<strong>in</strong>gle MCPG. The generator is derived<br />
from the sawtooth chaotic map implemented <strong>in</strong> a f<strong>in</strong>ite-state<br />
mach<strong>in</strong>e <strong>in</strong> the modular arithmetic. The benefit is that we can<br />
comb<strong>in</strong>e many source streams <strong>in</strong>to a s<strong>in</strong>gle sequence without<br />
significantly decreas<strong>in</strong>g the speed of produc<strong>in</strong>g pseudor<strong>and</strong>om<br />
<strong>number</strong>s.<br />
Let Sλ denotethesawtoothmap,namedalsotheRényimap,<br />
the Bernoulli shift, or the Bernoulli map. Map Sλ transforms<br />
the unit <strong>in</strong>terval I = [0, 1) ⊂ X, X ≡ R <strong>in</strong>to itself <strong>and</strong> has<br />
the follow<strong>in</strong>g form<br />
Sλ(x) = λx mod 1, (2)<br />
where λ is a real <strong>number</strong>. Comput<strong>in</strong>g successive values of<br />
expression<br />
sn = ⌊αxn⌋ , α ≥ 2 , n = 1, 2, ..., (3)<br />
where α is an <strong>in</strong>teger <strong>and</strong> xn = λxn−1 mod 1, we obta<strong>in</strong><br />
a sequence {sn} of <strong>in</strong>teger <strong>number</strong>s. Numbers sn can be<br />
regardedas <strong>in</strong>dices of sub<strong>in</strong>tervalsconta<strong>in</strong><strong>in</strong>g xn <strong>and</strong> obta<strong>in</strong>ed<br />
as the result of partition<strong>in</strong>g I <strong>in</strong>to α disjo<strong>in</strong>t, equal-sized<br />
sub<strong>in</strong>tervals Ij, j = 0, 1, 2, ..., α − 1, cover<strong>in</strong>g the whole<br />
set I. Through assign<strong>in</strong>g a unique <strong>number</strong> (symbol) from<br />
set Aα = {0, 1, ..., α − 1} to every Ij, the macroscopic<br />
behavior of the dynamical system (Sλ, I) can be studied.<br />
This macroscopic dynamics is called symbolic dynamics. It is<br />
knownthatsymbolicsequencesmaybetreatedastrulyr<strong>and</strong>om<br />
sequences <strong>in</strong> many aspects [7]–[10]. Assum<strong>in</strong>g <strong>in</strong>teger λ <strong>and</strong><br />
rational x0 = (p0)/(q0), where 0 < pn < q0, we obta<strong>in</strong> that<br />
[11] ⎧ ⎨<br />
⎩<br />
sn = ⌊αxn⌋<br />
xn = pn<br />
q0<br />
pn = λ · pn−1 mod q0<br />
n = 1, 2, . . .<br />
. (4)<br />
Because <strong>in</strong> a f<strong>in</strong>ite-state mach<strong>in</strong>e the <strong>number</strong> of bits encod<strong>in</strong>g<br />
the values of all variables is limited to l, where l is f<strong>in</strong>ite,<br />
expression (4) can be written as<br />
⎧<br />
⎪⎨ sn = ⌊α · xn⌋ � �<br />
pn<br />
xn = truncl n = 1, 2, . . . , (5)<br />
q0<br />
⎪⎩<br />
pn = λpn−1 mod q0<br />
where truncl denotes the truncation operation, leav<strong>in</strong>g l the<br />
most significant bits of quotient (pn)/(q0). If α = 2k , 1 ≤<br />
k ≤ l, then sequence {sn} consists of <strong>number</strong>s encoded by<br />
the k most significant bits of xn. If additionally q0 = 2l or<br />
q0 = 2l − 1, these bits are the same as the most significant<br />
bits of pn (see [11] for examples). Then (5) is reduced to<br />
�<br />
sn = trunck(pn)<br />
. (6)<br />
pn = λpn−1 mod q0<br />
The second formula <strong>in</strong> (6) describes the multiplicative congruential<br />
pseudor<strong>and</strong>om generator (1) with a = λ <strong>and</strong> b = q0.<br />
For α = 2 k , 1 ≤ k ≤ l <strong>and</strong> q0 = 2 l or q0 = 2 l − 1,<br />
sequence {sn} is the same as the output sequence of the<br />
truncated multiplicative congruentialpseudor<strong>and</strong>omgenerator.<br />
To improvethe statistical propertiesof {pn}, successive pn are<br />
first written <strong>in</strong>to Table T with L cells, addressed from 0 to<br />
L − 1. Next, we read off K <strong>number</strong>s T1, T2, ..., TK from T<br />
per one iteration of equation (6), where it is assumed that<br />
L ≥ αK. The addresses of T1, T2, ..., TK depend on sn.<br />
Numbers T1, T2, ..., TK are treated as vectors encoded by l<br />
bits. The elements of K vectors are summed modulo 2 <strong>and</strong><br />
added modulo 2 to current <strong>number</strong> pn, denoted for clarity as<br />
T0, form<strong>in</strong>g a s<strong>in</strong>gle vector Un. Its elements can encode an<br />
<strong>in</strong>teger <strong>number</strong> from <strong>in</strong>terval (0, 2 l ) or a real <strong>number</strong> from<br />
unit <strong>in</strong>terval I = (0, 1). The pseudocode of an algorithm<br />
proposed for produc<strong>in</strong>g {Un} has the follow<strong>in</strong>g form:<br />
Algorithm 1 Algorithm CPRNG<br />
Initialization:<br />
Choose k, p0 ∈ (0, q0) <strong>and</strong> the size L of Table T;<br />
Write p0 <strong>in</strong>to the first cell of Table T, i.e. T [0] := p0;<br />
for n := � 1 to L − 1 do<br />
pn := λpn−1 mod q0, n = 1, 2, ...L − 1<br />
(7)<br />
T [n] := pn<br />
end for<br />
Computations:<br />
for n := 1 to N do<br />
⎧<br />
pn+L−1 := λpn+L−2 mod q0<br />
⎪⎨<br />
j := n mod L, L ≥ αK, α = 2<br />
⎪⎩<br />
k , 1 ≤ k ≤ l<br />
T [j] := pn+L−1<br />
s ′ n+L−1 := 1 + trunck(pn+L−1)<br />
Un := T [j] ⊕ T [ � j + s ′ �<br />
n+L−1 mod L]<br />
⊕ · · · ⊕ T [ � j + Ks ′ (8)<br />
�<br />
n+L−1 mod L]<br />
end for<br />
In (8) it is that s ′ n+L−1 = 1+sn+L−1. The comb<strong>in</strong>ed pseudor<strong>and</strong>om<br />
<strong>number</strong> generator CPRNG repeatedly uses the “bit<br />
stripp<strong>in</strong>g”,known from the shuffl<strong>in</strong>galgorithmsof Gebhardor<br />
Bays<strong>and</strong>Durham(seep.10<strong>in</strong>[2]).Numbers pn written<strong>in</strong>to T<br />
can be regarded as digits encod<strong>in</strong>g a certa<strong>in</strong> <strong>number</strong> p, written<br />
<strong>in</strong> the fixed-po<strong>in</strong>t <strong>number</strong> system with base q0. If {pn} is a<br />
r<strong>and</strong>om sequence, then all sequences composed of digits chosen<br />
from digits encod<strong>in</strong>g p are <strong>in</strong>dependent [2]. The addresses<br />
of <strong>number</strong>s T0, T1, .., TK differ by a constant value s ′ n+L−1 .<br />
Numbers s ′ n+L−1 are the elements of symbolicsequence {sn}<br />
produced by chaotic Sλ <strong>and</strong> realized <strong>in</strong> computer <strong>in</strong> the<br />
modulararithmetic – shifted by unity. The same algorithm can<br />
be used for other values q0 but symbols s ′ n+L−1 have to be<br />
computed from formula s ′ n+L−1 = 1 + trunck(pn+L−1/q0),<br />
i.e., they cannot be the most significant digits of pn+L−1<br />
<strong>in</strong>creased by 1. Chang<strong>in</strong>g the method of address<strong>in</strong>g Table T,<br />
we can obta<strong>in</strong> different comb<strong>in</strong>ed generators.<br />
The period mu ofsequence {Un} dependson theperiod mp<br />
of sequence {pn} <strong>and</strong> the size L of Table T. Table T is filled<br />
with L elements of sequence {pn} dur<strong>in</strong>g the Initialization.<br />
After n = LCM(mp, L) iterations of expression (8), where<br />
LCM(mp, L) is the least common multiple of <strong>number</strong>s mp<br />
<strong>and</strong> L, Table T is filled with the same <strong>number</strong>s as after the<br />
Initialization. For n > LCM(mp, L), we obta<strong>in</strong><br />
U n+LCM(mp,L) = Un. (9)
MIECZYSŁAW JESSA: IMPROVING STATISTICAL PROPERTIES OF NUMBER SEQUENCES 53<br />
For n < LCM(mp, L) Table T does not conta<strong>in</strong> the same<br />
elements as dur<strong>in</strong>g the Initialization. If some element Un is<br />
repeated for j = n, where n < LCM(mp, L), it is not<br />
repeated for all n be<strong>in</strong>g a multiple of j, which results directly<br />
fromthemethodofcomput<strong>in</strong>gof Un.Consequently,theperiod<br />
of {Un} cannot be smaller than LCM(mp, L). Chang<strong>in</strong>g the<br />
size L of Table T, we can <strong>in</strong>fluence the period of generated<br />
sequences. If L is relatively prime to mp, the period of {Un}<br />
is L times greater than the period of m-sequence produced by<br />
the MCPG, implemented <strong>in</strong> the same f<strong>in</strong>ite-state mach<strong>in</strong>e.<br />
III. THE RESULTS OF NIST STATISTICAL TESTS<br />
To verify the hypothesis that the statistical properties of<br />
{pn} can be improved by the proper choice of α, K, <strong>and</strong> L,<br />
the st<strong>and</strong>ard NIST statistical test suite v. 1.8 for cryptographic<br />
applications was applied. It conta<strong>in</strong>s 15 tests, designed for analyz<strong>in</strong>g<br />
different statistical properties of generated sequences,<br />
turned <strong>in</strong>to b<strong>in</strong>ary streams [4]. The goal of the tests is to<br />
detect non-r<strong>and</strong>omness <strong>in</strong> b<strong>in</strong>ary sequences produced us<strong>in</strong>g<br />
r<strong>and</strong>om <strong>number</strong> or pseudor<strong>and</strong>om <strong>number</strong> generators. The<br />
tested sequences are composed of bits encod<strong>in</strong>g successive<br />
Un. The null hypothesis is that any sequence be<strong>in</strong>g tested is<br />
r<strong>and</strong>om. Associated with this null hypothesis is an alternative<br />
hypothesis, which, for the NIST tests, is that any tested<br />
sequence is not r<strong>and</strong>om. The tests search for deviations from<br />
the properties of truly r<strong>and</strong>om b<strong>in</strong>ary sequences <strong>in</strong> b<strong>in</strong>ary sequences<br />
produced by a source under test. If a b<strong>in</strong>ary sequence<br />
passes thetests, thereis noreasonto reject thenull hypothesis.<br />
The empirical results can be <strong>in</strong>terpreted <strong>in</strong> many ways. In<br />
this paper two approaches proposed by NIST were used: (1)<br />
the exam<strong>in</strong>ation of the proportion R of sequences that pass a<br />
statistical test <strong>and</strong> (2)the distributionof the so called P-values<br />
computed by software. In the first case, we f<strong>in</strong>d the proportion<br />
of sequences that pass a given test. The second approach,<br />
adopted by NIST, measures the distribution of P-values <strong>in</strong><br />
<strong>in</strong>terval [0, 1] divided <strong>in</strong>to ten equal-sized sub<strong>in</strong>tervals. The<br />
P-value is the probability (under the null hypothesis of r<strong>and</strong>omness)thatthechosentest<br />
statistic will assumevaluesequal<br />
to or worse than the test statistic value observed when consider<strong>in</strong>g<br />
the null hypothesis. The P-value is frequently called<br />
the “tail probability”. When the sequences are r<strong>and</strong>om b<strong>in</strong>ary<br />
sequences, the P-values obta<strong>in</strong>ed for these sequences have to<br />
be uniformlydistributed<strong>in</strong> [0, 1] [4]. As the result of apply<strong>in</strong>g<br />
a χ 2 test <strong>and</strong> an additional function, we obta<strong>in</strong> a new P-value<br />
(PT ), correspond<strong>in</strong>g to the Goodness-of-Fit Distribution Test<br />
onthe P-valuesobta<strong>in</strong>edforanarbitrarystatistical test(i.e.the<br />
P-value of the P-values).If PT ≥ 0.0001,then the sequences<br />
can be considered to be uniformly distributed. The details of<br />
comput<strong>in</strong>g PT can be found <strong>in</strong> [4].<br />
The statistical tests were performed on 1000 different sequences<br />
of length 10 6 . The sequences were successive fragments<br />
of sequence {pn} or {Un}, produced for the smallest λ<br />
for which {pn} was the m-sequence. Modulus q0 was a prime<br />
<strong>number</strong> equal to 2 31 − 1 (l = 31) <strong>and</strong> p0 was equal to unity.<br />
The size of Table T was constant dur<strong>in</strong>g all experiments <strong>and</strong><br />
equal to L = 32. Because the least common multiple of mp<br />
<strong>and</strong> L is equal to 34359738336, the period mu of {Un} is 16<br />
TABLE I<br />
THE RESULTSOF NIST TESTSFOR MCPG WITH λ = 7<br />
Type of the test R(> 0.981) PT (> 0.0001) F<strong>in</strong>al result<br />
Block Frequency 0.0000 0.00000 fail<br />
Serial* 0.9780 0.05642 fail<br />
Approximate Entropy 0.9750 0.00711 fail<br />
L<strong>in</strong>ear Complexity 0.9900 0.7944 pass<br />
Universal 0.9120 0.00000 fail<br />
Overlapp<strong>in</strong>g<br />
Templates<br />
0.5490 0.00000 fail<br />
Non-overlapp<strong>in</strong>g<br />
Templates<br />
0.9640 0.00000 fail<br />
Cumulative Sums* 0.9670 0.00000 fail<br />
Runs 0.9950 0.01570 pass<br />
Longest Runs of Ones 0.9640 0.00000 fail<br />
Rank 0.9880 0.43543 pass<br />
Spectral DFT 0.0000 0.00000 fail<br />
R<strong>and</strong>om Excursions* 0.9836 0.07375 pass<br />
R<strong>and</strong>om Excursions<br />
Variant**<br />
0.9800 0.01526 pass<br />
Frequency 0.9760 0.00000 fail<br />
*This test consists of several subtests: the worst result is shown.<br />
**The m<strong>in</strong>imum pass rate for this test for a st<strong>and</strong>ard set of parameters is<br />
approximately 0.978.<br />
times longer than the period mp = 2 31 − 2 = 2 147 483 646<br />
of {pn} produced by the MCPG. The results of the st<strong>and</strong>ard<br />
NIST test suite performed for b<strong>in</strong>ary sequences, composed<br />
of bits encod<strong>in</strong>g successive pn, generated by MCPG with<br />
λ = 7, are shown <strong>in</strong> TABLE I. The results of the same tests<br />
for b<strong>in</strong>ary sequences, composed of bits encod<strong>in</strong>g successive<br />
Un, are presented <strong>in</strong> TABLE II. Parameter α was equal to 4.<br />
Numbers from TABLE II were obta<strong>in</strong>ed for the smallest K<br />
forwhichsequencesproducedbyCPRNG passedall statistical<br />
tests.<br />
IV. CONCLUSION<br />
A new method for improv<strong>in</strong>g the quality of a multiplicative<br />
congruential pseudor<strong>and</strong>om generator was proposed <strong>in</strong> this<br />
paper. The method uses symbols produced by the sawtooth<br />
map realized <strong>in</strong> a f<strong>in</strong>ite-state mach<strong>in</strong>e <strong>and</strong> <strong>number</strong>s produced<br />
by a multiplicative congruential generator, obta<strong>in</strong>ed as the<br />
result of implement<strong>in</strong>g the same map <strong>in</strong> the same mach<strong>in</strong>e<br />
<strong>in</strong> the modular arithmetic. Although the proposed algorithm<br />
improves the statistical properties of sequences produced by<br />
a known pseudor<strong>and</strong>om generator, it can be treated as a new<br />
generator, derived from a chaotic map. The basic weakness<br />
of this generator is the lack of theory which could simplify<br />
the choice of α, K <strong>and</strong> L. Simulation experiments, performed<br />
by the author for many λ <strong>and</strong> q0 = 2 31 − 1, show that it is<br />
always possible to choose relatively small K (of the order of<br />
8) which yields sequences pass<strong>in</strong>g all tests from the st<strong>and</strong>ard<br />
NIST statistical test suite v. 1.8. The speed of produc<strong>in</strong>g {Un}<br />
with α = 4, L = 32 <strong>and</strong> K = 3 is only about 25% smaller<br />
than the speed of produc<strong>in</strong>g {pn} on the same hardware <strong>and</strong><br />
software platform.<br />
Access to a pseudor<strong>and</strong>om generator produc<strong>in</strong>g long period<br />
<strong>number</strong> sequences that pass all NIST tests for many multi-
54 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
TABLE II<br />
THE RESULTSOF NIST TESTSFOR CPRNG WITH λ = 7, α = 4,<br />
K=3<br />
Type of the test R(> 0.981) PT (> 0.0001) F<strong>in</strong>al result<br />
Block Frequency 0.9900 0.86288 pass<br />
Serial* 0.9870 0.13728 pass<br />
Approximate Entropy 0.9920 0.13112 pass<br />
L<strong>in</strong>ear Complexity 0.9920 0.68902 pass<br />
Universal 0.9850 0.00737 pass<br />
Overlapp<strong>in</strong>g<br />
Templates<br />
0.9900 0.16170 pass<br />
Non-overlapp<strong>in</strong>g<br />
Templates*<br />
0.9820 0.02979 pass<br />
Cumulative Sums* 0.9840 0.67661 pass<br />
Runs 0.9860 0.04198 pass<br />
Longest Runs of Ones 0.9930 0.89348 pass<br />
Rank 0.9950 0.96019 pass<br />
Spectral DFT 0.9880 0.26757 pass<br />
R<strong>and</strong>om Excursions* 0.9865 0.31094 pass<br />
R<strong>and</strong>om Excursions<br />
Variant**<br />
0.9828 0.09676 pass<br />
Frequency 0.9870 0.93900 pass<br />
*This test consists of several subtests: the worst result is shown.<br />
**The m<strong>in</strong>imum pass rate for this test for a st<strong>and</strong>ard set of parameters is<br />
approximately 0.978.<br />
pliers λ enables us to construct a high-speed pseudor<strong>and</strong>om<br />
generatorwithlongperiodsofgeneratedstreams. Thesimplest<br />
method uses a field programmable gate array (FPGA). In this<br />
circuit, we implement r CPRNGs with different values of λ<br />
that work <strong>in</strong> parallel. In each step of generation, we obta<strong>in</strong> r<br />
pseudor<strong>and</strong>om<strong>number</strong>s.Consequently,the speedof produc<strong>in</strong>g<br />
pseudor<strong>and</strong>om <strong>number</strong>s <strong>in</strong>creases r times. This property can<br />
be used <strong>in</strong> cryptography <strong>and</strong> <strong>in</strong> multi-core processors for fast<br />
generation of high-quality pseudor<strong>and</strong>om <strong>number</strong>s with long<br />
periods.<br />
REFERENCES<br />
[1] P. Bratley, B. L. Fox, <strong>and</strong> L. E. Schrage, A Guide to Simulation. New<br />
York: Spr<strong>in</strong>ger-Verlag, 1987, ch. 6.<br />
[2] J. E. Gentle, R<strong>and</strong>om Number Generation <strong>and</strong> Monte Carlo Methods.<br />
New York: Spr<strong>in</strong>ger, 2003, ch. 1.<br />
[3] D. E. Knuth, The Art of Computer Programm<strong>in</strong>g, 2nd ed. Addison<br />
Wesley, 1981, vol. 2, ch. 3.<br />
[4] [onl<strong>in</strong>e], http://csrc.nist.gov/rng/.<br />
[5] F. Gebhard, “Generat<strong>in</strong>g pseudo-r<strong>and</strong>om <strong>number</strong>s by shuffl<strong>in</strong>g a Fibonacci<br />
sequence,” Mathematics of Computation, vol. 21, pp. 708–709,<br />
1967.<br />
[6] C. Bays <strong>and</strong> S. D. Durham, “Improv<strong>in</strong>g a poor r<strong>and</strong>om <strong>number</strong> generator,”<br />
ACM Trans. on Mathematical Software, vol. 2, pp. 59–64, 1976.<br />
[7] M. P. Kennedy, R. Rovatti, <strong>and</strong> G. Setti, Chaotic <strong>Electronics</strong> <strong>in</strong> Telecommunications.<br />
Boca Raton: CRC Press, 2000, ch. 3.<br />
[8] L. Kocarev, G. Jakimoski, <strong>and</strong> Z.Tasev, Chaos <strong>and</strong> Pseudo-R<strong>and</strong>omness<br />
<strong>in</strong> Chaos Control, 2003, pp. 247–263.<br />
[9] T.Kohda <strong>and</strong> A. Tsuneda, “Statistics of chaotic b<strong>in</strong>ary sequences,” IEEE<br />
Trans. Inf. Theory, vol. 43, pp. 104–112, Jan. 1997.<br />
[10] T. Stojanovski <strong>and</strong> L.Kocarev, “Chaos-based r<strong>and</strong>om <strong>number</strong> generators<br />
– Part I: Analysis,” IEEE Trans. Circuits Syst. I, vol. 48, pp. 281–288,<br />
Mar. 2001.<br />
[11] M. Jessa, “Design<strong>in</strong>g security for <strong>number</strong> sequences generated by means<br />
of the sawtooth chaotic map,” IEEE Trans. Circuits Syst. I, vol. 53, pp.<br />
1140–1150, May 2006.<br />
Mieczyslaw Jessa was born <strong>in</strong> Pol<strong>and</strong> <strong>in</strong> 1961. He received the M.Sc. degree<br />
with honors from Poznan University of Technology <strong>in</strong> 1985 <strong>and</strong> the Ph.D.<br />
degree <strong>in</strong> 1992 from the same University. S<strong>in</strong>ce 1985 he has been employed<br />
at the Institute of <strong>Electronics</strong> <strong>and</strong> Telecommunications <strong>in</strong> Poznan. Now, he<br />
works with the Chair of Telecommunication Systems <strong>and</strong> Optoelectronics of<br />
the same University.<br />
Initially, his research <strong>in</strong>terest <strong>in</strong>cluded phase-locked loops <strong>and</strong> PDH/SDH<br />
network synchronization. In the years 1995-1997 he was an expert of Polish<br />
M<strong>in</strong>istry of Communications <strong>in</strong> the field of digital network synchronization.<br />
His current research concerns r<strong>and</strong>omness <strong>and</strong> pseudo-r<strong>and</strong>omness, the applications<br />
of the chaos phenomenon, <strong>and</strong> mathematical models of systems<br />
evolution. He is the author or co-author of over one hundred journal <strong>and</strong><br />
conference papers <strong>and</strong> fifteen patents.
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 55<br />
New Tailbit<strong>in</strong>g Convolutional Codes over R<strong>in</strong>gs<br />
Abstract—In this paper a method of us<strong>in</strong>g convolutional codes<br />
over r<strong>in</strong>gs for packet data transmission over additive white gaussian<br />
noise (AWGN) channel is proposed. The tailbit<strong>in</strong>g method<br />
is generalized <strong>and</strong> applied to convolutional codes based on r<strong>in</strong>g<br />
of <strong>in</strong>tegers modulo-M. The codes were named tailbit<strong>in</strong>g codes<br />
over r<strong>in</strong>g (TBR). This paper presents a method to des<strong>in</strong>g TBR<br />
codes obta<strong>in</strong>ed by the concatenation of feedback convolutional<br />
encoder over r<strong>in</strong>g <strong>and</strong> M-QAM modulator. The paper describes<br />
how a systematic r<strong>in</strong>g convolutional encoder with feedback can<br />
obta<strong>in</strong> the same start<strong>in</strong>g <strong>and</strong> end<strong>in</strong>g state. The best TBR codes<br />
with different <strong>number</strong> of encoder states for 16-QAM modulated<br />
symbol sequences of vary<strong>in</strong>g lengths are tabulated.<br />
Index Terms—Convolutional codes over r<strong>in</strong>gs, tailbit<strong>in</strong>g codes<br />
Piotr Remle<strong>in</strong> <strong>and</strong> Dawid Szłapka<br />
I. INTRODUCTION<br />
P ACKETdatatransmissionschemesareoftenused<strong>in</strong>wireless<br />
telecommunication systems. The convolutionalcodes<br />
are used <strong>in</strong> such systems as an efficient <strong>and</strong> powerful class of<br />
error correct<strong>in</strong>g codes [1]. To be able to use convolutional<br />
codes (of rate R = k/n <strong>and</strong> m memory elements) <strong>in</strong> the<br />
packet transmission, we must convert these codes to block<br />
codes. There are some methods for this conversion. One of<br />
such methods is called tailbit<strong>in</strong>g. In this method, no additional<br />
bits are appended to the codeword to drive the encoder to a<br />
known state [2]. The encoder starts <strong>and</strong> f<strong>in</strong>ishes the encod<strong>in</strong>g<br />
process <strong>in</strong> the same state but this state is not known by the<br />
decoder. In the paper [3] it is shown that the turbo-codes<br />
generally provide the best error rate performance for long<br />
blocks(over150bits), but forshort blocks(under150bits) the<br />
tailbit<strong>in</strong>g convolutional codes provide the best performance.<br />
The motivation for <strong>in</strong>vestigat<strong>in</strong>g the r<strong>in</strong>g convolutional codes<br />
was to explore a natural relation between M-ary modulation<br />
<strong>and</strong> codes over the r<strong>in</strong>gs of <strong>in</strong>tegers modulo-M [4]. Up to<br />
now, the best tailbit<strong>in</strong>g codes with the greatest m<strong>in</strong>imum<br />
Hamm<strong>in</strong>g distance were published <strong>in</strong> the literature [2], [5],<br />
[6]. In case of search for the best convolutional codes for the<br />
signals transmitted over the AWGN channel, the quality the<br />
criterion is Euclidean distance [1]. In this article we assumed<br />
the Euclidean distance as a parameter to estimate the quality<br />
of TBR codes. To f<strong>in</strong>d the best TBR codes, one can search the<br />
full space of the codes. Such method gives the certa<strong>in</strong>ty that<br />
the found codes are the best. The fault of this method is the<br />
exponentially grow<strong>in</strong>g complexity with the grow<strong>in</strong>g <strong>number</strong><br />
of memory cells of the encoder <strong>and</strong> the <strong>number</strong> of its <strong>in</strong>puts.<br />
In this paper we analysed the tailbit<strong>in</strong>g codes over r<strong>in</strong>gs<br />
encoded by systematic r<strong>in</strong>g convolutional encoders with feedback.<br />
We present the results of the search for the best convolutional<br />
encoders over r<strong>in</strong>g modulo-4 with code rate R = 1/2<br />
Piotr Remle<strong>in</strong> <strong>and</strong> Dawid Szłapka are with Poznan University of Technology,<br />
Faculty of <strong>Electronics</strong> <strong>and</strong> Telecommunications, Piotrowo 3a; 60-965<br />
Poznan (Pol<strong>and</strong>).<br />
used with 16-QAM modulation, with label<strong>in</strong>g as <strong>in</strong> [7]. We<br />
found new TBR codes for 16-QAM modulation with the best<br />
Euclidean distance.<br />
This paper is organised as follows. Section 2 describes<br />
the procedure of encod<strong>in</strong>g TBR by us<strong>in</strong>g the systematic<br />
convolutionalencoderswith feedback.In Section 3 we present<br />
the results of computersearch for the best TBR codes. F<strong>in</strong>ally,<br />
Section 4 gives the conclud<strong>in</strong>g remarks.<br />
II. TAILBITING CODES OVER RINGS – ENCODING<br />
METHOD<br />
In this article we generalize the tailbit<strong>in</strong>g method onto<br />
convolutional codes over r<strong>in</strong>gs of <strong>in</strong>tegers modulo-m [2], [3]<br />
<strong>and</strong> we name the result<strong>in</strong>gcodestailbit<strong>in</strong>g convolutionalcodes<br />
over r<strong>in</strong>gs (TBR). In the proposed method we encode <strong>and</strong><br />
decode a block of N (M-ary) symbols without a known tail,<br />
thus keep<strong>in</strong>g the effective rate of transmission equal to the<br />
code rate. This is done by lett<strong>in</strong>g the encoder start <strong>and</strong> end<br />
<strong>in</strong> the same state, unknown for the decoder. The encod<strong>in</strong>g<br />
procedure to achieve this is not difficult if the structure of the<br />
encoder is feedforward. In this case, the start<strong>in</strong>g state depends<br />
on the m last <strong>in</strong>formation symbols <strong>in</strong> the transmited packet,<br />
wherem is the <strong>number</strong>of memorycells <strong>in</strong> the encoder.Incase<br />
of convolutional encoder with feedback Fig. 1, the start<strong>in</strong>g<br />
state depends on all the <strong>in</strong>formation symbols <strong>in</strong> the packet.<br />
F<strong>in</strong>d<strong>in</strong>g the <strong>in</strong>itial state where<strong>in</strong> the encoder should start<br />
encod<strong>in</strong>g <strong>and</strong> – after N symbol <strong>in</strong>tervals – end encod<strong>in</strong>g <strong>in</strong><br />
the same state is complex. One of the methods for f<strong>in</strong>d<strong>in</strong>g this<br />
<strong>in</strong>itial state was proposed <strong>in</strong> [8] <strong>and</strong> extended for multilevel<br />
codes <strong>in</strong> [9].<br />
InFig.1,weshowtherealizationofthesystematicfeedback<br />
convolutional encoder over r<strong>in</strong>g of <strong>in</strong>tegers modulo-M [4], [6]<br />
with the code rate R = k/n, n = k + 1.<br />
At time t, the <strong>in</strong>formation vector Ut with M-ary elements<br />
u (i)<br />
t belong<strong>in</strong>g to the r<strong>in</strong>g ZM = 0, 1, 2, ..., M − 1, (ℜ = ZM)<br />
<strong>in</strong>puts the encoder.<br />
Ut = (u (1)<br />
t , u (2)<br />
t , ..., u (k)<br />
t ) (1)<br />
The convolutional encoder produces a coded sequence of<br />
symbols which belong to the same r<strong>in</strong>g ZM<br />
Vt = (v (1)<br />
t , v (2)<br />
t , ..., v (n)<br />
t ) (2)<br />
where n = k + 1.<br />
The coefficients <strong>in</strong> the encoder Fig. 1 are taken from the<br />
set 0, ..., M − 1. The memory cells are capable of stor<strong>in</strong>g r<strong>in</strong>g<br />
elements. Multipliers <strong>and</strong> adders perform multiplication <strong>and</strong><br />
addition, respectively, <strong>in</strong> the r<strong>in</strong>g of <strong>in</strong>tegers modulo-M.<br />
The encod<strong>in</strong>g process can be described as mapp<strong>in</strong>g of the<br />
<strong>in</strong>formation vector (1) <strong>in</strong>to the encoded vector (2)
56 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 1. Systematic feedback convolutional encoder over r<strong>in</strong>g of <strong>in</strong>tegers modulo-M.<br />
Vt = UtG (3)<br />
where G denotes the generator matrix of the encoder [8].<br />
The state of the encoder at time t is determ<strong>in</strong>ed by the<br />
content of memory elements<br />
Xt = (x (1)<br />
t , x (2)<br />
t , ..., x (m)<br />
t ) T , (4)<br />
where m is the <strong>number</strong> of encoder memory elements.<br />
In case of packet transmission without tail, where the<br />
convolutional encoders with feedback are utilized, we have<br />
to calculate the <strong>in</strong>itial state X0 that must be the same as the<br />
f<strong>in</strong>d state XN of the encoder after N cycles. This is not quite<br />
easy. To f<strong>in</strong>d this start<strong>in</strong>g state, we used the method proposed<br />
<strong>in</strong> [8]. The correct start<strong>in</strong>g state can by calculated us<strong>in</strong>g the<br />
state space representation. The state of the encoder <strong>in</strong> time<br />
t + 1 can be described as:<br />
Xt+1 = AXt + BU T t<br />
, (5)<br />
where A is the (m × m) state matrix which def<strong>in</strong>es connections<br />
between memory elements, B is the (m × k) control<br />
matrix which def<strong>in</strong>es connections between encoder <strong>in</strong>puts <strong>and</strong><br />
memory elements.<br />
The vector Vt at the encoder output <strong>in</strong> time t can be<br />
described as <strong>in</strong> [8]:<br />
V T<br />
t = CXt + DU T t , (6)<br />
where: C is the (n × m) observation matrix which def<strong>in</strong>es<br />
connections between encoder outputs <strong>and</strong> memory elements,<br />
D is the (n × k) transition matrix which def<strong>in</strong>es connections<br />
between encoder entries <strong>and</strong> outputs.<br />
In the paper [8] it was also shown that the state (Xt) <strong>in</strong><br />
time t, of the systematic convolutional encoder with feedback<br />
can be described as the superposition of two vectors X [zi]<br />
t <strong>and</strong><br />
which def<strong>in</strong>e the end<strong>in</strong>g state of the encoder<br />
X [zs]<br />
t<br />
where X [zi]<br />
t<br />
Xt = X [zi]<br />
t<br />
+ X[zs] t<br />
is the vector which def<strong>in</strong>es the encoder state<br />
achieved after t cycles if the encod<strong>in</strong>g process started <strong>in</strong> state<br />
(7)<br />
X0 <strong>and</strong> all <strong>in</strong>puts symbols are zero, X [zs]<br />
t<br />
is the vector which<br />
def<strong>in</strong>estheencoderstateachievedafter tcyclesiftheencod<strong>in</strong>g<br />
stared <strong>in</strong> the all zero state (X0 = 0) <strong>and</strong> the <strong>in</strong>formation<br />
symbol sequence is encoded.<br />
From the equations (5) <strong>and</strong> (7) we can write that:<br />
Xt = X [zi]<br />
t<br />
+ X [zs]<br />
t<br />
�<br />
= A t t−1<br />
X0 + A (t−1)−τ BU T τ . (8)<br />
τ =0<br />
If we assume that the state <strong>in</strong> time t = N is equal to the <strong>in</strong>itial<br />
state X0, we obta<strong>in</strong> from (8):<br />
(Im − A N )X0 = X [zs]<br />
N , (9)<br />
Thisequationcanbewrittenforconvolutionalencodersover<br />
r<strong>in</strong>g ℜ = ZM as:<br />
(Im + A N )X0 = X [zs]<br />
N , (10)<br />
where Im is the (m × m) identity matrix. As it is seen<br />
from (10), we can calculate the correct <strong>in</strong>itial state X0 of the<br />
encoder if the matrix (Im + AN ) is <strong>in</strong>vertible.<br />
The matrix A from equation (10) for the systematic convolutional<br />
encoder with feedback is described as [8], [9]:<br />
⎡<br />
⎢<br />
A = ⎢<br />
⎣<br />
0 · · · 0<br />
1<br />
. ..<br />
1<br />
�<br />
�<br />
�<br />
�<br />
�<br />
�<br />
�<br />
�<br />
�<br />
fm<br />
fm−1<br />
.<br />
.<br />
f1<br />
⎤<br />
⎥<br />
⎦<br />
(11)<br />
Us<strong>in</strong>g the mathematical relations (9) <strong>and</strong> (10), obta<strong>in</strong>ed above<br />
we can describe the encod<strong>in</strong>g process for TBR codes as<br />
follows: at first, we have to calculate the vector X [zs]<br />
N for a<br />
given <strong>in</strong>formation data packet. Accord<strong>in</strong>gly, the encoder starts<br />
<strong>in</strong> the all zero state. All the N · k <strong>in</strong>formation symbols are<br />
encoded but the output symbols are ignored. After N cycles<br />
the encoder will be <strong>in</strong> the state X [zs]<br />
N . Then, form (10) we can<br />
calculate the correct <strong>in</strong>itial state X0, the encoder can start the<br />
proper encod<strong>in</strong>g process <strong>and</strong> a valid codeword results. After<br />
N cycles the encoder ends its work, reaches the state which<br />
is the same as its start<strong>in</strong>g state.
REMLEIN AND SZŁAPKA: NEW TAILBITING CONVOLUTIONAL CODES OVER RINGS 57<br />
�<br />
Fig. 2. Encoder of the convolutional code G(D) = 1<br />
from the example.<br />
3+2D+D 2<br />
1+3D+3D 2<br />
Fig. 3. Tree diagram when the zero state response is obta<strong>in</strong>ed X [zs]<br />
4 .<br />
Follow<strong>in</strong>g this description, we show an example of TBR<br />
encod<strong>in</strong>g procedure with feedback systematic convolutional<br />
encoder over r<strong>in</strong>g Z4.<br />
1) Example: A packet of four symbols is encoded. The<br />
symbols belong to the r<strong>in</strong>g Z4. The encoder is a systematic<br />
convolutional encoder over r<strong>in</strong>g Z4 with feedback, with code<br />
rate R = 1/2 <strong>and</strong> two memory elements m = 2. In Fig. 1 we<br />
show the structure of this encoder. We encode the <strong>in</strong>formation<br />
block U = (U0, U1, � U2,U3) � = (1, 0, 3, 3). The state matrix<br />
0 3<br />
is given as A = . Therefore, N = 4, k = 1, <strong>and</strong><br />
1 3 � � � �<br />
4<br />
0 3<br />
from equation (9) we can calculate I2 −<br />
X0 =<br />
1 3<br />
X [zs]<br />
� �<br />
2 1<br />
4 . From this formula we obta<strong>in</strong>: X0 = X<br />
3 3<br />
[zs]<br />
4 .<br />
Therefore, we have to calculate the state X [zs]<br />
4 .<br />
From Fig. 2 we can see that this state is equal to (3, 1) T<br />
<strong>and</strong> the correct state from which � we�must � �start<br />
�the encod<strong>in</strong>g �<br />
2 1 3 3<br />
process is equal to X0 =<br />
= . From<br />
3 3 1 0<br />
Fig. 3 we can see that, if we start to encode the sequence U<br />
from state (3, 0) T , then after N = 4 cycles we reach the same<br />
state <strong>and</strong> obta<strong>in</strong> valid codeword V = (13, 02, 31, 30).<br />
III. SEARCH RESULTS<br />
In this section we present the results of computer search for<br />
the best tailbit<strong>in</strong>g codes over r<strong>in</strong>gs modulo-M for transmission<br />
over AWGN channel. As the quality criterion we take the<br />
m<strong>in</strong>imum Euclidean distance de_m<strong>in</strong>. We compute the m<strong>in</strong>imum<br />
Euclidean distance as the m<strong>in</strong>imum distance over all<br />
pairs of dist<strong>in</strong>ct codewords [10]. Each coded sequence must<br />
be comparedto all the other coded sequences. The codes were<br />
generated by the feedback systematic convolutional encoder<br />
�<br />
Fig. 4. Tree diagram for proper encod<strong>in</strong>g process for tailbit<strong>in</strong>g codes over<br />
r<strong>in</strong>g Z4.<br />
overr<strong>in</strong>g.An exhaustivesearchwas used to f<strong>in</strong>d TBR codes<strong>in</strong><br />
Fig. 4. The object of search <strong>in</strong> this article were tailbit<strong>in</strong>g codes<br />
over r<strong>in</strong>g Z4, generated by concatenation of the systematic<br />
encoders with feedback with code rate R = 1/2 <strong>and</strong> 16-<br />
QAM modulator.Thefoundencodershave m memorycells, S<br />
states <strong>and</strong> k <strong>in</strong>puts. N denotes the length of the <strong>in</strong>put symbol<br />
sequence of k <strong>in</strong>formation bits per symbol. For codes over<br />
r<strong>in</strong>g, feedback coefficients f0 ∼ fm <strong>and</strong> the coefficients <strong>in</strong> the<br />
systematic branches g k 0 ∼ gk m<br />
are written as a sequence of<br />
decimal <strong>number</strong>s.<br />
The coefficients equal to zero at the beg<strong>in</strong>n<strong>in</strong>g of the<br />
sequence are skipped <strong>in</strong> the description. All TBR codes over<br />
r<strong>in</strong>g found for 16-QAM are presented <strong>in</strong> Table I. We found<br />
the best TBR codes for encoders with 16, 64 <strong>and</strong> 256 states.<br />
All of these TBR codes are the new codes that have not been<br />
published yet.<br />
IV. CONCLUSION<br />
In this paper we generalized the tailbit<strong>in</strong>g techniques onto<br />
the tailbit<strong>in</strong>g codes over r<strong>in</strong>gs of <strong>in</strong>tegers modulo-M. We<br />
described how the systematic r<strong>in</strong>g convolutional encoder with<br />
feedback can have the same start<strong>in</strong>g <strong>and</strong> end<strong>in</strong>g state. We<br />
presented the search results of the best tailbit<strong>in</strong>g codes over<br />
r<strong>in</strong>g Z4 for the transmission over AWGN channel. As the<br />
optimization criterion of the we took the Euclidean distance.<br />
A table of the best new tailbit<strong>in</strong>g convolutional codes over<br />
r<strong>in</strong>g Z4 with rate R = 1/2 for 16-QAM modulation was<br />
obta<strong>in</strong>ed by computer search. All TBR codes shown <strong>in</strong> Fig. 4<br />
have not been presented <strong>in</strong> the literature known to the authors.<br />
REFERENCES<br />
[1] A. Dholakia, Introduction to Convolutional Codes with Applications.<br />
Kluwer Academic Publishers, 1994.<br />
[2] H. Ma <strong>and</strong> J. Wolf, “On tailbit<strong>in</strong>g convolutional codes,” IEEE Trans.<br />
Commun., vol. 34, pp. 104–111, Feb. 1986.<br />
[3] S. Crozier, A. Hunt, K. Gracie, <strong>and</strong> J. Lodge, “Performance <strong>and</strong><br />
complexity comparison of block turbo-codes, hyper-codes <strong>and</strong> tail-bit<strong>in</strong>g<br />
convolutional codes,” <strong>in</strong> Proceed<strong>in</strong>gs of 19-th, Biennial Symposium on<br />
Communications, K<strong>in</strong>gston Ontario, Canada, May 1998, pp. 84–88.<br />
[4] J. L. Massey <strong>and</strong> T. Mittelholzer, “Convolutional codes over r<strong>in</strong>gs,”<br />
<strong>in</strong> Proceed<strong>in</strong>gs of 4th Jo<strong>in</strong>t Swedish-USSR Int. Workshop Information<br />
Theory, 1989, pp. 14–18.<br />
[5] P. Ståhl, J. Anderson, <strong>and</strong> R. Johannesson, “A note on tailbit<strong>in</strong>g codes<br />
<strong>and</strong> their feedback encoders,” IEEE Trans. Inf. Theory, vol. 48, pp. 529–<br />
534, Feb. 2002.
58 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
TABLE I<br />
TAILBITING CODES OVER RING Z4 WITH CODE RATE R=1/2 FOR 16-QAM MODULATION(NATURAL LABELING AS IN [7]).<br />
S 16 TBR 64 TBR 256 TBR<br />
N f, g de_m<strong>in</strong> f, g de_m<strong>in</strong> f, g de_m<strong>in</strong><br />
4 130,100 12,000 1300,1000 12,000 11100,10000 12,000<br />
5 113,210 14,128 1130,2100 14,128 10132,10000 14,128<br />
6 102,111 14,141 1121,1100 14,828 12330,11000 14,828<br />
7 111,123 16,944 1312,1313 16,970 11331,20110 16,970<br />
8 111,221 16,970 1121,120 18,129 - -<br />
[6] I. Bocharova, R. Johannesson, B. Kudryashov, <strong>and</strong> P. Ståhl, “Tail-bit<strong>in</strong>g<br />
codes: Bounds <strong>and</strong> search results,” IEEE Trans. Inf. Theory, vol. 48, pp.<br />
597–610, Apr. 2000.<br />
[7] R. Carrasco <strong>and</strong> P. Farrell, “R<strong>in</strong>g-TCM for fixed <strong>and</strong> fad<strong>in</strong>g channels:<br />
l<strong>and</strong>-mobile satellite fad<strong>in</strong>g channels with QAM,” IEE Proceed<strong>in</strong>gs-<br />
Communications, vol. 143, no. 5, pp. 281–288, Oct. 1996.<br />
[8] C. B. Weiß, “Code construction <strong>and</strong> decod<strong>in</strong>g of parallel concatenated<br />
tail-bit<strong>in</strong>g codes,” IEEE Trans. Inf. Theory, vol. 47, pp. 366–386, Jan.<br />
2001.<br />
[9] P.Remle<strong>in</strong>, “Theencoders with the feedback for thepacked transmission<br />
without tail symbols,” <strong>in</strong> VIII-th Poznan Workshop on Telecommunication,<br />
PWT ‘03, Poznan, Dec. 2003, pp. 165–169, (<strong>in</strong> polish).<br />
[10] J. Anderson, T. Aul<strong>in</strong>, <strong>and</strong> C. Sundberg, Digital PhaseModulation. NY<br />
Plenum Press, 1986.<br />
Piotr Remle<strong>in</strong> received the M. Sc. <strong>and</strong> Ph. D. degrees <strong>in</strong> telecommunications<br />
from the Poznan University of Technology, Pol<strong>and</strong>, <strong>in</strong> 1991 <strong>and</strong> 2002,<br />
respectively. S<strong>in</strong>ce 1992 he has been work<strong>in</strong>g at the Faculty of <strong>Electronics</strong> <strong>and</strong><br />
Telecommunications, Poznan University of Technology, where he currently is<br />
an Assistant Professor.<br />
His scientific <strong>in</strong>terests cover wireless networks, communication theory,<br />
error control cod<strong>in</strong>g, cryptography, digital modulation, trellis coded modulation,<br />
cont<strong>in</strong>uous phase modulation, mobile communications, digital circuits<br />
design. He is author <strong>and</strong> co-author of over 50 publications <strong>and</strong> unpublished<br />
reports. He is a member of IEEE.<br />
Dawid Szłapka received the M. Sc. degree from the Faculty of <strong>Electronics</strong><br />
<strong>and</strong> Telecommunications, Poznan University of Technology, <strong>in</strong> 2005.
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 59<br />
Model<strong>in</strong>g Step Index Fiber to Soliton Propagation<br />
Abstract—Step <strong>in</strong>dex fiber model<strong>in</strong>g process is carried out<br />
through numerical solv<strong>in</strong>g of eigenvalue equation to calculate<br />
propagation constant for fundamental mod. Input data <strong>in</strong> the<br />
process is only <strong>in</strong>dex of refraction calculated from Sellmeier<br />
dispersive formula for appropriate mol percentage dop<strong>in</strong>g of<br />
germanium dioxide <strong>in</strong> silica glass fiber. Output data <strong>in</strong> the<br />
model<strong>in</strong>g process is optimal value of the normalized frequency,<br />
which guarantees that s<strong>in</strong>gle mode operation region is equal to<br />
brightsolitonpropagation region.F<strong>in</strong>alverificationof theprocess<br />
is soliton generation up to sixth-order <strong>in</strong>side such modeled fiber.<br />
In this end nonl<strong>in</strong>ear Schöd<strong>in</strong>ger equation is solved numerically<br />
for <strong>in</strong>itial condition of hyperbolic secant form. Maximization of<br />
s<strong>in</strong>gle mode operation <strong>and</strong> bright soliton propagation region is<br />
essential <strong>in</strong> wavelength division multiplex<strong>in</strong>g technique.<br />
IndexTerms—eigenvalue equation,nonl<strong>in</strong>earSchöd<strong>in</strong>gerequation,<br />
solitons<br />
Tomasz Kaczmarek<br />
I. INTRODUCTION<br />
THE word soliton refers to special k<strong>in</strong>ds of wave packets<br />
that can propagate undistorted over long distances. In the<br />
context of optical fibers solitons have found practical applications<br />
<strong>in</strong> the field of fiber-optic communications. Solitons<br />
results from a balance between group-velocity dispersion <strong>and</strong><br />
self-phase modulation, both of which can be calculated <strong>in</strong><br />
effect of step <strong>in</strong>dex fiber model<strong>in</strong>g process.<br />
Propagation of soliton <strong>in</strong> s<strong>in</strong>gle-mode optical fiber is described<br />
by the nonl<strong>in</strong>ear Schröd<strong>in</strong>ger equation [1]–[4]<br />
j ∂A<br />
∂z<br />
− β2<br />
2<br />
∂ 2 A<br />
∂T 2 + γ |A|2 A = 0, (1)<br />
where A is the slowly vary<strong>in</strong>g envelope of the pulse, γ<br />
is nonl<strong>in</strong>ear parameter of the fiber, β2 is group velocity<br />
dispersion, z <strong>and</strong> T are spatial <strong>and</strong> time variable, respectively.<br />
Group velocity dispersion expressed <strong>in</strong> ps 2 /km is def<strong>in</strong>ed as<br />
the second derivative of mode propagation constant β with<br />
respect to frequency ω i.e. β2 = d 2 β/dω 2 , <strong>and</strong> is related to<br />
dispersion parameter D expressed <strong>in</strong> ps/(km · nm) through<br />
the relation D = −2πcβ2/λ 2 where c is the speed of light <strong>in</strong><br />
vacuum. Nonl<strong>in</strong>ear parameter is def<strong>in</strong>ed as follows [1], [4]<br />
γ = nNLk<br />
, (2)<br />
Aeff<br />
where nNL is nonl<strong>in</strong>ear refractive <strong>in</strong>dex, Aeff is known as<br />
effective core area. For pulses as short as 1 ps <strong>and</strong> <strong>in</strong> case of<br />
s<strong>in</strong>gle mode fiber, which core is made of silica glass doped<br />
by germanium dioxide, value of nNL is approximately equal<br />
to nNL = 2.2 · 10 −20 m 2 /W [1]. Effective core area is related<br />
to the transverse component of electric field vector E0 <strong>and</strong><br />
T. Kaczmarek is with the Institute of Telecommunications, Photonics <strong>and</strong><br />
Nanomaterials, Kielce University of Technology, Al. 1000-lecia P.P.7, 25-314<br />
Kielce, Pol<strong>and</strong> (e-mail: tkaczmar@tu.kielce.pl).<br />
effective core radius ωeff through the relations [1], [4]<br />
� ∞�<br />
2π |E0 (r)|<br />
0<br />
Aeff =<br />
2 �2 rdr<br />
∞�<br />
|E0 (r)| 4 = πω<br />
rdr<br />
2 eff , (3)<br />
0<br />
where r is radial coord<strong>in</strong>ate <strong>in</strong> the cyl<strong>in</strong>drical coord<strong>in</strong>ate<br />
system. Absolute value of E0 is related to the transverse<br />
components of electric field vector Er <strong>and</strong> Eφ through well<br />
known formula |E0| = (|Er| 2 + |Eφ| 2 ) 1/2 . The transverse<br />
components are determ<strong>in</strong>ed by the use of axial component<br />
of electric Ez <strong>and</strong> magnetic Hz field vectors through the<br />
follow<strong>in</strong>g relations [3], [5], [6]<br />
Er1 = −j<br />
χ 2<br />
Hr1 = −j<br />
χ 2<br />
�<br />
β ∂Ez1<br />
∂r<br />
Eφ1 = −j<br />
χ2 �<br />
β<br />
r<br />
�<br />
β ∂Hz1<br />
Hφ1 = −j<br />
χ 2<br />
∂Ez1<br />
∂φ<br />
+ ωµ0<br />
r<br />
− ωµ0<br />
∂r − ωε0n2 1<br />
r<br />
�<br />
β ∂Hz1<br />
r ∂φ + ωε0n 2 1<br />
�<br />
∂Hz1<br />
, (4)<br />
∂φ<br />
∂Hz1<br />
∂r<br />
∂Ez1<br />
∂φ<br />
∂Ez1<br />
∂r<br />
�<br />
, (5)<br />
�<br />
, (6)<br />
�<br />
, (7)<br />
for the core. In case of cladd<strong>in</strong>gsubscript 1 should be changed<br />
to 2 <strong>and</strong>, moreover, variable χ 2 should be replaced with –<br />
σ 2 . Equations from (4) to (7) are essential for comput<strong>in</strong>g an<br />
average power curried by the core [5], [6]<br />
�<br />
P1 = π<br />
<strong>and</strong> cladd<strong>in</strong>g [5], [6]<br />
0<br />
a<br />
�<br />
P2 = π<br />
�<br />
Er1H ∗ φ1 − Eφ1H ∗ �<br />
r1 rdr, (8)<br />
+∞<br />
�<br />
Er2H ∗ φ2 − Eφ2H ∗ r2<br />
a<br />
� rdr, (9)<br />
where for example H∗ φ1 means complex conjugate to Hφ1.<br />
Averagepowerpropagated<strong>in</strong>sidethe core P1 canbe expressed<br />
as percentage through the relation P1% = [P1/(P1 + P2)] ·<br />
100%. The expressions for Ez <strong>and</strong> Hz are given by [3], [5],<br />
[6]<br />
Ez1 = AEJm (χr) exp [j (mφ + ωt − βz)] , (10)<br />
Hz1 = AHJm (χr) exp [j (mφ + ωt − βz)] , (11)<br />
for the core <strong>and</strong> [3], [5], [6]<br />
Ez2 = BEKm (σr) exp [j (mφ + ωt − βz)] , (12)<br />
Hz2 = BHKm (σr) exp [j (mφ + ωt − βz)] , (13)<br />
for the cladd<strong>in</strong>g of the step <strong>in</strong>dex fiber, where AE, AH, BE<br />
<strong>and</strong> BH are arbitrary constants, Jm(χr) is the Bessel function
60 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
ofthe first k<strong>in</strong>doforder m <strong>and</strong> Km(σr) isthe modifiedBessel<br />
function of the second k<strong>in</strong>d of order m. The constant m must<br />
be an <strong>in</strong>teger s<strong>in</strong>ce the fields must be periodic <strong>in</strong> φ with a<br />
period of 2π. Inside the core factor χ 2 is given by [3], [5], [6]<br />
while outside the core<br />
χ 2 = k 2 n 2 1 − β 2 , (14)<br />
σ 2 = β 2 − k 2 n 2 2. (15)<br />
Time coord<strong>in</strong>ate T from equation (1), which describes pulse<br />
evolution <strong>in</strong>side a s<strong>in</strong>gle-mode fiber, is related to t from<br />
equations (9), (10), (11) <strong>and</strong> (12) <strong>in</strong> the follow<strong>in</strong>g way [1],<br />
[4]<br />
T = t − z/vg = t − β1z, (16)<br />
where vg is the group velocity at which the frame of reference<br />
is mov<strong>in</strong>g with the pulse, β1 is the first derivative of β with<br />
respect to ω <strong>and</strong> isrelatedto groupvelocitydispersionthrough<br />
well known relation β2 = dβ1/dω.<br />
The solution for β frompermissible rangefor guidedmodes<br />
kn2 ≤ β ≤ kn1, (17)<br />
must be determ<strong>in</strong>ed from the boundary conditions, which<br />
require that the tangential components Eφ <strong>and</strong> Ez of electric<br />
field vector � E <strong>in</strong>side <strong>and</strong> outside of the dielectric <strong>in</strong>terface<br />
at r = a must be the same <strong>and</strong> similarly for the tangential<br />
components Hφ <strong>and</strong> Hz of magnetic field vector � H. By<br />
requir<strong>in</strong>g the cont<strong>in</strong>uity of Ez,Hz, Eφ, <strong>and</strong> Hφ at r = a,<br />
one can obta<strong>in</strong> a set of four homogeneous equations satisfied<br />
by AE, AH, BE <strong>and</strong> BH. These equations have a nontrivial<br />
solution only if the determ<strong>in</strong>ant of the coefficient matrix<br />
vanishes. After considerable algebraic details, this condition<br />
leads to the follow<strong>in</strong>g eigenvalue equation for β(EV(β) = 0)<br />
[5], [6]: � J |<br />
m (u)<br />
uJm(u)<br />
+ K|<br />
m (w)<br />
−<br />
wKm(w)<br />
� βm<br />
k<br />
� � J |<br />
m (u)<br />
uJm(u) n21 + n2 K<br />
2<br />
|<br />
m (w)<br />
wKm(w)<br />
�2 �<br />
1<br />
u2 + 1<br />
w2 II. METHOD<br />
�<br />
� 2 = 0. (18)<br />
Step <strong>in</strong>dex fiber model<strong>in</strong>g <strong>in</strong> order to soliton propagation<br />
can be divided <strong>in</strong>to two stages. In the fist stage the optimal<br />
value of the normalized frequency Vopt is calculated. In this<br />
end, eigenvalue equation (18) for step <strong>in</strong>dex fiber is solved<br />
numerically. The optimal value of the normalized frequency<br />
guarantees that the cut off wavelength λC for T E01 mode is<br />
equal to the zero dispersion wavelength λZD, furthermore, if<br />
λC = λZD thenalso ∆λC = ∆λZD, where ∆λC = λopt−λC<br />
<strong>and</strong> similarly ∆λZD = λopt − λZD (λopt = 1.55µm is<br />
optimal operat<strong>in</strong>g wavelength). In this special case, s<strong>in</strong>gle<br />
mode condition λoper > λC is <strong>in</strong> full agreement with bright<br />
soliton propagation condition λoper > λZD, where λoper<br />
is operat<strong>in</strong>g wavelength. If V > Vopt then λC > λZD<br />
which means that ∆λC < ∆λZD <strong>and</strong> simultaneousfulfillment<br />
of s<strong>in</strong>gle mode <strong>and</strong> bright soliton propagation condition is<br />
only possible for λoper > λC. Similarly if V < Vopt, then<br />
λZD > λC (∆λZD < ∆λC) <strong>and</strong> simultaneous fulfillment of<br />
TABLE I<br />
SELLMEIER COEFFICIENTS VALUES FOR APPROPRIATE GERMANIUM<br />
DIOXIDE MOL % DOPING OF SILICA GLASS AND FOR PURE SILICA GLASS<br />
[6], [7]<br />
100m%<br />
SiO2<br />
3.1m%<br />
GeO2<br />
5.8m%<br />
GeO2<br />
7.9m%<br />
GeO2<br />
13.5m%<br />
GeO2<br />
a1 0.69616 0.70285 0.70888 0.71368 0.71104<br />
a2 0.40794 0.41463 0.42068 0.42548 0.45188<br />
a3 0.89749 0.89745 0.89565 0.89642 0.70404<br />
λ1 [µm] 0.06840 0.07277 0.06090 0.06171 0.06427<br />
λ2 [µm] 0.11624 0.11430 0.12545 0.12708 0.12940<br />
λ3 [µm] 9.89616 9.89616 9.89616 9.89616 9.42547<br />
TABLE II<br />
FOUR CASES OF CORE AND CLADDING CHEMICAL COMPOSITION OF STEP<br />
INDEX FIBER<br />
Case Core Cladd<strong>in</strong>g<br />
1 3.1mol% GeO2 & 96.9mol% SiO2 100mol% SiO2<br />
2 5.8mol% GeO2 & 94.2mol% SiO2 100mol% SiO2<br />
3 7.9mol% GeO2 & 92.1mol% SiO2 100mol% SiO2<br />
4 13.5mol% GeO2 & 86.5mol% SiO2 100mol% SiO2<br />
s<strong>in</strong>glemodework<strong>in</strong>gregime<strong>and</strong>pulselike solitonpropagation<br />
condition is possible if <strong>and</strong> only if λoper > λZD (TABLE III).<br />
If one starts from value 2.4 for normalized frequency <strong>and</strong><br />
tries to calculate the optmal value of core radius of the fiber<br />
which cladd<strong>in</strong>g is made of pure SiO2 <strong>and</strong> its core is doped by<br />
different mol % GeO2, one has to use the follow<strong>in</strong>g relation<br />
[3], [5], [6]<br />
� �<br />
a = V/ k(λ) n2 1 (λ) − n22 (λ)<br />
�<br />
, (19)<br />
where V = 2.4 is the normalized frequency, k = 2π/λ is<br />
the wave <strong>number</strong>, n1 <strong>and</strong> n2 are refractive <strong>in</strong>dices of the<br />
core <strong>and</strong> cladd<strong>in</strong>g, respectively. The values of both <strong>in</strong>dices<br />
are determ<strong>in</strong>ed through Sellmeier dispersive formula [3], [6],<br />
[7]<br />
�<br />
�<br />
� 3�<br />
n = � aiλ<br />
1 +<br />
2<br />
, (20)<br />
λ<br />
i=1<br />
2 − λ2 i<br />
where ai is the oscillator strength, λi is the oscillator resonance<br />
wavelength. Both coefficients values for appropriate<br />
GeO2 mol % dop<strong>in</strong>g of SiO2 are presented <strong>in</strong> TABLE I.<br />
By the assumption that the cladd<strong>in</strong>g is made of pure silica<br />
glass there are four cases <strong>in</strong> the model<strong>in</strong>g of step <strong>in</strong>dex fiber<br />
for four types of germanium dioxide dop<strong>in</strong>g, which can be<br />
<strong>number</strong>ed <strong>in</strong> <strong>in</strong>creas<strong>in</strong>g GeO2 dop<strong>in</strong>g order (TABLE II).<br />
After suitable rearrang<strong>in</strong>g of equation (19) to the follow<strong>in</strong>g<br />
form λ = 2πa � n 2 1 (λ) − n2 2 (λ)� 1/2 /V, it is possible to calculate<br />
cut off wavelength λC for the T E01 mode. Obta<strong>in</strong><strong>in</strong>g of<br />
zero dispersion wavelength λZD can be done <strong>in</strong> two ways. By<br />
the use of group velocity dispersion β2 = f(λ) or dispersion<br />
parameter D = f(λ) characteristic. In each case the result<br />
should be the same.<br />
In the second stage, nonl<strong>in</strong>ear Schröd<strong>in</strong>ger equation is<br />
solved numerically by the use of split-step Fourier (SSF)<br />
method, for each case of the optimized step <strong>in</strong>dex fiber
KACZMAREK: MODELING STEP INDEX FIBER TO SOLITON PROPAGATION 61<br />
TABLE III<br />
INTERMIDIET AND FINAL RESULTS OF THE FIRST STAGE MODELING<br />
PROCESS<br />
Normalized<br />
Frequency<br />
Case 1<br />
λ[µm]<br />
V = 2.4 λC=1.547<br />
λZD=1.287<br />
V=2.3 λC=1.483<br />
λZD=1.291<br />
Case 2<br />
λ[µm]<br />
λC=1.547<br />
λZD=1.295<br />
λC=1.483<br />
λZD=1.303<br />
Case 3<br />
λ[µm]<br />
λC=1.547<br />
λZD=1.310<br />
λC=1.483<br />
λZD=1.323<br />
Case 4<br />
λ[µm]<br />
λC=1.547<br />
λZD=1.380<br />
λC=1.480<br />
λZD=1.408<br />
Vopt=2.231 λC=λZD=<br />
=1.434<br />
V=2.2 λC=1.420<br />
λZD=1.296<br />
λC=1.419<br />
λZD=1.314<br />
λC=1.419<br />
λZD=1.340<br />
Vopt=2.107 λC=λZD=<br />
=1.360<br />
V=2.1 λC=1.356<br />
λZD=1.302<br />
λC=1.354<br />
λZD=1.327<br />
Vopt=2.065 λC=λZD=<br />
=1.333<br />
Vopt=2.024 λC=λZD=<br />
=1.308<br />
V=2.0 λC=1.292<br />
λZD=1.310<br />
λC=1.291<br />
λZD=1.345<br />
λC=1.355<br />
λZD=1.362<br />
λC=1.414<br />
λZD=1.448<br />
<strong>in</strong> the first stage, for soliton pulses up to the sixth order.<br />
Split-step Fourier is a pseudospectral method, which has<br />
been extensively used to solve the pulse-propagation problem<br />
<strong>in</strong> nonl<strong>in</strong>ear dispersive media. In this method approximate<br />
solution is obta<strong>in</strong>ed by the assumption that <strong>in</strong> propagat<strong>in</strong>g<br />
the optical field over a small distance h, the dispersive <strong>and</strong><br />
nonl<strong>in</strong>ear effects act <strong>in</strong>dependently. It can be understood if<br />
Eq. (1) is rewritten <strong>in</strong> the follow<strong>in</strong>g form [1], [4]<br />
∂A<br />
= (D + N) A, (21)<br />
∂z<br />
where D = −(jβ2/2)(∂ 2 /∂T 2 ) is a differential operator that<br />
accounts for dispersion <strong>in</strong> a l<strong>in</strong>ear medium <strong>and</strong> N = jγ|A| 2 is<br />
a nonl<strong>in</strong>ear operator that governs the effect of fiber nonl<strong>in</strong>earities<br />
on pulse propagation. So <strong>in</strong> case of SSF method optical<br />
field propagation from zto z + h is carried out <strong>in</strong> two steps. In<br />
the first step D = 0 <strong>in</strong> Eq. (21) <strong>and</strong> nonl<strong>in</strong>earity acts alone, <strong>in</strong><br />
the second step N = 0 <strong>in</strong> Eq. (21) <strong>and</strong> dispersion acts alone.<br />
Mathematically it can be prescribed as follows [1], [4]<br />
A(z+h, T )≈F −1 {exp [hD(jω)] F [exp(hN)A(z, T )]} , (22)<br />
where F denotes the Fourier-transform operation, D(jω) =<br />
jω 2 β2/2 is obta<strong>in</strong>ed from a differential operator by replac<strong>in</strong>g<br />
∂/∂T with jω, where ω is the frequency <strong>in</strong> the Fourier<br />
doma<strong>in</strong>.<br />
III. RESULTS<br />
Search<strong>in</strong>g the optimal value of the normalized frequency<br />
Vopt was started from V = 2.4 <strong>and</strong> closed for V = 2.0 (V ∈<br />
{2.4, 2.3, 2.2, 2.1, 2.0}). Intermediate (λC �= λZD) <strong>and</strong> f<strong>in</strong>al<br />
(λC = λZD) results are presented <strong>in</strong> TABLE III.<br />
Summarized results for the first stage of step <strong>in</strong>dex fiber<br />
model<strong>in</strong>g process for λopt = 1.55 µm <strong>and</strong> for HE11 mode<br />
are presented <strong>in</strong> TABLE IV.<br />
In order to solve Eq. (1) numerically for <strong>in</strong>itial condition<br />
of the form [1]–[4] A(z = 0, T ) = A0 sech(T/T0), it is<br />
TABLE IV<br />
SUMMARIZED RESULTS FOR THE FIRST STAGE OF STEP INDEX FIBER<br />
MODELING PROCESS<br />
Parameter Case 1 Case 2 Case 3 Case 4<br />
Vopt 2.024 2.065 2.107 2.231<br />
a [µm] 4.293 3.178 2.762 2.200<br />
P 1% [%] 60.96 62.74 64.49 67.26<br />
ωeff [µm] 5.228 3.815 3.270 2.510<br />
Aeff [µm 2 ] 86.86 45.73 33.60 19.79<br />
γ [1/W km] 1.039 1.950 2.654 4.507<br />
λC = λZD [µm] 1.308 1.333 1.360 1.434<br />
∆λC = ∆λZD = [nm] 242.3 217.2 189.9 115.8<br />
D [ps/km nm] 16.86 13.34 10.61 5.693<br />
β2 [ps 2 /km] -21.51 -17.02 -13.53 -7.263<br />
TABLE V<br />
FOUR PARAMETERS VALUE CALCULATED FOR FOUR CASES OF STEP<br />
INDEX FIBER FOR FUNDAMENTAL SOLITON INITIAL WIDTH T0 = 1 ps.<br />
Parameter Case 1 Case 2 Case 3 Case 4<br />
P0 [W] 20.71 8.725 5.098 1.612<br />
A0 4.551 2.954 2.258 1.269<br />
LD [m] 46.49 58.77 73.89 137.7<br />
z0 [m] 73.03 92.32 116.1 216.3<br />
necessary to calculate peak amplitude value A0 (which is<br />
proportional to peak power P0) for appropriate soliton order<br />
N from the follow<strong>in</strong>g relation [1]–[4] N 2 = γP0LD, where<br />
LD = T 2 0 /|β2| is the dispersion length <strong>and</strong> T0 is the measure<br />
of the impulse width. For fundamental (N = 1) <strong>and</strong> higher<br />
order solitons (N = 2, 3, 4, . . .), it is possible to calculate<br />
solitonperiod z0 fromthedispersionlengthvalue LD obta<strong>in</strong>ed<br />
earlier because [1]–[4] z0 = (π/2)LD.<br />
Only fundamental soliton (N = 1) can be used as <strong>in</strong>formation<br />
bits <strong>in</strong> soliton-based communication systems <strong>and</strong> only<br />
when <strong>in</strong>dividual solitons are well isolated (RZ format). The<br />
last requirement can be used to relate the soliton width T0 to<br />
the bit rate B as follows [2]–[4] B = 1/TB = 1/(2q0TB),<br />
where TB is the duration of the bit slot <strong>and</strong> 2q0 = TB/T0<br />
is the separation between neighbor<strong>in</strong>g solitons <strong>in</strong> normalized<br />
units. For T0 = 1 ps <strong>and</strong> q0 = 5, bit rate B <strong>in</strong> soliton based<br />
communication system is equal to B = 100 Gbit/s. Table V<br />
showscalculationresultsforfournecessaryparametersneeded<br />
to solve numerically Eq. (1), for <strong>in</strong>itial width T0 = 1 ps <strong>and</strong><br />
for fundamental soliton (N = 1).<br />
IV. DISCUSSION<br />
Fig. 1 shows lack of the shape variation of the pulse as a<br />
function of the propagationdistance (one soliton period which<br />
is equal to z0 = 216.3 m) for the fundamental soliton <strong>in</strong> case<br />
of <strong>number</strong> 4. It means that first-order soliton (N = 1) can be<br />
generated for peak amplitude value A0 = 1.269 (column 5 of<br />
TABLE V).<br />
V. CONCLUSIONS<br />
On the basis of the performedcalculationsit has been found<br />
thatifmol%dop<strong>in</strong>gofgermaniumdioxideis<strong>in</strong>creas<strong>in</strong>g<strong>in</strong>side
62 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 1. Evolution of the first-order soliton (N = 1) over one soliton period.<br />
the core, then the optimal value of the normalized frequency<br />
Vopt ofthemodeledstep<strong>in</strong>dexfiberisalso<strong>in</strong>creas<strong>in</strong>g.Increase<br />
of Vopt implies <strong>in</strong>crease of zero dispersion wavelength λZD<br />
<strong>and</strong> cut off wavelength λC, which are equal <strong>in</strong> case of<br />
normalized frequency optimization. Additionally, growth of<br />
Vopt value is responsible for rise of the average power curried<br />
by the core P1. There is only one more parameter which<br />
value is <strong>in</strong>creas<strong>in</strong>g when mol % dop<strong>in</strong>g of germaniumdioxide<br />
is <strong>in</strong>creas<strong>in</strong>g. It is nonl<strong>in</strong>ear parameter γ, which <strong>in</strong> turn is<br />
responsible for decreas<strong>in</strong>g the peak power needed to generate<br />
fundamental soliton <strong>in</strong> each case of step <strong>in</strong>dex fiber model<strong>in</strong>g<br />
process. Furthermore, decrease of dispersion parameter D<br />
<strong>and</strong> absolute value of group velocity dispersion parameter<br />
β2 is responsible for <strong>in</strong>crease of dispersion length LD <strong>and</strong><br />
value of the soliton period z0. Fundamental disadvantage<br />
of <strong>in</strong>creas<strong>in</strong>g λZD <strong>and</strong> λC is decreas<strong>in</strong>g of bright soliton<br />
generation region ∆λZD <strong>and</strong> s<strong>in</strong>gle mode operation region<br />
∆λC, which are essential <strong>in</strong> wavelength division multiplex<strong>in</strong>g<br />
technique application.<br />
REFERENCES<br />
[1] G. P.Agrawal, Nonl<strong>in</strong>ear Fiber Optics, third edition ed. Academic Press,<br />
2001.<br />
[2] ——, Applications of Nonl<strong>in</strong>ear Fiber Optics. Academic Press, 2001.<br />
[3] ——, Fiber-Optic Communication Systems. John Wiley & Sons, 2002.<br />
[4] E. Iannone, F. Matera, A. Mecozzi, <strong>and</strong> M. Settembre, Nonl<strong>in</strong>ear Optical<br />
Communication Networks. John Wiley & Sons, 1998.<br />
[5] G. Keiser, Optical Fiber Communications. McGraw-Hill, 1991.<br />
[6] A. Majewski, Teoria i projektowanie ´Swiatłowodów. WNT, Warszawa,<br />
1991, (<strong>in</strong> Polish).<br />
[7] M. J. Adams, An Introduction to Optical Waveguides. John Wiley &<br />
Sons, 1981.<br />
Tomasz Kaczmarek received the M.Sc. degree <strong>in</strong> electrical eng<strong>in</strong>eer<strong>in</strong>g from<br />
Kielce University of Technology <strong>in</strong> 1994 <strong>and</strong> the Ph.D. degree <strong>in</strong> electronic<br />
eng<strong>in</strong>eer<strong>in</strong>g from Warsaw University of Technology <strong>in</strong> 2002. Currently he<br />
is the Head of Laboratory of Optical Fiber Technology of the Institute of<br />
Telecommunication, Photonics <strong>and</strong> Nanomaterials at the Kielce University of<br />
Technology. He authored <strong>and</strong> co-authored over 30 publications. His current<br />
research <strong>in</strong>terests <strong>in</strong>clude fiber optics <strong>and</strong> nonl<strong>in</strong>ear fiber optics.
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 63<br />
Are Carrier Transport Effects Important for Chirp<br />
Model<strong>in</strong>g of Quantum-Well Lasers?<br />
Abstract—The paper <strong>in</strong>vestigates the impact of carrier transport<br />
effects on the chirp model<strong>in</strong>g of quantum-well lasers.<br />
Particularly, the difference between the full model<strong>in</strong>g based on<br />
quantum-well laser rate equations is compared with model<strong>in</strong>g<br />
based on formulas derived for bulk lasers. As it was shown,<br />
the relations between chirp <strong>and</strong> <strong>in</strong>tensity modulation are quite<br />
similar <strong>in</strong> both cases.<br />
Index Terms—laser chirp, laser model<strong>in</strong>g<br />
Przemysław Krehlik<br />
I. INTRODUCTION<br />
THE quantum-, or multi-quantum-well (QW, or MQW)<br />
structure <strong>in</strong>troduced to the semiconductor laser design<br />
implies some new phenomena <strong>in</strong> the device operation, when<br />
compared with the bulk laser design. Among them the<br />
transport of <strong>in</strong>jected carriers across the separate-conf<strong>in</strong>ementheterostructure(SCH)<strong>and</strong>captur<strong>in</strong>gthem<strong>in</strong>totheQWregions<br />
<strong>in</strong>troduce some delay <strong>in</strong> the carriers flow. Consequently, noticeable<br />
variationsof theconcentrationof carriersaccumulated<br />
<strong>in</strong> SCH region occur. Because a large fraction of the optical<br />
mode lies <strong>in</strong> the SCH, this carrier density variations affect the<br />
las<strong>in</strong>g frequency i.e. <strong>in</strong>troduces a new chirp component.<br />
There are plenty of papers <strong>in</strong> which significant differences<br />
<strong>in</strong> chirp characteristics of bulk <strong>and</strong> QW lasers are po<strong>in</strong>ted out<br />
[1]–[4].On the other h<strong>and</strong>, there are some papers <strong>in</strong> which the<br />
QW laser chirp is modeled us<strong>in</strong>g equations derived for bulk<br />
device. In some of them the considerations are verified by<br />
experiments, which seems to proof such chirp treatment [5]–<br />
[7]. The aim of the work presented here<strong>in</strong> is to clarify this<br />
confus<strong>in</strong>g <strong>in</strong>consistency <strong>and</strong> to po<strong>in</strong>t out the area <strong>in</strong> which the<br />
simple chirp model may be used for QW lasers.<br />
II. THEORETICAL BASICS<br />
The basic mathematical model of semiconductorlaser is the<br />
set of rate equations, which describes the dynamics of carrier<br />
<strong>and</strong> photon densities, <strong>and</strong> relate them to the laser frequency<br />
chirp <strong>and</strong> the output optical power.<br />
A. Bulk laser model<strong>in</strong>g<br />
For the bulk laser the rate equations may be written <strong>in</strong> the<br />
follow<strong>in</strong>g form:<br />
dN<br />
dt<br />
I<br />
= −<br />
eVa<br />
N<br />
τe<br />
dS<br />
dt = Γg0(N − NT )<br />
S −<br />
1 + εgS<br />
S<br />
− g0(N − NT )<br />
S (1)<br />
1 + εgS<br />
τP<br />
+ ΓβN<br />
τe<br />
P. Krehlik is with the Institute of <strong>Electronics</strong>, AGH University of Science<br />
<strong>and</strong> Technology, Mickiewicza 30, 30-059 Kraków, Pol<strong>and</strong>; e-mail:<br />
krehlik@agh.edu.pl.<br />
(2)<br />
∆ν = α<br />
4π Γg0(N − NT H) (3)<br />
P = ηVahν0<br />
S (4)<br />
Γτp<br />
where N is the carrier concentration <strong>in</strong> the active region,<br />
S is the photon concentration, I is the <strong>in</strong>jected current, e<br />
is the electron charge, Va is the active region <strong>volume</strong>, τe<br />
is the carrier lifetime, g0 is the differential ga<strong>in</strong>, εg is the<br />
ga<strong>in</strong> compression factor, NT is the carrier concentration for<br />
transparency, NT H is threshold carrier concentration, Γ is<br />
the conf<strong>in</strong>ement factor, τp is the photon lifetime, β is the<br />
spontaneous emission coefficient, ∆ν is the optical frequency<br />
deviation(i.e.the chirp), α isthe l<strong>in</strong>eenhancementfactor, P is<br />
the output power, h is Planc’s constant, <strong>and</strong> ν0 is the nom<strong>in</strong>al<br />
optical frequency.<br />
As may be noticed, the frequency chirp is described by (3),<br />
which shows that the frequency deviation is proportional to<br />
the concentration of carriers <strong>in</strong> the laser active region.<br />
A serious practical drawback of the (3) is that it relates the<br />
chirp to the unobservable carrier concentration, which cannot<br />
be predicted without the precise knowledge about all the rate<br />
equationsparameters. Thus, it is very useful to relate the chirp<br />
to the measurable laser output power. Calculat<strong>in</strong>g the carrier<br />
concentration N from (2) <strong>and</strong> putt<strong>in</strong>g it <strong>in</strong>to (3), the frequency<br />
chirp may be related to the photon concentration. Ignor<strong>in</strong>g<br />
some negligible terms <strong>and</strong> us<strong>in</strong>g (4), we may f<strong>in</strong>ally relate the<br />
chirp to the laser output power:<br />
∆ν(t) = α<br />
4π<br />
�<br />
1 dP (t)<br />
+ κP (t)<br />
P (t) dt<br />
where κ = Γεg/(ηVahν0) is the so called adiabatic chirp<br />
coefficient. The part of the chirp <strong>in</strong>duced by the time derivate<br />
of power is called the dynamic chirp, <strong>and</strong> the part directly<br />
proportional to the power is called the adiabatic one.<br />
In case of small signal laser modulation, the frequency<br />
modulation (FM) efficiency may be determ<strong>in</strong>ed us<strong>in</strong>g (5). In<br />
the frequency doma<strong>in</strong> it takes the form:<br />
� �<br />
δν(ωm) α jωm δP (ωm)<br />
= + κ (6)<br />
δI(ωm) 4π 〈P 〉 δI(ωm)<br />
where δ(·) denotes the small signal component of each quantity,<br />
ωm is the angular frequency of laser modulation, 〈P 〉<br />
isthemeanopticalpower,<strong>and</strong> δP (ωm)/δI(ωm)isthe<strong>in</strong>tensity<br />
modulation (IM) efficiency.<br />
Thus, hav<strong>in</strong>g the knowledge about the laser IM behavior<br />
(some k<strong>in</strong>d of model or measured data) we need only two<br />
parameters (α <strong>and</strong> κ) to accurate chirp characterization. Some<br />
relatively simple measurement methods for determ<strong>in</strong><strong>in</strong>g these<br />
parameters are described <strong>in</strong> many papers [8].<br />
�<br />
(5)
64 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
B. QW laser model<strong>in</strong>g<br />
In the QW lasers the carrier concentrations <strong>in</strong> SCH <strong>and</strong><br />
QW regions should be dist<strong>in</strong>guished, <strong>and</strong> thus two separate<br />
rate equations for the carriers are <strong>in</strong>troduced:<br />
dNw<br />
dt<br />
dNb<br />
dt<br />
I<br />
= −<br />
eVw<br />
Nb<br />
−<br />
τcap<br />
Nb<br />
+<br />
τe<br />
Nw<br />
τesc<br />
Nb<br />
= −<br />
τcap<br />
Nw<br />
−<br />
τesc<br />
Nw<br />
τe<br />
(7)<br />
− g0(Nw − NT )<br />
S (8)<br />
1 + εgS<br />
where Nw is the carrier concentration <strong>in</strong> the quantum wells,<br />
Nb is some equivalent concentrationrelated with the real SCH<br />
carrier concentration Ns by the relation: Nb = NsVs/Vw,<br />
<strong>in</strong> which Vs <strong>and</strong> Vw are the <strong>volume</strong>s of SCH <strong>and</strong> QW,<br />
respectively. The captur<strong>in</strong>g of the carriers from SCH to QW is<br />
characterized by capture time τcap, <strong>and</strong> (much less efficient)<br />
escap<strong>in</strong>g <strong>in</strong> the opposite direction by τesc. The photon density<br />
depends only on the Nw concentration, thus:<br />
dS<br />
dt = Γg0(Nw − NT )<br />
S −<br />
1 + εgS<br />
S<br />
τP<br />
+ ΓβNw<br />
τe<br />
The frequency chirp depends on both QW <strong>and</strong> SCH carrier<br />
densities, because the optical field lies <strong>in</strong> both regions undergo<strong>in</strong>g<br />
carrier concentration variations. Thus, the chirp may be<br />
expressed as follows [1]:<br />
∆ν = α<br />
4π Γg0(Nw − NwT H) + (1 − Γ)gb(Nb − NbT H) (10)<br />
where NwT H <strong>and</strong> NbT H arethresholdcarrierconcentrations<strong>in</strong><br />
QW <strong>and</strong> SCH, respectively, gb is the coefficient characteriz<strong>in</strong>g<br />
the efficiency of <strong>in</strong>fluence of Nb on the laser frequency.<br />
Unfortunately, this time the chirp cannot be easily related<br />
to the <strong>in</strong>tensity modulation, as it was made <strong>in</strong> (5) <strong>and</strong> (6)<br />
for the bulk lasers. Large signal relation, analogous to (5), is<br />
quite complicated, <strong>and</strong> even after many simplifications needs<br />
at least four parameter values to be determ<strong>in</strong>ed <strong>in</strong> some way.<br />
Similarly, the small signal relation analogous to (6) is also<br />
troublesome <strong>and</strong> needs a large set of parameters [1].<br />
Thus, the question of practical importance arises whether<br />
a relatively simple model of the laser IM <strong>and</strong> FM properties,<br />
based on the bulk laser rate equations, may be adopted for behavioral<br />
(i.e. not strictly connected with physical phenomena)<br />
model<strong>in</strong>g of the QW lasers.<br />
In case of IM characteristics, is was shown <strong>in</strong> [7] that the<br />
effects aris<strong>in</strong>g from the carrier accumulation <strong>in</strong> the SCH may<br />
be simply modeled by a first order low-pass filter with time<br />
constant equal to τcap, preced<strong>in</strong>g the bulk model of the <strong>in</strong>ner<br />
QW structure. It may be also shown that for QW lasers with<br />
any low capture time the difference <strong>in</strong> the IM properties of<br />
models described by Eqs. (1), (2) <strong>and</strong> (7) ... (9) practically<br />
vanishes.<br />
III. SMALL-SIGNAL CONSIDERATIONS<br />
First, the small-signal chirp characteristics aris<strong>in</strong>g from the<br />
QW laser model based on the rate equations (7) ... (10)<br />
will be analyzed. Us<strong>in</strong>g this model <strong>and</strong> start<strong>in</strong>g from two<br />
experimentally verified sets of its parameters, taken from [9],<br />
the laser FM efficiency versus modulation frequency was<br />
obta<strong>in</strong>ed. In some <strong>in</strong>itial <strong>in</strong>vestigations it was observed that<br />
(9)<br />
Fig. 1. . IM efficiency |δP/δI| versus modulation frequency <strong>and</strong> capture<br />
time.<br />
Fig. 2. FM efficiency |δν/δI| versus modulation frequency <strong>and</strong> capture<br />
time.<br />
under the reasonable assumption that τcap
PRZEMYSŁAW KREHLIK: ARE CARRIER TRANSPORT EFFECTS IMPORTANT FOR CHIRP MODELING OF QUANTUM-WELL LASERS? 65<br />
Fig. 3. Comparison of FM efficiency obta<strong>in</strong>ed from full QW model <strong>and</strong><br />
from (6).<br />
κ was trimmed to obta<strong>in</strong> a desired value of the low frequency<br />
chirp for each value of the taken capture time. It should be<br />
also po<strong>in</strong>ted out that the IM response δP (ωm)/δI(ωm) was<br />
modified each time by tak<strong>in</strong>g the actual one obta<strong>in</strong>ed from<br />
full QW rate equationsmodel.As may be noticed,averygood<br />
agreement between the chirp obta<strong>in</strong>ed from the full model <strong>and</strong><br />
from 6) was obta<strong>in</strong>ed, even for frequencies far above the laser<br />
relaxation frequency.<br />
Conclud<strong>in</strong>g, the QW laser small-signal chirp may be accuratelydeterm<strong>in</strong>edbythesimpleformulagiven<strong>in</strong>(6).However,<br />
the accurate IM response (known from any k<strong>in</strong>d of model or<br />
measured data) is crucial for good accuracy.<br />
IV. LARGE-SIGNAL CONSIDERATIONS<br />
The small-signal FM response is a basic laser property <strong>in</strong><br />
any transmission system based on frequency/phase modulation,<br />
as some coherent or dispersion-supported systems. But<br />
also <strong>in</strong> case of systems based on direct <strong>in</strong>tensity modulation,<br />
the laser chirp may be important when it <strong>in</strong>teracts with the<br />
transmission channel chromatic dispersion. This time, however,<br />
rather large signal chirp properties should be analyzed.<br />
Natural extension of the above presented small-signal considerations<br />
would be that also large-signal relation between<br />
bulklaserFM<strong>and</strong>IMmaybeadoptedtoQWlasers.Follow<strong>in</strong>g<br />
the previous strategy, the large-signal laser chirp was determ<strong>in</strong>ed<br />
by simulat<strong>in</strong>g the full QW rate equations model, <strong>and</strong><br />
nextcomparedwiththechirpobta<strong>in</strong>edfrom(5).Aspreviously,<br />
the adiabatic chirp coefficient was trimmed to obta<strong>in</strong> the best<br />
agreement with the full model. The results are illustrated <strong>in</strong><br />
Fig. 4 for various capture time values. The laser model was<br />
driven by the 200 ps long, nearly-rectangular current pulse.<br />
One may notice that the chirp obta<strong>in</strong>ed from (5) is extremely<br />
close to that result<strong>in</strong>g from the full model. Only for very large<br />
capture time, as 50 ps, some quite small delay (about 8 ps)<br />
may be observed <strong>in</strong> the chirp obta<strong>in</strong>ed from (5).<br />
AverygoodagreementoftheQWlaserchirpcharacteristics<br />
obta<strong>in</strong>ed from the full model with that determ<strong>in</strong>ed from (5)<br />
<strong>and</strong> (6) is somewhat surpris<strong>in</strong>g when we have <strong>in</strong> m<strong>in</strong>d that<br />
they are derived from the bulk laser model. However, some<br />
<strong>in</strong>tuitive explanation may be proposed. First, it should be<br />
noticed that us<strong>in</strong>g the “bulk” equations (5) <strong>and</strong> (6), the chirp<br />
Fig. 4. Comparison of time doma<strong>in</strong> chirp evolution obta<strong>in</strong>ed from full QW<br />
model <strong>and</strong> from (6); capture time equal to 5 ps (a), 15 ps (b) <strong>and</strong> 50 ps (c).<br />
In the <strong>in</strong>sets correspond<strong>in</strong>g power waveforms.<br />
<strong>in</strong>duced <strong>in</strong> SCH region is “pushed” <strong>in</strong>to the adiabatic chirp of<br />
the active region. This way the changes of the SCH carrier<br />
density (which <strong>in</strong> fact make the SCH chirp component) were<br />
<strong>in</strong> the model “substituted” by the changes of laser optical<br />
power, which <strong>in</strong> case of high-speed modulation would not<br />
exactlyfollowtheSCH carrierdensity.Consider<strong>in</strong>gthecaseof<br />
large capture time first, we whould recall its low-pass filter<strong>in</strong>g<br />
feature. The 50 ps capture time <strong>in</strong>duces about 3 GHz cut-off,<br />
which depresses fast changes <strong>in</strong> the SCH carrier density. In<br />
thissituationthe“<strong>in</strong>ner”laserisfastenough,<strong>and</strong>so theoptical<br />
power nearly exactly follows the SCH carrier density, which<br />
expla<strong>in</strong>s the simple models accuracy.<br />
For lower values of capture time the optical power may be<br />
more mismatched from SCH carrier density. But, on the other<br />
h<strong>and</strong>, small capture time results <strong>in</strong> small carrier accumulation<br />
<strong>in</strong> the SCH <strong>and</strong> so small chirp component caused by The<br />
SCH region. This way even less accurate model<strong>in</strong>g of this<br />
component has no significant <strong>in</strong>fluence on total chirp, <strong>and</strong> the<br />
simplified model is still quite accurate.<br />
V. EXPERIMENTAL VERIFICATION<br />
Directmeasurementoflarge-signaltime-resolvedlaserchirp<br />
is quite complicated <strong>and</strong> usually suffers from <strong>in</strong>herent b<strong>and</strong>width<br />
limitation <strong>in</strong>troduced by the frequency response of<br />
FM/IM convert<strong>in</strong>g optical filters. Some <strong>in</strong>direct but quite<br />
precise verification of chirp model<strong>in</strong>g may be, however, performed<br />
based on the optical fiber chromatic dispersion. The<br />
<strong>in</strong>teraction of the laser chirp with the fiber dispersion causes<br />
serious distortions <strong>in</strong> the time evolution of optical power<br />
detected at the fiber end. Compar<strong>in</strong>g the distortions of the<br />
measured signal with that calculated based on the taken chirp<br />
model, its adequacy may be verified. The results of such<br />
experimentare shown <strong>in</strong> Fig. 5. The high-speedIM modulated<br />
signal (a piece of 10 Gb/s data stream) outgo<strong>in</strong>g the MQW<br />
DFB laser (PT3563 type) is illustrated <strong>in</strong> Fig. 5(a). Tak<strong>in</strong>g<br />
the chirp model <strong>in</strong> the form of (5), with parameters α <strong>and</strong><br />
κ obta<strong>in</strong>ed <strong>in</strong> other measurements, the chirp caused signal
66 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 5. Modulated laser output power (a), <strong>and</strong> fiber output power corrupted<br />
by <strong>in</strong>terplay of the laser chirp <strong>and</strong> the fiber chromatic dispersion (b).<br />
distortions after the 20 km long fiber were calculated, <strong>and</strong><br />
compared with the measurement. As it is visible <strong>in</strong> Fig. 5(b),<br />
the calculated<strong>and</strong>measuredfiberoutputsignalsare practically<br />
identical, which proves the adequate chirp model<strong>in</strong>g.<br />
VI. CONCLUSIONS<br />
The <strong>in</strong>fluence of the carrier transport between the SCH <strong>and</strong><br />
QW regions is analyzed <strong>in</strong> the paper <strong>in</strong> the context of chirp<br />
model<strong>in</strong>g. It was shown that even for high values of carrier<br />
capture time, when the transport effects seriously affect the<br />
laser IM <strong>and</strong> FM characteristics, the simple relations coupl<strong>in</strong>g<br />
<strong>in</strong>tensity modulation with chirp, derived for bulk lasers, may<br />
beused.Itisofseriouspracticalimportancebecauseitallowed<br />
us to determ<strong>in</strong>e the chirp from IM characteristics, us<strong>in</strong>g the<br />
model requir<strong>in</strong>g only two parameters: the l<strong>in</strong>e enhancement<br />
factor <strong>and</strong> the adiabatic chirp coefficient. Namely, the time<br />
doma<strong>in</strong> evolution of chirp may be obta<strong>in</strong>ed from the measured<br />
(or somehow modeled) time doma<strong>in</strong> evolution of the laser<br />
output power, by means of (5). Alternatively, the frequency<br />
doma<strong>in</strong> FM transfer function may be obta<strong>in</strong>ed from the<br />
frequency doma<strong>in</strong> IM transfer function, us<strong>in</strong>g (6). This way<br />
<strong>in</strong> many cases the troublesome full QW laser model<strong>in</strong>g may<br />
be omitted without sacrific<strong>in</strong>g the accuracy of considerations.<br />
REFERENCES<br />
[1] R. Ribeiro, J. da Rocha, A. Cartaxo, H. da Silva, B. Franz, <strong>and</strong><br />
B. Wedd<strong>in</strong>g, “FM response of quantum-well lasers tak<strong>in</strong>g <strong>in</strong>to account<br />
carrier transport effects,” IEEE Photon. Technol. Lett., vol. 7, no. 8, 1995.<br />
[2] E. Peral, W. Marshall, <strong>and</strong> A. Yariv, “Precise measurement of semiconductor<br />
laser chirp us<strong>in</strong>g effect of propagation <strong>in</strong> dispersive fiber<br />
<strong>and</strong> application to simulation of transmission through fiber grat<strong>in</strong>gs,” J.<br />
Lightw. Technol., vol. 16, no. 10, 1998.<br />
[3] E. Peral <strong>and</strong> A. Yariv, “Measurement <strong>and</strong> characterization of laser chirp<br />
ofmultiquantum-well distributed-feedback lasers,” IEEEPhoton. Technol.<br />
Lett., vol. 11, no. 3, 1999.<br />
[4] O. Nobuyuki, K. Masahiro, I. Masato, <strong>and</strong> M. Yasushi, “1.5-µm Stra<strong>in</strong>ed-<br />
Layer MQW-DFB Lasers with High Relaxation-Oscillation Frequency<br />
<strong>and</strong> Low-Chirp Characteris-tics,” IEEE J. Quantum Electron., vol. 32,<br />
no. 7, 1996.<br />
[5] L.Bjerkan, A.Royset, L.Hafskjaer, <strong>and</strong>D.Myhre,“Measurement oflaser<br />
parameters for simulation of high-speed fiberoptic systems,” J. Lightw.<br />
Technol., vol. 14, no. 5, 1996.<br />
[6] J. Morgado <strong>and</strong> A. Cartaxo, “Directly modulated laser parameters optimization<br />
for metropolitan area networks utiliz<strong>in</strong>g negative dispersion<br />
fibers,” IEEE J. Sel. Topics Quantum Electron., vol. 9, no. 5, 2003.<br />
[7] K. Czotscher, S. Weisser, A. Leven, <strong>and</strong> J. Rosenzweig, “Intensity<br />
Modulation <strong>and</strong> Chirp of 1.55-µm Multiple-Quantum-Well Laser Diodes:<br />
Model<strong>in</strong>g <strong>and</strong> Experimental Verification,” IEEE J. Sel. Topics Quantum<br />
Electron., vol. 5, no. 3, 1999.<br />
[8] P. Krehlik, “Characterization of semiconductor laser frequency chirp<br />
based on signal distortion <strong>in</strong> dispersive optical fiber,” Opto-Electron. Rev.,<br />
vol. 14, no. 2, 2006.<br />
[9] H. da Silva <strong>and</strong> M. Freire, “Multi-quantum well laser parameters for<br />
simulation of optical transmission systems up to 40 gbit/s,” <strong>in</strong> IEEE<br />
Global Telecommun. Conf., 1998.
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong> 67<br />
Precise Measurements of Highly Attenuated Optical<br />
Eye Diagrams<br />
Przemysław Krehlik, Łukasz ´Sliwczyński, <strong>and</strong> Grzegorz Sikorski<br />
Abstract—The idea <strong>and</strong> practical realization of a measurement<br />
system dedicated for highly attenuated eye diagrams diagnostics<br />
is presented <strong>in</strong> the paper. It is specially oriented on high-speed<br />
modulated optical data transmission signals which amplification<br />
is difficult <strong>and</strong>/or undesired. The presented measurements displayed<br />
the usefulness of proposed solution.<br />
Index Terms—eye diagram, optical measurements, noise reduction<br />
I. INTRODUCTION<br />
A NALYSISoftheeyediagram(calledalsotheeyepattern)<br />
is a simple but powerful method of digital transmission<br />
channel diagnostics. The eye diagram arises from overlapp<strong>in</strong>g<br />
many differentdata patterns time-shiftedby an <strong>in</strong>teger <strong>number</strong><br />
of unit <strong>in</strong>tervals (i.e. serial clock cycles) – see Fig. 1a.<br />
Degradation of the digital signal, caused by the transmission<br />
channel, may be thus recognized <strong>and</strong> measured. Some well<br />
known cases of signal distortions are illustrated on Fig. 1b.<br />
The simplest way to obta<strong>in</strong> the eye diagram is to register<br />
the data signal with an oscilloscope hav<strong>in</strong>g long persistence,<br />
dur<strong>in</strong>gthesynchronizationofthetimebasefromthedataclock<br />
signal (alternatively divided by any <strong>in</strong>teger factor).<br />
In case of fast optical signals, the best choice is to use<br />
the sampl<strong>in</strong>g oscilloscope with the optical-to-electrical (O/E)<br />
converter <strong>in</strong>tegrated with the sampl<strong>in</strong>g unit. This solution<br />
offers the outst<strong>and</strong><strong>in</strong>g equivalent b<strong>and</strong>width up to 70 GHz,<br />
with flat frequency response <strong>and</strong> low group delay dispersion<br />
[1]. However, it suffers from relatively high noise, <strong>in</strong> range<br />
of 10 ... 20 µWRMS of equivalent optical power. The noise<br />
disturbs or even completely blurs the observed eye diagram<br />
when the measured signal is strongly attenuated by long fibers<br />
or other optical devices. In some cases the problem may be<br />
overcome by us<strong>in</strong>g an optical amplifier, or external O/E converter<br />
followed by electronic amplifier. Unfortunately,<strong>in</strong> some<br />
situationsthose solutionscouldnot be usedor are suspectedof<br />
<strong>in</strong>troduc<strong>in</strong>g some artefacts affect<strong>in</strong>g the measurement results.<br />
Therefore, some method of noise reduction<strong>in</strong> the eye diagram<br />
measurements is desired.<br />
II. THE IDEA OF EYE NOISE REDUCTION<br />
A well known method of noise reduction, used <strong>in</strong> the measurements<br />
of periodic signals on digitis<strong>in</strong>g oscilloscopes, is to<br />
P. Krehlik is with the Institute of <strong>Electronics</strong>, AGH University of Science<br />
<strong>and</strong> Technology, Mickiewicza 30, 30-059 Kraków, Pol<strong>and</strong>; e-mail:<br />
krehlik@agh.edu.pl.<br />
Ł. ´Sliwczyński is with the Institute of <strong>Electronics</strong>, AGH University of<br />
Science <strong>and</strong> Technology, Mickiewicza 30, 30-059 Kraków, Pol<strong>and</strong>; e-mail:<br />
sliwczyn@agh.edu.pl.<br />
G. Sikorski graduated from AGH University of Science <strong>and</strong> Technology <strong>in</strong><br />
2007.<br />
Fig. 1. The idea of the eye diagram construction (a), <strong>and</strong> common eye<br />
distortions (b).<br />
average many registrations of the same trace (so called boxcar<br />
averag<strong>in</strong>g). When the noise is zero mean <strong>and</strong> uncorrelated<br />
<strong>in</strong> subsequent measurements, the root-mean-square (RMS) of<br />
the noise is reduced accord<strong>in</strong>gly to the square root of the<br />
<strong>number</strong> of averaged registrations. In the ord<strong>in</strong>ary eye diagram<br />
measurement, however, the overlapp<strong>in</strong>g of different patterns<br />
on the scope screen prohibits direct averag<strong>in</strong>g.<br />
The presented idea changes the manner of collect<strong>in</strong>g signal<br />
samples to allow averag<strong>in</strong>g-basednoise reduction. The pattern<br />
generator, connected to the <strong>in</strong>put of transmission l<strong>in</strong>k under<br />
test, outputs a set of different data sequences. Each sequence<br />
is repeated a <strong>number</strong> of times to allow the averag<strong>in</strong>g of<br />
particular patterns measured at the tested l<strong>in</strong>k output. F<strong>in</strong>ally,<br />
all stored averaged patterns are overlapped <strong>and</strong> shown on<br />
“virtual” oscilloscope display – see Fig. 2.<br />
It should be realized that the described method of the eye<br />
diagram construction changes <strong>in</strong> some way the <strong>in</strong>formation<br />
gathered <strong>in</strong> the eye diagram. By reduc<strong>in</strong>g the measurement<br />
noise it clarified all pattern dependent signal distortions (<strong>in</strong>tersymbol<br />
<strong>in</strong>terferences (ISI), nonl<strong>in</strong>ear distortions, pattern dependent<br />
jitter <strong>and</strong> so on). At the other h<strong>and</strong>, however, the<br />
averag<strong>in</strong>g reduces not only measurement noise but also any<br />
possibler<strong>and</strong>omevents<strong>in</strong>thereceivedsignal,suchastransmitter<br />
relative <strong>in</strong>tensity noise (RIN), optical amplifier amplified<br />
spontaneous emission (ASE), adjacent signals crosstalks <strong>in</strong><br />
multichannel systems etc.<br />
III. MEASUREMENT SYSTEM IMPLEMENTATION<br />
Based on the idea presented above, a measurement system<br />
dedicated for measur<strong>in</strong>g highly attenuated optical eye<br />
diagrams was built. The system (see Fig. 3) is based on
68 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 2. The idea of eye noise reduction.<br />
Fig. 3. Block diagram of measurement system.<br />
HP83480A sampl<strong>in</strong>g oscilloscope with HP83485B optical<br />
plug-<strong>in</strong>, offer<strong>in</strong>g 30 GHz measurement b<strong>and</strong>width. The oscilloscope<br />
is connected, via the GPIB <strong>in</strong>terface, with a system<br />
softwarerunonpersonalcomputer(PC).Thesoftwarecontrols<br />
also the data sequence generator. The generator repeats the<br />
current pattern until it receives a new one from the PC.<br />
The actually implemented sequence generator operates with<br />
10 Gb/s output data rate, <strong>and</strong> produces 16-bit patterns. The<br />
tested optical l<strong>in</strong>k consists of a transmitter <strong>and</strong> arbitrary set<br />
of optical components, as fibers, optical amplifiers, dispersion<br />
compensators, filters etc. Optionally, it may be term<strong>in</strong>ated by<br />
O/E receiver for electrical eye diagram measurement.<br />
The entiremeasurementprocessiscontrolledbyadedicated<br />
software,written<strong>in</strong> Matlabenvironment[2].After def<strong>in</strong><strong>in</strong>gthe<br />
set of data patterns to be used <strong>in</strong> the measurement, <strong>and</strong> sett<strong>in</strong>g<br />
some parameters (as the <strong>number</strong> of averages of each pattern),<br />
the measurement process may be <strong>in</strong>itialized by the operator.<br />
Then, subsequent patterns are sent to the sequence generator.<br />
Each sequence is repeated at its output for the time needed for<br />
the averag<strong>in</strong>g process, performed by the oscilloscope. Next,<br />
the result<strong>in</strong>g averaged output pattern is acquired by GPIB<br />
<strong>in</strong>terface <strong>and</strong> stored on the PC. Then the next pattern is sent<br />
to the generator,the oscilloscope averag<strong>in</strong>g memory is cleared<br />
<strong>and</strong> <strong>in</strong>itialized, <strong>and</strong> so on. F<strong>in</strong>ally, all the patterns got from the<br />
oscilloscope are overlapped to form the eye diagram, which is<br />
displayed on “virtual” oscilloscope display, emulated by the<br />
Fig. 4. Align<strong>in</strong>g of patterns shifted by transmission delay variation.<br />
software on the PC monitor.<br />
When test<strong>in</strong>g the system, a generally proper behavior was<br />
observed. However, <strong>in</strong> some cases some malfunction, manifested<br />
<strong>in</strong> the horizontal eye smear was detected. It was found<br />
that the problem arises when the tested optical l<strong>in</strong>k <strong>in</strong>troduces<br />
seriousatransmissiondelay,i.e.it <strong>in</strong>cludeslongfiber. Because<br />
the oscilloscope is triggered by the signal com<strong>in</strong>g from the<br />
sequence generator, any drift of the transmission delay results<br />
<strong>in</strong> horizontal w<strong>and</strong>er of the received signal observed on the<br />
oscilloscope. As the measurement procedure takes significant<br />
time (<strong>in</strong> the range of a few m<strong>in</strong>utes up to an hour), the subsequentlyreceivedpatternsmaybemutuallyshifted,whichblurs<br />
the result<strong>in</strong>geyepattern.In the case offiber optictransmission<br />
the common reason for the transmission delay drift is fiber<br />
chromatic dispersion <strong>in</strong>teract<strong>in</strong>g with temperature dependent<br />
laser wavelength. As it was observed, for transmitters with<br />
uncooled laser operat<strong>in</strong>g <strong>in</strong> dispersive 1.55 µm w<strong>in</strong>dow, even<br />
a few kilometers of fiber may <strong>in</strong>troduce unacceptable delay<br />
<strong>in</strong>stability.<br />
Toovercometheproblem,anoptionalprocedureperform<strong>in</strong>g<br />
auto-alignment of received patterns is added. The idea of the<br />
alignment algorithm is illustrated <strong>in</strong> Fig. 4.<br />
Inthisoptionthefirsthalfofthe16-bitpatternsoutgo<strong>in</strong>gthe<br />
sequence generator is reserved for constant reference pattern,<br />
consist<strong>in</strong>g of four “1” <strong>and</strong> four “0” symbols. The rema<strong>in</strong><strong>in</strong>g 8<br />
bits are chang<strong>in</strong>g <strong>and</strong> used for eye pattern construction. The<br />
software automatically recognizes the reference transition <strong>and</strong><br />
aligns all received patterns before overlapp<strong>in</strong>g.<br />
IV. EXPERIMENTAL RESULTS<br />
To illustrate the system abilities, some examples are presented<br />
<strong>in</strong> this section. In the first one the tested transmission<br />
l<strong>in</strong>k consists of the 10 Gb/s transmitter, based on directly<br />
modulated laser operat<strong>in</strong>g at 1.55 µm wavelength, two pieces<br />
of st<strong>and</strong>ard s<strong>in</strong>gle-mode fiber with dispersion compensat<strong>in</strong>g<br />
fiber between them. The total fiber length was 110 km. The<br />
fiber l<strong>in</strong>k presents some residual chromatic dispersion (about<br />
600 ps/nm), caused by <strong>in</strong>sufficient length of the compensat<strong>in</strong>g<br />
fiber. Because of high attenuation of the set of fibers, the<br />
received signal was very weak, <strong>and</strong> so the eye diagram<br />
measured directly on the oscilloscope was completely hidden
KREHLIK et al.: PRECISE MEASUREMENTS OF HIGHLY ATTENUATED OPTICAL EYE DIAGRAMS 69<br />
Fig. 5. Eye diagram of weak signal register directly on the oscilloscope (a), <strong>and</strong> us<strong>in</strong>g the described system (b).<br />
Fig. 6. Eye diagram obta<strong>in</strong>ed at the end of same optical l<strong>in</strong>k for vary<strong>in</strong>g<br />
EDFA amplification.<br />
<strong>in</strong> oscilloscope noise, as shown <strong>in</strong> Fig. 5a. Us<strong>in</strong>g the presented<br />
system, the clear eye was obta<strong>in</strong>ed, as shown <strong>in</strong> Fig. 5b. Now<br />
some signaldistortions,causedbytheresidualdispersion,may<br />
be precisely observed. In the experiment the laser cooler was<br />
turned off, so the ambient temperature variations affected the<br />
transmission delay. The ma<strong>in</strong> eye displayed <strong>in</strong> Fig. 5b was<br />
taken with the auto-align<strong>in</strong>g option turned on, <strong>and</strong> the <strong>in</strong>set<br />
shows the smeared eye diagram obta<strong>in</strong>ed without align<strong>in</strong>g.<br />
The eye diagrams presented <strong>in</strong> Fig. 6 were obta<strong>in</strong>ed for a<br />
l<strong>in</strong>k consist<strong>in</strong>g of the transmitterdescribedabove,the boost<strong>in</strong>g<br />
erbium doped fiber amplifier (EDFA) <strong>and</strong> 70 km of dispersion<br />
compensated fiber. Three measurements were performed for<br />
various EDFA ga<strong>in</strong>s. The eye shown <strong>in</strong> Fig. 6a was obta<strong>in</strong>ed<br />
for low amplification, result<strong>in</strong>g <strong>in</strong> 5 dBm power at fiber<br />
<strong>in</strong>put. This time the output eye was clearly opened, with only<br />
small over- <strong>and</strong> undershoots observed. For higher amplification<br />
the signal distortions become evident (Fig. 6b), <strong>and</strong><br />
f<strong>in</strong>ally, for even higher amplification, the output eye pattern<br />
was completely destroyed (Fig. 6c). This way an evident<br />
manifestation of fiber nonl<strong>in</strong>earities was observed. (The eye<br />
diagram measured at EDFA output had still the same shape.)<br />
It should be po<strong>in</strong>ted out that the reference eye diagram, taken<br />
with the lowest amplification, could not be obta<strong>in</strong>ed without<br />
the presented measurement system, because of the weakness<br />
of the fiber output signal.<br />
V. SUMMARY<br />
A solution for measur<strong>in</strong>g the highly attenuated optical eye<br />
diagrams is presented <strong>in</strong> the paper. It is dedicated for use <strong>in</strong><br />
situations when the optical or electrical amplification of the<br />
received weak optical signal is impossible or is suspected of<br />
<strong>in</strong>troduc<strong>in</strong>g some undesired artifacts.<br />
The ma<strong>in</strong> idea is to repeatedly transmit each data pattern<br />
to allow measurement noise reduction by means of averag<strong>in</strong>g,<br />
<strong>and</strong> to overlap all registered patterns afterwards. The idea of<br />
cop<strong>in</strong>g with the possible transmission delay w<strong>and</strong>er is also<br />
proposed. The practical implementation of the measurement<br />
system <strong>and</strong> realized experiments verify the usefulness of the<br />
solution.<br />
REFERENCES<br />
[1] “DSA8200 Digital Sampl<strong>in</strong>g Oscilloscopes,” [onl<strong>in</strong>e],<br />
http://www.tek.com/products/oscilloscopes/.<br />
[2] G. Sikorski, “Stanowisko do automatycznego sterowania i akwizycji<br />
danych dla cyfrowego oscyloskopu sampl<strong>in</strong>gowego,” Master’s thesis,<br />
AGH, Kraków, 2007.
70 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Bit Error Rate Tester for 10 Gb/s Fibre Optic L<strong>in</strong>k<br />
Łukasz ´Sliwczyński <strong>and</strong> Przemysław Krehlik<br />
Abstract—The bit error rate tester suitable for operation<br />
<strong>in</strong> 10 Gb/s fibre optic l<strong>in</strong>ks is described <strong>in</strong> the paper. The<br />
BER tester was built from commercially available components.<br />
Generation <strong>and</strong> reception of 10 Gb/s data stream is performed<br />
with help of high-speed serialiser <strong>and</strong> deserialiser by Maxim. The<br />
ma<strong>in</strong> functions of the BER tester are implemented <strong>in</strong> the field<br />
programmable gate array (FPGA) Spartan3 device by Xil<strong>in</strong>x.<br />
The part of the FPGA runs with the clock speed equal to 622<br />
MHz. Some measurement results obta<strong>in</strong>ed <strong>in</strong> the fibre optic l<strong>in</strong>ks<br />
operated with 10 Gb/ data rate are also presented.<br />
Index Terms—bit error rate, fibre optic l<strong>in</strong>ks, field programmable<br />
gate arrays<br />
I. INTRODUCTION<br />
BIT error rate (BER) is one of the most important parameters<br />
describ<strong>in</strong>g the performance of transmission <strong>in</strong> the<br />
digital l<strong>in</strong>k. It is usually def<strong>in</strong>ed as:<br />
BER = ne<br />
, (1)<br />
N<br />
where ne is the total <strong>number</strong> of received bits <strong>and</strong> N is the<br />
<strong>number</strong> of bits be<strong>in</strong>g <strong>in</strong> error. Because of r<strong>and</strong>om nature of<br />
the phenomenon, BER is also regarded as the probability of<br />
errors occurr<strong>in</strong>gdur<strong>in</strong>g data transmission. BER <strong>in</strong> the order of<br />
10 −9 or even 10 −12 is often considered as be<strong>in</strong>g characteristic<br />
for modern fibre optic systems. Because of that, measur<strong>in</strong>g<br />
BER accord<strong>in</strong>gly to equation (1) is <strong>in</strong>convenient as it would<br />
require us<strong>in</strong>g a counter with huge capacity (generally, greater<br />
than 1/BER. Thus, it is better to transform equation (1) <strong>in</strong>to:<br />
BER = 1 ne<br />
, (2)<br />
B ∆t<br />
where B is the bit rate <strong>and</strong> ∆t is the measurementtime. When<br />
us<strong>in</strong>g equation (2), it is convenient to express ∆t <strong>in</strong> seconds,<br />
<strong>and</strong> the bit rate is only a scal<strong>in</strong>g factor.<br />
Nowadays 10 Gb/s transmission rate is <strong>in</strong>creas<strong>in</strong>gly common<br />
<strong>in</strong> fibre optic l<strong>in</strong>ks. Commercial BER testers capable<br />
of operation with such fast signals are often very advanced<br />
(e.g. [1]–[3]). They allow the test<strong>in</strong>g of a transmission system<br />
more comprehensively (for example to check its immunity to<br />
jitter or pathological data patterns), not just to simply measure<br />
BER. Unfortunately, the cost of such test systems is very<br />
high, which make them rarely available for most universities<br />
research/students labs. Thus, an idea was born to develop 10<br />
Gb/s BER tester (BERT), which would be possible to be built<br />
from commercially available components, with most of its<br />
functions be<strong>in</strong>g implemented <strong>in</strong> the FPGA circuit. Below a<br />
Łukasz ´Sliwczyński is with the AGHUniversity ofScience <strong>and</strong> Technology,<br />
Mickiewicza 30 Ave., 30-059 Krakow, Pol<strong>and</strong> (phone: +48 12-617-27-40, fax:<br />
+48-12-633-23-98, e-mail: sliwczyn@agh.edu.pl)<br />
Przemysław Krehlik iswiththeAGHUniversity ofScience <strong>and</strong> Technology,<br />
Mickiewicza 30 Ave., 30-059 Krakow, Pol<strong>and</strong> (e-mail: krehlik@agh.edu.pl)<br />
Fig. 1. Generic block diagram of the BER tester.<br />
design of such BERT is presented, along with a theory of its<br />
operation.<br />
II. IDEA OF OPERATION OF THE BER TESTER<br />
Each BERT is composed of two ma<strong>in</strong> parts: the transmitter<br />
(that <strong>in</strong>cludes the generator of the test sequences) <strong>and</strong> the<br />
receiver (that <strong>in</strong>cludes the error detector <strong>and</strong> analyser) [4].<br />
The block diagram of the BER tester is presented <strong>in</strong> Fig. 1.<br />
The purposeof the test sequence generatoris to producethe<br />
stream of the data bits accord<strong>in</strong>g to some rule that must be<br />
known for the receiver as well. The most often the pseudo<br />
r<strong>and</strong>om bit sequence (PRBS) generators are used for this<br />
purpose. There are a <strong>number</strong> of st<strong>and</strong>ard polynomialsdef<strong>in</strong><strong>in</strong>g<br />
different PRBS, developed by st<strong>and</strong>ardisation bodies (e.g. [5])<br />
for test<strong>in</strong>g telecommunication equipment. Alternatively, some<br />
bit sequence def<strong>in</strong>ed by the user <strong>and</strong> stored <strong>in</strong> the tester<br />
memory may be periodically generated.<br />
In the receiver, the error detector compares the received<br />
bits with the orig<strong>in</strong>al pattern <strong>and</strong>, <strong>in</strong> case of <strong>in</strong>compatibility,<br />
<strong>in</strong>creasesthe errorcounter.Theresult ofthemeasurementmay<br />
be presented <strong>in</strong> many different ways: simply as a <strong>number</strong>, or<br />
<strong>in</strong> the form of detailed diagram, display<strong>in</strong>g the <strong>number</strong> of bits<br />
be<strong>in</strong>g <strong>in</strong> error dur<strong>in</strong>g each second of the measurement.<br />
Because of the delay <strong>in</strong>troduced by the tested transmission<br />
l<strong>in</strong>k, the measurement process must be preceded by the synchronisationofthelocaltest<br />
sequencegenerator<strong>in</strong>thereceiver<br />
with the generator <strong>in</strong>cluded <strong>in</strong> the transmitter. Details of this<br />
process are described <strong>in</strong> [4] <strong>and</strong> [6] <strong>and</strong> will not be discussed<br />
here.<br />
It should be mentioned that the BER measurement must be<br />
performed on the formed, digital signal with clearly def<strong>in</strong>ed<br />
logical levels. In particular, the transmission clock is required<br />
to be either recovered or supplied externally to the receiver.<br />
III. BER TESTER FOR 10 GB/S SYSTEM<br />
The idea described <strong>in</strong> the previous section may be applied<br />
to the signal with any bit rate, at least <strong>in</strong> pr<strong>in</strong>ciple. However,<br />
at Gb-per-second data rates some special techniques <strong>and</strong>
´SLIWCZYŃSKI AND KREHLIK: BIT ERROR RATE TESTER FOR 10 GB/S FIBRE OPTIC LINK 71<br />
Fig. 2. Simplified block diagram of 10 Gb/s BER tester.<br />
modifications of the basic idea must often be used, accord<strong>in</strong>g<br />
to the available technology. One of the most important th<strong>in</strong>gs<br />
when design<strong>in</strong>g the BER tester is the necessity to generate<br />
the serial data stream runn<strong>in</strong>g with 10 Gb/s rate. Hav<strong>in</strong>g no<br />
access to highly advanced technology of mak<strong>in</strong>g <strong>in</strong>tegrated<br />
circuit, it is practically impossible to build a classical PRBS<br />
generator based on the serial shift register with feedback. This<br />
difficulty may be overcome by design<strong>in</strong>g a generator <strong>and</strong> the<br />
error detector to operate on parallel words rather than on<br />
<strong>in</strong>dividual bits. Parallel data may then be easily converted <strong>in</strong>to<br />
the serial stream bymeansof properserialiser <strong>and</strong>deserialiser.<br />
This way the speed of the clock necessary to operate the<br />
tester may be reduced substantially. In the solution described<br />
here<strong>in</strong>, it was assumed at first that generation <strong>and</strong> further data<br />
process<strong>in</strong>g would be performed with 622 MHz clock us<strong>in</strong>g<br />
Xil<strong>in</strong>x’s Spartan3 FPGA (see Fig. 2). MAX3952/MAX3953<br />
serialiser/deserialiserbyMaximareresponsibleforperform<strong>in</strong>g<br />
serial/parallel conversion.<br />
Althoughsome<strong>in</strong>itialanalysissuggestedthatitwaspossible<br />
to build BERT accord<strong>in</strong>g to the diagram presented <strong>in</strong> Fig. 2,<br />
it turned out f<strong>in</strong>ally that full parallel architecture cannot<br />
be implemented <strong>in</strong> the Spartan3 device. Because of that, a<br />
modified <strong>and</strong> simplified architecture was developed, that fits<br />
<strong>in</strong>to chosen FPGA circuit. The most important features of this<br />
architecture will be presented <strong>in</strong> the next chapters.<br />
IV. TRANSMITTER WITH THE TEST SEQUENCES<br />
GENERATOR<br />
The full parallel PRBS architecture (e.g. as described <strong>in</strong><br />
[7]) proved to be too complex to operate with 622 MHz clock<br />
signal after implementation <strong>in</strong> Spartan3 FPGA. It was thus<br />
assumed that BER would be measured only on a few chosen<br />
bits (called the measurementchannels)fromthe 16-bitparallel<br />
word. It was also taken that the transmitter would repeat each<br />
16-bit sequence four times, which effectively lowers its clock<br />
speed to 155 MHz. This simplified greatly the test sequence<br />
generator.<br />
The structure of the parallel words sent to the serialiser<br />
is presented <strong>in</strong> Fig. 3. Inside this word two bits, D8 <strong>and</strong><br />
D15, have their values fixed to “zero” <strong>and</strong> “one”, respectively.<br />
The BER of the l<strong>in</strong>k under test is determ<strong>in</strong>ed based on these<br />
two bits only. The rema<strong>in</strong><strong>in</strong>g 14 bits are divided <strong>in</strong>to two<br />
unequal fields, with length 8 <strong>and</strong> 6 bits. These fields are filled<br />
Fig. 3. Structure of the s<strong>in</strong>gle word of the test sequence.<br />
Fig. 4. 10 Gb/s BERT transmitter.<br />
with PRBS hav<strong>in</strong>g period 2 8 − 1 <strong>and</strong> 2 6 − 1, respectively.<br />
Such structure of the test word is justified by the requirement<br />
of hav<strong>in</strong>g the serial data stream as “r<strong>and</strong>om” as possible,<br />
simultaneouslypreserv<strong>in</strong>gitsDCbalance.Becauseofdifferent<br />
PRBS periods, the period of the result<strong>in</strong>g sequence is much<br />
longer than <strong>in</strong> the case of two PRBS with the same period.<br />
The structure of the test word proposed here<strong>in</strong> posseses<br />
some shortcom<strong>in</strong>gs, however. The longest run of the same<br />
consecutive symbols is limited to 9 “ones” <strong>and</strong> 7 “zeros”.<br />
It limits BERT capabilities when test<strong>in</strong>g the immunity of<br />
transmission system to the low frequencyspectral components<br />
conta<strong>in</strong>ed <strong>in</strong> the signal. Further, the <strong>number</strong> of bits that could<br />
result <strong>in</strong> <strong>in</strong>tersymbol <strong>in</strong>terference (ISI) is also limited: for<br />
“one” there are 6 bits before <strong>and</strong> 8 bits after, for “zero” there<br />
is the reverse. These limitations, however, seem not to be a<br />
big problem, especially if the BERT is used to evaluate errors<br />
caused by fibre dispersion or laser chirp.<br />
A completeBERT transmitter<strong>in</strong>cludesalso a few additional<br />
blocks: pattern synchronisator, <strong>in</strong>verter <strong>and</strong> error <strong>in</strong>serter. The<br />
<strong>in</strong>verter is useful if the transmission l<strong>in</strong>k under test <strong>in</strong>verts<br />
the signal itself. This may be easily done even accidentally<br />
because I/O <strong>in</strong>terfaces of the BERT use differential signal<strong>in</strong>g.<br />
The error <strong>in</strong>serter allows for perform<strong>in</strong>g some k<strong>in</strong>d of BERT<br />
self-test. If activated, it periodically changes the polarity of<br />
signals<strong>in</strong> the measurementchannelsforone clockperiod,thus<br />
forc<strong>in</strong>g errors. Because the rate of these errors is known, it<br />
may be used to check for possible BERT or l<strong>in</strong>k under test<br />
malfunction. The pattern synchronization is necessary to align<br />
the bits at the output of the MAX3953 deserialiser with <strong>in</strong>puts<br />
of MAX3952 serialiser. When the deserialiser acquires serial<br />
synchronism with the data stream produced by the serialiser,<br />
thepositionofbitsatitsoutputisnotnecessarilycorrect.Thus,<br />
some k<strong>in</strong>d of a barrel shifter, capable of the rotation of bits<br />
appear<strong>in</strong>g at the output of the transmitter, is necessary to set<br />
the proper order of the bits. Although this circuit is associated<br />
rather with the deserialiser than serialiser, it appeared much<br />
easier to implement it <strong>in</strong>side the BERT transmitter.
72 ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS, VOL. 1, NO. 2, NOVEMBER <strong>2010</strong><br />
Fig. 5. 10 Gb/s BERT receiver.<br />
V. BER DETECTOR AND ANALYSER<br />
Because of the structure of the test word used <strong>in</strong> the<br />
presented design, the detection of errors is a straightforward<br />
task. To do this, it is enough to count the clock periods where<br />
the bits <strong>in</strong> the measurement channels differ from that set <strong>in</strong><br />
the transmitter.<br />
It is crucial for the BERT operation to run error counters<br />
at the clock speed equal to 622 MHz. To facilitate operation<br />
with such speed, the count<strong>in</strong>g of errors is divided <strong>in</strong>to a few<br />
tasks (see Fig. 5). At the <strong>in</strong>put of each measurement channel a<br />
3-bit fast counter is implemented. The Johnson’s counters are<br />
used there because of their potential for high-speed operation.<br />
Simulations performed us<strong>in</strong>g ISE7 <strong>and</strong> ModelSim XE software<br />
packages (available form Xil<strong>in</strong>x <strong>and</strong> Mentor Graphics,<br />
respectively) showed that the counter composed of maximum<br />
of three D flip-flops (F/F) is capable to operate with required<br />
speed. The capacity of such Johnson counter equals 6. This<br />
allows the lower<strong>in</strong>g of the clock<strong>in</strong>g speed of the rest of the<br />
circuit four times (blocksoperat<strong>in</strong>gwith lower clock speed are<br />
marked with additional dashed border <strong>in</strong> Fig. 5). The counter<br />
used <strong>in</strong> the design is the synchronousone, with <strong>in</strong>put from the<br />
measurement channel connected to the Clock Enable <strong>in</strong>puts of<br />
the F/F.<br />
To obta<strong>in</strong> the <strong>number</strong> of bits be<strong>in</strong>g <strong>in</strong> error dur<strong>in</strong>g the<br />
four consecutive clock cycles, it is necessary to calculate the<br />
difference between the current state of the Johnson’s counter<br />
<strong>and</strong> its delayed state. To facilitate this operation, the output<br />
from the counter is converted <strong>in</strong>to the natural b<strong>in</strong>ary format.<br />
After subtraction, the partial results are totaled <strong>in</strong> the 16-bit<br />
b<strong>in</strong>ary counter. The totalizer has two 3-bit <strong>in</strong>puts, one for each<br />
measurementchannel(process<strong>in</strong>gthe circuitryfor onechannel<br />
only is shown <strong>in</strong> Fig. 5).<br />
VI. EDITORIAL POLICY<br />
After total<strong>in</strong>g the errors, the result is passed to the software<br />
PicoBlaze [8] processor implemented <strong>in</strong> the FPGA. This<br />
processor is responsible for calculat<strong>in</strong>g BER, display<strong>in</strong>g the<br />
result <strong>and</strong> communicat<strong>in</strong>g with the user.<br />
BER calculation is made accord<strong>in</strong>g to the formula similar<br />
to that given <strong>in</strong> equation (2):<br />
BER = L<br />
NA<br />
1 ne<br />
, (3)<br />
B ∆t<br />
where L is the length of the parallel word <strong>and</strong> NA is the<br />
<strong>number</strong>of channelsmeasur<strong>in</strong>g BER. Modificationof the basic<br />
formula (2) results from the fact that BERT described <strong>in</strong> the<br />
paper does not count all errors occurr<strong>in</strong>g dur<strong>in</strong>g the transmission.<br />
It rather samples errors that degrade transmission<br />
on two chosen bits only. Tak<strong>in</strong>g the assumption that the<br />
probability of errors affect<strong>in</strong>g the rest of bits is the same<br />
<strong>and</strong> that errors are <strong>in</strong>dependent, one may correct the result<br />
by simply <strong>in</strong>creas<strong>in</strong>g the error rate, as it is done <strong>in</strong> equation<br />
(3).<br />
In our case L = 16, NA = 2, B = 10 · 10 9 <strong>and</strong> ∆t was<br />
chosen to be measured <strong>in</strong> seconds. Putt<strong>in</strong>g all these <strong>number</strong>s<br />
<strong>in</strong>to equation (3), it may be simplified as:<br />
BER = 4 ne<br />
5 ∆t 10−9 . (4)<br />
Us<strong>in</strong>g equation (4), BER may be quite easily calculated because<br />
all required mathematical operationsare performedwith<br />
the natural <strong>number</strong>s. Multiplicationby 4 <strong>in</strong> the numeratormay<br />
be carried out by the logical left shift of ne by two positions,<br />
whereas multiplication by 5 <strong>in</strong> the denom<strong>in</strong>ator requires two<br />
shifts <strong>and</strong> one more addition. The division operation must<br />
be performed hav<strong>in</strong>g <strong>in</strong> m<strong>in</strong>d a possibly very wide, be<strong>in</strong>g <strong>in</strong><br />
orders of magnitude, dynamic range of the result. However,<br />
becausethereisalotoftimetoobta<strong>in</strong>theresult(1second),the<br />
entire operation may be executed without resort<strong>in</strong>g to the full<br />
float<strong>in</strong>g po<strong>in</strong>t arithmetic. A simple procedure implemented <strong>in</strong><br />
the design, exploits only multiplication by 10 <strong>and</strong> subtraction<br />
<strong>and</strong> allows calculate BER directly <strong>in</strong> the decimal x.xx · 10−y format.Thecodeforthisprocedurerealized<strong>in</strong>24-bitprecision<br />
occupies about 150 PicoBlaze assembler <strong>in</strong>structions <strong>and</strong><br />
executes <strong>in</strong> a small fraction of second.<br />
VII. EXPERIMENTAL RESULTS<br />
Us<strong>in</strong>g the BERT described above, some experimental data<br />
were taken <strong>in</strong> l<strong>in</strong>ksoperat<strong>in</strong>gwith 10Gb/s transmissionspeed.<br />
The results are presented <strong>in</strong> Fig. 6.<br />
In Fig. 6a BER measured <strong>in</strong> the l<strong>in</strong>k composed of the laser<br />
transmitter followed by erbium doped fibre amplifier (EDFA)<br />
booster <strong>and</strong> 40 km of the st<strong>and</strong>ard s<strong>in</strong>glemode fibre (SSM)<br />
is presented. The two curves are plotted for two different<br />
values of EDFA ga<strong>in</strong>. Based on the plot, the power penalty<br />
may be determ<strong>in</strong>ed. For the case presented this penalty is<br />
quite <strong>in</strong>dependent of the <strong>in</strong>put power <strong>and</strong> is about -2 dB.<br />
The negative value of the penalty results probably from a<br />
constructive <strong>in</strong>teraction of fibre nonl<strong>in</strong>earity/dispersion with<br />
the chirp of directly modulated laser.<br />
When perform<strong>in</strong>g BER measurements, some care must be<br />
taken, however. In Fig. 6b the results of back-to-back BER<br />
measurement with neither fibre nor EDFA <strong>in</strong>serted between<br />
the transmitter <strong>and</strong> the receiver are presented. Two different<br />
results were obta<strong>in</strong>ed <strong>in</strong> exactly the same experimental setup.<br />
Between two measurements, only the connector <strong>in</strong> the optical<br />
path was disconnected <strong>and</strong> connected aga<strong>in</strong>. The difference is<br />
probably caused by the light backreflected from the connector<br />
tothelaser. Thisgeneratessomenoise<strong>in</strong>thelaserthatstrongly<br />
depends on the quality of the optical connection. This is<br />
evident, thus, that any conclusions concern<strong>in</strong>g the penalties<br />
<strong>in</strong> the order of 1 dB should be drawn very carefully. It would<br />
be best to perform measurements a few times, observ<strong>in</strong>g the<br />
consistence of the results.
´SLIWCZYŃSKI AND KREHLIK: BIT ERROR RATE TESTER FOR 10 GB/S FIBRE OPTIC LINK 73<br />
Fig. 6. BER versus received optical power <strong>in</strong> a few experimental setups<br />
–desription <strong>in</strong> the text.<br />
VIII. CONCLUSION<br />
The bit error rate tester designed for operation <strong>in</strong> 10 Gb/s<br />
fibre optic l<strong>in</strong>ks is described <strong>in</strong> the paper. The ma<strong>in</strong> purpose<br />
of this BERT was to evaluate the degradation of the signal<br />
quality, caused by a <strong>in</strong>teraction of directly modulated laser<br />
chirp with fibre dispersion. This, however, does not limit the<br />
applications of the BERT to these cases only.<br />
The architecture of the BERT described here<strong>in</strong> was tailored<br />
to the abilities of Spartan3 FPGA, that is used to implement<br />
most of the design. The usual operat<strong>in</strong>g idea of the BERT<br />
was found to be unsuitable for the design, therefore some<br />
special solutions were proposed. Us<strong>in</strong>g high-speed SiGe serialiser/deserialiser<br />
<strong>and</strong> exploit<strong>in</strong>g extensively parallel architecture<br />
with pipel<strong>in</strong><strong>in</strong>g, it was possible to overcome <strong>in</strong>herent<br />
speed limits of Spartan3 FPGA <strong>and</strong> build functional BERT<br />
operat<strong>in</strong>g at 10 Gb/s data rate.<br />
The tester built accord<strong>in</strong>g to the idea presented <strong>in</strong> the paper<br />
was tested <strong>in</strong> the laboratory <strong>and</strong> proved its usefulness for<br />
research <strong>and</strong> <strong>in</strong>vestigation purposes. The design lacks some<br />
features, however, that should be added <strong>in</strong> the next version.<br />
Because BER measurements are relatively time-consum<strong>in</strong>g, it<br />
would be very helpful to log past values of BER for further<br />
analysis. This way, it would be possible to tell if the measured<br />
BER is <strong>in</strong>herent for the system under test, or if it was caused<br />
by some external <strong>in</strong>terference. In addition, the capacity of<br />
the totalizer (16 bits) proved to be too small <strong>and</strong> should be<br />
extended to 24 bits.<br />
REFERENCES<br />
[1] “BERTScope S,” [onl<strong>in</strong>e], http://www.bertscope.com.<br />
[2] “ParBERT,” [onl<strong>in</strong>e], http://www.agilent.com.<br />
[3] “J-BERT N4903A,” [onl<strong>in</strong>e], http://www.agilent.com.<br />
[4] C. Coombs, Electronic Instrument H<strong>and</strong>book. McGraw-Hill, 1995.<br />
[5] “ITU-T Recommendations O151, O152 <strong>and</strong> O153,” Tech. Rep.<br />
[6] A. Liwak <strong>and</strong> L. ´Sliwczyński, “Laboratoryjny miernik bitowej stopy<br />
bł˛edu,” <strong>in</strong> Proc. of Poznańskie Warsztaty Telekomunikacyjne, 2004, pp.<br />
75–80, (<strong>in</strong> Polish).<br />
[7] L. ´Sliwczyński, “PRBS generator runs at 1.5 Gbps,” <strong>in</strong> Proc. of EDN,<br />
Mar. 2007, pp. 76–80.<br />
[8] PicoBlaze 8-bit Embedded Microcontroller User Guide for Spartan-3,<br />
Virtex-II, <strong>and</strong> Virtex-II Pro FPGAs, Xil<strong>in</strong>x, 2005.
About the journal<br />
ADVANCES IN ELECTRONICS AND TELECOMMUNICATIONS is a peer-reviewed journal published by Poznan University of<br />
Technology. It publishes scientific papers devoted to several problems <strong>in</strong> the area of contemporary electronics <strong>and</strong> telecommunications.<br />
Its scope is focused on, but not limited to the follow<strong>in</strong>g issues:<br />
• electronic circuits <strong>and</strong> systems,<br />
• microwave devices <strong>and</strong> systems,<br />
• DSP structures <strong>and</strong> algorithms for wireless <strong>and</strong> wirel<strong>in</strong>e communication systems,<br />
• digital modulations,<br />
• data transmission techniques,<br />
• multiple access techniques <strong>and</strong> MAC issues,<br />
• <strong>in</strong>formation <strong>and</strong> channel cod<strong>in</strong>g theory <strong>and</strong> its applications,<br />
• software def<strong>in</strong>ed radio <strong>and</strong> cognitive radio technologies,<br />
• wireless local area networks (WLANs),<br />
• satellite communication,<br />
• navigation <strong>and</strong> localization,<br />
• synchronization subsystems,<br />
• time <strong>and</strong> tim<strong>in</strong>g,<br />
• model<strong>in</strong>g techniques of package & on-chip <strong>in</strong>terconnects,<br />
• radiation & <strong>in</strong>terference, electromagnetic compatibility,<br />
• propagation aspects <strong>in</strong> wireless communication,<br />
• UWB channel model<strong>in</strong>g,<br />
• measurements <strong>and</strong> wireless sensor networks,<br />
• web technologies,<br />
• e-learn<strong>in</strong>g,<br />
• multimedia communication,<br />
• audio <strong>and</strong> speech process<strong>in</strong>g,<br />
• image <strong>and</strong> video process<strong>in</strong>g,<br />
• software <strong>and</strong> hardware system implementation,<br />
• advanced A/D <strong>and</strong> D/A conversion techniques <strong>and</strong> their applications,<br />
• SDI - Software Def<strong>in</strong>ed Instruments,<br />
• effective measurement, estimation <strong>and</strong> computation of signal parameters,<br />
• consumer electronics.<br />
Detailed <strong>in</strong>formation about the journal can be found at: www.advances.et.put.poznan.pl.<br />
The Editorial Board <strong>in</strong>vites paper submissions on the above topics for Open Call.