Techniques to Enhance Spectral

Eciency of OFDM Wireless

Systems
by
Suvra Sekhar Das, B.Eng.
Dissertation
Presented to the International Doctoral School of Technology and Science
in Partial Fulllment of the Requirements for the Degree of
Doctor of Philosophy
Aalborg University
7
th.
September 2007
Supervisors:
Professor Ramjee Prasad
Associate Professor Elisabeth de Carvalho
The Assessment Committee:
Professor Preben Mogensen, Aalborg University, Denmark
Professor Lajos Hanzo, University of Southampton, UK
Professor Shinsuke Hara, Osaka City University Japan
Moderator:
Associate Professor Fleming B. Frederiksen, Aalborg University, Denmark
ISBN: 87-92078-07-9
ISSN: 0908-1224
Copyright
c
September, 2007 by
Suvra Sekhar Das
Center for TeleInFrastruktur (CTiF)
Aalborg University
Niels Jernes Vej 12
9220 Aalborg Ost
Denmark
e-mail: [email protected]
All rights reserved by the author. No part of the material protected by this
copyright notice may be reproduced or utilized in any form or by any means,
electronics or mechanical, including photocopying, recording, or by any information
storage and retrieval system, without written permission from the author.
Dedicated to my Parents, my Sister and Madhulipa.
iv
Center for TeleInFrastruktur (CTIF), Aalborg University
Abstract
In recent years Orthogonal Frequency Division Multiplexing (OFDM) based tech-
nologies are in wide use for wireless communication systems. This is because OFDM
elegantly overcomes the adverse eects of frequency selective fading channels and of-
fers high spectral eciency. Investigation of techniques to further enhance the spectral
eciency of OFDM based wireless systems is the prime objective of this thesis.
As a rst step, a comparison is made between OFDM and Multi Carrier Spread
Spectrum (MC-SS) scheme. OFDM is found to oer relatively better performance
than MC-SS under channel estimation and synchronization errors. This is the mo-
tivation to select OFDM for further investigation and performance enhancement in
this thesis.
Though OFDM has many advantages, yet it is severely aected by Inter Carrier
Interference (ICI), which is caused by residual phase error, carrier frequency oset
and Doppler frequency spread. To track the residual phase error, pilot sub carriers
are embedded between the data sub carriers. To reduce the pilot overhead, it is
proposed in this thesis to load data bits on pilot sub carriers without degrading
system performance. It is found that up to 15% increase in spectral eciency can be
obtained by using this technique.
To mitigate the impact of ICI, due to Doppler frequency spread, a novel tech-
nique of using adaptive sub carrier bandwidth is proposed in this work. This technique
enhances the spectral eciency in the range of 10% to 30% over OFDM systems which
use xed sub carrier bandwidth.
OFDM systems need a large Guard Interval (GI) to overcome the eect of Inter
v
vi
Symbol Interference (ISI). In order to decrease the overhead due to GI, an algorithm
to dynamically select the GI duration is derived in this thesis. By adaptively selecting
the GI duration, it is found that the spectral eciency can be increased up to 20%.
The performance of OFDM based wireless systems is also limited by the time
variations of the propagation channel. In such situations Link Adaptation (LA) tech-
niques using adaptive bit rate transmission achieves very high spectral eciency by
exploiting the channel variations. OFDM provides a suitable framework for LA.
However, the combination of LA and OFDM results in increased implementation
complexity. Another aim of this thesis is to provide low complexity techniques to in-
crease spectral eciency. Hence, low complexity, low overhead LA - OFDM schemes
which have near optimal spectral eciency are proposed in this work. The proposed
schemes reduce overhead by 50% as well as signicantly bring down the implementa-
tion complexity.
The impact of non linear signal distortion caused by the high power amplier
and frequency synchronization errors on the performance of LA-OFDM systems are
also analyzed in this work. Methods to overcome the eects of these impairments by
suitable adjustments to the LA algorithms are presented in this thesis.
As a result of this work it can be concluded that signicant cumulative gain in
spectral eciency can be obtained by using the proposed transmission schemes. The
techniques and guidelines for spectral eciency improvement presented in this thesis
are promising enough for future OFDM based wireless systems.
Center for TeleInFrastruktur (CTIF), Aalborg University
Dansk Resume
I de seneste ar er Orthogonal Frequency Division Multiplexing (OFDM) baserede
systemer taget mere og mere i brug i forbindelse med tradlst kommunikationsudstyr.
Det skyldes, at OFDM pa en eektiv made kan hindre den delggende virkning af
frequency-selective-fading, og at OFDM samtidig kan tilbyde hj spektral eektivitet.
Det primre formal med denne afhandling er undersgelse af teknikker, som
yderligere kan forbedre den spektrale eektivitet af OFDM baserede tradlse sys-
temer. Som det frste trin gennemfres en sammenligning mellem OFDM og dets
udvidelsessystem MC-SS. OFDM ses her at tilbyde forholdsvis bedre ydeevne end
MC-SS m.h.t. modtagereektivitet og -fejl. Derfor foretrkkes OFDM i denne afhan-
dling og undersges njere med henblik pa yderligere forbedring af ydeevnen.
OFDM har mange fordele men er umiddelbart strkt pavirket af Inter Carrier
Interference (ICI). ICI skyldes residuale fasefejl, carrier-frequency-oset og Doppler-
frequency-spread. For at spore de risiduale fasefejl er der indfrt pilot-sub-carriers
mellem de enkelte data-sub-carriers. Med sigte pa at mindske overheadet i forbindelse
med indfrelsen af pilot-sub-carriers er det i afhandlingen foreslaet at tilfre data bits
til de enkelte pilot-sub-carriers for dermed at undga reduktion af systemydeevnen.
Ved undersgelse ndes det, at der med denne teknik kan opnas op til 15% forbedring
i spektral eektivitet.
For at mindske den betydelige virkning, som ICI ved Doppler-frequency-spread
har pa OFDM systemer, foreslas der i nrvrende arbejde en ny teknik baseret pa
adaptiv sub-carrier bandbredde. Der opnas med denne teknik en forbedring pa 10%
- 30% af den spektrale eektivitet i forhold til OFDM systemer med fast sub-carrier
bandbredde.
Der behves i OFDM systemer et stort Guard Interval (GI) for at overvinde
virkningen af Inter Symbol Interference (ISI). Med sigte pa at fa mindsket overheadet
i forbindelse med GI er der i afhandlingen udviklet en algoritme, som dynamisk
fastlgger GI-varigheden. Det vises, at hvis GI-varigheden vlges adaptivt v.h.a.
vii
viii
algoritmen, kan den spektrale eektivitet ges op til 15%.
OFDM baserede tradlse systemer er ogsa begrnsede af den tidsvarierende
kanal. I sadanne situationer kan man ved hjlp af Link Adaptation (LA) teknik
med adaptiv bit-rate-transmission opna meget hj spektraleektivitet ved at udnytte
variationer i kanalen. OFDM tilvejebringer en passende ramme for LA. Kombina-
tionen af LA og OFDM resulterer dog i forget implementationskompleksitet. Et
andet sigte med denne afhandling er derfor at tilvejebringe en lav-kompleksitets LA-
OFDM teknik. I nrvrende arbejde foreslas der af den grund lav-kompleksitets,
lille-overheads LA-OFDM systemer med nsten optimale spectrale eektiviteter. Det
beregnes, at de foreslaede systemer kan reducere overheadet med 50% samtidig med,
at kompleksiteten reduceres signicant.
Virkningen af en praktisk funktionsbegrnsning - f.eks. i forbindelse med
ikke-liner signalforvrngning forarsaget af eektforstrkere og givet ved resultatet
af frekvens-synkroniseringsfejl i forbindelse med driften af LA-OFDM systemer - er
ogsa analyseret, og metoder til at fa bugt med disse forringelser og fejl ved passende
justering af LA algoritmerne er prsenteret i afhandlingen.
Som et resultat af arbejdet kan det konkluderes, at signicant kumulativ
forbedring af den spektrale eektivitet kan opnas ved brug af de foreslaede adap-
tive transmissionsteknikker. De adaptive teknikker og retningslinier for forbedring
af den spektrale eektivitet, som er prsenteret i afhandlingen, vurderes at vre
tilstrkkeligt lovende m.h.p. yderligere overvejelser i forbindelse med fremtidige
OFDM baserede tradlse systemer.
Center for TeleInFrastruktur (CTIF), Aalborg University
Acknowledgements
I am deeply indebted to my supervisor Prof. Ramjee Prasad, for giving me the
opportunity to work under his supervision and learn from his vast experience which
spans beyond technical elds. His continuous motivation helped me sail through
this voyage and reach the nal destination. I express my gratitude to Elisabeth De
Carvalho who has been my co-supervisor for a major duration of my work and has
provided important feedback in several aspects of my work. I am also highly obliged
to Frank H.P. Fitzek and Ole Olsen who were my supervisors during the early part
of my research at Aalborg University without whose support I could not have begun
this work.
I am grateful to Dr. Sunil Sherlekar, for taking the initiative for this project
and to the management team of Tata Consultancy Services Ltd., India, for funding
this work. I specially thank Mr. Debasis Bandopadhay, for providing all support
needed to succesfully complete this work without any hindrance. I would like to
thank Mr. Arpan Pal, Mr. P. Balamuralidhar and Mr. Prateep Mishra, for giving
me the prestigious opportunity to work on this project. I have learnt a lot from my
team members of the Embedded Systems group of TCS Kolkata. I would like to
thank them all.
I would like to mention my colleague Imadur specially here for he has been
a continuous discussion partner throughout the work. I express my sincere thanks
to Daniel and Soren who have provided continuous help in understanding technical
concepts. I would like to remember Akhilesh, Basuki, Anas , Petar , Huan, Hiro with
whom I had the chance of some good technical discussion.
I am thankful to the several MSc. students who have made contributions to
this work while working on their graduation projects or on internships. Fuad, Bayu,
Carlos, Faisal and Nidcha have been associated with this work is some time or the
other. I like to specially cite Yuyuane Wang for his interest in research activities and
the contributions he made in many ways to this work.
ix
x
I thank all my colleagues and secretaries of section of RATE, CTIF, former
WING group and former Department of KOM, provided all necessary help needed to
do my research smoothly.
Fleming B. Frederiksen was always keen about my progress and supported me
in several student projects. I am grateful to him for all his kindness. I would note the
help of Sanjay Kumar from BIT Mesra, Ranchi for providing important feedback to
make the nal version of the thesis. Mrs. Nisha Gupta from BIT Mesra, also helped
in making the document complete. I am grateful to her for the help she extended.
I also owe a lot my colleague Nicola Marchetti for reading through the thesis and
helping to improve it.
It is the sacrice and blessings of my parents that gave me the strength to
reach the completion of this work steadily. My sister kept continuously motivating
me to persevere to the end of this long project. It was the tremendous support from
my wife, Madhulipa, that helped me put all my eort in this work. She made many
sacrices so that I could devote my time to this thesis. My little daughter Aane,
(Debosmita) has been a new inspiration in my life since she was born. Finally it is
by Gods grace that I could do my research relentlessly to complete the program in
time.
Center for TeleInFrastruktur (CTIF), Aalborg University
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Dansk Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiii
Chapter 1 Introduction 1
1.1 Background to OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Problem Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Goal and Scope of the thesis . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Research Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6 Organization of the thesis . . . . . . . . . . . . . . . . . . . . . . . . 8
1.7 Contributions of this Thesis . . . . . . . . . . . . . . . . . . . . . . . 9
Chapter 2 Wireless Channel and Multi Carrier Systems 13
2.1 Wireless Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1 Channel Parametrization . . . . . . . . . . . . . . . . . . . . . 14
2.1.2 Propagation loss . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.3 Shadowing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.4 Small scale Fading . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Quadrature Amplitude Modulation . . . . . . . . . . . . . . . . . . . 23
2.3 OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.1 OFDM Fundamentals . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.2 Parameters values for OFDM based Standards . . . . . . . . . 34
2.4 Link Adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Chapter 3 Evaluation of Hybrid Multi Carrier Spread Spectrum 41
3.1 Sub Carrier Hopping Multi Carrier Spread Spectrum . . . . . . . . . 42
xi
xii CONTENTS
3.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.1.2 System Description . . . . . . . . . . . . . . . . . . . . . . . . 42
3.1.3 Analytical Model . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1.4 Simulation Results and Discussion . . . . . . . . . . . . . . . . 47
3.1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 MC-SS with receiver impairments . . . . . . . . . . . . . . . . . . . . 57
3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2.2 System Description . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2.3 Simulation Environment, Results and Discussion . . . . . . . . 59
3.2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Chapter 4 Bit loading on Pilot Sub Carriers 69
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2 System Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3 Analytical Framework and Algorithm . . . . . . . . . . . . . . . . . . 76
4.4 Simulation and Discussion . . . . . . . . . . . . . . . . . . . . . . . . 84
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Chapter 5 Adaptive Sub Carrier Bandwidth 91
5.1 Adaptive Sub Carrier Bandwidth in Time Division Multiplexing (TDM)-
OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.1.1 System Description . . . . . . . . . . . . . . . . . . . . . . . . 93
5.1.2 Analytical Model . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.1.3 Algorithm for Adaptive Bandwidth for Sub Carriers . . . . . . 96
5.1.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . 99
5.1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.2 OFDMA Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.2.1 Analytical Model . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.2.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . 111
5.2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Chapter 6 Variable Guard Interval 119
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.2 System Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.3 Required GI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Center for TeleInFrastruktur (CTIF), Aalborg University
CONTENTS xiii
6.4 Performance and Discussion . . . . . . . . . . . . . . . . . . . . . . . 130
6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
Chapter 7 Hybrid Link Adaptation 135
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.3 Hybrid LA strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.3.1 Dierent Link Adaptation Algorithms . . . . . . . . . . . . . . 140
7.3.2 LA with Dierent Sub-channel Sizes . . . . . . . . . . . . . . 143
7.3.3 Fixed Coding Rate . . . . . . . . . . . . . . . . . . . . . . . . 146
7.3.4 LA Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.3.5 Dierent LA & PC Rates . . . . . . . . . . . . . . . . . . . . 150
7.3.6 Interaction between Spatial Diversity and Link Adaptation . . 155
7.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
7.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Chapter 8 Link Adaptation under Transceiver Impairments 161
8.1 Inuence of Non Linear High Power Amplier (HPA) . . . . . . . . . 162
8.1.1 HPA Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
8.1.2 Eect of HPA on dierent Modulation and coding rates . . . . 166
8.1.3 Link Adaptation under HPA Impairments . . . . . . . . . . . 176
8.1.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
8.2 LA under ICI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
8.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
8.2.2 LA under undetected ICI . . . . . . . . . . . . . . . . . . . . . 193
8.2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
8.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Chapter 9 Conclusions and Future Work 199
9.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
9.2 Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Chapter A Selected Publications Related to the thesis 203
A.1 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
A.1.1 IPR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
A.1.2 Journal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
A.1.3 Conference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
c
Suvra Sekhar Das, 2007
xiv CONTENTS
A.2 Chapter wise Publications . . . . . . . . . . . . . . . . . . . . . . . . 206
Chapter B Link Adaptation 207
B.1 Bit and Power Loading Algorithm . . . . . . . . . . . . . . . . . . . . 207
Chapter C Hybrid Link Adaptation 213
Chapter D LA in OFDM systems under HPA 221
Bibliography 237
Vita 249
Center for TeleInFrastruktur (CTIF), Aalborg University
List of Figures
1.1 Wireless Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 Characterization of Fading . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Propagation Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Amplitude response . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 channel impulse response and transfer function relationship . . . . . . 19
2.5 Multipath propagation . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 A typical power delay prole . . . . . . . . . . . . . . . . . . . . . . . 21
2.7 Frequency domain channel response . . . . . . . . . . . . . . . . . . . 22
2.8 Power Spectral Density vs frequency of Jakes spectrum . . . . . . . . 24
2.9 Power Spectral Density vs frequency of typical Gauss spectrum . . . 24
2.10 Signal space diagram for rectangular 64-QAM . . . . . . . . . . . . . 24
2.11 Non orthogonal carriers . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.12 Orthogonal Sub carriers in Multi carrier systems (OFDM) . . . . . . 26
2.13 Time domain representation of the signal waveforms to show orthogo-
nality among the sub carriers . . . . . . . . . . . . . . . . . . . . . . 27
2.14 Base band modules of the OFDM transmitter . . . . . . . . . . . . . 28
2.15 Time Frequency representation of OFDM Signal . . . . . . . . . . . . 29
2.16 Frequency Selective and non Selective Fading . . . . . . . . . . . . . 30
2.17 The use of Cyclic Prex . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.18 Top level architecture of OFDM receiver circuitry showing important
signal processing modules in base band part. . . . . . . . . . . . . . . 31
2.19 Training sequence for WLAN [44] . . . . . . . . . . . . . . . . . . . . 31
2.20 SNR switching threshold points for LA System. M=4 indicates QPSK,
M=16 is 16-QAM, and M=64 is for 64-QAM, while C represents coding
rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.21 Link Adaptation basic framework . . . . . . . . . . . . . . . . . . . . 38
xv
xvi LIST OF FIGURES
2.22 Time Sequence of Events in Link Adaptation . . . . . . . . . . . . . . 38
2.23 Spectral Eciency Gain for LA System. . . . . . . . . . . . . . . . . 39
3.1 Transmitter for Sub-Carrier Hopped Multi Carrier Spread Spectrum
(SCH-MC-SS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Receiver for SCH-MC-SS . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3 Time frequency diagram of the sub carrier hopping scheme. It is shown
that a sub carrier allocated, which is represented by a particular colour
shade, moves in time and frequency grid in one packet. . . . . . . . . 44
3.4 Mean throughput Vs spreading gain at dierent SNRs with single sym-
bol detection, for channel model 2.
intr represents interleaved sub carrier arrangement and
blk implies block sub carrier arrangement.
OFDMA CDM is the parent MC-SS scheme, while SCH OFDMA CDM
is the proposed sub carrier hopping scheme. . . . . . . . . . . . . . . 48
3.5 10% outage throughput Vs spreading gain at dierent SNRs with single
symbol detection, for channel 2 . . . . . . . . . . . . . . . . . . . . . 49
3.6 Mean throughput Vs spreading gain at dierent SNRs with successive
interference cancelation, for channel 2 . . . . . . . . . . . . . . . . . . 50
3.7 10% outage throughput Vs spreading gain at dierent SNRs with suc-
cessive interference cancelation, for channel 2 . . . . . . . . . . . . . . 51
3.8 Mean throughput Vs spreading gain at dierent SNRs with single sym-
bol detection, for channel 6 . . . . . . . . . . . . . . . . . . . . . . . 52
3.9 10% outage throughput Vs spreading gain at dierent SNRs with single
symbol detection, for channel 6 . . . . . . . . . . . . . . . . . . . . . 53
3.10 Mean throughput Vs spreading gain at dierent SNRs with successive
interference cancelation, for channel 6 . . . . . . . . . . . . . . . . . . 54
3.11 10% outage throughput Vs spreading gain at dierent SNRs with suc-
cessive interference cancelation, for channel 6 . . . . . . . . . . . . . . 55
3.12 Eect of dierent spreading gain for ideal receiver conditions. The
numbers in the bracket, e.g. MC-SS(4) indicates the spreading gain. . 60
3.13 Eect of loading for spreading gain of 16 under ideal conditions . . . 61
3.14 Eect of spreading gain in full load under residual carrier frequency
oset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.15 Eect of loading for spreading gain of 16 under residual carrier fre-
quency oset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Center for TeleInFrastruktur (CTIF), Aalborg University
LIST OF FIGURES xvii
3.16 Eect of spreading gain in full load under Channel Estimation error . 63
3.17 Eect of loading for spreading gain of 16 under Channel estimation error 64
3.18 Eect of spreading gain on the 10% outage performance in full load for
ideal receiver conditions . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.19 Eect of loading for spreading gain of 16 on the 10% outage perfor-
mance for ideal receiver conditions . . . . . . . . . . . . . . . . . . . 65
3.20 Eect of spreading gain on the 10% outage performance in full load for
Channel Estimation error and synchronization error . . . . . . . . . . 66
3.21 Eect of loading for spreading gain of 16 on the 10% outage perfor-
mance for Channel Estimation error and synchronization error . . . . 66
4.1 OFDM Symbol Format . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2 16-QAM Data constellation . . . . . . . . . . . . . . . . . . . . . . . 74
4.3 QPSK Pilot constellation . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4 Piece-wise linear interpolation . . . . . . . . . . . . . . . . . . . . . . 82
4.5 BER Vs SNR, Data64-QAM, pilotBPSK.
alg3 completely data aided pilot based OFDM system, using the
exact algorithm.
alg2 the exact algorithm in the proposed semiblind environment.
alg1 the proposed low complexity, approximated algorithm. . . . . 85
4.6 BER Vs SNR, Data64-QAM, pilotQPSK . . . . . . . . . . . . . 85
4.7 BER Vs SNR, Data16-QAM, pilotBPSK . . . . . . . . . . . . . 86
4.8 BER Vs SNR, Data16-QAM, pilotQPSK . . . . . . . . . . . . . 87
4.9 BER Vs SNR, DataQPSK, pilotBPSK . . . . . . . . . . . . . . . 87
5.1 Time frequency diagram for the proposed TDM based ASB OFDM . 93
5.2 Time frequency diagram of the proposed Adaptive Sub Carrier Bandwidth
(ASB) with Band Division Multiplexing (BDM) OFDM . . . . . . . . 94
5.3 SINR vs sub carrier bandwidth at 15dB SNR . . . . . . . . . . . . . 96
5.4 Throughput vs sub carrier bandwidth at 15dB SNR . . . . . . . . . . 97
5.5 Sub carrier bandwidth selected by the proposed ASB system. . . . . . 100
5.6 Throughput comparison of the proposed ASB vs standard Fixed Sub
Carrier Bandwidth (FSB) OFDM systems, at 15 dB SNR. . . . . . . 101
5.7 Throughput comparison of the proposed ASB vs FSB OFDM systems,
at 25 dB SNR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.8 BER of the proposed ASB and FSB OFDM systems at 15 dB SNR,
when target BER is 0.01 . . . . . . . . . . . . . . . . . . . . . . . . . 102
c
Suvra Sekhar Das, 2007
xviii LIST OF FIGURES
5.9 Frequency domain conguration of Variable Sub Carrier Bandwidth
(VSB) OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.10 Downlink Transmitter for the proposed VSB Orthogonal Frequency
Division Multiple Access (OFDMA) at the base station . . . . . . . . 106
5.11 Up link Receiver for the proposed VSB OFDMA at the base station . 107
5.12 SINR of standard OFDM systems at 20dB SNR . . . . . . . . . . . . 113
5.13 Capacity of standard OFDM systems at 20dB . . . . . . . . . . . . . 114
5.14 Capacity of standard OFDM systems at 10dB . . . . . . . . . . . . . 114
5.15 Capacity of VSB OFDM at 20dB . . . . . . . . . . . . . . . . . . . . 115
5.16 Capacity comparison when users with dierent mobility conditions co-
exist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.1 Eect of small GI. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.2 Interference Power due to previous OFDM symbol vs Ratio of GI over
rms delay spread. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.3 SINR vs Ratio of GI over rms delay spread . . . . . . . . . . . . . . . 129
6.4 Ratio of GI Vs rms delay spread. Depicting variation of T
gi
with
respect to various

. . . . . . . . . . . . . . . . . . . . . . . . . . . 132
6.5 Performance with increasing SNR. . . . . . . . . . . . . . . . . . . . . 133
6.6 Cumulative Distribution Function of required GI. . . . . . . . . . . . 133
6.7 Gain in throughput of the proposed VGI over xed GI system for 2dB
extra SNR for 1s rms delay spread. . . . . . . . . . . . . . . . . . . 134
7.1 OFDM based link adaptation transceiver architecture . . . . . . . . . 138
7.2 Link adaptation frame structure. . . . . . . . . . . . . . . . . . . . . 139
7.3 Spectral Eciency for SISO with Fd=50Hz,
rms
= 0.5s . . . . . . . 140
7.4 Power Utilization for SISO with Fd=50Hz,
rms
= 0.5s . . . . . . . . 141
7.5 Spectral Eciency for SISO with Fd=250Hz,
rms
= 2s . . . . . . . 141
7.6 Power Utilization for SISO with Fd=250Hz,
rms
= 2s . . . . . . . . 142
7.7 Spectral Eciency with Doppler 50Hz, delay spread 0.5s . . . . . . 144
7.8 Spectral Eciency with Doppler 250Hz, delay spread 2s . . . . . . . 145
7.9 Block Error Rate (BLER) for SISO,LA per 1 frame(s),BLER=0.05,

rms
=0.5s, fd=50Hz. . . . . . . . . . . . . . . . . . . . . . . . . . . 146
7.10 Spectral Eciency with Doppler 50Hz, delay spread 0.5s . . . . . . 147
7.11 Spectral Eciency with Doppler 250Hz, delay spread 2s . . . . . . . 148
7.12 Spectral Eciency with Doppler 50Hz, delay spread 0.5s . . . . . . 149
7.13 Spectral Eciency with Doppler 250Hz, delay spread 2s . . . . . . . 149
Center for TeleInFrastruktur (CTIF), Aalborg University
LIST OF FIGURES xix
7.14 Combined slow LA with fast power control . . . . . . . . . . . . . . . 151
7.15 Spectral Eciency for Dierent LA & PC Rates, Fd=50Hz,
rms
= 0.5s153
7.16 Spectral Eciency for Dierent LA & PC Rates, fd=250Hz,
rms
= 2s 154
7.17 Spectral Eciency for Multi-antenna Schemes, LA Per 0.5ms . . . . . 155
7.18 Spectral Eciency for Multi-antenna Schemes, LA Per 2ms . . . . . 156
7.19 Spectral Eciency for Multi-antenna Schemes, LA Per 5ms . . . . . 157
7.20 Spectral Eciency for Multi-antenna Schemes, LA Per 10ms . . . . . 157
8.1 Power Back of (BO) representation in Rapps Model . . . . . . . . . . 165
8.2 Relation between Amplier Distortion and BO (in dB) . . . . . . . . 166
8.3 Spectrum plot of OFDM signal. BO indicates BO value in dB. . . . 167
8.4 SDNR plot for 4QAM modulation in AWGN Channel. . . . . . . . . 169
8.5 SDNR plot for 4QAM modulation in Fading Channel. The BO values
are given in dB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
8.6 BER vs SNR curve for uncoded and M=4 in fading channel . . . . . 172
8.7 BLER vs SNR curve for C =
1
2
and M=4 in fading channel . . . . . . 172
8.8 BER vs SNR curve for uncoded and M=16 in Fading channel . . . . 173
8.9 BLER vs SNR curve for C =
1
2
and M=16 in fading channel . . . . . 173
8.10 BER vs SNR curve for uncoded and M=64 in Fading channel . . . . 174
8.11 BLER vs SNR curve for C =
1
2
and M=64 in fading channel . . . . . 174
8.12 TD plot for FEC =
1
2
with BLER Threshold= 0.1 in Fading Channel 175
8.13 TD plot for FEC =
1
2
with BLER Threshold= 0.05 in Fading Channel 176
8.14 Flow chart of bit loading algorithm used for analyzing the inuence of
HPA in LA OFDM systems . . . . . . . . . . . . . . . . . . . . . . . 177
8.15 PAPR distribution for LA based OFDM system. . . . . . . . . . . . . 179
8.16 Performance of LA system with basic LUT when no power amplier
is applied. HPA-0 implies no HPA situation while HPA-1 implies that
HPA is used in the simulation. . . . . . . . . . . . . . . . . . . . . . . 180
8.17 Performance of LA system with basic LUT when power amplier is used181
8.18 Performance of LA system with revised LUT when power amplier is
used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
8.19 Spectral Eciency comparison for LA system with and without PAPR
consideration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.20 Performance of LA system with basic LUT for 6 dB of BO power . . 183
8.21 Performance of LA system with basic LUT for 4 dB of BO power . . 184
8.22 Performance of LA system with revised LUT for 6 dB of BO power . 186
c
Suvra Sekhar Das, 2007
xx LIST OF FIGURES
8.23 Performance of LA system with revised LUT for 4 dB of BO power . 186
8.24 Impact of frequency oset on 4-QAM in fading channel. . . . . . . . . 190
8.25 Impact of frequency oset on 4-QAM, FEC rate 1/2, in fading channel. 190
8.26 Impact of frequency oset on 16-QAM in fading channel. . . . . . . . 191
8.27 Impact of frequency oset on 16-QAM, FEC rate 1/2, in fading channel.191
8.28 Impact of frequency oset on 64-QAM in fading channel. . . . . . . . 192
8.29 Impact of frequency oset on 64-QAM, FEC rate 1/2, in fading channel.192
8.30 Bler performance without additional margin for coding rate 1/2 . . . 195
8.31 Bler performance with additional margin for coding rate 1/2 . . . . . 195
8.32 Spectral eciency performance with additional margin for coding rate
1/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
B.1 Flow diagram of the SAMPDA algorithm . . . . . . . . . . . . . . . . 208
B.2 Spectral eciency achievement of the adaptation algorithms . . . . . 211
B.3 Number of iterations required by dierent adaptation algorithms . . . 211
C.1 Throughput comparison of dierent Link adaptation algorithms at 0.5
s and 2.0 s rms delay spread and Doppler condition for sub-band
size of 8 sub carriers . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
C.2 Power utilization comparison of dierent Link adaptation algorithms
at dierent rms delay spread and Doppler condition for sub-band size
of 8 and 32 sub carriers . . . . . . . . . . . . . . . . . . . . . . . . . . 215
C.3 Throughput performance of dierent sub-band sizes for dierent rms
delay spread, Doppler velocity. . . . . . . . . . . . . . . . . . . . . . . 216
C.4 Throughput performance comparison for xed coding with adaptive
modulation Vs adaptive modulation and coding for sub-band size of 8
sub carriers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
C.5 Throughput comparison for dierent adaptation rates, for rms delay
spread of 0.5s at 20 kmph . . . . . . . . . . . . . . . . . . . . . . . . 219
D.1 Comparison of theoretical and simulated CDF of PAPR . . . . . . . . 223
D.2 Eect of BO of 6 dB on 16QAM constellation points . . . . . . . . . 223
D.3 Eect of BO on 16QAM constellation points . . . . . . . . . . . . . . 223
D.4 16QAM basic constellation points . . . . . . . . . . . . . . . . . . . . 224
D.5 Eect of dierent modulation scheme on CDF of PAPR when Fast
Fourier Transform (FFT) size is 512. M indicates the modulation level,
and C the coding rate. . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Center for TeleInFrastruktur (CTIF), Aalborg University
LIST OF FIGURES xxi
D.6 Eect of dierent coding rate on the CDF of PAPR when FFT size is
128. M indicates the modulation level, and C the coding rate. . . . . 224
D.7 SDNR plot for 16QAM modulation in AWGN Channel . . . . . . . . 225
D.8 SDNR plot for 16QAM modulation in Fading Channel . . . . . . . . 225
D.9 SDNR plot for 64QAM modulation in AWGN Channel . . . . . . . . 225
D.10 SDNR plot for 64QAM modulation in Fading Channel . . . . . . . . 225
D.11 BER vs SNR curve for uncoded and M=4 in AWGN channel . . . . . 226
D.12 BLER vs SNR curve for C =
1
2
and M=4 in AWGN channel . . . . . 226
D.13 BER vs SNR curve for uncoded and M=16 in AWGN channel . . . . 226
D.14 BLER vs SNR curve for C =
1
2
and M=16 in AWGN channel . . . . . 226
D.15 BER vs SNR curve for uncoded and M=64 in AWGN channel . . . . 226
D.16 BLER vs SNR curve for C =
1
2
and M=64 in AWGN channel . . . . . 226
D.17 BLER vs SNR curve for C =
1
3
and M=4 in AWGN channel . . . . . 227
D.18 BLER vs SNR curve for C =
2
3
and M=4 in AWGN channel . . . . . 227
D.19 BLER vs SNR curve for C =
1
3
and M=16 in AWGN channel . . . . . 227
D.20 BLER vs SNR curve for C =
2
3
and M=16 in AWGN channel . . . . . 227
D.21 BLER vs SNR curve for C =
1
3
and M=64 in AWGN channel . . . . . 227
D.22 BLER vs SNR curve for C =
2
3
and M=64 in AWGN channel . . . . . 227
D.23 TD plot for FEC =
1
2
with BLER Threshold= 0.1 in AWGN . . . . . 228
D.24 TD plot for FEC =
1
2
with BLER Threshold= 0.05 in AWGN . . . . 228
D.25 TD plot for FEC =
1
3
with BLER Threshold= 0.1 in AWGN . . . . . 228
D.26 TD plot for FEC =
1
3
with BLER Threshold= 0.05 in AWGN . . . . 228
D.27 TD plot for FEC =
1
3
with BLER Threshold= 0.1 in Fading Channel 228
D.28 TD plot for FEC =
1
3
with BLER Threshold= 0.05 in Fading Channel 228
D.29 TD plot for FEC =
2
3
with BLER Threshold= 0.1 in AWGN . . . . . 229
D.30 TD plot for FEC =
2
3
with BLER Threshold= 0.05 in AWGN . . . . 229
D.31 TD plot for FEC =
2
3
with BLER Threshold= 0.1 in Fading Channel 229
D.32 TD plot for FEC =
2
3
with BLER Threshold= 0.05 in Fading Channel 229
c
Suvra Sekhar Das, 2007
xxii LIST OF FIGURES
Center for TeleInFrastruktur (CTIF), Aalborg University
List of Tables
1.1 Wireless Systems using OFDM . . . . . . . . . . . . . . . . . . . . . 2
2.1 Value of parameters for urban terrain. . . . . . . . . . . . . . . . . . 17
2.2 Parameters in WLAN . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 WMAN system parameters . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Switching Threshold for Link Adaptation . . . . . . . . . . . . . . . . 39
4.1 SNR advantage of pilots in dierent modulation schemes . . . . . . . 73
4.2 Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3 Semi Blind Congurations and gains . . . . . . . . . . . . . . . . . . 88
5.1 List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.1 Bit Error Rate for SINR: 15 dB , rms delay spread: 1 s, Carrier
Frequency: 3.5 GHz, Bandwidth 20 Mhz bandwidth, Number of sub
carriers: 1024, Guard length for xed GI: 128. . . . . . . . . . . . . . 131
7.1 System Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.2 SISO,LA per 1 frame(s),BLER=0.05, tau
rms
=0.5s,fd=50Hz . . . . . 145
7.3 SNR Threshold for Coding Rate Switching . . . . . . . . . . . . . . . 150
7.4 Overhead in Mbps for Adapt Power LA . . . . . . . . . . . . . . . . . 155
7.5 Summary of Hyrbid Link Adaptation . . . . . . . . . . . . . . . . . . 159
8.1 Table for Calculation of Total Degradation in dB . . . . . . . . . . . 175
8.2 Variation in number of sub carriers Vs SNR. . . . . . . . . . . . . . 178
8.3 LUT with basic and updated values for system with FEC=
1
2
in AWGN
Channel(Values in dB) . . . . . . . . . . . . . . . . . . . . . . . . . . 179
8.4 LUT with reference values for system with FEC =
1
2
in Fading Chan-
nel(Values in dB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
xxiii
xxiv LIST OF TABLES
8.5 LUT with updated values for system with FEC =
1
2
and PRESNR
of 10 dB in Fading Channel(Values in dB) . . . . . . . . . . . . . . . 185
8.6 LUT with updated values for system with FEC =
1
2
and PRESNR
of 20 dB in Fading Channel(Values in dB) . . . . . . . . . . . . . . . 185
8.7 LUT with updated values for system with FEC =
1
2
and PRESNR
of 25 dB in Fading Channel(Values in dB) . . . . . . . . . . . . . . . 185
8.8 Look Up Table (LUT) for dierent Doppler spread conditions for code
rate of 1/2 at BLER threshold of 10
1
. . . . . . . . . . . . . . . . . 193
A.1 Chapter wise Publications . . . . . . . . . . . . . . . . . . . . . . . . 206
B.1 Parameter description . . . . . . . . . . . . . . . . . . . . . . . . . . 208
C.1 Average SNR thresholds (in dB) for switching coding rate for dierent
rms delay spread and Doppler condition . . . . . . . . . . . . . . . . 217
Center for TeleInFrastruktur (CTIF), Aalborg University
1
Introduction
1.1 Background to OFDM
Orthogonal Frequency Division Multiplexing (OFDM) based access/multiplexing schemes
are used in wireless applications such as Wireless Personal Area Network (WPAN),
Wireless Local Area Network (WLAN), Wireless Metropolitan Area Network (WMAN),
high quality digital radio (Digital Audio Broadcasting (DAB)) and television broad-
casting (Digital Video Broadcasting (DVB)) [1]. It is being considered for IEEE 802.20,
IEEE 802.16 and 3GPP-LTE [2] systems. OFDM will remain as the key enabling
technology for achieving higher data rates in wireless packet based communication
systems in the next few years to come [3]. Its extensions with time frequency domain
spreading are under investigation for use in future wireless systems [4]. OFDM tackles
the frequency selective wireless fading channel eectively by converting a wide band
signal into a set of parallel narrow band signals such that each stream of narrow band
signal experiences at fading. With the use of cyclic prex to eliminate Inter Symbol
Interference (ISI), there is need for only a simple one tap equalizer at the OFDM re-
ceiver. OFDM brings in unparalleled gains in bandwidth savings, which leads to very
high spectral eciency. These properties make OFDM systems extremely attractive
transmission technologies for wireless scenarios.
1
2 Chapter 1. Introduction
OFDM was initially used in military systems, such as KINEPLEX in 1958,
KATHRYN in 1967, and ANDEFT in 1968 [5]. A bank of conventional transmit-
ters/receivers with overlapping spectra were used in these systems. In 1971, Wein-
stein and Eberts proposal to use the Discrete Fourier Transform (DFT) to modu-
late/demodulate all the sub-carriers, with a single oscillator [6] was a pioneering eort.
With its implementation via FFT nally OFDM was realizable in commercial commu-
nication system. It started with a number of wireline standards. High bit-rate Dig-
ital Subcarrier Lines (HDSL) [7], Asynchronous Digital Subscriber Line (ADSL)[8],
and Very High speed Digital Subscriber Line (VDSL)[8] were a sequence of standards
which led to throughput of up to 100Mb/s. Then it was introduced into the wireless
arena through DAB[9] and WLAN [10, 11]. Then came DVB[12, 13] around 2004.
In the WMAN application, OFDM is considered for the Worldwide Interoperability
for Microwave Access (WiMAX) implementation via the IEEE 802.16d,a,e [14, 15]
standards. It is also being considered for the 3GPP Long Term Evolution, which in
under development.
Table 1.1 summarizes some wireless systems which use OFDM as the trans-
mission technology [1].
Table 1.1: Wireless Systems using OFDM
Application WMAN WLAN WPAN
Cell Radius 1km to 20km up to 300m few 10s of meter
Mobility High and low Low very low
Freq Band 2-66Ghz 2-5Ghz 5-10GHz
Data Rate Few Mbps upto 100Mbps upto 10 Mbps
Deployment IEEE 802.16a, d, e,
WiMAX, 3GPP-LTE
IEEE 802.11a, g,
HiperLAN2
IEEE 802.15,
ZigBee
1.2 Motivation
Now is the juncture where wireless internet access is taking over wire line internet
access in several countries. Edinholm, who was the chief technology ocer of Nortel
Networks, predicted the exponential growth of data rate in wireline and wireless
networks [16] and said that wireless data access would eventually catch up with it
wireline counterpart but not within 2008. However, the current scenario of Mobile
users is tending to shorten the time line.
Center for TeleInFrastruktur (CTIF), Aalborg University
1.2 Motivation 3
It is said that currently there are around 2.7 billion mobile phone users. The
use of mobile phones is changing the way of life for the next generation, which has
already been signicantly changed by the Internet.In Japan, South Korea and China,
the majority of web access now comes from mobile phones, not Personal Computers
(PCs). The need to support higher and higher data rate in wireless systems can be
easily understood in this context.
Fig. 1.1 represents a common view of current and future wireless systems. Only
two dimensions are present in this picture. Power consumption is also an important
dimension along with these metrics. In current systems, there is a tradeo between
mobility, coverage and data rate. The need for next generation systems is to provide
higher data rate at high mobility conditions, but at the same time implementation
complexity of the devices must be as minimum as possible to reduce power consump-
tion. The systems must also be able to cater to a whole range of mobility conditions,
and must consider that devices with dierent capabilities will coexist in the same
network. In other words future generation systems must be able to provide higher
data rates at all mobility conditions consuming minimum power and other available
resources.
This is supported by the visionary statement It is dangerous to put limits on
wireless data rates, considering economic constraints, by Professor Ramjee Prasad
in 1999 [1]. Wireless spectrum available for commercial use is limited and expensive.
One of the main ways to support the increasing demand of wireless data services is
to push spectral eciency to its limits.
Increasing the spectral eciency of wireless communication systems is one of
the greatest challenges faced by wireless communication engineers. The available
bandwidth is scarce and costly, where as, there is a huge demand for data rate cre-
ated by increasing number of subscribers and increase in multimedia application which
require large bandwidth. Increasing the spectral eciency is the answer to this grow-
ing demand of data rate when the available bandwidth is limited. OFDM already
provides very high spectral eciency but current implementations of OFDM do not
fully exploit the capabilities of OFDM. There are still several avenues which can be
explored to increase the spectral eciency of OFDM systems even further. Therefore
the necessity to increase the spectral eciency has been a prime motivating factor
for this work.
One of the approaches to increase the spectral eciency is to design high per-
formance receivers, which leads to increase in signal processing complexity. Increase
in receiver complexity contributes to higher power consumption, and costlier compo-
c
Suvra Sekhar Das, 2007
4 Chapter 1. Introduction
Fast
Medium
Slow
Moveable
Stationary
UserMobility
TransmissionRate(Mbits/s)
10.0 100.0 1.0 0.1
3G
WLAN
802.11a/g
Hiperlan/2
MMAC
2000
4G
2010
Mobile
WiMAX
Fixed
WiMAX
3GPP-LTE
WiBro
Figure 1.1: Wireless Systems
nents. With increase in multimedia applications, there is a further addition to the
processing complexity. With limited battery power available on portable devices to
support the full range of operations for long durations the need for low complexity low
power consuming algorithms can be easily understood. To target the mass market of
wireless modules, low cost solutions have to be found, keeping in mind the tradeo
between eciency and price. Therefore it is important that the techniques to improve
spectral eciency do not increase the signal processing complexity. Hence one of the
motivating factors for this work is the need for low complexity schemes to increase
the spectral eciency of wireless systems.
1.3 Problem Denition
It is known that OFDM is spectrally very ecient and robust in dealing with the
frequency selective wireless fading channels, yet some combinations of spread spec-
trum techniques and OFDM are being considered to develop even better systems [17].
Though some works give details of their performance, it is important to compare them
against OFDM in the same test conditions, which include among others non ideal re-
Center for TeleInFrastruktur (CTIF), Aalborg University
1.3 Problem Denition 5
ceiver operations. Thus the rst problem addressed is the performance evaluations
and enhancements to multi carrier spread spectrum techniques for indoor conditions.
This analysis has been extended to the conditions when the receiver cannot obtain
perfect frequency synchronization and there exists channel estimation error.
Though OFDM brings several benets, yet its performance depends on the
channel estimation accuracy and residual phase error due to remaining uncorrected
carrier frequency osets [18]. Therefore pilot sub carriers are embedded among data
sub carriers [11] so that good estimates of these errors can be obtained. Good es-
timation of the phase errors helps in better compensation of the errors which leads
to improved performance. Since pilot sub carriers do not carry information bits they
are overhead and causes loss in bandwidth eciency. Investigation of methods of
reducing this loss in bandwidth eciency to improve the spectral eciency of the
system is thus considered as a problem area in this thesis.
One of the strengths of OFDM is its closely packed sub carriers. The sub
carrier bandwidth is a primary design parameter in OFDM systems. Carrier fre-
quency oset and Doppler frequency spread cause ICI which severely degrade the
performance of OFDM based systems. There are several algorithms to estimate and
compensate the carrier oset [19]. However, the Doppler frequency spread consists
of multiple frequency osets [20] and cannot be compensated by carrier oset com-
pensation algorithms. Algorithms to address ICI due to Doppler are very complex
from implementation point of view [21, 22]. The ratio of the maximum uncorrectable
residual carrier oset and the maximum Doppler frequency spread to the sub carrier
spacing is an important factor on which the ICI depends. The maximum value of the
ratio is usually kept within 2% [23]. Thus once the maximum distortion is known
the maximum value of sub carrier bandwidth gets decided. The larger the maximum
distortions, the larger is the value of the sub carrier bandwidth needed in order to
keep the ratio within the limits mentioned above. The useful signal duration of the
symbol is inversely proportional to the sub carrier spacing. Therefore the larger the
sub carrier spacing the smaller is the useful signal duration. The OFDM symbol
eciency can be dened as
OFDM symbol eciency =
useful symbol duration
total symbol duration
=
useful symbol duration
useful symbol duration + guard interval duration
(1.1)
c
Suvra Sekhar Das, 2007
6 Chapter 1. Introduction
The GI duration is dependent on the Root Mean Square (RMS) delay spread of the
channel and not on the residual carrier oset or the maximum Doppler frequency
spread. Therefore the design of GI length is independent of the sub carrier band-
width value. Thus the OFDM symbol eciency decreases as the useful signal dura-
tion reduces, due to increasing sub carrier bandwidth, when the GI duration is xed.
This design is based on the worst case scenario which is not always encountered. In
other situations there is a large margin in the parameter and there is an unnecessary
wastage of resources. Thus one of the problems addressed in this thesis is to inves-
tigate adaptive techniques to overcome this situation and improve upon the spectral
eciency without increasing the complexity of the User Equipment (UE).
The GI is an important designed parameter for OFDM systems. Its contri-
bution to the overhead has already been seen in (1.1). A properly chosen length of
GI prevents ISI and ICI which enables the use of single tap equalizer per sub car-
rier. Ideally the length of GI should be larger than the maximum excess delay of the
channel. Now similar to the earlier situation, the operating conditions are not that
harsh in most occasions. In such scenarios there is wastage of power and bandwidth.
If the margins were reduced, then performance would be aected since there would
be introduction of irreducible ICI and ISI when worse conditions are encountered.
Therefore several attempts to address the situation have been made towards develop-
ing high performance receivers, which can reduce the interference but these schemes
unfortunately need heavy signal processing [24],[25]. Therefore a technique is pro-
posed in this work, which can reduce the GI overhead, yet does not incur penalty in
terms of performance or receiver complexity.
Link Adaptation (LA) through the use of adaptive modulation, coding and
power control has been under active investigation to overcome the time frequency
selective fading of the wireless channel in an eective manner using feedback of channel
state information from the receiver [26]. LA becomes more complex when applied
to multi carrier systems because the degrees of freedom available for performance
improvement increase. It is vital to investigate schemes with low implementation
complexity. It may not be necessary to use all of the degrees of freedom at the same
time. Some aspects of reduced LA rate are investigated in [27], however, a much wider
scenario and several levels of adaptation are analyzed in this thesis. The tradeo with
performance loss in such mechanisms needs to be identied which is also addressed
in this thesis.
Though LA schemes promise signicant gains for OFDM based systems, the
performance is limited by the handicaps of OFDM. There have been investigations
Center for TeleInFrastruktur (CTIF), Aalborg University
1.4 Goal and Scope of the thesis 7
verifying the performance of LA due to channel feedback delay and channel estimation
error [28]. However, their tolerance to synchronization error and non linear operation
of the power amplier, which becomes signicant for OFDM with high Peak to Av-
erage Power Ratio (PAPR) properties, need a critical study. Hence nding the true
performance of OFDM systems using LA under such conditions is a prime area of
investigation of this thesis.
1.4 Goal and Scope of the thesis
This project started as a collaboration between Tata Consultancy Services, India and
Aalborg University. The objective of this project is to investigate ecient technique
for broad band wireless communication system for indoor and outdoor scenarios. The
rst goal towards the objective is to nd a suitable multiplexing scheme among OFDM
and MC-SS systems. Therefore in order to make a high eciency system the second
goal is to nd transmission mechanisms to overcome the impairments which impede
OFDM based schemes. The research focus is in the physical layer.
The environments under consideration are both indoor and outdoor scenarios.
Such conditions, would cover WLAN and WMAN applications. Under indoor and low
mobility conditions, physical specications from IEEE 802.11a,g [11] based WLAN
standard have been considered for simulations. The parameters for these include a
bandwidth of 5 MHz - 20 MHz, with a carrier frequency in the range of 2GHz to
6GHz as per the situation. The channel model used for multi path propagation is
from [29]. For outdoor conditions, physical layer system parameters closely adhere
to the WIMAX and the developing 3GPP LTE [2] specications, where bandwidth is
5MHz.
1.5 Research Methodology
The impairments that aect general OFDM systems are rst identied. The problems
have been evaluated in specic scenarios where they are more prevalent. Each scenario
usually maps to either WLAN or WMAN. Accordingly either 802.11a/g parameters
or WiMAX/3GPP-LTE parameters have been considered in respective situations.
The analysis presented in this work consists of analytical as well as computer
simulation. The system model is usually built on an analytical framework. The
appropriate channel conditions are simulated following standard channel models such
as [29],[2],[30],[31],[32], etc as applicable. Analysis is made for uncoded systems and
c
Suvra Sekhar Das, 2007
8 Chapter 1. Introduction
also considering forward error control coding. As found suitable, capacity analysis is
also done to nd the potential of a scheme. Key performance indicators have been
Bit Error Rate (BER), Block Error Rate (BLER), Frame Error Rate (FER), spectral
eciency and outage in dierent situations.
1.6 Organization of the thesis
Chapter 2: This chapter presents the technical background and introduction to
multi carrier techniques needed to explain the work in the following chapters.
It details the fundamental description of standard OFDM systems and explains
important concepts of multi carrier systems, such as orthogonality of sub car-
riers and the use of cyclic prex among others. It also describes the wireless
channel model used in the analysis.
Chapter 3: The analysis of MC-SS systems is provided, and the necessity of sub
carrier hopping for indoor conditions is evaluated in this chapter. The eect
of varying spreading gain, and symbol loading in indoor conditions is evalu-
ated here. The importance of using Successive Interference Cancelation (SIC)
receivers under such conditions is also veried. The performance comparison
of MC-SS against basic OFDM system under ICI and channel estimation error
conditions is also included in the chapter.
Chapter 4 In this chapter, the method of using semi blind pilot sub carriers to en-
hance spectral eciency of OFDM systems in WLAN environment is presented.
It brings out the possible option of loading the pilot sub carriers with informa-
tion bits without degrading bit error rate performance of the system. It also
discusses a low complexity residual phase tracking algorithm for implementing
the scheme.
Chapter 5 This chapter proposes a novel mechanism to deal with inter carrier inter-
ference generated due to Doppler eect and carrier frequency oset. Instead of
using a compensation mechanism, it proposes adaptive sub carrier bandwidth to
dynamically minimize the impact of Doppler. The chapter consists of two parts,
where the rst part deals with evaluating the system with Orthogonal Frequency
Division Multiplexing - Band Division Multiplexing (OFDM-BDM) and Orthog-
onal Frequency Division Multiplexing - Time Division Multiplexing (OFDM-
TDM) systems. The second part of the chapter presents capacity analysis for
Center for TeleInFrastruktur (CTIF), Aalborg University
1.7 Contributions of this Thesis 9
a proposed model of Orthogonal Frequency Division Multiple Access - Time
Division Multiple Access (OFDMA-TDMA) systems which use ASB.
Chapter 6 This chapter contains the proposal of using adaptive guard interval to
reduce the overhead of GI. The algorithm presented here selects the optimal
GI length based on the channel condition, residual carrier oset, required SNR
margin and received SNR condition. The analysis can be applied to both WLAN
and WMAN scenario.
Chapter 7 This chapter presents the proposal of hybrid link adaptation mechanism
in OFDM system for spectral eciency enhancement. Link adaptation dis-
cussed in this chapter includes dynamic selection of modulation, coding, and
power for the transmitted symbols. The impact of channel condition on the
selection of sub band size, modulation and coding rate adaptation interval are
investigated. It presents the performance of a proposed low complex system
using fast power adaptation along with slow modulation adaptation. Detailed
performance analysis is presented which provides guidelines for implementation
in a typical outdoor environment.
Chapter 8 Investigation of LA system in presence of non linear distortion due to
the HPA and the eect of ICI is made in this chapter. Method to overcome
these impairments is also discussed in this chapter.
Chapter 9 This is the concluding chapter of the thesis which summarizes the con-
clusions of each contributing chapters and lists the possible future works.
1.7 Contributions of this Thesis
1. Multi Carrier Spread Spectrum (MC-SS) techniques enhance the performance
of OFDM systems by increasing the frequency diversity gain through spreading
over frequency domain. However, for indoor conditions where the coherence
bandwidth and coherence time are quite large, then a set of sub carriers will
remain in deep fade for a long duration. In such a situation the proposal for
Sub-Carrier Hopped Multi Carrier Spread Spectrum (SCH-MC-SS) for indoor
systems is made in this work. The proposed SCH-MC-SS can improve the
outage performance for indoor scenarios by increasing diversity using fast sub
carrier frequency hopping over the entire system bandwidth.
c
Suvra Sekhar Das, 2007
10 Chapter 1. Introduction
2. Performance evaluation and comparison of OFDM and MC-SS for various prac-
tical operating conditions such as channel estimation error and frequency syn-
chronization errors have been made. These analysis bring out the details of
performance dierence between OFDM and MC-SS. This helps in selecting the
best multiplexing scheme in dierent situations. For this project since increas-
ing the spectral eciency is a major target, OFDM stands out as the clear
winner.
3. Pilot sub carriers used for residual phase tracking in WLAN type systems carry
pre-dened symbols and hence are overhead. Semi blind pilots have been pro-
posed which can carry data symbols and can still be used for estimating the
phase rotation due to the high SNR content of these sub carriers. This technique
of overloading the pilot sub carriers increase the spectral eciency by 5-15% by
reducing the pilot overhead in IEEE 802.11a/g type WLAN environment.
4. A low complexity algorithm for implementing the semi blind pilots is presented
in this thesis. The low complexity algorithm does not compute the phase angle
using the highly complex inverse tangent function. Instead it uses the com-
plex coecients directly for estimating the phase coecients and uses them for
compensation.
5. Proposal for Adaptive Sub carrier Bandwidth (ASB) to overcome ICI due to
Doppler frequency spread in Orthogonal Frequency Division Multiplexing -
Time Division Multiple Access (OFDM-TDMA) systems is made in this work.
6. Proposal for adaptive sub carrier bandwidth to overcome ICI due to Doppler
frequency spread in OFDMA systems is made in this thesis. The improvement
in spectral eciency obtained from these schemes is up to 30%.
7. Algorithm for dynamically variable Guard Interval (GI) is developed which
reduces the GI overhead up to 60% and contributes towards spectral eciency
enhancement by as much as 15%.
8. Proposal for hybrid link adaptation to suitably use the many degrees of free-
dom available for link adaptation by combining fast and slow adaptation of
the dierent parameters so as to gain high eciency while using less complex
system is made. The feedback overhead is reduced up to 50% by the proposed
techniques. Though primarily parameters from 3GPP LTE have been used, an
Center for TeleInFrastruktur (CTIF), Aalborg University
1.7 Contributions of this Thesis 11
investigation of these parameters has been made which provides important in-
sights into possible alternative to the ones selected till now. These results can
thus serve as important inputs to future wireless systems.
9. Problem analysis and solution for implementing Link Adaptation (LA)-OFDM
under the eect of non linear distortions introduced by High Power Amplier
(HPA) is made in this work.
10. Analysis and solution for implementing LA under practical non ideal conditions
of Inter Carrier Interference (ICI) due to Doppler frequency spread is also made.
c
Suvra Sekhar Das, 2007
12 Chapter 1. Introduction
Center for TeleInFrastruktur (CTIF), Aalborg University
2
Wireless Channel and Multi
Carrier Systems
2.1 Wireless Channel
In 1860s James Clerk Maxwell developed the fundamental laws of electromagnetic
theory and Heinrich Hertz proved the existence of electromagnetic waves in 1880s.
In the early 1890s Nicola Tesla demonstrated radio telegraphy and Alexander Popov
build his rst radio receiver in mid 1890s. Then Acharya Jagadish Chandra Bose gave
his rst public demonstration of electromagnetic waves (at millimeter wavelengths),
by using them to ring a bell remotely and to explode some gunpowder. It was
reported in The Daily Chronicle of England in 1896: The inventor (J.C. Bose)
has transmitted signals to a distance of nearly a mile and herein lies the rst and
obvious and exceedingly valuable application of this new theoretical marvel. This was
followed by the rst successful wireless signalling experiment by Marconi on Salisbury
Plain in England in May 1897 [33]. Since then through several developmental stages
we have reached an age of near ubiquitous wireless communication network.
To establish any communication system the knowledge of the channel is very
important. Its characteristics drive signal design for the system. Professor Ramjee
13
14 Chapter 2. Wireless Channel and Multi Carrier Systems
Prasads statement is worth mentioning here that understanding of the channel is
bread and butter for the communication engineer [1]. A proper understanding of
the environment leads to correct parameterizations and optimization of the target
solution. Characterization and modeling (for simulation) of the wireless channel is
necessary in situation where the error probability computation is too complicated or
might not yet have been solved. Through computer simulation we can get an idea
about the performance of the system under test in the environment it is supposed to
work in. The model to be used for computer simulation must be simple enough for
easy and fast implementation. The correctness of the model is also very important.
An erroneous characterization and modeling of the channel would lead to improper
estimates of the performance of the system. So it is essential that a proper model of
the wireless channel is used. In this chapter the wireless channel models commonly
used for system evaluation are presented. An introduction on OFDM and LA is also
included.
2.1.1 Channel Parametrization
WirelessChannelFading
Frequency
Selective
LargeScaleFading
Time
Selective
Frequency
Selective
Space
Selective
Fast Slow
SmallScaleFading
Rich
Scattering
Poor
Scattering
PathLoss
Shadowing
Flat
Figure 2.1: Characterization of Fading
The systems analyzed in this thesis are concerned with wide band for both the
indoor wireless channel and outdoor channel between 2GHz and 6GHz. The received
signal power varies as a function of space, frequency and time in the entire region of
described environment. The variation in general is classied as either sarge scale or
small scale fading. The dierent fading conditions can be largely classied as shown
in Fig. 2.1.
Center for TeleInFrastruktur (CTIF), Aalborg University
2.1 Wireless Channel 15
Large Scale Fading is dealt by propagation model that predicts the mean received
signal strength for an arbitrary transmitter receiver separation. The large scale fading
model gives such an average with measurements across 4 to 40 [34], where is the
wavelength. This is useful for estimating coverage area. Large Scale fading can be
broadly classied as path loss and shadowing. Path loss deals with the propagation
loss due to distance between transmitter and receiver while shadowing describes vari-
ation in the average signal strength due to varying environmental clutter at dierent
locations.
Small Scale Fading deals with signal strength characteristics within small distance
of the receiver location. In such region of space the average signal strength remains
constant. Multi path propagation of the electromagnetic waves is the main cause
of such eects. It includes the eect of time, space and frequency selective fading
characteristics. For each domain, there are broadly two kinds of conditions, one is
when the variability is high and the other when the variability is very small over the
observation interval.
Thus, the signal strength at a particular location depends on the large scale
fading and the small scale fading. As the receiver moves, the instantaneous power of
the received signal varies rapidly giving rise to small scale fading. In such a situation
the received power may vary by as much as 20-40dB over a range a few order of a
fraction of a wavelength depending upon the particular environment. As the distance
between the Transmitter and the Receiver increases, the local averaging of the received
signal power decreases gradually, this is predicted by the large scale fading statistics.
This phenomenon of a combined slow and fast fading is briey explained in Fig. 2.2.
P
o
w
e
r
Distance
PathLoss
Shadowing
SmallScaleFastFading
Figure 2.2: Propagation Loss
Three main factors which inuence the radio wave propagation are Reection,
c
Suvra Sekhar Das, 2007
16 Chapter 2. Wireless Channel and Multi Carrier Systems
Diraction and Scattering. Reection is caused when the Electromagnetic Waves
(EM) impinge upon surface having dimensions much larger than the wavelength of
the impinging wave. Diraction is caused due to eects of sharp edges in the path
of the radio waves between the Transmitter and the Receiver. Scattering is caused
when the EM waves encounter objects of dimension much smaller than the wave in
the propagation medium.
Most radio propagation models use a combination of empirical and analytical
methods. The empirical approach is based on tting curves or analytical expressions
that recreate a set of measured data. This has the advantage of implicitly taking into
account all the propagation factors. However the validity of an empirical model at
transmitter frequencies or environments other than those used to derive the model
can only be established by additional measured data in the new environment and
frequencies. Propagation models and multi path reection models have emerged over
time to enable easy simulation of the wireless channel [34].
2.1.2 Propagation loss
There are many models for predicting the path loss such as Hata-Okumura and
COST231-Hata model [34]. However, both these models are for frequency ranges
up to 2000 MHz. To modify this a model is proposed in [30]. For a given close in
distance
1
, d
0
of 100m, the median path loss (P
L
), in dB, is given by (2.1) [30].
P
L
= A + 10n
p
log
10
d
d
o
+ s d > d
o
, (2.1)
where
A = 20log
10
(4d
o
/), (2.2)
and n
p
is the path loss exponent and is given by,
n
p
= a bh
b
+ c/h
b
, 10m < h
b
< 80m. (2.3)
The value of a, b and c are dierent for dierent terrain types. Values for urban
area are given in Table 2.1 [32]. The shadowing factor is s and follows a log normal
1
The Friis free space model, which is the basis for large scale propagation models, is valid for
values of d which are in the far eld of the transmitting antenna, i.e. does not hold for d = 0.
Therefore, large-scale propagation models [34] use a close-distance, d
0
, as known as received power
reference point. The received power, P
r
(d), at any distance d d
0
may be calculated in relation
to that received at d
0
. The reference distance is chosen so that it lies in the distance used in the
mobile communication system.
Center for TeleInFrastruktur (CTIF), Aalborg University
2.1 Wireless Channel 17
Table 2.1: Value of parameters for urban terrain.
Parameters Urban terrain Unit
a 4.6
b 0.0075 m
1
c 12.6 m
distribution, with a typical value around 6 dB [30]. This model is proposed for a
receiver antenna height of 2 m and operating frequency of 2 GHz, and a correction
factor for other frequencies and antenna heights is proposed [32]. The modied path
loss in (2.1) is:
PL
mod
= PL + PL
f
+ PL
h
, (2.4)
where, PL is path loss given by (2.1), PL
f
is frequency correction term given by
6 log
10
(f/2000), f is the frequency in MHz and PL
h
is receiver antenna height cor-
rection term given by 10.8log
10
(h/2), where h is the new receiver antenna height
(m) such that 2 < h < 8.
The propagation model used for 3GPP-LTE system can be found in [2] where
dierent parameters have been used for dierent channel condition and cell orienta-
tion.
2.1.3 Shadowing
The path loss model does not capture the varying environmental clutter at dierent
locations. However, measurements have shown that at any value d, the path loss
PL(d) at a particular location is random and distributed log-normally (normal in
dB) about the mean distance dependent values. Since the surrounding environmental
clutter may be dierent at dierent locations the path loss will be dierent than
the average value predicted by (2.4). This variation is mainly due to refraction and
diraction o Interfering Objects (IO) in the path of the traveling signal, and is
an additive term to the path loss, with random values. This phenomenon is called
shadowing. It has a log-normal distribution about the mean path loss value [34].
Therefore the modied path loss expression is,
Pl(d) = PL(d) + X

(2.5)
= PL(d
o
) + (10n
p
)Log
10
(
d
d
o
) + X

(2.6)
Pr(d) = Pt(d) Pl(d); antenna gains included in Pl(d), (2.7)
c
Suvra Sekhar Das, 2007
18 Chapter 2. Wireless Channel and Multi Carrier Systems
where X

is zero mean Gaussian distributed random variable (in dB) with standard
deviation also in dB.
2.1.4 Small scale Fading
In small scale fading, the signal varies rapidly over a short distance. The variation
is caused by the multipath propagation of the received signal and the Doppler fre-
quency shift. The channel impulse response h(t, ) is a function of two variables, time
t and delay [35]. Due to some reecting objects such as buildings, hills, trees, etc
some delayed versions of the transmitted signal, each with dierent amplitudes (A
np
),
phases (
n
) arrive at the receiver at dierent delays (
n
). The parameters (amplitude,
phase, and delay) are random variables, and can be characterized by a channel im-
pulse response. If unit impulse is transmitted and there are N
SE
scattering elements,
then the receiver would receive N
SE
dierent signals. Therefore, the channel impulse
response would be the sum of these N
SE
scattered signals as given below [36].
h(t, ) =
N
SE

m=1
A
np,m
(t
m
) exp(j
m
). (2.8)
The channel impulse response is a function of time frequency and space [37]. A typical
channel impulse amplitude response over a region is shown in Fig. 2.3.
Amplitude
Delay
S
p
a
c
e

/
T
i
m
e
RmsDelaySpread
MeanExcessDelay
MinimumSensitivityLevel
MaximumExcessDelay
Figure 2.3: Amplitude response
The relationship between the impulse response and the transfer function of the
Center for TeleInFrastruktur (CTIF), Aalborg University
2.1 Wireless Channel 19
S(f, )
d
t
h(t, ) t
T(t,f)
H(f,f)
d
F
F
-1
F
-1
F
-1
F
-1
F F
F
Figure 2.4: channel impulse response and transfer function relationship
channel is shown in Fig. 2.4, where f
d
is Doppler frequency, is delay, t is time and
f is the Fourier domain representation of the delay.
2.1.4.1 Multipath Fading
Multipath propagation as shown in Fig. 2.5 gives rise to small scale fading in time
and frequency domain. The multipath properties of a given environment are usually
characterized by the power delay prole. Power delay prole denotes the average
power of each multipath. Figure 2.6 shows a typical power delay prole. When the
rst multipath component has the highest power then it is a Ricean channel. Where as
when, the rst path does not have the highest power which usually happens in non line
of sight scenario, then it is usually a Rayleigh channel and this is used in this thesis.
The power delay prole of a typical Rayleigh multipath propagation is shown to have
a exponential decay prole which is a commonly used model. [31]. There are several
other models which consider the cluster eect. i.e. there is a double exponential
decay, where each multipath is followed by a sequence of multipath during a very
short interval with a steeper decay constant. Another model for the delay prole has
the rst few taps with same average power followed by exponential decay [38].
The channel impulse response is an instantaneous realization of the power delay
prole. A typical method of implementing it is via the Clarks methods as in [34].
Another method is the Rice method [39]. There exists other methods such as ray
c
Suvra Sekhar Das, 2007
20 Chapter 2. Wireless Channel and Multi Carrier Systems
Figure 2.5: Multipath propagation
tracing models. The Clarks method has been mostly followed in this work. In some
cases the rayleighchan function of Matlab

has been used. In these models, each

multipath component is generated so that they follow a desired Doppler spectrum.
The Doppler spectrum can be easily integrated into the system. Though for indoor
channels most of the taps in the models are supposed to have Jakes Doppler spectrum,
the outdoor multipath channel model taps can have mixed Doppler spectrum, i.e.
while some of the taps are advocated to use Jakess spectrum the one which are
towards the tail of the power delay prole may have Gauss spectrum, details which
can be found in [39]. Due to the multipath reections a transmitted impulse gets time
dispersed, i.e. spread in time domain. A measure of this time spread phenomenon is
the mean excess delay, which is dened as [34]

m
=
_

max
0
E
|h()|
2 d
_

max
0
E
|h()|
2 d
(2.9)
where E denotes expectation,
max
is the maximum delay of the arriving multi paths,
h() is the component of the arriving multi path at a delay of . The rms delay spread
Center for TeleInFrastruktur (CTIF), Aalborg University
2.1 Wireless Channel 21
P
o
w
e
r
Delay
MaximmDelay
First
multipath
RmsDelaySpread
MeanExcessDelay
SensitivityLevel
Exponentially
decayingenvelope
Figure 2.6: A typical power delay prole
of the channel is dened as

rms
=
_

m
2
(2.10)
where

2
=
_

max
0

2
E
|h()|
2 d
_

max
0
E
|h()|
2 d
. (2.11)
The exponential power delay prole denes,
E
|h()|
2 =
e

0
_

max
0
e

0
, for 0 < <
max
= 0, elsewhere. (2.12)
In the above
0
is the characteristic of the power delay. The rms delay spread is
the average information of a certain environment, but it is expected to have a local
variation over a few hundred nano seconds [31, 40].
The small scale channel model does not generate or absorb any power. i.e.
_

max
0
[h()[
2
d = 1. (2.13)
This ensures that we are concerned only with short term multi path fading scenario.
c
Suvra Sekhar Das, 2007
22 Chapter 2. Wireless Channel and Multi Carrier Systems
The Fourier transform of the channel impulse response is the channel transfer
function, as shown in Fig. 2.7. The signal experiences dierent levels of fading for dif-
30
25
20
15
10
5
0
5
Frequency
G
a
i
n

i
n

d
B
Figure 2.7: Frequency domain channel response
ferent frequencies in the fading channel. With such characteristics, a fading channel
could be either frequency or non-frequency selective. This depends on the bandwidth
of the system compared with the channel coherence bandwidth, B
c
. The coherence
bandwidth is dened as the frequency separation f such that the correlation coe-
cient falls below a dened real value between 0 and 1. B
c
is inversely proportional to
the rms delay spread
RMS
[41, 37].
B
c

1

RMS
(2.14)
If the system bandwidth is much smaller compared to the coherence bandwidth,
then the channel is said to be frequency non selective. In this case, the correlation
coecient
2
of the sub carrier channel transfer function is almost 1 for the frequencies
within the system bandwidth. Physically, the frequency response within the system
bandwidth is almost at, so it is also called at fading. On the other hand, if the
2
E
H(f),H

(f+f)
, where H is the channel frequency response at frequency f.
Center for TeleInFrastruktur (CTIF), Aalborg University
2.2 Quadrature Amplitude Modulation 23
system bandwidth is larger than the coherence bandwidth, then the channel is said
to be frequency selective. In this case, the frequencies within the channel would
experience dierent level of fading.
2.1.4.2 Time varying channel
Similar to the channel characteristics in frequency domain, there are also signal varia-
tions in the time domain due to the time varying nature of the wireless channel. This
time varying channel is mainly caused by the movement of either the transmitter,
receiver, or the reectors, which results in the Doppler eect. With a velocity vthe
maximum Doppler shift is f
m
=
v

. The coherence time of a time varying channel is

inversely proportional to the maximum Doppler frequency. The coherence time is
dened as [34]
T
c
=
9
16f
(2.15)
where f is the maximum Doppler frequency. The above model is true for Jakes
spectrum. The coherence time for Gauss spectrum for the same velocity is much
more than in the case of Jakes Spectrum. A Typical Jakes spectrum is given in
Fig. 2.8 and a typical Gauss spectrum is given in Fig. 2.9. Ideally the correlation for
a tap of the multipath channel model is given by [42] as,
R
h(,t)h(,t+t)
= F
1
(, S(f)), (2.16)
where F
1
mean inverse Fourier Transform, S(f) is the Doppler frequency power
spectrum.
2.2 Quadrature Amplitude Modulation
As Quadrature Amplitude Modulation (QAM) is used in this work, the QAM is
described in this section. Digital Modulation is a process that maps a digital symbol
onto a signal suitable for transmission [43]. This is done in two steps: at rst, the k
bits of the digital signal are mapped to one symbol of the baseband signal and are
represented as complex constellation points. In second step the resultant baseband
signal is then up converted to the transmitting frequency via Radio Frequency (RF)
modulation. When the amplitude of the modulated signal is varied to carry the
source information, the modulation is called Pulse Amplitude Modulation (PAM).
c
Suvra Sekhar Das, 2007
24 Chapter 2. Wireless Channel and Multi Carrier Systems
300 200 100 0 100 200
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Figure 2.8: Power Spectral Density vs fre-
quency of Jakes spectrum
300 200 100 0 100 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Figure 2.9: Power Spectral Density vs fre-
quency of typical Gauss spectrum
-7-5-3-11357
-
7

-
5

-
3

-
1

7
I
Q
Figure 2.10: Signal space diagram for rectangular 64-QAM
On the other hand, the term Phase Shift Keying (PSK) implies the phase of the
modulated waveform has the source information. QAM uses the combination of both
these techniques and carries information in the phase as well as the amplitude of the
modulated waveform. This can also be seen as embedding two simultaneous sequences
of k bits information signal on two quadrature carriers cos2f
c
t and sin2f
c
t. The
corresponding modulated waveform can be written as [35]:
s
m
(t) = (A
mc
+ jA
ms
)g(t)e
j2f
c
t
m = 1, 2, ...M (2.17)
where A
mc
and A
ms
are the information-bearing signal amplitudes of the quadra-
ture carrier and g(t) is the signal pulse.
Center for TeleInFrastruktur (CTIF), Aalborg University
2.2 Quadrature Amplitude Modulation 25
The signal space diagram of rectangular QAM for dierent values of M is
shown in Figure 2.10 [35], where M = 2
k
and k is the number of information bits per
modulated symbol. It is common practice to have rectangular QAM where M = 2
2j
,
with each symbol representing 2j information bits, because it has the advantage of
being generated as superposition of two PAM signal on quadrature carriers.
c
Suvra Sekhar Das, 2007
26 Chapter 2. Wireless Channel and Multi Carrier Systems
2.3 OFDM
2.3.1 OFDM Fundamentals
OFDM is an advanced form of Frequency Division Multiplexing (FDM) where the
frequencies multiplexed are orthogonal to each other and their spectra overlap with
the neighboring carriers. In a standard FDM system the sub carriers do not overlap
as shown in Fig. 2.11 which represents the amplitude frequency response of such
systems. OFDM is built on the principle of overlapping orthogonal sub carriers. The
frequency domain view of the signal is shown in Fig. 2.12. The peak of one sub
carrier coincides with the nulls of the other sub carriers due to the orthogonality.
Thus there is no interference from other sub carriers at the peak of a desired sub
carrier even though the sub carrier spectrums overlap. It can be understood that
A
m
p
l
i
t
u
d
e
Frequency
GuardBand
Subcarrierbandwidth Subcarrierbandwidth
Figure 2.11: Non orthogonal carriers
SubCarrierPeaks
SubcarrierNulls
SubCarrierSpacing/
Bandwidth
Frequency
A
m
p
l
i
t
u
d
e
Figure 2.12: Orthogonal Sub carriers in Multi carrier systems (OFDM)
OFDM systems avoid the loss in bandwidth eciency prevalent in system using non
orthogonal carrier set. This brings in huge benet in spectral eciency for OFDM
systems over earlier systems.
Center for TeleInFrastruktur (CTIF), Aalborg University
2.3 OFDM 27
The orthogonality among sub carriers can be viewed in time domain as shown
in Fig. 2.13. Each curve represents the time domain view of the wave for a sub carrier.
The gure shows that integer number of cycles of each sub carrier is present in the
symbol duration. If such is not the case, then orthogonality is not ensured. The
mathematical explanation is that two functions f
1
and f
2
are orthogonal in an interval
(t1,t2) if the following condition is satised.
_
t=t
2
t=t
1
f
1
(t)f
2
(t) dt = 0, (2.18)
i.e. the area under the product of the two functions in the region is zero. In case of
OFDM, the frequencies are chosen so that all of them form a mutually orthogonal
set.
Wavecyclesfordifferentsubcarriers
PulseDuration
A
m
p
l
i
t
u
d
e
Figure 2.13: Time domain representation of the signal waveforms to show orthogonality
among the sub carriers
OFDM is well known for eectively combating the frequency selective fading
which arise due to multi path reections of the wireless channel. To understand this,
the OFDM transmitter architecture needs to be studied. Fig. 2.14 shows a typical
transmitter structure for the system. The Inverse Fast Fourier Transform (IFFT)
replaces the bank of modulators needed in a conventional multi carrier system. It can
be considered that the input data symbols to the IFFT modulates the sub carriers,
each of which has a pulse period which is the product of the sampling period of the
system times the number of sub carriers in the system. In other words, each data
symbol modulates one sub carrier.
During the OFDM system design, parameters are chosen in a way such that
the sub carrier bandwidth is smaller than the coherence bandwidth of the channel
so that each sub carrier experiences at fading. This together with the use of cyclic
prexed Guard Interval (GI) helps in using a one tap equalizer at the receiver.
c
Suvra Sekhar Das, 2007
28 Chapter 2. Wireless Channel and Multi Carrier Systems
BinaryInput Serial
to
Parallel
Symbol
Mapping
QPSK/
16-QAM
64-QAM
Withorwithout
FEC
IFFT
Parallel
to
Serial
Add
Cyclic
Prefix
ToRFchain
...00101110110...
(0011)
(1010)
Complexvalued
inputs
Complexvalued
outputs
Figure 2.14: Base band modules of the OFDM transmitter
The time frequency diagram of the OFDM signal in Fig. 2.15 shows the dif-
ference between single and multi carrier systems with respect to the symbol duration
when compared against the channel impulse response. Single carrier system have a
symbol duration which is decided by the sampling period of the system. When the
channel impulse response is larger than this period, there is ISI. The whole band-
width is split into a set of parallel orthogonal sub streams each of which has a long
symbol duration. The symbol duration becomes signicantly greater than the channel
impulse response length. This makes each stream, i.e. each sub carrier, experiencing
at fading as is depicted in Fig. 2.16.
The one tap equalizer can be realized if there is no ISI. ISI between consecutive
OFDM symbols is avoided by the use of GI, which is discarded at the receiver to reject
the ISI. Using Cyclic Prex (CP) in the GI is a very eective technique to preserve
the orthogonality among sub carriers which can be understood from Fig. 2.17. The
CP is an extension of the last part of the time domain signal. The length of CP
is designed to be larger than the maximum delay of the channel. After the OFDM
symbol with the CP passes through a time dispersive channel, the interference from
the previous symbol is limited to be within the CP. The contaminated CP is rejected
at the receiver. The continuity of the signal waveforms in time domain inside the GI
in the form of CP ensures that the remaining portion contains full cycles of the sub
carriers as can be seen in the picture. This prevents orthogonality loss among the
sub carriers and between consecutive OFDM symbols.
The long symbol duration which brings in several benets is limited by the
Doppler condition of the channel. When the number of sub carriers is made large,
Center for TeleInFrastruktur (CTIF), Aalborg University
2.3 OFDM 29
ChannelImpulseResponse
InterSymbolInterference
B
a
n
d
w
i
d
t
h
LargeSuymbolduration,narrowbandwidthparallelcarriers
Flatfadingovereachsubcarrier,noISI
Figure 2.15: Time Frequency representation of OFDM Signal
the system bandwidth is sampled at a higher rate in the frequency domain, which in
turn makes the sub carrier bandwidth smaller. This leads to a higher pulse duration.
To maintain orthogonality among the sub carriers OFDM systems must have a static
channel during the pulse period, i.e. the coherence time of the channel must be much
larger than the pulse period of the OFDM symbol.
The sub carrier bandwidth to be selected is also limited by the tolerance of
frequency oset due to imperfect carrier synchronization. The ratio of residual carrier
oset (due to uncompensated carrier oset because of imperfect carrier synchroniza-
tion and Doppler frequency spread) to the sub carrier bandwidth must be less than
a certain value [23] to get an acceptable BER performance. Usually this value is
considered to be around 0.02.
The receiver for OFDM follows similar sequence of operations as in the trans-
mitter but in the reverse direction. There is a front end to this part that is needed
usually. This part consists of the synchronization and channel estimation modules. A
block diagram of the OFDM receiver is given in Fig. 2.18. The synchronization and
channel estimation is usually done with the help of training sequence and embedded
pilots. A typical training sequence as used in IEEE 802.11a is given in Fig. 2.19.
c
Suvra Sekhar Das, 2007
30 Chapter 2. Wireless Channel and Multi Carrier Systems
Frequency
C
h
a
n
n
e
l

G
a
i
n

Wideband Frequency Selective Fading
Narrow Band Flat Fading
Figure 2.16: Frequency Selective and non Selective Fading
IntegerCycles
CyclicPrefix
IntegerCycles
UsefulpartofOFDMsymbol
Figure 2.17: The use of Cyclic Prex
Center for TeleInFrastruktur (CTIF), Aalborg University
2.3 OFDM 31
TimeSynch
Frequency
Synch
FFT
Channel
Equalization
Phase
Tracking
LowNoise Amplifier
L.O.
LowPass
Filter
ADC
FECDecoder
Symbol
Demapper
OutputBits
Backend(OuterReceiver)
Frontend(InnerReceiver)
AGC
AutomaticGain
Control
Signal
Figure 2.18: Top level architecture of OFDM receiver circuitry showing important signal
processing modules in base band part.
t1t2t3t4t5t6t7t8t9t10
GI2 T1 T2 GISIGNAL GIData GIData GIData
10x0.8=8 s m 2x0.8+2x3.2=8 s m
8+8=16 s m
0.8+3.2=4 s m 0.8+3.2=4 s m 0.8+3.2=4 s m 0.8+3.2=4 s m
SignalDetect
AGC,Diversity
Selection
CoarseFreq&
TimingSynch
Channel&FineFreq
Estimation
RATE
LENGTH
DATA DATA DATA
Figure 2.19: Training sequence for WLAN [44]
c
Suvra Sekhar Das, 2007
32 Chapter 2. Wireless Channel and Multi Carrier Systems
The analytical model for the OFDM system is presented in this part. The s
th
baseband transmitted OFDM symbol can be expressed from [45]
x
s
(t) =
1

T
f
N
f
2
1

k=
N
f
2
X
s
[k]e
j2
k
T
f
(tsT
s
T
gi
)

T
s
(t sT
s
), (2.19)
with
x(t) =
s=

s=
x
s
(t),
where T
f
is the duration of OFDM symbol except the guard interval, k is the sub
carrier index, N
f
denotes the number of sub carriers, X[s, k] is the modulated data
symbol on the sub carrier, T
s
is the symbol duration which is the sum of T
f
and the
guard interval T
gi
.
T
s
(t sT
s
) is the gate pulse of duration T
s
starting from t = sT
s
,
which can be implemented in digital domain [44].
After passing through the channel, the signal can be represented as,
r(t) =
_

max
0
h()e
j2f
d
(t)
x
s
(t ) d +(t) (2.20)
where
max
is the maximum tail of the channel impulse response h(), (t) is the
noise component and f
d
is the doppler frequency for delay . With perfect timing
synchronization, but residual carrier frequency oset of f
c
(Hz) , the received OFDM
symbol is
r
s
(t) = r(t)e
j2f
c
t

T
f
(t sT
s
T
gi
) (2.21)
the signal portion without the noise part is
_

max
0
h()x
s
(t )e
j2f
d
(t)
e
j2f
c
t

T
f
(t sT
s
T
gi
) d
Since h()e
j2f
d

cannot be separated from h(), we re-write the above as

r
s
(t) =
_

max
0
h()x
s
(t )e
j2(f
c
+f
d
)t

T
f
(t sT
s
T
gi
) d. (2.22)
Considering ideal conditions of no synchronization error, i.e. f
c
= 0 and near static
Center for TeleInFrastruktur (CTIF), Aalborg University
2.3 OFDM 33
channel , i.e. (f
d
0), the received sub carrier can be expressed from [45],[46] as,
R
s,k
=
1

T
f
_
sT
s
+T
gi
+T
f
sT
s
+T
gi
(r
s
(t) + (t)) e
j2
k

T
f
(tsT
s
T
gi
)
dt (2.23)
which becomes
R
s,k
= X[s, k

]H[s, k

] +[s, k

] (2.24)
where [s, k

] is the frequency domain noise component and H[s, k

] is the channel
coecient for k
th
sub carrier of s
th
OFDM symbol. The above expressions hold when
the CP is larger than the maximum tail of the channel impulse response.
Thus the OFDM framework has been developed above, and it is also discussed
that OFDM has very high spectral eciency due to close packing of the sub carriers.
It is shown that OFDM elegantly handles multipath fading eect, i.e. the frequency
selective fading eect and also needs only one tap equalizer at the receiver due to the
long symbol duration and the use of CP.
In spite of these facts OFDM has some weaknesses. The choice of GI duration
is very critical. A small value can introduce ISI and ICI thereby destroying the
orthogonality and jeopardizing the system performance. Therefore usually a large
value is selected which is greater than the largest possible channel impulse response
length expected to be encountered in the environment. This adds a high overhead
in the system. In one of the chapters of this thesis an algorithm to overcome this
problem is addressed. OFDM is also vulnerable to carrier oset errors. The choice of
sub carrier bandwidth is very critical in this case. Residual carrier oset and Doppler
frequency spread introduces ICI. For large values of the ratio of these osets to the sub
carrier bandwidth, the performance is severely aected. The sub carrier bandwidth
is therefore made large enough to sustain a certain amount of such errors. Making
the sub carrier bandwidth large in turn implies a smaller useful symbol duration.
Now, for a xed GI duration, the ratio of the useful symbol duration to the total
symbol duration (sum of GI period and useful symbol period) becomes smaller thereby
aecting the system eciency. This is also addressed in this thesis and a solution is
also proposed. Since an OFDM signal is generated as an addition of several parallel
orthogonal carriers, there is high PAPR in the signal. The non linear operation of
the power amplier in the transmitter introduces distortion. Therefore a large back
o is usually used and hence a large power ineciency has to be tolerated in the
transmitter. This also has an impact when adaptive modulation and coding are
used in OFDM systems. A method to adjust for such distortions in case of adaptive
c
Suvra Sekhar Das, 2007
34 Chapter 2. Wireless Channel and Multi Carrier Systems
modulation and coding is also addressed in this thesis.
2.3.2 Parameters values for OFDM based Standards
Some parameters for OFDM system design as used in standards are discussed here.
IEEE 802.11a,g [11] is a WLAN standard which is designed for indoor conditions. The
path loss is not a very important consideration in this situation since the propagation
distance is within 300m. The channel rms delay spread varies between 10ns and
250ns [29]. The user velocity is limited to 5 kmph. The physical layer parameter
values related to OFDM used in this standard are in Table 2.2. The sub carrier
Table 2.2: Parameters in WLAN
Parameter Values
System Bandwidth 20MHz
Carrier Frequency 2-5 GHz
Number of points in FFT 64
Sub carrier bandwidth 312.5KHz
Number of useful carriers 52
Number of null carriers 12
Number of pilot sub carriers 4
OFDM symbol duration 4 s
GI duration 0.8 s
Table 2.3: WMAN system parameters
Parameter WiMAX 3GPPLTE
Carrier frequency 2-11GHz 2GHz
Bandwidth 1.25MHz,5MHz,10MHz,20MHz1.25MHz,5MHz,10MHz,20MHz
Number of point in
FFT
128,512,1024,2048 128,512,1024,2048
Sub carrier bandwidth 10.94KHz 15KHz
GI ratio
1
4
,
1
8
,
1
16
1
4
,
1
8
,
1
16
useful part OFDM
symbol duration
91.4s 66.67s
Max transmit power 43dBm (10MHz) 38dBm (10MHz)
Minimum Frame size 2ms 0.5ms
Inter site distance in
macro cell
2.8Km 1.7Km
Number of sectors 3 3
bandwidth is very wide, and the symbol duration is also very small which allows for
Center for TeleInFrastruktur (CTIF), Aalborg University
2.3 OFDM 35
large tolerance of carrier oset and Doppler spread. It can also tolerate the other
eect of Doppler i.e. fast fading channel conditions since the symbol duration is very
small, though these conditions are not likely to occur in the indoor conditions. The
coherence bandwidth can be calculated using the formula B
c
=
1
5
rms
as 0.8MHz which
is much more than sub carrier bandwidth. Therefore each sub carrier can be expected
to experience a at fading channel.
The values for the standards IEEE 802.16a,e (WiMAX) and the 3GPP-LTE
(under development) are in Table 2.3. It can be seen that both systems have very
similar parameter values but these are drastically dierent from the WLAN standard.
These standards are more mature than WLAN system and have dierent channel
conditions. The rms delay spread of the channel can be up to 2-3s. Therefore the
GI duration and the OFDM symbol duration is chosen accordingly. The coherence
bandwidth is around 66KHz and hence the sub carrier bandwidth is kept smaller
than this which is signicantly dierent compared to WLAN systems. The maximum
transmit power at the base stations is related to the cell radius, and hence the two
systems have dierent values since the cell radius is dierent.
c
Suvra Sekhar Das, 2007
36 Chapter 2. Wireless Channel and Multi Carrier Systems
2.4 Link Adaptation
The link between the transmitter and receiver is the wireless channel. Taming the
wireless channel to gain high spectral eciency has been one of the greatest challenges
of the wireless system design engineer. There has been several ways of approaching
the problem. OFDM has been one such proposal. The wireless channel varies in the
time and frequency domain. Though OFDM is very eective, yet it is not optimized
to handle the dynamically changing conditions. The channel gain of each sub carrier
varies in time, and the channel gain varies from one sub carrier to another in one
given time as has been seen before. Therefore each sub carrier cannot have the
same performance in terms of bit error rate under these circumstances for a certain
modulation and coding rate.
The bit error rate performance of dierent modulation and coding rates are
given in Fig. 2.20. The x-axis is the SNR. In a time frequency selective fading channel,
if the same modulation and coding rate are used then the BER performance will vary
as per the channel gains. In systems using OFDM, the sub carriers experience at
fading. This implies that to each sub carrier the channel appears like Additive White
Gaussian Noise (AWGN) but with dierent Signal to Noise Ratio (SNR) levels.
In practical multi carrier link adapted systems a set of consecutive sub carriers
are usually grouped to form a sub-band, and this forms the basic unit of operation. In
order to maintain a certain BLER threshold at the receiver, (a value of 10
1
has been
shown in the gure) the modulation and coding rates need to be dynamically changed
as per the measured SNR at each sub band for each OFDM symbol. For example,
if the received SNR is below 8 dB, it is better not to transmit any signal since the
BLER will be higher than the required level, and when the SNR is between 8 dB and
11 dB, the ideal choice would be 4-QAM, rate 1/2 coding and so on. The objective
is to select the bit loading such that the required BLER is satised and maximum
spectral eciency is obtained. To realize this, a feedback system is necessary. The
system framework is explained in Fig. 2.21.
Usually there are pilot symbols transmitted which are used to estimate the
channel coecients. The channel coecients are then encoded into a Channel Quality
Index (CQI) parameter which is sent back to the transmitter via a feedback channel.
The CQI information is then used to adapt the transmit parameters which are used
in the next few time slots. The time sequence of these operations is indicated in
Fig. 2.22.
The spectral eciency of each modulation and coding rate reaches a saturation
Center for TeleInFrastruktur (CTIF), Aalborg University
2.4 Link Adaptation 37
0 5 10 15 20 25 30
2
1.5
1
0.5
0
SNR (dB)
l
o
g
1
0
(
B
L
E
R
)


M=4, C=1by2
M=16, C=1by3
M=16, C=1by2
M=16, C=2by3
M=64, C=1by2
BLER
Threshold=10
1
No
Transmission
Switching
Threshold
Figure 2.20: SNR switching threshold points for LA System. M=4 indicates QPSK, M=16
is 16-QAM, and M=64 is for 64-QAM, while C represents coding rate
point after a certain SNR. This is because the probability of being in error becomes
very small after a certain SNR, and any further decrease in BER/BLER does not
improve the spectral eciency notably. The adaptive bit loading systems select the
modulation and coding scheme which has the highest spectral eciency while sat-
isfying the target BER/BLER. Fig. 2.23 shows the spectral eciency performance
of LA. The curves with dierent markers represent the SE performance of dierent
modulation and coding. Now, if Link Adaptation (LA) is used then with the increas-
ing SNR the system will choose the modulation and coding rate, which has the best
Spectral Eciency (SE) performance at that SNR. For example, from 0 to 2.5 dB,
the only option is 4QAM with FEC =
1
2
rate and the system will transmit using
this scheme. However, after this, and until 5 dB there are two options, one is 4QAM
with FEC =
1
2
and other is 16QAM with FEC =
1
3
. The system will opt for the
previous one since it has the better spectral eciency. At the value close to 8 dB the
SE performance of 16QAM with FEC =
1
3
crosses that of 4QAM with FEC =
1
2
and
immediately the system will switch to 16QAM with FEC =
1
3
. The dashed curve
with maroon color represents the SE performance of this system.
However, if some Quality of Service (QoS) constraint is imposed for particular
c
Suvra Sekhar Das, 2007
38 Chapter 2. Wireless Channel and Multi Carrier Systems
Link Adaptation
ControlUnit
Transmitter Receiver
P
o
w
e
r
&
b
i
t

l
o
a
d
i
n
g
R
F
c
o
m
p
o
n
e
n
t
D
e
t
e
c
t
o
r
C
h
a
n
n
e
l

E
s
t
i
m
a
t
i
o
n
ChannelQualityIndicator
FeedbackUnit
CQIFeedback
TimeFrequency
SelectiveFading
Channel
Figure 2.21: Link Adaptation basic framework
ChannelEstimateatReceiver
CQIFeedback
Link AdaptedDataSymbols
PilotSymbol PilotSymbol
Link AdaptedDataSymbols
PilotSymbol
Figure 2.22: Time Sequence of Events in Link Adaptation
application which requires a certain BLER criteria to be met, then the system needs
to look at BLER rst and if the condition is met, then it will check for the option
which will maximize the SE performance. The Fig. 2.20 shows the threshold points
when the system can change from one modulation and coding scheme to another
while maintaining the BLER constraint of 10
1
. To maintain this QoS, the system
will not transmit until 8 dB SNR. The spectral eciency curve for such a case will
following the solid blue curve in Fig. 2.23
The analytical framework for adaptive bit loading is given below. The bit load
Center for TeleInFrastruktur (CTIF), Aalborg University
2.4 Link Adaptation 39
0 5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y

(
b
/
s
/
H
z
)


M=4, C=1by2
M=16, C=1by3
M=16, C=1by2
M=16, C=2by3
M=64, C=1by2
LA without QoS
LA with QoS
Figure 2.23: Spectral Eciency Gain for LA System.
Table 2.4: Switching Threshold for Link Adaptation
Modulation
& FEC
M=0,
C=0
M=4,
C=
1
2
M=16,
C=
1
3
M=16,
C=
1
2
M=16,
C=
2
3
M=64,
C=
1
2
SNR (dB) 0 8.2 10.9 14.7 18.4 20.4
estimate per sub carrier can be expressed as [26]
b
L
= 2
1
2
log
2
_
1
1.6
ln(
b
o req
0.2
)

rx
_
| (2.25)
The above expression is valid for square constellation, where the operation .| is the
oor operation. In the above, b
o req
is the target BER which is to be satised. The
BER associated with the chosen bit load is
b
o
= 0.2e

1.6
rx
2
b
L1
(2.26)
which is tight for high SNR and has been claimed to be valid within 1.5dB for 4-QAM
to 1024-QAM for bit error rate BER 10
3
. In the above expression
rx
is taken as

rx

P
X[k]
P
H[k]

(2.27)
where P
X[k]
is the transmitted power per sub carrier, P
H[k]
is the channel power per
c
Suvra Sekhar Das, 2007
40 Chapter 2. Wireless Channel and Multi Carrier Systems
sub carrier and
2

is the noise power. However, for practical purpose forward error

control coding is used in systems. Added to this, the fading channels vary over a
wide range of conditions. It is very dicult to obtain the closed form expression of
the BLER in these conditions and hence extensive simulation based evaluations are
done. The details of which are presented in later chapters. The algorithm used for
LA is given in Appendix B.
Center for TeleInFrastruktur (CTIF), Aalborg University
3
Evaluation of Hybrid Multi Carrier
Spread Spectrum
The objective of this chapter is to investigate a suitable multiple access/ multiplexing
technique for indoor channel conditions. Keeping this in view the performance of a
Multi Carrier Spread Spectrum (MC-SS) technique for indoor channel conditions is
investigated and it is proposed to use Subcarrier hopping (SCH) along with the MC-SS
technique to enhance the outage performance of the system. The MC-SS technique
is evaluated for dierent spreading gains and loading factors. The performance of
interleaved and block subcarrier grouping and Successive Interference Cancelation
(SIC) receiver is also investigated in this chapter.
Further, the performance of MC-SS technique is compared against OFDM. The
comparison is carried out under channel estimation and frequency synchronization
errors. The proposed sub subcarrier hopping MC-SS technique improves the outage
performance in indoor conditions and OFDM is found to be more robust to receiver
impairments when compared against MC-SS under full load conditions.
41
42 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
3.1 Sub Carrier Hopping Multi Carrier Spread Spec-
trum
3.1.1 Introduction
The search for a suitable access/ multiplexing scheme for next generation wireless
networks has produced several combinations of spread spectrum and multi carrier
techniques, details of which can be found in [17],[47],[48],[49],[50],[51]. Each of them
is specialized to deal with a particular channel condition. The motivation to combine
these two is to harness the benet of two very successful physical layer technologies
of the recent years. Both spread spectrum and OFDM techniques are suitable for
multi path environment. Spread spectrum techniques such as Code Division Multiple
Access (CDMA) combines the frequency selective fading from multi path reections
while OFDM avoids it. Further, CDMA oers the exibility of fractional loading
which is eective in achieving better BER at the cost of throughput, while OFDM
provides a very high spectral eciency. Again, on one hand CDMA has provided very
good low data rate connectivity for highly mobile users (voice), on the other hand
OFDM has been known to provide very high data rate for low mobility applications
such as WLAN systems [11]. This leads to investigation of possible combinations
which can be very robust yet have high throughput along with several other exibility
options.
In this section SCH-MC-SS is proposed and described for use in indoor down
link environment. It is compared against its parent MC-SS scheme(Orthogonal Fre-
quency Division Multiple Acess with Code Division Multiplexing (OFDMA-CDM)
which can also be called Multi Carrier Spread Spectrum Multiple Access (MC-SS-
MA) [17]), which is known to be better than other such combinations as it avoids multi
user interference [52]. The outage throughput and the mean throughput as experi-
enced by a User Equipment (UE) for both schemes are compared through simulation.
Results are presented for dierent channel conditions following COST259 [29]. Based
on the simulation results, guidelines for system implementation are also discussed.
3.1.2 System Description
Primary characteristics of indoor channels are the rich multipath environment and the
quasi static nature, i.e. channels remain almost constant over long duration of time.
Multi carrier schemes are very eective in such scenarios and have been adopted in
wireless local area networks [11] and metropolitan area network [14]. To support mul-
Center for TeleInFrastruktur (CTIF), Aalborg University
3.1 Sub Carrier Hopping Multi Carrier Spread Spectrum 43
tiple users, either frequency division multiple access or code division multiple access
can be used in conjunction with multi carrier systems. Some important contributions
are [47][51]. OFDMA-CDM [17] which is a special MC-SS scheme, is supposed to
be one of the optimal solutions. It avoids multi user interference by using frequency
division multiple access while it uses frequency diversity by taking advantage of code
division multiplexing by spreading one users data over the allocated sub carrier set.
However, this system will suer from low outage throughput particularly in quasi
static channel conditions. This is because a user is given a set of consecutive sub
carriers which span only a part of the entire bandwidth. If the set of sub carriers
allocated to a user during one packet (number of consecutive OFDM symbols) are
in deep fade, they will remain in deep fade for a long duration. To improve the
situation, a highly ecient dynamic channel allocation scheme for multiple users is
needed. Such schemes are very complex, require high amount of signalling overhead
and still might not be optimal.
As an alternative to overcome the problem of outage, the use sub carrier hop-
ping over a large bandwidth in addition to the MC-SS [17] is proposed in this work.
The sub carrier hopping can follow a very elementary cyclic step increment. It will be
shown here, that the use of the proposed SCH-MC-SS scheme signicantly improves
the outage performance over the non hopping system. The transmitter architecture
for such a scheme can be represented as in Fig. 3.1. As is shown in the diagram,
P/S
Add
CP
User1
S/P
S/P
Spreader,
code1
Spreader,
code2
Spreader,
code Vu
+
I
n
t
e
r
l
e
a
v
e
r
I
F
F
T
S/P
S/P
d
1
1
d
1
V
+
s
u
b
c
a
r
r
i
e
r
h
o
p
p
i
n
g
Figure 3.1: Transmitter for SCH-MC-SS
data symbols from a user are rst converted from serial to parallel. Then each sym-
c
Suvra Sekhar Das, 2007
44 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
bol is spread. Then chips of the spread symbol are added and put on a sub carrier
after interleaving. The interleaver is followed by the IFFT, as is used in OFDM sys-
tems. The output is converted from parallel to serial. This is followed by the usual
addition of guard interval before up conversion. Fig. 3.2 shows the corresponding
receiver architecture, where the reverse chronology of events of the transmitter occur.
In addition there is a frequency domain channel equalizer [53]. The time frequency
User1
P/S
D
e
-
I
n
t
e
r
l
e
a
v
e
r
S/P
d
1
1
d
1
V
P/S
P/S
P/S
De-Spreader,
code1
De-Spreader,
code2
De-Spreader,
codev
s
y
n
c
h
r
o
n
i
z
e
d

d
e
-
r
o
t
a
t
i
o
n
C
h
a
n
n
e
l

E
q
u
a
l
i
z
a
t
i
o
n
F
F
T
Figure 3.2: Receiver for SCH-MC-SS
S
u
b

c
a
r
r
i
e
r
s
OneOFDMPacket
Time,OFDMsymbols
DataSubcarriersofauser Pilot Tones
Figure 3.3: Time frequency diagram of the sub carrier hopping scheme. It is shown that a
sub carrier allocated, which is represented by a particular colour shade, moves in time and
frequency grid in one packet.
diagram for the proposed SCH for a particular user is presented in Fig. 3.3.
Center for TeleInFrastruktur (CTIF), Aalborg University
3.1 Sub Carrier Hopping Multi Carrier Spread Spectrum 45
3.1.3 Analytical Model
OFDM is the basic system on which the code division multiplexing and sub carrier
hopping is used. For the s
th
OFDM symbol of u
th
user device a set of sub carriers is
allocated, which is indicated by u
s
, where [u[, denotes the number of sub carriers
allocated to the user device. /
u,s
n is a mapping function that maps sequential index
from 0 to [u[ 1 onto all sub carrier indices in u
s
. It is dened as,
/
u,s
n u
s
; n 0, 1, . . . , [u[ 1 (3.1)
/
u,s
n ,= /
u,s
n

, for n ,= n

. (3.2)
Let v denote the data symbol index in the range 0 to V
u
-1 where V
u
denote the
number of data symbols transmitted by user u in one OFDM symbol, that is kept
constant during one transmission burst. Let c
u,v
[n] be the n
th
chip of v
th
code of u
th
user. W
u
is the code length of user u. The n
th
chip of u
th
user, for s
th
OFDM symbol
can be written as

u,
[n] =
V
u
1

v=0
d
u
[sV
u
+ v]c
u,v
[n]. (3.3)
The n
th
chip of u
th
user, for s
th
OFDM symbol
u,
[n] maps to the sub carrier symbol
X
s
[k] and hence
u,
[n] = X
s
[k]. The system needs to be designed such that the guard
interval is larger than the delay spread of the channel, this is to avoid inter symbol
interference. Another design criteria for OFDM systems is that each sub carrier
experiences a at fading channel. In such situation the expression for a received
sub carrier in baseband assuming ideal synchronization can be written as Y
s
[k] =
H
s
[k]X
s
[k] + N [38] where N is the noise in one sub carrier bandwidth. With Z
s
[k] as
the equalizer coecient for the k
th
sub carrier for s
th
OFDM symbol, the estimated
subcarrier is
R
s
[k] = Z
s
[k]H
s
[k]
V
u
1

v=0
d
u
[sV
u
+ v]c
u,v
[(/
1
u,s
k)] + Z
s
[k]N. (3.4)
The recovered v
th
data symbol of s
th
OFDM symbol of u
th
user is given as

d
u,s
[ v] =

k:ku
s
Z
s
[k]H
s
[k]
V
u
1

v=0
d
u
[sV
u
+v]c
u,v
[(/
1
u,s
k)]c
u, v
[(/
1
u,s
k)]

+ .
(3.5)
c
Suvra Sekhar Das, 2007
46 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
where represents the processed noise (due to combining after multiplying with the
spreading code). The above expression can be broken into two parts, the useful data
symbol part and the inter code interference part. The useful part is,

d
u,s
[ v]
I
= d
u
[sV
u
+ v]
W
u
1

n=0
Z
s
[/
u,s
n]H
s
[/
u,s
n], (3.6)
and the total noise term, which includes the processed noise and interference due to
orthogonality loss of the codes in the frequency selective fading channel is

d
u,s
[ v]
II
=

k
Z
s
[k]H
s
[k]
V
u
1

v= v
d
u
[sV
u
+ v] .c
u,v
[(/
1
u,s
k)]c
u, v
[(/
1
u,s
k)]

. .
=inter code interference
+,
(3.7)
where
=

ku
s
Z
s
[k]Nc
u, v
[(/
1
u,s
k)]

. (3.8)
The proposed sub carrier hopping is involved in the above expressions by the reference
to the OFDM symbol index s. By choosing dierent u
s
for dierent OFDM symbols,
sub carrier hopping can be ensured.
Assuming BPSK transmit symbol, given that transmitted data symbol is d
u
[sV
u
+
v] =

c
s
, the probability of error is Pr(error[Z
s
[k]H
s
[k]
ku
s
)
= Pr
_

d
u,s
[ v]
I
<

d
u,s
[ v]
II
_
, (3.9)
= Q
_
c
s

W
2
u

k:ku
s
Z
s
[k]H
s
[k]
_
, (3.10)
where Q(x) =
1
2
erfc(
x

2
), where erfc is the complementary error function [35]. It can
be easily extended to QPSK and other modulations. The performance of the sys-
tem depends on the type of equalizationcombining. Dierent options are presented
in [54]. The expression for
2

(which is the variance of in (3.7)) and the term in

the summation changes with each type of channel equalization Z
s
[k] and chip com-
bining technique. The performance will also depend on the frequency correlation of
the channel. For realistic frequency selective correlated fading channel conditions, it
is dicult to assess the performance of the system without using real channel mod-
Center for TeleInFrastruktur (CTIF), Aalborg University
3.1 Sub Carrier Hopping Multi Carrier Spread Spectrum 47
els. Extensive simulations are done with COST259 channel parameters to obtain
the mean and the 10% outage throughput for each user for the hopping and the non
hopping systems. The expression used for computing the normalized throughput is
1 Ber.
3.1.4 Simulation Results and Discussion
Down link indoor COST259 channel models with rms delay spread of 50 ns (referred
as channel 2) and 250 ns (referred as channel 6) are used. The channel is assumed
static during one packet. This is a valid assumption knowing that indoor velocities
do not exceed 3 kmph and the packet length is limited to 1000 octets. It is considered
that each packet experiences independent channel fading.
The total number of sub carriers is set to 64. WalshHadamard orthogonal
spreading codes are used. The spreading gain is kept equal to the number of sub
carriers in one group. Number of allocated subcarriers per user device tested are 4, 8
and 16.
Successive Interference Cancelation (SIC) receiver is also considered in some
cases. For such situations loading is kept as 75%. A loading of 75% means that,
when there are 16 sub carriers in a group, i.e. the spreading gain is 16, then only
12 data symbols are loaded, i.e. only 12 spreading codes are used. At the receiver,
Minimum Mean Square Error (MMSE)-Equal Gain Combining (EGC) is used for chip
combining.
Two kind of sub carrier grouping are considered. One is block which groups
a set of consecutive sub carriers together, i.e. the chips of a code are spread over a
set of consecutive sub carriers. The other type of sub carrier grouping considered is
interleaved. In interleaved sub carrier grouping, the set of sub carriers which form a
group are distributed in the entire bandwidth with equal distance between neighboring
sub carriers of the same group.
In all cases uncoded system is used. Modulation used is QPSK. The following
are to be noted for the gures below,
Sub-Carrier Hopped Orthogonal Frequency Division Multiple Access with Code
Division Multiplexing (SCH-OFDMA-CDM) is the proposed sub carrier hop-
ping MC-SS scheme
OFDMA-CDM is the parent MC-SS scheme
intr represents interleaved sub carrier arrangement
c
Suvra Sekhar Das, 2007
48 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
blk implies block sub carrier arrangement
SIC represents successive interference cancelation receiver
3.1.4.1 Channel model 2, rms delay spread 50 ns
In this channel the 50% coherence bandwidth spans on an average up to 13 sub
carriers for system using 20 MHz bandwidth.
Figure 3.4 shows the mean throughput of the sub carrier hopped and non
hopped system. It can be seen from the gure that the mean throughput is almost
4 6 8 10 12 14 16
0.7
0.75
0.8
0.85
0.9
0.95
1
spreading gain
M
e
a
n

t
h
r
o
u
g
h
p
u
t


(
b
i
t
s
/
s
e
c
/
H
z
)
ofdma cdm intr 8dB
SCH ofdma cdm intr 8dB
ofdma cdm blk 8dB
SCH ofdma cdm blk 8dB
ofdma cdm intr 14dB
SCH ofdma cdm intr 14dB
ofdma cdm blk 14dB
SCH ofdma cdm blk 14dB
Figure 3.4: Mean throughput Vs spreading gain at dierent SNRs with single symbol
detection, for channel model 2.
intr represents interleaved sub carrier arrangement and
blk implies block sub carrier arrangement.
OFDMA CDM is the parent MC-SS scheme, while SCH OFDMA CDM is the proposed sub
carrier hopping scheme.
same for both the proposed SCH and the parent non hopping schemes. The mean
throughput for the block and interleaved sub carrier grouping is also same at a given
SNR. This is because the BER shown in the gure is the average over several possible
instantaneous channel realizations. It is observed that with increasing spreading
Center for TeleInFrastruktur (CTIF), Aalborg University
3.1 Sub Carrier Hopping Multi Carrier Spread Spectrum 49
gain the mean throughput is decreasing. It must be remembered that since loading
fraction is kept constant (100%), with the increase in spreading gain, the number of
data symbols loaded also increases. With increasing spreading gain the number of
codes in use is also increasing since full loading (100%) is considered in this case.
This performance degradation is because of increasing inter code interference as the
the orthogonality of the codes are lost due to the frequency selective fading channel.
4 6 8 10 12 14 16
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
spreading gain
1
0
%

o
u
t
a
g
e

t
h
r
o
u
g
h
p
u
t

(
b
i
t
s
/
s
e
c
/
H
z
)
ofdma cdm intr 8dB
SCH ofdma cdm intr 8dB
ofdma cdm blk 8dB
SCH ofdma cdm blk 8dB
ofdma cdm intr 14dB
SCH ofdma cdm intr 14dB
ofdma cdm blk 14dB
SCH ofdma cdm blk 14dB
Figure 3.5: 10% outage throughput Vs spreading gain at dierent SNRs with single symbol
detection, for channel 2
Figure 3.5 shows the 10% outage throughput
1
for the proposed SCH scheme
and the parent non hopping scheme for both interleaved and block sub carrier ar-
rangement. It can be seen that for block sub carrier assignment at small spreading
gain, the improvement with sub carrier hopping is about 15%. Therefore the proposed
SCH brings the benet it has been used for, i.e. to improve the outage. The pro-
posed SCH increases the frequency diversity of the system which provides the benet
of improved outage performance. However it is seen that if interleaved sub carrier
grouping is considered the proposed SCH and the non hopped scheme have similar
1
X% outage throughput is dened as the throughput below which the system is for X% of the
time [55].
c
Suvra Sekhar Das, 2007
50 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
4 6 8 10 12 14 16
0.965
0.97
0.975
0.98
0.985
0.99
0.995
1
1.005
spreading gain
M
e
a
n

t
h
r
o
u
g
h
p
u
t

(
b
i
t
s
/
s
e
c
/
H
z
)
ofdma cdm intr 20dB
SIC ofdma cdm intr 20dB
ofdma cdm blk 20dB
SIC ofdma cdm blk 20dB
ofdma cdm intr 30dB
SIC ofdma cdm intr 30dB
ofdma cdm blk 30dB
SIC ofdma cdm blk 30dB
Figure 3.6: Mean throughput Vs spreading gain at dierent SNRs with successive interfer-
ence cancelation, for channel 2
performance. The interleaved sub carrier grouping attains the maximum possible
frequency diversity and hence SCH does not increase the outage any further.
Figure 3.6 compares the mean throughput of using a SIC and one using a
single symbol detector. Both systems were used with sub carrier hopping. The
loading considered is 75%. It is seen that SIC is eective for both interleaved and
block sub carrier arrangement. It is also seen that SIC is useful for large spreading
gains. This is because the diversity order increases with spreading gain.
Figure 3.7 compares the 10% outage performance of using SIC against using
a single symbol detector with block and interleaved assignment of sub carriers. The
performance is very similar to that of the mean throughput.
Center for TeleInFrastruktur (CTIF), Aalborg University
3.1 Sub Carrier Hopping Multi Carrier Spread Spectrum 51
4 6 8 10 12 14 16
0.93
0.94
0.95
0.96
0.97
0.98
0.99
1
spreading gain
1
0
%

o
u
t
a
g
e

t
h
r
o
u
g
h
p
u
t

(
b
i
t
s
/
s
e
c
/
H
z
)
ofdma cdm intr 20dB
SIC ofdma cdm intr 20dB
ofdma cdm blk 20dB
SIC ofdma cdm blk 20dB
ofdma cdm intr 30dB
SIC ofdma cdm intr 30dB
ofdma cdm blk 30dB
SIC ofdma cdm blk 30dB
Figure 3.7: 10% outage throughput Vs spreading gain at dierent SNRs with successive
interference cancelation, for channel 2
Recommendations that can be extracted from these results are
If block assignment is used, at low spreading gain, the proposed sub carrier
hopping is necessary to improve performance. Gain of about 15% in outage
throughput is achieved in this case.
If interleaved sub carrier allocation is used, there is no need for the proposed
sub carrier hopping as they have same performance.
At high spreading gains, hopping is not necessary.
SIC receiver improves performance especially for the high spreading gains, it is
a matter of complexity that may decide its use.
For low spreading gains, SIC does not give much improvement and hence SIC
is not recommended. At high spreading gains, if interleaved allocation is used,
SIC is suggested for better performance. But, if block assignment is used, SIC
has performance only slightly better than single symbol detector and hence may
not be used.
c
Suvra Sekhar Das, 2007
52 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
4 6 8 10 12 14 16
0.7
0.75
0.8
0.85
0.9
0.95
spreading gain
M
e
a
n

t
h
r
o
u
g
h
p
u
t


(
b
i
t
s
/
s
e
c
/
H
z
)
ofdma cdm intr 8dB
SCH ofdma cdm intr 8dB
ofdma cdm blk 8dB
SCH ofdma cdm blk 8dB
ofdma cdm intr 14dB
SCH ofdma cdm intr 14dB
ofdma cdm blk 14dB
SCH ofdma cdm blk 14dB
Figure 3.8: Mean throughput Vs spreading gain at dierent SNRs with single symbol
detection, for channel 6
3.1.4.2 Channel model 6, rms delay spread 250 ns
In this channel the 50% coherence bandwidth spans on an average up to 3 sub carriers.
Figure 3.8 shows the mean throughput of the proposed sub carrier hopped and non
hopped system for interleaved and block sub carrier allocation. It shows that at a
given spreading gain, they have almost same performance. This result is almost same
as was observed for channel 2.
Figure 3.9 shows the 10% outage throughput for the same conditions as above.
It can be seen that that the performance is very similar to that of model 2.
Figure 3.10 compares the performance of a SIC and nonSIC scheme for sub
carrier hopping scheme. Once again the earlier conclusions from the previous channel
condition hold here.
Figure 3.11 shows outage performance for the same scenario as the previous
case. Here again, it is seen that the channel with rms delay spread of 250 ns also have
similar eect as does the channel with 50 ns.
Center for TeleInFrastruktur (CTIF), Aalborg University
3.1 Sub Carrier Hopping Multi Carrier Spread Spectrum 53
4 6 8 10 12 14 16
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
spreading gain
1
0
%

o
u
t
a
g
e

t
h
r
o
u
g
h
p
u
t


(
b
i
t
s
/
s
e
c
/
H
z
)
ofdma cdm intr 8dB
SCH ofdma cdm intr 8dB
ofdma cdm blk 8dB
SCH ofdma cdm blk 8dB
ofdma cdm intr 14dB
SCH ofdma cdm intr 14dB
ofdma cdm blk 14dB
SCH ofdma cdm blk 14dB
Figure 3.9: 10% outage throughput Vs spreading gain at dierent SNRs with single symbol
detection, for channel 6
c
Suvra Sekhar Das, 2007
54 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
4 6 8 10 12 14 16
0.95
0.955
0.96
0.965
0.97
0.975
0.98
0.985
0.99
0.995
1
spreading gain
M
e
a
n

t
h
r
o
u
g
h
p
u
t

(
b
i
t
s
/
s
e
c
/
H
z
)
ofdma cdm intr 20dB
SIC ofdma cdm intr 20dB
ofdma cdm blk 20dB
SIC ofdma cdm blk 20dB
ofdma cdm intr 30dB
SIC ofdma cdm intr 30dB
ofdma cdm blk 30dB
SIC ofdma cdm blk 30dB
Figure 3.10: Mean throughput Vs spreading gain at dierent SNRs with successive inter-
ference cancelation, for channel 6
Center for TeleInFrastruktur (CTIF), Aalborg University
3.1 Sub Carrier Hopping Multi Carrier Spread Spectrum 55
4 6 8 10 12 14 16
0.92
0.93
0.94
0.95
0.96
0.97
0.98
0.99
1
spreading gain
1
0
%

o
u
t
a
g
e

t
h
r
o
u
g
h
p
u
t

(
b
i
t
s
/
s
e
c
/
H
z
)
ofdma cdm intr 20dB
SIC ofdma cdm intr 20dB
ofdma cdm blk 20dB
SIC ofdma cdm blk 20dB
ofdma cdm intr 30dB
SIC ofdma cdm intr 30dB
ofdma cdm blk 30dB
SIC ofdma cdm blk 30dB
Figure 3.11: 10% outage throughput Vs spreading gain at dierent SNRs with successive
interference cancelation, for channel 6
c
Suvra Sekhar Das, 2007
56 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
Recommendations that can be made for these conditions are as follows:
The gain in outage throughput with hopping in block assignment is about 15%.
At low spreading gain, block assignment with the proposed hopping has same
performance as interleaved assignment.
With high spreading gain, sub carrier hopping may be avoided, but interleaved
sub carrier assignment is suggested.
For low spreading gain, SIC is not needed for sub carrier hopped systems.
For higher spreading gains, SIC gives improvement in both mean and outage
throughput.
3.1.5 Conclusion
The goal of the study was to analyze the performance of the proposed SCH MC-SS
and compare it against its parent nonhopped scheme for indoor conditions. It is
found that the proposed hopping scheme improves the outage performance for block
sub carrier grouping by about 15%. As part of the analysis the need for SIC receivers
was also made. It is seen that SIC improves the performance for higher spreading
gain while it is not necessary for small spreading gains. It can also be concluded from
the analysis that using interleaved sub carriers rather than block assignment of sub
carriers is shown to provide a benet of about 8% in outage throughput. It is seen
that none of the above conclusions are universal. Each situation of spreading gain,
sub carrier grouping demands individual attention. Therefore it can be said that
the schemes need to be adaptive and choose the best option considering the system
parameters and channel condition.
Center for TeleInFrastruktur (CTIF), Aalborg University
3.2 MC-SS with receiver impairments 57
3.2 MC-SS with receiver impairments
3.2.1 Introduction
Factors which inuence the performance of the MC-SS schemes, other than the ones
considered in the previous section, are the non ideal receiver characteristics. Several
works mentioned in the previous section have proposed many new schemes under
synchronized environment and with perfect channel estimate at the receiver. Other
works analyze the performance of the individual access/ multiplexing schemes with
receiver impairments with independent assumptions [56, 45]. Since both channel
conditions and receiver impairments inuence the performance of the access / mul-
tiplexing schemes, it is highly important to evaluate them in a unied framework to
decide on the appropriate scheme for the particular situation.
The objective of this section is to analyze the potential multiplexing scheme
among OFDM and MC-SS, for dierent spreading gains and load conditions under
carrier frequency oset (due to mismatch in local oscillators) and channel estimation
error in the realistic quasi static indoor environment. A comparison of OFDM and
MC-SS was made in [57, 58], however it does not delve into the system design anal-
ysis of the impact of spreading gain, loading as made in this chapter. Performance
comparison of OFDM and MC-SS schemes was made at the same time as this work
in this chapter in [59] but the article concentrates on cellular scenario and the focus
is not on receiver impairments.
3.2.2 System Description
The MC-SS scheme has been described in the previous section in details. Further
important considerations are made here in continuation to the earlier description.
Loading is the ratio of the number of codes used in a group of sub carriers, to the
length of the spreading gain; i.e. the ratio of the number of modulated symbol in a
code group to the spreading gain.
The MC-SS techniques use OFDM as the basic system and apply code division
multiplexing on it, as described earlier. The general system model for all the schemes
is developed in Section 3.1.1.
The time domain signal of the s
th
transmitted OFDM symbol is dened in
(2.19) through (2.22). Now, considering that channel coecients remain static over a
time period less than the coherence time period of the channel, f
c
+f
d
can be termed
as eective carrier oset and represented as f. The relative oset, i.e. the ratio of
c
Suvra Sekhar Das, 2007
58 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
the eective oset to the sub carrier spacing can be dened as
f
f
sc
, where f
sc
is the sub carrier bandwidth. The change in channel coecients can be considered
over a period greater than the 90% coherence time. The received sub carrier can be
computed as [45],[46],
R
s
_
k

_
=
1

T
f
_
sT
s
+T
gi
+T
f
sT
s
+T
gi
(r(t)
T
f
(t sT
s
T
gi
)
+ (t)) e
j2
k

T
f
(tsT
s
T
gi
)
e
j2
t
T
f

dt
=
1

T
f
_

T
f
(...)
__

max
0
h(; t)x(t ) d
+(t)) e
j2
k

T
f
(tsT
s
T
gi
)
e
j2
t
T
f

dt d. (3.11)
With Z
s
[k] as the equalizer coecient for the k
th
sub carrier for s
th
OFDM symbol,
the estimated subcarrier is

X
s
[k] = Z
s
[k]H
s
[k]R
s
_
k

_
. (3.12)
The received sub carrier has a useful component and a noise component, which is
made of processed additive noise and inter carrier interference (ICI). The useful part
of the sub carrier is
Z
s
[k

]H
s
[k

]X
s
[k

]e
(
j2
T
f
(sT
s
+T
gi
)
k

,k

,
)
e
(
k

,k

,
)
sinc(
k

,k

,
), (3.13)
where
k,k

,
= k k

+. The inter carrier interference component is

k,k=k

Z
s
[k]H
s
[k]X
s
[k]e
(
j2
T
f
(sT
s
+T
gi
)
k,k

,
)
e
(
k,k

,
)
sinc(
k,k

,
). (3.14)
The recovered v
th
data symbol of s
th
OFDM symbol of u
th
user is given in (3.5). The
useful data symbol part (decision variable) can be written as

d
u,s
[ v]
I
=d
u
[s] v

,k

u
u
Z
s
[k

]H
s
[k

]e
(
j2
T
f
(sT
s
+T
gi
)
k

,k

,
)
e
(
k

,k

,
)
sinc(
k

,k

,
). (3.15)
The combined inter code interference and noise is given in (3.7). Sub carrier hopping
can be ensured by appropriately choosing the sub carrier mapping for each OFDM
symbol in the above equations. Assuming BPSK transmit symbol, given that trans-
mitted data symbol is d
u
[s]v =

c
s
the probability of error computation will follow
Center for TeleInFrastruktur (CTIF), Aalborg University
3.2 MC-SS with receiver impairments 59
that as has been mentioned in (3.9) and (3.10). Simulations are done to obtain the
mean bit error rate (BER) and the 10% outage throughput for all schemes under
dierent test conditions.
3.2.3 Simulation Environment, Results and Discussion
Simulation results are presents for down link indoor environment. COST259 channel
model with 50 ns of rms delay spread is used. The channel is assumed to be static
during one packet. This is a valid assumption knowing that indoor velocities do not
exceed 3 kmph and that packet length is limited. The channel impulse response for
each burst of packet transmission is considered to be independent.
Total number of sub carriers is set to 256 as per a test parameter set used in the
6
th
framework EU funded project MAGNET. WalshHadamard orthogonal spreading
codes are used for the MC-SS schemes. The spreading gain is varied between 4 and 64.
The loading for MC-SS is varied between 50% and 100%. The OFDM system is always
considered to be fully loaded, i.e. all sub carriers have data symbols. There is no
spreading gain for OFDM. Orthogonality restoration combining is used. Consecutive
sub carrier assignment is done. In all cases uncoded system, i.e. no error correcting
code is considered. The modulation used in QPSK.
Since a relative carrier frequency oset of 0.02 does not introduce signicant
distortion, therefore to study its impact, the residual carrier frequency oset relative
to the sub carrier bandwidth is kept at 0.04. The channel estimation error is modeled
as gaussian noise added to the true channel coecients. The noise added to the
channel coecients to model the channel estimation error is as per the SNR condition.
Ideal Conditions Figure 3.12 presents the performance comparison of OFDM
against MC-SS scheme with dierent spread gains. The loading is taken as 100%, i.e.
when the spreading gain is 16, there are 16 sub carriers, and the number of data sym-
bols spread over these 16 sub carriers is 16. It can be seen from the gure that OFDM
has the lowest BER performance. The MC-SS scheme with spreading gain of 4, has
almost same performance as that of OFDM. When the spreading is 16, the perfor-
mance degrades but only slightly and when the spreading gain is 64 the performance
is the worst. It is mentioned that the coherence bandwidth spans approximately 13
sub carriers for the 50ns rms delay spread channel condition. Therefore as long as
the spreading gain is near to this number, the codes do not loose the orthogonality
due to the atness of the channel in the frequency domain over which the chips are
spread. At the same time there is not much frequency diversity gain as there is hardly
c
Suvra Sekhar Das, 2007
60 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
0.001
0.01
0.1
1
0 5 10 15 20 25
B
E
R
SNR in dB
OFDM
MC-SS(4)
MC-SS(16)
MC-SS(64)
Figure 3.12: Eect of dierent spreading gain for ideal receiver conditions. The numbers in
the bracket, e.g. MC-SS(4) indicates the spreading gain.
any variation in the channel gain over the bandwidth the data symbol is spread. The
performance deterioration at high spreading gain is due to the loss in orthogonality
of the codes because of frequency selective fading. However up to a spreading gain
of 16, i.e. which covers 16 sub carriers, there is not much loss in orthogonality as
the coherence bandwidth is very high for the channel under consideration. When the
spreading gain is 64, one hand there is diversity gain due to spreading, on the other
hand there is orthogonality loss between the codes. When the loss in orthogonality
is higher than the diversity gain, there is performance loss.
Figure 3.13 shows the performance comparison of the schemes, under dierent
load conditions for ideal channel estimation and synchronization. OFDM is always
considered to be fully loaded, i.e. all its sub carriers carry data symbols. However
for the MC-SS when 50% loading is considered, it means that if the spreading gain
is 16, then only 8 data symbols are loaded i.e. only 8 code sequences out of 16 are
used. It can be seen that MC-SS with 50% loading is the best in terms of BER. The
improvement in performance comes from the redundant data transmitted by using
spreading with a less than 100% loading. This has some eect similar to the Forward
Error Correction (FEC) coding. However , the performance of OFDM is as good as
that of MC-SS with 75% loading. Therefore the BER performance of MC-SS schemes
are found to improve with decreasing load condition, but only upto a loading of 75%.
If the loading is increased, it is expected to perform worse than OFDM. Though the
Center for TeleInFrastruktur (CTIF), Aalborg University
3.2 MC-SS with receiver impairments 61
0.001
0.01
0.1
1
0 5 10 15 20 25
B
E
R
SNR in dB
OFDM
MC-SS(16,75%)
MC-SS(16,50%)
Figure 3.13: Eect of loading for spreading gain of 16 under ideal conditions
BER performance can be improved with low loading, there is a loss in throughput
as percentage utilization of the resource is less. Following the previous case, it is to
be noted that the BER performance with a certain loading percentage is expected to
decrease with increasing spreading gain.
Impact of Carrier Oset: Figure 3.14 shows the performance of the schemes
for residual frequency oset of 0.04, under full loading for dierent spreading gains.
The carrier frequency oset introduces ICI as described in (3.13). It can be seen that
OFDM has almost identical performance as MC-SS under full loading, with spreading
of 4. MC-SS with spreading gain of 16 is only slightly worse compared to the smaller
spreading gain while MC-SS with spreading gain of 64 has the worst performance.
The performance in this situation is similar to when receiver is assumed to be at
condition. Therefore it can be seen that the carrier frequency oset does not inuence
the relative nature of performance of the schemes under consideration. However, it
must be noted that only Quadrature Phase Shift Keying (QPSK) modulation has
been used and the residual carrier oset simulated is not very high. The performance
could have been dierent had the residual carrier oset been larger or higher order
modulation were used.
Figure 3.15 shows performance of the schemes for residual frequency oset of
0.04, for dierent amount of loading and for a spreading gain of 16. As observed
before, the MC-SS scheme with 50% loading has the best performance, which is
c
Suvra Sekhar Das, 2007
62 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
0.001
0.01
0.1
1
0 5 10 15 20 25
B
E
R
SNR in dB
OFDM
MC-SS(4)
MC-SS(16)
MC-SS(64)
Figure 3.14: Eect of spreading gain in full load under residual carrier frequency oset
0.001
0.01
0.1
1
0 5 10 15 20 25
B
E
R
SNR in dB
OFDM
MC-SS(16,75%)
MC-SS(16,50%)
Figure 3.15: Eect of loading for spreading gain of 16 under residual carrier frequency oset
Center for TeleInFrastruktur (CTIF), Aalborg University
3.2 MC-SS with receiver impairments 63
0.001
0.01
0.1
1
0 5 10 15 20 25
B
E
R
SNR in dB
OFDM
MC-SS(4)
MC-SS(16)
MC-SS(64)
Figure 3.16: Eect of spreading gain in full load under Channel Estimation error
followed by the 75% loading while OFDM has the poorest performance.
Impact of Channel Estimation Error Figures 3.16 and 3.17 shows the perfor-
mance of the schemes for channel estimation error. The estimated channel coecients
have been assumed to be aected by noise. It can be distinctly observed that channel
estimation error aects the performance of the schemes most severely. In Figure 3.16,
the observation regarding the relative performance of the dierent schemes under
varying spreading gain, still holds as made for the ideal case earlier in this section.
However, it must be noted that MC-SS with spreading gain of 4 now has worse per-
formance than OFDM. In contrast, when ideal channel estimation was assumed, then
MC-SS scheme using a spreading gain up to 16 has very close performance to that
of OFDM. The channel estimation error degrades the performance of MC-SS scheme
more because the noisy channel estimates destroys the orthogonality of the codes and
inter code interference becomes higher. A data symbol is recovered by summing the
chips which have been spread over the sub carriers. Now, errors due to noisy channel
estimates creep into the chips. When these chips are combined then the noise also
gets amplied and hence there is no gain from the spreading, rather there is worse
performance.
It can be seen from Figure 3.17 that the spreading schemes with 50% load
conditions has the best performance but 75% loading has performance which is same
as OFDM. Therefore it can be concluded that spreading schemes are severely aected
c
Suvra Sekhar Das, 2007
64 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
0.001
0.01
0.1
1
0 5 10 15 20 25
B
E
R
SNR in dB
OFDM
MC-SS(16,75%)
MC-SS(16,50%)
Figure 3.17: Eect of loading for spreading gain of 16 under Channel estimation error
by the channel estimation error.
Outage Performance: Figure 3.18 shows the eect of spreading gain on the 10%
outage performance in full load for ideal receiver conditions. It can be observed
that the outage throughput of OFDM is best compared to the MC-SS schemes with
dierent spreading gains. The reason is as explained before, i.e. the spread spectrum
schemes loose the orthogonality of the codes due to the frequency selective fading
channel.
Figure 3.19 shows the eect of loading for spreading gain of 16 on the 10%
outage throughput for ideal receiver conditions. It can be seen that MC-SS with
50% loading has the best performance while with 75% loading the performance is
similar to that of OFDM, i.e. loading above 75% does not provide any benet from
spreading. The reason for this has been discussed before that there is redundant data
information transmitted due to less than 100% loading which provides the extra SNR
gain.
Figure 3.20 shows the eect of spreading gain on the 10% outage throughput
in full load for channel estimation error and synchronization error. It is seen that
OFDM performs better than MC-SS with any spreading gain.
Figure 3.21 shows the eect of loading for spreading gain of 16 on the 10%
outage throughput under channel estimation error and synchronization error. The
channel estimation error and synchronization error are considered to be in the same
Center for TeleInFrastruktur (CTIF), Aalborg University
3.2 MC-SS with receiver impairments 65
0.1
1
0 5 10 15 20 25
N
o
r
m
a
l
i
z
e
d

T
h
r
o
u
g
h
p
u
t
SNR in dB
OFDM
MC-SS(4)
MC-SS(16)
MC-SS(64)
Figure 3.18: Eect of spreading gain on the 10% outage performance in full load for ideal
receiver conditions
0.1
1
0 5 10 15 20 25
N
o
r
m
a
l
i
z
e
d

T
h
r
o
u
g
h
p
u
t
SNR in dB
OFDM
MC-SS(16,75%)
MC-SS(16,50%)
Figure 3.19: Eect of loading for spreading gain of 16 on the 10% outage performance for
ideal receiver conditions
c
Suvra Sekhar Das, 2007
66 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
0.1
1
0 5 10 15 20 25
N
o
r
m
a
l
i
z
e
d

T
h
r
o
u
g
h
p
u
t
SNR in dB
OFDM
MC-SS(16)
MC-SS(64)
Figure 3.20: Eect of spreading gain on the 10% outage performance in full load for Channel
Estimation error and synchronization error
0.1
1
0 5 10 15 20 25
N
o
r
m
a
l
i
z
e
d

T
h
r
o
u
g
h
p
u
t
SNR in dB
OFDM
MC-SS(16,75%)
MC-SS(16,50%)
Figure 3.21: Eect of loading for spreading gain of 16 on the 10% outage performance for
Channel Estimation error and synchronization error
Center for TeleInFrastruktur (CTIF), Aalborg University
3.3 Summary 67
order as in the earlier cases of non ideal receiver performance. It can be seen that
OFDM has better resilience to such errors compared to the MC-SS schemes.
3.2.4 Conclusion
In this study, the performance of OFDM and MC-SS schemes have been evaluated
for down link channel in realistic indoor conditions in presence of receiver impair-
ments. For fully loaded conditions it is found that OFDM has the best mean BER
performance, while for loading less than 75%, MC-SS outperforms OFDM. It has
been distinctly found that increasing spreading gain degrades the performance, but
provides the exibility of changing the loading. Changing the loading can give im-
proved performance as compared to OFDM. Under such conditions the throughput
is adversely aected, but BER can be improved. Therefore it can be concluded that,
operating the spreading schemes under full load is not desirable, as OFDM outper-
forms them. Synchronization error has similar eect on the spreading schemes as on
the OFDM schemes, for the range of errors considered.
Channel estimation error aects the MC-SS more than OFDM. The exibility
of improving the bit error rate performance of the spreading schemes over OFDM is
almost lost for channel estimation error conditions as veried in the work. Similar
observations as above has been made for the 10% outage throughput results.
Finally it is important to note that the choice of the scheme depends on the
required criteria of performance and hence must be chosen accordingly. It is not pos-
sible to claim one scheme to have better performance over other without considering
the application it is being considered for. For example, even though it has been seen
that OFDM has relatively better performance than MC-SS under full load conditions,
yet MC-SS schemes can provide a better BER by simply reducing the loading, which
is otherwise not possible in OFDM systems. Thus the results presented in this work
can be used to choose the scheme after deciding the performance criteria for particular
applications.
3.3 Summary
A proposal to enhance a MC-SS scheme by introducing SCH has been made.
The proposed scheme is found to have improved outage performance.
Detailed guidelines regarding spreading gain, loading, use of SIC are discussed.
c
Suvra Sekhar Das, 2007
68 Chapter 3. Evaluation of Hybrid Multi Carrier Spread Spectrum
In the second part of the chapter non ideal receiver operation has been consid-
ered.
It is found that channel estimation error severely degrades the performance of
MC-SS schemes.
It is also seen that channel estimation error has a greater inuence on the
performance of MC-SS schemes than the carrier oset errors.
Under fully loaded condition, it is found that OFDM has better resilience to
receiver impairments over MC-SS schemes.
If it is considered that full loading will be used for higher spectral eciency,
then it can be concluded that OFDM should be used.
If full loading is not considered, then MC-SS has better performance.
The studies made in this chapter did not consider Forward Error Correction
(FEC) coding. In OFDM system use of FEC is expected to further enhance the
performance by increasing the frequency diversity gain and also time diversity
gains if designed carefully. It is already seen that OFDM has better performance
in the uncoded set up. Therefore it can be expected with the use of FEC coding
the performance of OFDM will be further improved.
Center for TeleInFrastruktur (CTIF), Aalborg University
4
Bit loading on Pilot Sub Carriers
The aim of this chapter is to increase the spectral eciency by reducing the pilot
overhead. In WLAN type system where a preamble (Training sequence) is used in
the front for the packet being transmitted for synchronization and channel estimation,
pilots sub carriers are embedded among the data sub carriers to track the residual
phase errors [53]. Residual phase errors are the combined eect of residual uncorrected
carrier frequency oset error together with the sampling clock oset error. The pilot
sub carriers are pre-dened waveforms, does not carry any data information and hence
are necessary overhead as they use up some of the sub carriers. In order to reduce the
overhead due to these pilot sub carriers a semi blind pilot scheme is proposed in this
chapter. Semi blind pilots carry data bits instead of a priori known symbol sequence
by exploiting the extra SNR available at the pilots. The term semi blind is used
because a preamble is used for initial acquisition of the synchronization and channel
estimation while the pilot sub carriers are used for tracking the residual phase errors.
Further in the proposed scheme, the pilot sub carriers are loaded with a modulation
which is less than that of the data sub carriers, i.e. when the data sub carriers carry
16-QAM and 64-QAM, the pilot sub carriers are made to carry QPSK signal. Since
the power of the data and pilot sub carriers are quite similar, (Pilots sometime may
have higher power) therefore clearly under the above conditions the data symbols in
69
70 Chapter 4. Bit loading on Pilot Sub Carriers
the pilots enjoy much more power than required. Thus there is a higher probability of
detecting the data on the pilot sub carriers correctly. These correctly decoded data
symbols can then be considered as pilots and hence used for estimating the phase
errors.
To track the residual phase error a low complexity algorithm is proposed in
this chapter. A highly complex computation of the inverse tangent function is needed
to estimate the phase errors. The proposed algorithm reduces complexity by avoiding
such computation.
The performance of semi blind pilot is investigated with existing and the pro-
posed algorithm. It is observed that the semi blind pilot scheme enhances the spectral
eciency in the range of 5% to 15% without degradation in the BER performance
as compared to the data aided pilot based systems. The proposed phase tracking
algorithm reduces the implementation complexity signicantly at the cost of 1-2 dB
loss in BER performance in the semi blind environment, however it is found to be
robust against high residual phase errors.
4.1 Introduction
OFDM is spectrally ecient, but high accuracy is needed in synchronization since
coherent demodulation of OFDM is extremely sensitive to synchronization errors. In
a real environment the synchronization blocks placed at the receiver front end are
not able to estimate the exact carrier frequency oset due to thermal noise and xed
word length eects. Moreover due to sampling frequency oset there is a slowly
increasing timing oset. The receiver has to continuously track and compensate for
these eects in order to keep the BER low. Residual Carrier Frequency Oset (CFO)
and Sampling Frequency Oset (SFO) tracking are thus very critical for successful
implementation of OFDM systems. Residual CFO and SFO errors are jointly termed
as Residual Phase Errors.
In WLAN type systems, in order to estimate the channel and get initial ac-
quisition of the frequency oset, a completely data aided Training Sequence (TS)
is used as described in Fig. 2.19. Pilot sub carriers, which are a known sequence
of waveforms, are embedded in the data part of the packets to enable tracking of
residual phase oset. Pilot sub carriers do not carry any information content and
thus add signicant overhead to the system. To avoid the overhead due to pilot sub
carriers blind algorithms [60], [61] can be used but they add very high complexity to
the receivers. Therefore a semiblind system is proposed in this chapter.
Center for TeleInFrastruktur (CTIF), Aalborg University
4.1 Introduction 71
It is proposed in this work to use the pilot sub carriers embedded in the OFDM
symbols to carry data-information, instead of known symbol sequence. This would
increase system eciency by reducing the pilot overhead, while not deteriorating the
BER performance, for WLAN systems.
High order modulations such as 16-QAM and 64-QAM are often used for high
data rate transmission. Pilot sub carriers used in these systems, which are known a
priory at the receiver, are similar to BPSK waveform, i.e sub-carriers have normalized
values of 1. BPSK modulated signal requires between 12 dB and 16 dB less SNR
as compared to 64-QAM, depending on the FEC coding rate to achieve the desired
Packet Error Rate (PER) of in a typical WLAN system, which is about 10% for IEEE
802.11a [44]. This additional SNR advantage (Table 4.1) of the pilot may be exploited
to transmit additional data bit, thus, increasing the throughput of the system without
any increase in bandwidth or SNR requirement or any loss in performance. That is,
while transmitting 64-QAM modulated waveform, the pilots can be BPSK or QPSK
modulated to carry information bits instead of known a priori symbol sequences.
The increase in eciency achieved via this scheme though is within modest
values yet it has several advantages, such as, utilizing the existing frame format with
very little changes. Further the existing demodulation schemes and receiver baseband
and front end algorithms can function as usual and do not require any additional
hardware extension. Only the phase correction algorithm needs to be changed.
Algorithms used for phase tracking and correction use the search function
argmax [45, 62, 63, 64] after complexconjugatemultipyadd operations. They also
need to compute the inverse tangent [62],[38],[53],[65],[66]. A series implementation
of the inverse tangent function, will need up to seven term for appropriate accuracy.
This operation needs approximately more than 60 arithmetic operations. In other
words the implementation complexity of the phase tracking algorithm is very high. An
algorithm for residual phase tracking which does not use complex arithmetic functions
is also presented in this chapter. Instead of computing the phase angles using inverse
tangent function, the proposed algorithm computes the complex exponential of the
phase angle at the pilot-tone locations to minimize the implementation complexity.
The design of the algorithm is such that it can be very easily applied to any coherent
OFDM based WLAN type of networks.
c
Suvra Sekhar Das, 2007
72 Chapter 4. Bit loading on Pilot Sub Carriers
Figure 4.1: OFDM Symbol Format
4.2 System Description
The case for Wireless Local Area Networks (WLAN) systems using a frame format
similar to the one used by IEEE 802.11a is considered here. Such systems have a
training sequence or preamble (Fig. 2.19) [44]. The initial ten short training sequences
are used for packet detection, automatic gain control, Symbol Timing Oset (STO)
synchronization and coarse Carrier Frequency Estimation; it is followed by two long
training sequences for ne carrier frequency oset (CFO) synchronization and channel
estimation [53].
Packet detection and symbol timing synchronization are done by the Time
Synch block of the receiver. Initial carrier frequency estimation and compensation
is done by the Frequency Synch block. Channel Estimation and compensation is
done by the Channel Equalization block (Once per packet). Then follows the residual
phase correction done by Phase Tracking block. The residual phase occurs because
of nonexact carrier frequency compensation and sampling frequency oset (SFO).
The residual phase increases with OFDM symbol index [45]. Pilot subcarriers are
embedded in the OFDM symbol among the data subcarriers to enable the tracking
of this residual phase.
Pilot Structure Every OFDM symbol is made of data and pilot subcarriers as
shown in Figure 4.1. Data subcarriers are modulated using either Binary Phase Shift
Keying (BPSK), Quadrature Phase Shift Keying (QPSK) or high order Quadrature
Amplitude Modulation (QAM). Pilot subcarriers are a pseudo random sequence of
1 [44], the sequence being known at the receiver. Pilots are the same irrespective of
the modulation scheme of the data subcarriers and are transmitted with the same
maximum power as that of data subcarriers. The minimum SNR required by the
Center for TeleInFrastruktur (CTIF), Aalborg University
4.2 System Description 73
Table 4.1: SNR advantage of pilots in dierent modulation schemes
Modulation Scheme Pilots SNR advantage
64-QAM BPSK 16dB
64-QAM QPSK 12dB
16-QAM BPSK 12dB
16-QAM QPSK 7dB
highest order QAM (64-QAM) modulation scheme is about 16 dB more than that
required by BPSK modulation scheme to achieve a PER of 10% [44],[53],[35]. Hence
we know that, the pilots carry more power when a 64-QAM modulated symbol is
transmitted than in the case of a BPSK modulated symbol. The additional SNR the
pilots enjoy with dierent modulations is shown in Table 4.1 [44].
Proposed Scheme It is proposed that the pilot sub carriers present in OFDM
based WLANs as in IEEE 802.11a can be modulated to transmit data without loss
of performance to increase the data rate. The data and pilot sub carrier constellation
can be seen in Fig. 4.2 & Fig. 4.3.
c
Suvra Sekhar Das, 2007
74 Chapter 4. Bit loading on Pilot Sub Carriers
Figure 4.2: 16-QAM Data constellation Figure 4.3: QPSK Pilot constellation
Table 4.2: Symbols
Symbol Explanation
R
l,k
Received OFDM subcarrier
l Index of OFDM symbol index
k Index Subcarrier symbol index
H
l,k
Channel coecient at k
th
subcarrier
T Sampling duration
N Number of points of DFT
N
g
Number of guard samples
N
s
=N + N
g
total number of samples in a transmitted
OFDM symbol
T
u
DFT duration = T.N
T
g
Guard Interval = T.N
g
T
s
Transmitted OFDM symbol duration = T.N
s
X
l,k
Transmitted symbol

k,i
= (1 + )(fT
u
+ k) i, is the residual phase error

k
=
k,k
= fT
u
+ k
Relative Sampling frequency oset between Transmitter
and receiver
f residual carrier frequency oset
n
l,k
noise term
F
sc
Sub carrier spacing =
1
T
u
N
o
Thermal noise power density

2
s
Signal power, variance of signal

r
required signal to noise power

s
signal to thermal noise ratio of signal
Center for TeleInFrastruktur (CTIF), Aalborg University
4.2 System Description 75
To prove this statement theoretically, we must consider the reason why pilots
are used in OFDM based WLAN systems for indoor communication. Several litera-
tures [38],[53], [67] discuss that pilots are used for residual phase correction. Residual
phase eect has been explained briey before. It happens because of the nonideal
frequency synchronization by the initial carrier frequency oset correction block which
uses the training sequence to obtain an initial acquisition. Added to this there is the
sampling frequency oset. In some systems the local oscillator and the sampling clock
are tied together. This means that they drift between the transmitter and the receiver
in the same direction. To correct these eects, the literatures referenced above and
several others available treat these two together as residual phase. The algorithms
used to track them estimate the progressive phase dierence between the subcarriers
of two consecutive OFDM symbols. The eects of the residual phase are many fold
as is evident from the equation below. The received subcarrier after initial carrier
frequency oset correction, frame and symbol synchronization and DFT is
R
l,k
= e
j
k
e
j2
lN
s
+N
g
N

k
sin(
k
)
sin(

k
N
)
X
l,k
H
l,k
+
N1

i=0,i=k
e
j
k,i
e
j2
lN
s
+N
g
N

i
sin(
k,i
)
sin(

k,i
N
)
X
l,i
H
l,i
+ n
l,k
(4.1)
where the symbols are as dened in Table 4.2. The rst term is the desired signal
component, while the second term is the ICI term [23] due to subcarrier misalign-
ment, which is caused by the residual carrier frequency error while the third term
is the noise term. As can be seen from ( 4.1) above, the rst term has a reduced
amplitude and a phase rotation, as well as added to this is the ICI component. Hence
the two important eects of the residual phase are rotation of received constellation
points and reduction in SNR. Pilots are used to track and compensate for these ef-
fects. The principle of operation of these algorithms has been discussed in the above
mentioned literature, such that the relative phase rotation between the subcarriers
of consecutive OFDM symbols is computed.
For the proposed semi-blind pilot scheme to work using the above mentioned
algorithms for phase compensation, it is desired that the residual phase rotation be-
tween two OFDM symbols, satisfy two limiting conditions: i) The angle of rotation
of the rst component (the desired signal) is less than that required for correct de-
modulation, ii) The ICI is suciently small, so that the BER of the pilot subcarriers
is less than that of the data subcarriers.
c
Suvra Sekhar Das, 2007
76 Chapter 4. Bit loading on Pilot Sub Carriers
4.3 Analytical Framework and Algorithm
Analysis In this section the two limiting conditions mentioned above for successful
deployment of the scheme will be analyzed. First the expression for the maximum
residual carrier frequency oset will be given. This will be done assuming initial data
aided acquisition using training sequence by commonly used algorithms [38, 62, 53].
Then the expression and limiting allowable phase rotation in the light of constellation
rotation of the desired signal will be shown. Next the maximum allowable phase
rotation in the light of ICI will be given. The minimum of these two criteria will
give the allowable phase rotation for successful demodulation and detection of pilot
symbol. This will be equated with the maximum residual carrier frequency oset and
the feasibility of the system will be validated.
Residual Carrier Frequency Oset With reference to a generic receiver architec-
ture, there is an initial carrier frequency oset estimation block. This block performs
an initial carrier frequency oset acquisition using the training sequence. The esti-
mate is never perfect since the incoming signal is contaminated by noise. The variance
of the estimate is dependent on the SNR of the training sequence used to estimate
the oset. From [38] it can be written that the variance of the estimate is

f
=
1
2
N
L

M
s
(4.2)
where L is the distance between the corresponding subcarriers across which phase
dierence is being measured, M is the number of summations, and
s
is the signal
to noise ratio. The residual carrier frequency appears as residual phase error. The
maximum carrier frequency oset can be considered as 3

f
. Accordingly the residual
phase error at the k
th
subcarrier of the l
th
OFDM symbol can be written as

l,k
=
2lN
s
N

k
=
2lN
s
N
(fT
u
+ k)
=
2lN
s
N
(3
f
T
u
+ k) (4.3)
The algorithms operating on pilots, estimate the residual phase between two con-
secutive OFDM symbols. Therefore, the residual phase error between the same two
Center for TeleInFrastruktur (CTIF), Aalborg University
4.3 Analytical Framework and Algorithm 77
subcarriers of consecutive OFDM symbol can be written as

k
=
l,k

l1,k
=
2N
s
N
(3
f
T
u
+ k) (4.4)
Constellation Rotation It has been proposed in this work, to modulate the pilot
tones with BPSK or QPSK. Accordingly, the largest constellation rotation allowable
is /2 and /4 for BPSK and QPSK respectively. The residual phase error must be
less than this value. i.e. the phase rotation between the same two pilots of consecutive
OFDM symbols must be less than these values. Accordingly the equations can be set
up as

k

2
2
b+1
2N
s
N

k

2
2
b+1

k

N
N
s
2
b+1
(4.5)
Where b is the number of bits per symbol used in the modulation scheme. And it can
be continued that

min
1
=
N
N
s
2
b+1
(4.6)
Inter Carrier Interference The phenomenon of ICI has been explained previ-
ously. Since the symbol in the pilots need to be detected before data in datasub
carriers, and since the pilots are also needed for correcting the phase of the datasub
carriers, they are hard decision detected in this work. The pilots must achieve a Bit
Error Rate (BER) which is lower than the tolerable limits of the datasubcarriers.
The ICI can be modeled as additive noise [45]. The Signal to Interference plus Noise
Ratio (SINR) of the pilots must be such that the BER under ICI conditions must be
better than the required BER performance of the data subcarriers. In IEEE 802.11a
type systems with PER of 10% the BER equates to a range between 10
4
and 10
3
.
Hence the BER of pilots can be set to be less than this range. The ICI power which
is given by the second term of (4.1) can be approximated [45] to

2
ICI,k

2
s

2
3

2
k
(4.7)
c
Suvra Sekhar Das, 2007
78 Chapter 4. Bit loading on Pilot Sub Carriers
Accordingly the signal to noise ratio can be written
=

2
s
sinc
2
(
k
)

2
s

2
ICI,k
+ N
o
F
sc
(4.8)
This can be reduced with approximations for small angle
k
as


2
s

2
s

2
ICI,k
+ N
o
F
sc
(4.9)
The required PER of 10% for a particular modulation scheme can be mapped to a
certain minimum require SNR level. Using
r
as the minimum required SNR, the
second condition is

r
(4.10)
Replacing with the expression for it can be written as

2
s

2
s

2
ICI,k
+ N
o
F
sc

r

2
s

r

2
s

2
ICI,k
+ N
o
F
sc
1

r

2
ICI,k
+
N
o
F
sc

2
s
1

r

2
ICI,k
+
1

s
1

s

2
ICI,k
(4.11)
Replacing the value of
2
ICI,k
it can be written as
1

2
3

2
k
3

2
_
1

s
_

2
k

2
_
1

s
_

k
(4.12)
In other words it can be written as

min
2
=

2
_
1

s
_
(4.13)
Center for TeleInFrastruktur (CTIF), Aalborg University
4.3 Analytical Framework and Algorithm 79
Final Limits on Residual Error Now that the two limits on the residual phase
error have been found, the minimum between them needs to be chosen. In other
words,

k,max
min
min
1
,
min
2

= min
_
N
N
s
1
2
b+1
,
_
3

2
_
1

s
_
_
(4.14)
With parameters from IEEE 802.11a [44], i.e. N = 64, N
s
= 80,
r
= 15dB,
s
= 20
dB, and b=1 and 2 for modulating pilots with BPSK and QPSK respectively, it can
be found that, for this particular situation

k,max

min
2
(4.15)
Therefore it can be written that

k,max

_
3

2
_
1

s
_
3
1
2
N
L

M
+ k
_
3

2
_
1

s
_
(4.16)
The factor k is very small and can be neglected from the calculations, thus it can
be continued that, for

s
1
1 +
3
4
1
M
_
N
L
_
2
>
r
(4.17)
the system must be feasible to implement. For large values of M, this leads to

s
>
r
(4.18)
To state in words, as long as the required SNR of the pilot modulation is less than
that of the datasubcarriers, the scheme is feasible theoretically. An algorithm for
implementing the above scheme will be presented in the next section. Its performance
is also analyzed via simulation.
Algorithms for the proposed scheme Channel estimation and equalization can
be assumed ideal for now, though imperfections can be modeled with equivalent SNR
loss [45], [62]. Some SNR loss can be used in the model to account for channel
estimation inaccuracies. Further, the constant carrier phase oset can be included
c
Suvra Sekhar Das, 2007
80 Chapter 4. Bit loading on Pilot Sub Carriers
into the channel eect and thus after channel equalization the received signal of (4.1)
becomes
R

l,k
= e
j2
lN
s
N

k
sin(
k
)
sin(

k
N
)
X
l,k
+
N1

i=0,i=k
e
j
k,i
e
j2
lN
s
N

i
sin(
k,i
)
sin(

k,i
N
)
X
l,i
H
l,i
H
l,k
+ n

l,k
(4.19)
Where n

l,k
is the noise component changed due to channel equalization; this will be
dierent depending upon the type of equalization scheme used. Zero Forcing algo-
rithm causes noise enhancement at low SNR conditions. However, for the situation
under analysis it is assumed that the data will be modulated using higher order QAM
modulation, i.e. 256-QAM, 64-QAM, 16-QAM, etc. These require high values of SNR
are greater than 15dB, and hence even with Zero Forcing channel equalization the
noise enhancement will not be signicantly large.
The proposed low complexity residual phase tracking algorithm in the semi
blind system is presented next and compared against the performance of an exact
algorithm. The exact algorithm will be described later. It computes the exact phase
angle using the inverse tangent function. Whereas, the proposed low complexity
algorithm avoids the computation of the complex inverse tangent function.
The following symbols shall be used in the algorithm:
l,k
= e
j
l,k
= e
j2l
lN
s
N

k
and
l,k
= e
j
k
= e
j2
Ns
N

k
. It is further dened that
0,k
=
0,k
= 1, k = pilot
subcarrier index. In this work pilot subcarrier indices will be indicated by p instead
of the usual k. It is also to be noted that p
i
denotes the i
th
pilot tone. The algorithm
can now be written as
Step 1:
R

l,p
.

l1,p
=
_
e
j2
lN
s
N

p
sin(
p
)
sin(

k
N
)
X
l,p
+
N1

i=0,i=p
e
j
p,i
e
j2
lN
s
N

i
sin(
p,i
)
sin(

p,i
N
)
X
l,i
H
l,i
H
l,k
+ n

l,p
_
e
j2(l1)
lN
s
N

p
(4.20)
Under high SNR the second and the third term can be considered small and the
multiplication of the complex exponential does not change their power, thus the
equation can be approximated to
R

l,p
.

l1,p
e
j2
N
s
N

p
sin(
p
)
sin(

p
N
)
X
l,p
(4.21)
Center for TeleInFrastruktur (CTIF), Aalborg University
4.3 Analytical Framework and Algorithm 81
With a further assumption, that residual frequency oset is small, it can be approxi-
mated as
R

l,p
.

l1,p
e
j2
N
s
N

p
X
l,p
= e
j
p
X
l,p
= R

l,p
(4.22)
Where R

l,p
denotes the estimate of the modulated pilot. This can be demodulated
depending on the scheme used for modulating the pilot tones. Accordingly if a BPSK
is used

X
l,p
=
_
+1 if Real(R

l,p
) > 0
1 if Real(R

l,p
) < 0
(4.23)
The demodulation algorithm for QPSK can be derived similarly. Next it can be
written that
R

l,p
.

l,p
= e
j
p
X
l,p

l,p
=
l,p
since [X
l,p
[ = 1 (4.24)
Continuing further with the algorithm, the following steps needs to executed
next:

l,p
=
l1,p
+ (1 )R

l,p

X
l,p
R

l1,p

l1,p
(4.25)

l,p
=

l1,p

l,p
+ (1 )R

l,p

l,p
(4.26)
where
l,p
=

l,p
[
l,p
[
,

l,p
=

l,p
[

l,p
[
(4.27)
where is the memory of the algorithm, which is a design parameter. For the en-
vironment under discussion, a suitable value is
3
4
which is found from simulation.
The values of

are obtained at the pilot locations. With the help of the piecewise
linear interpolation, the compensating complex factors can be computed. It has to be
noted that instead of computing the phase angle, rather the complex exponential of
the phase angle is computed in this algorithm. This will become more evident when
compared with the steps of the exact algorithm which is explained in the next section.
The straight solid line in Fig. 4.4 is the actual phase that needs to be estimated, while
the curve is the sinusoid of the phase. In Fig. 4.4, = f, and C is a constant term.
Estimation of the sinusoid at the pilot locations is being done here. There will be a
sine and a cosine term, only one component is shown to retain the simplicity of the
gure. They are approximated to be piecewiselinear for small angles. Then the
c
Suvra Sekhar Das, 2007
82 Chapter 4. Bit loading on Pilot Sub Carriers
Figure 4.4: Piece-wise linear interpolation
mean slope of each of them is estimated as given below in (4.28).
m
l
=
1
2
_

l,p
21

l,p
7
21 (7)
+

l,p
7

l,p
21
7 (21)
_
(4.28)
The subscripts of the pilot indexes denote the sub carrier index where pilots sub
carrier are present. It needs to be noted that

X
l,p
is used in the algorithm.

X
l,p
is the estimated pilot symbol, reconstructed after detection, whereas in data aided
algorithms, P
l,p
is used instead, which is the known pilot. The values in (4.28) are in
reference to the IEEE 802.11a/g OFDM PHY standard.
Piecewiselinearinterpolation is done to nd the complex multiplication fac-
tor for each subcarrier for compensating the residual phase error as states below.
y
l,k
=

l,p
m
l
(p k) (4.29)
Where p indicates the nearest pilot index to k
th
subcarrier. Then for compensation
we use R

l,k
= R

l,k
y

l,k
. The maximum errors that may occur will be at the farthest
sub carriers (due to sampling frequency oset) where the dierence is largest because
of larger sub-carrier index (see (4.19) & Table 4.2).
The exact algorithm The exact algorithm which has been mentioned previously
is described here. To follow the exact algorithm, (4.19)is taken as the starting point.
Center for TeleInFrastruktur (CTIF), Aalborg University
4.3 Analytical Framework and Algorithm 83
With the initial value of
0,p
= 0, the algorithm can be started as
R

l,p
= R

l,p
.e
j
l1,p
(4.30)
R

l,p
denotes the estimate of the modulated pilot which is demodulated as explained
in the previous section. Continuing further with the exact algorithm, the following
can be stated.

l,p
=
l1,p
+ (1 )angle(R

l,p

X
l,p
R

l1,p

l1,p
) (4.31)

l,p
=(
l1,p
+
l,p
) + (1 )angle(R

l,p

l,p
) (4.32)
where angle(x) = arctan(
Imagimary(x)
Real(x)
) The phase angles estimated above are at the
pilot locations. It is to be noted, that for this method, the inverse tangent function
has to be executed two times, which becomes extremely computation intensive. Table
look up is often used to avoid the complexity of its implementation, but this procedure
is also limited when high resolution is required for small sized, low cost and low
powered devices. Cordic implementation is also limited under such conditions of
operation since they are slow. For compensation, the phase angles are interpolated
using piece wise linear interpolation to all the data sub carrier indexes. The phase
angles at the pilot locations can be explained as is depicted in Fig. 4.4. The mean
slope of the phases at the pilot locations can be computed as

l
=
1
2
_

l,p
21

l,p
7
21 (7)
+

l,p
7

l,p
21
7 (21)
_
(4.33)
Piecewiselinearinterpolation is done to nd the compensating phase angle for each
subcarrier as:

l,k
=
l,p

l
(p k) (4.34)
Where p indicates the nearest pilot index to k
th
subcarrier. For compensation the
following needs to be used
R

l,k
= R

l,k
e
j

l,k
(4.35)
In a data aided, non blind situation, the

X
l,p
is replaced by P
l,p
(known pilot
sequence at receiver) in the steps of the above two algorithms described.
c
Suvra Sekhar Das, 2007
84 Chapter 4. Bit loading on Pilot Sub Carriers
4.4 Simulation and Discussion
Simulation Parameters The parameters used for simulation are similar to that
in IEEE 802.11g [11]. The carrier frequency is taken as 5.4 GHz. Sampling frequency
oset is taken as 20 ppm, with 20 MHz of sampling frequency. Dierent values of
residual carrier frequency oset are used and are mentioned in the simulation result
curves. The number of sub carriers N is taken as 64. Number of zero sub carriers
= 12, number of data sub carriers were 48 and number pilot sub carriers were 4. The
GI T
g
= 0.8s, while the symbol duration T
s
= 4s and the useful symbol duration
T
u
= 3.2s is used. Dierent combinations of data sub carrier modulation and pilot
modulation is used. They are described on the simulation result curves.
The proposed low complexity algorithm is represented as alg1 (solid lines) in
the gures. The exact algorithm is denoted by alg2 (dotted lines) which is run in
the proposed semiblind environment. A completely data aided OFDM based system,
following the IEEE 802.11a/g frame format, using the exact algorithm is marked as
alg3 in the gures.
In WLAN environment, the channel is assumed to be quasi static during the
entire period of a packet. Therefore Doppler frequency spread can be neglected. The
proposed semi blind pilot scheme is to be used when data sub carriers will have
high order modulation. This occurs only when the average SNR is quite high. The
channel estimation using the long training sequence can be considered to suer from
little error under these conditions. Further the residual phase error is independent of
the channel properties. Therefore for simplicity AWGN channel has been used in the
analysis here. FEC is not used.
Results and Discussion The simulation related to Fig. 4.5 has data sub carriers
modulated with 64QAM, while the pilot sub carriers modulated with BPSK. The
CFO mentioned in the gures indicates the residual carrier frequency oset in ppm.
The result in Fig. 4.6 shows performance of the algorithms similar the previous case
but with QPSK modulation on the piots. An important observation to be made in
these two results is that, the performance of the exact algorithm, be it the proposed
semi blind case or the commonly used completely data aided case(alg 2 and alg 3),
is almost the same. Therefore, it is now seen that the known sequence of pilot as
is transmitted in WLAN systems based on IEEE 802.11a/g frame format is not a
necessity. It is also seen that the exact algorithms (alg 2 and alg 3), perform slightly
better (on an average about 2dB) than the proposed low complexity algorithm (alg1).
Center for TeleInFrastruktur (CTIF), Aalborg University
4.4 Simulation and Discussion 85
Figure 4.5: BER Vs SNR, Data64-QAM, pilotBPSK.
alg3 completely data aided pilot based OFDM system, using the exact algorithm.
alg2 the exact algorithm in the proposed semiblind environment.
alg1 the proposed low complexity, approximated algorithm.
Figure 4.6: BER Vs SNR, Data64-QAM, pilotQPSK
c
Suvra Sekhar Das, 2007
86 Chapter 4. Bit loading on Pilot Sub Carriers
Viewed in another way, the degradation of the low complexity algorithm is less than
one order in terms of Bit Error Rate (BER), at a given Signal to Noise Ratio (SNR).
Therefore it can also be said that the approximate algorithm, can be used in low
complexity, low powered devices due to signicant amount of reduction in complexity
without much loss in performance.
The result in Fig. 4.7 has data sub carriers modulated with 16QAM, while
BPSK is used for the pilots. It has similar characteristics as previous results for lower
residual phase error. An important observation to be made is that for high residual
phase errors (near and above CFO = 0.5 ppm), the exact algorithms (2 and 3) fail,
while the proposed approximate low complexity one continues to perform well. The
reason for this behavior is that the residual phase error increases with OFDM symbol
index, i.e. with time. If the accumulated residual phase reaches values near /2,
the exact algorithm using inverse tangent function becomes unstable. This is not
the case for the proposed algorithm, which completely avoids the computation of the
inverse tangent function. This is one of the added benets of the proposed algorithm.
Results in gures 4.8 and 4.9 are in accordance with the above.
Figure 4.7: BER Vs SNR, Data16-QAM, pilotBPSK
Center for TeleInFrastruktur (CTIF), Aalborg University
4.4 Simulation and Discussion 87
Figure 4.8: BER Vs SNR, Data16-QAM, pilotQPSK
Figure 4.9: BER Vs SNR, DataQPSK, pilotBPSK
c
Suvra Sekhar Das, 2007
88 Chapter 4. Bit loading on Pilot Sub Carriers
The additional data rate gained in this scheme can be computed as
N
p
.N
bp
T
s
. N
p
represents the number of pilots, N
bp
indicates number of bits carried by each pilot.
The percentage gain can be computed by the formula 100.
N
p
.N
bp
N
bd
.N
d
, where N
bd
implies,
number of bits per data sub carrier and N
d
implies, number of data sub carriers. The
increase in data rate is summarized in Table 4.3. These values are based on IEEE
802.11a/g frame structure, which is a single user system. Much higher gain can be
achieved in systems which need more pilots, especially for outdoor and multiuser
OFDMA schemes.
Table 4.3: Semi Blind Congurations and gains
data-rate
(mbps)
data-
modulation
pilot-
modulation
increase
(Mbps)
% gain Enhanced
rate
(Mbps)
12/18 QPSK BPSK 1 5-8 19
12/18 QPSK QPSK 2 11-16 20
24/36 16-QAM BPSK 1 3-4 26
24/36 16-QAM QPSK 2 5-8 38
48/54 64-QAM BPSK 1 2-3 50(51)
48/54 64-QAM QPSK 2 4-6 56(57)
4.5 Conclusion
In this chapter it has been shown that the proposed semiblind pilot scheme is viable,
i.e. it is possible to transmit data information on the pilot tones, without loss in
performance or requiring additional resource. This proposed technique provides up
to two Mbps increase in data rate. An algorithm has been described to implement the
proposed scheme. Depending on the data rate chosen, pilot sub carrier can be loaded
with QPSK or BPSK adaptively based on the SNR margin available. In systems
where more pilots are necessary, the proposed scheme can yield much more benet.
Since with the proposed semi blind pilot scheme, the overhead due to pilot
sub carriers can be reduced, therefore, instead of transmitting additional information
bits, the proposed semiblind pilot technique can be exploited to add more pilots to
the system for better estimation of the residual phase error which will lead to better
BER performance. Adding more pilot sub carriers would require a higher overhead,
however since the pilots can carry some data bits now, the pilot overhead can be
compensated by overloading the pilots with the bits as described in this Chapter.
Center for TeleInFrastruktur (CTIF), Aalborg University
4.5 Conclusion 89
In addition to the above, a low complexity residual phase tracking algorithm
in a semi blind system has been proposed to avoid the use of the highly complex
inverse tangent function. The algorithm has also been compared against the exact
algorithm, which uses the inverse tangent function to compute the residual phase,
operating in a completely data aided system. The results show, that the performance
degradation of the low complexity algorithm is within 12 dB, for very small residual
phase. For comparatively larger residual phase errors the proposed low complexity
algorithm is found to be more stable as compared to the exact algorithm. The reduc-
tion in complexity achieved by the proposed algorithm for residual phase tracking is
quite signicant when small, low powered, low complexity hand held and body worn
next generation wireless communicating devices are considered. The algorithm re-
duces complexity by avoiding the computation of the highly complex inverse tangent
function, which requires more than sixty arithmetic operations for series expansion
implementation.
Finally it can be concluded that the proposed semi blind system performs as
good as the currently used data aided phase tracking system. This gives a scope of
increasing the eciency of a standard OFDM based WLAN system by up to sixteen
percent for IEEE 802.11a/g type systems. Much more gain is expected for schemes
which need more pilots such as outdoor multiuser OFDMA systems. The good
performance of the proposed low complexity algorithm in the semi blind environment
makes it even more attractive.
c
Suvra Sekhar Das, 2007
90 Chapter 4. Bit loading on Pilot Sub Carriers
Center for TeleInFrastruktur (CTIF), Aalborg University
5
Adaptive Sub Carrier Bandwidth
In the previous chapter, a technique to overload the pilot sub carriers in order to
increase the spectral eciency of the system has been discussed. The focus of this
chapter is to present another novel transmission technique which has the potential
to enhance the spectral eciency of OFDM in Doppler frequency spread channel
conditions. The proposed scheme does not demand a signicant increase in the signal
processing complexity of User Equipment (UE) to improve the spectral eciency.
OFDM based wireless systems support very high spectral eciency, but their
performance is severely impaired by Inter Carrier Interference (ICI). ICI is introduced
by frequency synchronization errors. Doppler frequency spread, due to terminal mo-
bility, is one of the major contributors to ICI in OFDM systems. Many schemes to
reduce ICI have been proposed, but they are very complex and some reduce system
eciency. ICI is proportional to the received signal strength and Doppler frequency
spread. In a wide area since users will experience various signal strength due to path
loss and shadowing, and since the amount of Doppler will be dierent for each user
because of independent velocities, the ICI experienced by the users will vary over a
wide range. For a certain amount of frequency shift, the ICI is inversely proportional
to the sub carrier bandwidth. Therefore, it is proposed in this work to dynamically
adapt the sub carrier bandwidth along with adaptive bit loading to overcome the
91
92 Chapter 5. Adaptive Sub Carrier Bandwidth
eect of ICI. This scheme avoids the need for complex interference algorithms at
the receiver while making the system optimum for each situation. The focus of this
chapter is to describe the architectures and analyze the performance of the proposed
Adaptive Sub Carrier Bandwidth (ASB)-OFDM systems. Results show that ASB can
provide higher throughput than its Fixed Sub Carrier Bandwidth (FSB) counterpart
when both may use adaptive bit loading per sub carrier.
This chapter contains two sections. Each section analyzes a dierent imple-
mentation architecture for adaptive sub carrier bandwidth.
5.1 Adaptive Sub Carrier Bandwidth in TDM-OFDM
Frequency synchronization errors come from residual carrier frequency oset, sam-
pling frequency oset, phase noise and Doppler frequency spread. Residual carrier
oset is present due to non ideal performance of the carrier synchronization algo-
rithm at the receiver. Sampling frequency oset is present due to clock frequency
mismatch [45]. Phase noise arise because of imperfections in the local oscillators [68].
Doppler frequency spread is caused due to user mobility and movement of reectors.
Doppler frequency spread consists of multiple of frequency osets as each multi path
may have a dierent frequency shift [20]. Since carrier synchronization algorithms
are designed to track only one frequency oset, it becomes dicult to eliminate the
frequency spread due to Doppler. These impairments cause loss in orthogonality be-
tween the sub carriers and give rise to ICI and which severely limits the performance
of OFDM systems [69],[70]. ICI is proportional to the received signal strength. It is
also proportional to the square of the residual carrier oset plus Doppler frequency
spread and is inversely proportional to the sub carrier bandwidth. Most approaches
to combat ICI are towards using frequency synchronization or interference cancela-
tion [19, 21]. They are usually very complex and sometimes there is loss in bandwidth
eciency. An important aspect which is not considered by these methods is that the
amount of ICI is not same in all receivers under one coverage area. This is because
the signal strength and mobility conditions of users, on which the amount of ICI is
dependent, vary across a wide range. Another consideration is that dierent users
will have dierent BER requirements depending on the service. These algorithms do
not address this issue.
In all existing systems, the sub carrier bandwidth is kept constant. The band-
width is chosen large enough to tolerate a certain amount of Doppler spread. In
a dierent approach to mitigate ICI, Adaptive Sub Carrier Bandwidth (ASB) can
Center for TeleInFrastruktur (CTIF), Aalborg University
5.1 Adaptive Sub Carrier Bandwidth in TDM-OFDM 93
be used since ICI is inversely proportional to the sub carrier bandwidth for certain
maximum frequency oset. Such methods are expected to relieve the receiver from
implementing the complex ICI cancelation schemes. Authors in [71] shows the impact
of varying the number of sub carriers
1
but do not discuss its dynamic realization. To
address the target of satisfying a required bit error rate (BER) while maximizing the
throughput, adaptive bit loading [26] can be used in conjunction with ASB.
This kind of a novel OFDM system, which can adaptively select the sub carrier
bandwidth and bit loading based on Doppler condition, signal strength and BER
requirement is being proposed and analyzed here.
5.1.1 System Description
One implementation of ASB can be based on Time Division Multiplexing (TDM)
system. The number of sub carriers in dierent slots may be changed to generate
dierent sub carrier bandwidths. Users with similar requirement of sub carrier band-
width may share a time slot. The time frequency diagram can appear as in Figure 5.1.
This model is used for analysis via simulation in this work. Among other possible
T
gi
T
f1
T
gi
T
f2
T
gi
T
f3
f
sc3
f
sc2
f
sc1
Frameduration Frameduration Frameduration
S
y
s
t
e
m
B
a
n
d
w
i
d
t
h
Figure 5.1: Time frequency diagram for the proposed TDM based ASB OFDM
implementations, another one can be Band Division Multiplexing (BDM) based as
in Fig. 5.2. The entire available bandwidth may be divided into subbands with
dierent sub carrier bandwidth in each sub band for example a 100 MHz may be
divided into chunks of 20 MHz, 10 MHz or 5 MHz. Each sub band can be operated
on by an IFFT with dierent number of sub carriers. The User Equipment (UE) is
assumed to require only one type of sub carrier bandwidth and hence will operate
on only one subband. Therefore only one Programable FFT [72] is needed for the
user equipment. With a changing requirement of the sub carrier bandwidth the clock
1
For a given system bandwidth varying the number of sub carriers implies a variation in the sub
carrier bandwidth.
c
Suvra Sekhar Das, 2007
94 Chapter 5. Adaptive Sub Carrier Bandwidth
T
g
i
T
f
1
T
g
i
T
f
2
T
g
i
T
f
3
f
s
c
3
f
s
c
2
f
s
c
1
F
r
a
m
e

d
u
r
a
t
i
o
n
SubSystem1
SubSystem 2
Sub System 3
TotalSystemBandwidth
Figure 5.2: Time frequency diagram of the proposed ASB with BDM OFDM
and the FFT size of the programable FFT may be dynamically congured. At the
base station, as many FFTs may be used as there are dierent types of sub carrier
bandwidths.
As mentioned earlier, adaptive bit loading is done along with ASB. Though
joint bit and power adaption is optimal, it has been observed that the gain obtained
in keeping power constant while varying the rate is very close to being optimal [26].
Therefore the power per sub carrier is xed and equally distributed on all data sub
carrier. The rate is varied on each sub carrier by means of adaptive modulation. The
study will be restricted to systems without forward error control (FEC) coding, as
was done in [26] to analyze potential of the scheme.
5.1.2 Analytical Model
The time domain signal of the s
th
transmitted OFDM symbol is dened in (2.19)
(2.22). The received signal is given in (3.11) (3.14). In continuation to the above
(k, k

, ) = k k

+ , [s, k

] is taken as the frequency domain noise component

per sub carrier bandwidth. The symbol H[s, k

] is used as the channel coecient for

k
th
sub carrier of s
th
OFDM symbol instead of H
s
[k

] which was used earlier. For the

simplicity of notation, the index s is dropped where ever it is not needed. X[s, k

] is
zero mean, which implies that the interference term is also zero mean. The power of
Center for TeleInFrastruktur (CTIF), Aalborg University
5.1 Adaptive Sub Carrier Bandwidth in TDM-OFDM 95
the interference term due to ICI is then the same as its variance, which becomes,

2
ICI
X[k

]
= E[[X[k][
2
]
N
f
2
1

k=
N
f
2
,
k=k

[H[k][
2
sinc
2
_
(
f
f
sc
+ k k

)
_
, (5.1)
where E is the expectation operator. Replacing E[[X[k][
2
] by P
X[k]
which is the
average power per sub carriers, and, [H[k

][
2
by P
H[k

]
, the ICI power at the receiver
on sub carrier k

is:

2
ICI
X[k

]
P
X[k]
P
H[k

]
N
f
2
1

k=
N
f
2
,
k=k

sinc
2
_
(
f
f
sc
+ k k

)
_
.
(5.2)
It is assumed that the coherence bandwidth is large enough, so that the channel
coecients are the same for the most signicant (neighboring) sub carriers [45]. Also,
for small values of ff
sc
[45],

2
ICI
X[k

2
ICI
X[k

1
3
P
X[k]
P
H[k

]
(
f
f
sc
)
2
. (5.3)
Therefore signal to interference plus noise ratio (SINR) is

rx
[k

]
P
X[k

]
P
H[k

]
sinc(
f
c
f
sc
)
2

+
1
3
P
X[k]
P
H[k

]
(
f
f
sc
)
2
. (5.4)
Figure 5.3 shows the plot of the average SINR vs sub carrier bandwidth for
dierent velocity conditions. Each curve is for a particular velocity. It can be seen
from this gure, that SINR improves with increasing sub carrier bandwidth for a given
Doppler spread. But this does not ensure a monotonically increasing throughput with
increasing sub carrier bandwidth. This can be seen from Figure 5.4. This gure shows
that the throughput, for a given Doppler velocity, is maximum for a certain sub carrier
bandwidth only. The throughput curves have been obtained by considering adaptive
bit loading per sub carrier, which is discussed later. The decrease in throughput
after a certain sub carrier bandwidth can be attributed to the increasing sub carrier
bandwidth which causes the OFDM symbol duration to decrease thereby increasing
the overhead due to guard interval. This can be understood by referring to (1.1). The
c
Suvra Sekhar Das, 2007
96 Chapter 5. Adaptive Sub Carrier Bandwidth
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
x 10
4
4
2
0
2
4
6
8
10
12
14
16
Sub carrier bandwidth in Hz
S
I
N
R

i
n

d
B


10 kmph
80 kmph
160 kmph
240 kmph
Figure 5.3: SINR vs sub carrier bandwidth at 15dB SNR
Guard Interval (GI) which is decided based on the channel delay spread is mostly xed
and should not change with the change in sub carrier spacing. As long as the length
of GI is suciently large so that ISI is within tolerable limits the choice of sub carrier
bandwidth and choice of GI can be treated independently. The next chapter of this
thesis presents an analysis on the choice of GI. For the above it can be seen that
by choosing the appropriate sub carrier bandwidth there is potential for signicant
improvement in throughput.
5.1.3 Algorithm for Adaptive Bandwidth for Sub Carriers
An algorithm is proposed here to dynamically select the appropriate sub carrier band-
width and bit load per sub carrier to maximize the throughput while satisfying a
required BER constraint. The sub carrier bandwidth can be chosen as
f
chosen
= arg max
f
m
[Thpt(f
m
)], (5.5)
such that
f
m
< B
c
, where B
c
is the coherence bandwidth, and (5.6)
T
s
< T
c
, where T
c
is the coherence time. [34] (5.7)
Center for TeleInFrastruktur (CTIF), Aalborg University
5.1 Adaptive Sub Carrier Bandwidth in TDM-OFDM 97
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
x 10
4
0
0.5
1
1.5
2
2.5
Sub carrier bandwidth in Hz
T
h
r
o
u
g
h
p
u
t
(
b
/
s
/
H
z
)


10 kmph
80 kmph
160 kmph
240 kmph
Figure 5.4: Throughput vs sub carrier bandwidth at 15dB SNR
The index m, runs through the allowable sub carrier bandwidths (out of a nite
number of options) while meeting the constraints of coherence bandwidth and coher-
ence time as mentioned above. The estimated throughput Thpt(f
m
) in (5.5) can be
written as
1
B
w
(
1
f
m
+ T
gi
)
N
f
2
1

k=
N
f
2
b
L
(k, f
m
)(1 b
o
(k, f
m
)). (5.8)
In the above expression B
w
denotes the system bandwidth. The bit load estimate
per sub carrier used in (5.8), assuming square constellation (QAM) can be expressed
from [26] as,
b
L
(k, f
m
) = 2
1
2
log
2
_
1
1.6
ln(
b
o req
0.2
)

rx
(k, f
m
)
_
, | (5.9)
where .| is the oor operation, which has been introduced as a square constellation
such as QAM in considered to be used. In the above, b
o req
is the target BER which
is to be satised. The BER associated with the chosen bit load is [26]
b
o
(k, f
m
) = 0.2e

1.6
rx
(k,f
m
)
2
b
L
(k,f
m
)
1
, (5.10)
c
Suvra Sekhar Das, 2007
98 Chapter 5. Adaptive Sub Carrier Bandwidth
which is tight for high SNR and has been claimed in [26] to be valid within 1.5dB
for 4-QAM to 1024-QAM for bit error rate BER 10
3
. The expression
rx
(k, f
m
)
used above is taken as

rx
(k, f
m
)
P
X[k]
P
H[k]
sinc(
f
f
m
)
2

+
1
2
(
f
f
m
)
2
P
X[k]
P
H[k]
. (5.11)
Executing of the Algorithm for ASB: The following steps may be executed
in sequence to implement the above algorithm.
1. Select one sub carrier bandwidth from the available options.
2. Evaluate (5.11), i.e. SINR at each sub carrier for the selected sub carrier spacing.
For the calculations, f has to be estimated using advanced schemes such as [73].
3. Use the above in nding bit load for this chosen value of sub carrier spacing
following (5.9).
4. Calculate the associated BER for each sub carrier for the chosen bit load using
(5.10).
5. Use the above calculations of bit load and related BER for each sub carrier in
calculating the throughput for the chosen sub carrier bandwidth following (5.8).
6. Store the value of the estimated throughput along with the value of sub carrier
bandwidth and associated bit loads per sub carrier.
7. Repeat the above steps for all possible values of sub carrier bandwidth.
8. Finally execute (5.5) to select the sub carrier bandwidth and bit loads per sub
carrier which has the highest estimated throughput.
9. The rate of change of Doppler condition and average channel quality is much
slower compared to the rate of change of channel coecients. Therefore it can
be considered to adapt the sub carrier spacing, which is related to the Doppler
and average SNR condition, at a slower rate than adapting the bit load which is
related to the instantaneous channel gains. The bit loading should be done once
per coherence time of the channel coecients. i.e. once a sub carrier spacing
is selected, it may be used until the Doppler condition or the average signal
strength changes signicantly and hence step 1 and step 7 may be skipped, and
step 8 may be modied to read: Finally execute (5.5) to select bit loads per sub
Center for TeleInFrastruktur (CTIF), Aalborg University
5.1 Adaptive Sub Carrier Bandwidth in TDM-OFDM 99
carrier which has the highest estimated throughput for the chosen sub carrier
bandwidth.
The overhead for signalling the bit load and chosen sub carrier bandwidth depends on
the rate of signalling, number of bits used and also on the feedback mechanism. For
simplicity of analysis but to keep the results realistic, it is assumed that BPSK is used
for signalling the feedback information. It is also assumed that number of bits per
feedback is taken as log
2
[n(X)], where n(X) denotes the number of signalling levels
for X where X is the feedback parameter. The rate of feedback of the bit loading
parameter which depends on the channel gains is done once per coherence time per
coherence bandwidth, while that for the maximum Doppler frequency oset which
aects the selection of the sub carrier spacing is done once per second. This overhead
which will vary as a function of the channel statistics is considered in the simulations.
An example of a typical overhead calculations is shown in the later chapters on Link
Adaptation.
5.1.4 Results and Discussion
Each coecient of the time domain channel impulse response is taken as Rayleigh
distributed with Jakes [34] spectrum. Exponential power delay prole with rms
delay spread of 2 micro seconds is used. Bandwidth of 5 MHz at carrier of 3.6 GHz
is considered. The target BER is kept at 10
2
. To implement Adaptive Sub Carrier
Bandwidth (ASB) in the simulations the TDM mode (Figure 5.1) is taken. Number
of bits that can be loaded on a sub carrier are 0,2,4,6,8 and 10, where 0 means
no transmission. The curves labeled 2048, 1024,...,128, are for systems with xed
sub carrier bandwidth using as many sub carriers for the 5 MHz channel bandwidth.
These numbers correspond to 2.4 KHz to 39.063 KHz of sub carrier bandwidth. The
options for number of sub carries (sub carrier bandwidth) for the proposed ASB
system is selected from this range. The curve labeled with ASB, is for the proposed
adaptive sub carrier bandwidth system.
Figure 5.5 shows the average sub carrier spacing selected by ASB scheme for
dierent values of velocity at a received SNR of 15 dB. The increase in average value
of sub carrier bandwidth selected with increasing velocity can be easily seen.
Now the throughput of ASB system is compared against FSB systems. The
throughput of FSB system is computed considering adaptive bit loading per sub
carrier as is considered for the ASB system. If f
sc
is the sub carrier bandwidth,
then the throughput is calculated using (5.8)(5.11) where f
m
is replaced by f
sc
.
c
Suvra Sekhar Das, 2007
100 Chapter 5. Adaptive Sub Carrier Bandwidth
0 50 100 150 200 250 300
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
x 10
4
Doppler Velocity in Kmph
S
u
b

C
a
r
r
i
e
r

B
a
n
d
w
i
d
t
h
(
H
z
)


ABS
Figure 5.5: Sub carrier bandwidth selected by the proposed ASB system.
Figure 5.6 compares the throughput for ASB scheme against standard OFDM systems
with xed number of sub carrier bandwidth for dierent values of the number of
number carriers at 15 dB SNR. It is clearly seen that a given sub carrier bandwidth
has the highest throughput over a small range of velocity, while ASB has the highest
throughput over the entire range. The system with 2048 sub carriers (f
sc
= 2.4KHz)
is best for mobility less than 10kmph. Between 10 and 40 kmph system with 1024 sub
carriers (f
sc
= 4.88KHz) has the highest performance. Between 40 kmph and 140
kmph, the one with 512 sub carriers (f
sc
= 9.77KHz) is the most ecient, beyond
which the system with 256 sub carriers (f
sc
= 19.531KHz) is the best. It can be
noted that the ASB system rides the envelope, i.e. it has the highest throughput over
all velocities.
In the low mobility region ASB is better than one using 512 sub carrier by
about 12% and about 25% better than one using 256 sub carriers. In the high
mobility region (near 200 kmph) ASB is better than system with 512 sub carriers by
about 25% and by more than 30% over system with 1024 sub carriers, when the one
with 2048 sub carriers which is best for low mobility almost fails.
Figure 5.7 shows similar curves as above but for a higher SNR (25 dB). Com-
paring with the previous gure, one can clearly nd that the dierence in received
signal strength has caused the optimal range for each OFDM system (with dierent
but xed sub carrier bandwidth) to change, but ASB is still the most ecient in the
Center for TeleInFrastruktur (CTIF), Aalborg University
5.1 Adaptive Sub Carrier Bandwidth in TDM-OFDM 101
0 50 100 150 200 250 300
0
0.5
1
1.5
2
2.5
Doppler Velocity in Kmph
T
h
r
o
u
g
h
p
u
t

i
n

b
i
t
s
/
s
/
H
z


2048
1024
512
256
128
ABS
Figure 5.6: Throughput comparison of the proposed ASB vs standard FSB OFDM systems,
at 15 dB SNR.
entire range. Between 0 and 10 kmph, it is seen that 2048 has the highest eciency,
between 10 and 50 kmph, the system with 1024 sub carriers is found to be optimal,
between 50 kmph and 90 kmph, the system with 512 sub carriers is optimal, beyond
which it is seen that the system with 256 is optimal. The ASB scheme is better than
the xed sub carrier bandwidth system by about 10% to 30% in dierent velocity
regions. Finally Figure 5.8 shows the BER curves for all systems. It can be observed
that by use of Adaptive Bit Loading (ABL), the BER is maintained below the target
level (10
2
) for all system and for all velocities. Though the target BER is satised
by both schemes (ASB and FSB) using ABL, it can be said that ABL alone is not
sucient for FSB OFDM system to be ecient in all Doppler conditions. ASB with
ABL improves the throughput by a signicant amount.
c
Suvra Sekhar Das, 2007
102 Chapter 5. Adaptive Sub Carrier Bandwidth
0 50 100 150 200 250 300
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Doppler Velocity in Kmph
T
h
r
o
u
g
h
p
u
t

i
n

b
i
t
s
/
s
/
H
z


2048
1024
512
256
128
ABS
Figure 5.7: Throughput comparison of the proposed ASB vs FSB OFDM systems, at 25
dB SNR.
50 100 150 200 250 300
2
3
4
5
6
7
8
9
10
x 10
3
Doppler Velocity in Kmph
B
E
R


2048
1024
512
256
128
ASB
Figure 5.8: BER of the proposed ASB and FSB OFDM systems at 15 dB SNR, when
target BER is 0.01
Center for TeleInFrastruktur (CTIF), Aalborg University
5.1 Adaptive Sub Carrier Bandwidth in TDM-OFDM 103
5.1.5 Conclusion
It can be concluded that the proposed Adaptive Sub Carrier Bandwidth (ASB) system
in Doppler frequency spread scenario with varying received signal strength conditions
has the potential to improve the performance of Fixed Sub Carrier Bandwidth (FSB)-
OFDM system by 10% to 30% in dierent situations. While the latter system with
a chosen but xed sub carrier bandwidth is optimum only over a small range of
velocities and received signal strength, the proposed ASB has optimum performance
over all conditions. ASB avoids the complex compensation or interference cancelation
mechanism at the receiver, thereby allowing lower complexity receivers. Thus the
advantage of increased throughput with possibility of low complexity receivers makes
the proposed ASB a potential candidate for consideration in future systems. The
promising results pave the path for further investigation with realistic impairments
such as channel information feedback delay, channel estimation error, along with the
use of forward error control coding.
c
Suvra Sekhar Das, 2007
104 Chapter 5. Adaptive Sub Carrier Bandwidth
5.2 OFDMA Framework
In this section a dierent system architecture for implementing ASB will be described.
This kind of architecture can be very well suited for OFDMA systems. The derivations
presented in this chapter are unique as they provide the analysis of the interference
generated OFDM systems when two OFDM system with dierent symbol durations
are lie next to each other on orthogonal sub carriers. The analysis is very generic as
it covers several dierent combination of the ratio of symbol durations. Interestingly
when this condition occurs then the orthogonality of neighboring sub carriers is lost
however since a set of sub carriers use similar symbol duration, the impact is not
severe as detailed in the signal to interference plus noise ratio analysis detailed in
this Section. For analysis we shall consider a system whose baseband transmitter
architecture at the Base Station (BS) is given in Figure 5.10. The system of the
proposed ASB OFDM is also referred to as Variable Sub Carrier Bandwidth (VSB)
OFDM equivalently. The system is assumed to support multiple users simultaneously
across dierent sub carriers. There are dierent groups of sub carriers, where the
groups have dierent sub carrier bandwidths. Each IFFT spans the entire system
bandwidth by using the same sampling period. In each IFFT, N1, N2, N3, etc.
denotes the number of sub carriers, where larger numbers generate narrower sub
carriers bandwidths. It can be noted that only a fraction of the sub carriers in each
IFFT is activated (the number of active sub carriers in each IFFT can be made to
vary), and that the active sub carriers of the dierent groups are selected such that
the frequency band spanned by the dierent groups do not overlap. The frequency
domain view of such a conguration is given in Fig. 5.9. A user must be allocated to
IFFT (N1points)
IFFT (N2points)
ActiveSubcarriers
Figure 5.9: Frequency domain conguration of VSB OFDM
a particular sub carrier group, whose sub carrier bandwidth suits the requirement of
the user conditions optimally. At the receiver a exible FFT can be implemented [72]
as a user will need only one type of sub carrier at a time, while at the base station as
Center for TeleInFrastruktur (CTIF), Aalborg University
5.2 OFDMA Framework 105
many IFFTs are used as sub carrier types. The time frequency diagram of the signal
can be represented as in Figure 5.2 except that there need not be any guard band
between the sub-bands. It can be seen that since sub bands are next to each other,
and since the symbol duration are not the same, the orthogonality will no longer
be maintained. Therefore the system can also be called non-orthogonal frequency
division multiplexing. However, not all sub carriers will be non orthogonal. Some
sub carriers whose frequencies are integer multiple of another will still be orthogonal.
The receiver at the BS can be as in Figure 5.11.
c
Suvra Sekhar Das, 2007
106 Chapter 5. Adaptive Sub Carrier Bandwidth
BitStream
forauser
Symbol
Modulator
ModulatedSymbols
BPSK/QPSK/QAM16
/QAM64..
BitStream
forauser
Symbol
Modulator
ModulatedSymbols
BPSK/QPSK/QAM16
/QAM64..
Scheduler/
Multiplexer
(withparallel
to
sarallel
Converter)
ToIFFT
Block1
ToIFFT
Block1
IFFT
Block1
(NSub
carriers)
1
QAM
Modulated
Symbols
TimeDomain
Signal
Bw
Active
SubCarriers
Parallel
to
Serial
ToIFFT
BlockM
IFFT
BlockM
(N Sub
carriers)
M
TimeDomain
Signal
Bw
Active
SubCarriers
to
Serial
Parallel
Add
CyclicPrefix
ToIFFT
BlockM
Ineachtimeinstant
addthecorresponding
timesamplesfromall
IFFT outputs
Add
CyclicPrefix
+
Figure 5.10: Downlink Transmitter for the proposed VSB OFDMA at the base station
Center for TeleInFrastruktur (CTIF), Aalborg University
5.2 OFDMA Framework 107
BitStream
forauser
SymbolDe-
Modulator
ModulatedSymbols
BPSK/QPSK/QAM16
/QAM64..
BitStream
forauser
ModulatedSymbols
BPSK/QPSK/QAM16
/QAM64..
De-
Multiplexer
(withparallel
to
sarallel
Converter)
FromFFT
Block1
FFT
Block1
(NSub
carriers)
1
Modulated
Symbols
TimeDomain
Signal
Bw
Active
SubCarriers
FFT
BlockM
(N Sub
carriers)
M
TimeDomain
Signal
Bw
Active
SubCarriers
to
Serial
Parallel
Remove
CyclicPrefix
FromFFT
BlockM
Remove
CyclicPrefix
to
Serial
Parallel
SymbolDe-
Modulator
FromFFT
Block1
FromFFT
BlockM
SendaCopy
toeachbranch
Figure 5.11: Up link Receiver for the proposed VSB OFDMA at the base station
c
Suvra Sekhar Das, 2007
108 Chapter 5. Adaptive Sub Carrier Bandwidth
5.2.1 Analytical Model
Table 5.1: List of Symbols
Symbol Explanation
g Index of the group, each group indicates a type of sub carrier
k
g
Sub carrier index for group
s
g
OFDM symbol index for group
X
k
g
,s
g
sub carrier symbol
P
k
g
,s
g
Power on sub carrier
T
s
g
OFDM symbol period
T
f
g
DFT period for a group
T
gi
g
Guard period for a group
f
c
Carrier Frequency
T Sampling period
t Time

T
s
(t) Gate Function [46]
H[t, k
g
] Channel Transfer Function for sub carrier k
g
(t) Noise Term
N
gi
g
Number of samples in the guard period
N
f
g
Number of samples in the FFT
N
s
g
Number of samples in an OFDM symbol
The band pass signal for OFDM transmitted through a system as shown in
Figure 5.10 can be written as
x (t) = Re

s
g

k
g
X
k
g
,s
g
_
P
k
g
,s
g
e
j2
k
g
N
f
g
T
(ts
g
T
s
g
T
gi
g
)
.
T
s
(t s
g
T
s
g
)e
j2f
c
t
, (5.12)
where the symbols are dened in Table 5.1. After passing through the channel, the
signal can be represented as [20],
r(t) = Re

s
g

k
g
_
P
k
g
,s
g
X
k
g
,s
g
_

max
0
h(t, )
e
j2(f
c
+f
d
)(t)
e
j2
k
g
N
f
g
T
(ts
g
T
s
g
T
gi
g
)

T
s
(t s
g
T
s
g
) d + (t), (5.13)
where
max
is the maximum tail of the channel impulse response, (t) is the noise
component and f
d
is the Doppler frequency for delay . To simplify the above
Center for TeleInFrastruktur (CTIF), Aalborg University
5.2 OFDMA Framework 109
expression, we use f
max
, the maximum Doppler frequency instead of f
d
in all delays
in the above equation. This will also give us the worst case scenario. Then (5.13)
becomes,
r(t) = Re

s
g

k
g
_
P
k
g
,s
g
X
k
g
,s
g
_

max
0
h(t, )
e
j2(f
c
+f
max
)(t)
e
j2
k
g
N
f
g
T
(ts
g
T
s
g
T
gi
g
)

T
s
(t s
g
T
s
g
) d + (t). (5.14)
If the guard interval is larger than maximum channel impulse response tail and con-
sidering that h(t, )e
j2(f
c
+f
max
)()
is inseparable from h(t, ), then the above can be
written as
r(t) = Re

s
g

k
g
_
P
k
g
,s
g
X
k
g
,s
g
H[t, k
g
]
e
j2(f
c
+f
max
)(t)
e
j2
k
g
N
f
g
T
(ts
g
T
s
g
T
gi
g
)

T
s
(t s
g
T
s
g
) d + (t). (5.15)
At the receiver the signal is down converted by multiplying the above by e
j2f

c
, where
f

c
is the frequency of the local oscillator at the receiver and f

c
= f
c
+f
c
, where f
c
is
the carrier frequency oset error. After passing through a low pass lter the signal
can be written as
r(t) =

s
g

k
g
_
P
k
g
,s
g
X
k
g
,s
g
H[t, k
g
]e
j2(f
c
+f
max
)(t)
e
j2
k
g
N
f
g
T
(ts
g
T
s
g
T
gi
g
)

T
s
(t s
g
T
s
g
) d + (t). (5.16)
The received signal is then truncated, with a gate function as below
r
s
g
(t) = r(t)
T
f
g
(t s
g
T
s
g
T
gi
g
). (5.17)
At the receiver the signal is then sampled. We assume transmit and receive clock
synch error is negligible. The sampled signal can be written as [45]
r
s
g
(n) = r(n)
N
f
g
(n s
g
N
s
g
N
gi
g
). (5.18)
The sub carrier at the receiver constructed by an FFT operation (considering ideal
c
Suvra Sekhar Das, 2007
110 Chapter 5. Adaptive Sub Carrier Bandwidth
timing synch) is
R
s
g
,k

g
=
1
N
f
g
(s
g
+1)N
s
g

n=s
g
N
s
g
+N
gi
g
r
s
g
(n)e
j2
k

g
N
f
g
(ns
g
N
s
g
N
gi
g
)
. (5.19)
Considering that the channel is almost static over one coherence time, which is larger
than the largest OFDM symbol in the group, one can write the noiseless part of the
received sub carrier as [45]
R
s
g
,k

g
=

s
g

k
g
_
P
k
g
,s
g
X
k
g
,s
g
H[k
g
, s
g
]e
j2(k

g
,k
g
)
e
j2N
f
g
(f
c
+f
max
,k

g
,k
g
)
sinc(N
f
g
(f
c
+ f
max
, k

g
, k
g
)), (5.20)
where
(k

g
, k
g
) =
k
g
N
f
g
(s
g
N
s
g
+ N
gi
g
)
k

g
N
f
g
(s
g
N
s
g
+ N
gi
g
) (5.21)
(f
c
+ f
max
, k

g
, k
g
) = (f
c
+ f
max
)T +
k
g
N
f
g

g
N
f
g
. (5.22)
Therefore the desired signal is
A(k

g
, s
g
) =
_
P
k

g
,s
g
X
k

g
,s
g
H[k

g
, s
g
]e
j2N
f
g
(f
c
+f
max
)
sinc(N
f
g
(f
c
+ f
max
)). (5.23)
Thus the power of the desired signal is
E
|A(k

g
,s
g
)|
2
= P
k

g
,s
g
[H[k

g
, s
g
][
2
sinc
2
(N
f
g
(f
c
+ f
max
)). (5.24)
We assume E
|X
k
g
,s
g
|
2 = 1. The interference from the sub carriers of similar type (g = g,
but k
g
,= k

g
) is
B(k

g
, s
g
) =

k
g
=k

g
_
P
k
g
,s
g
X
k
g
,s
g
H[k
g
, s
g
]e
j2(k

g
,k
g
)
.e
j2N
f
g
(f
c
+f
max
,k

g
,k
g
)
sinc(N
f
g
(f
c
+ f
max
, k

g
, k
g
)). (5.25)
Center for TeleInFrastruktur (CTIF), Aalborg University
5.2 OFDMA Framework 111
The power of this interference is
E
|B(k

g
,s
g
)|
2
=

k
g
=k

g
P
k
g
,s
g
[H[k
g
, s
g
][
2
sinc
2
(N
f
g
(f
c
+ f
max
, k

g
, k
g
)) (5.26)
The orthogonality loss due to the structure of the transmitter being analyzed
can be understood by noting that even if f
c
+f
max
is made zero in the above expression,
the term sinc
2
(N
f
g
(f
c
+ f
max
, k

g
, k
g
)) does not vanish as seen from (5.22) since the
number of sub carriers in the IFFTs is dierent, i.e. the symbol periods are dierent.
The interference from sub carriers of other types (g ,= g) is
C(k

g
, s
g
) =

g= g

k
g
_
P
k
g
,s
g
X
k
g
,s
g
H[k
g
, s
g
]e
j2(k

g
,k
g
)
e
j2N
f
g
(f
c
+f
max
,k

g
,k
g
)
sinc(N
f
g
(f
c
+ f
max
, k

g
, k
g
)) (5.27)
The power of this interference is E
|C(k

g
,s
g
)|
2
=

g= g

k
g
P
k
g
,s
g
[H[k
g
, s
g
][
2
sinc
2
(N
f
g
(f
c
+ f
max
, k

g
, k
g
)). (5.28)
The received SINR can thus be written as

rx
(k

g
, s
g
) =
E
|A(k

g
,s
g
)|
2
E
|B(k

g
,s
g
)|
2
+ E
|C(k

g
,s
g
)|
2
+ N
o
f
N
f
g
. (5.29)
The capacity expression for the above SINR is,
Cap(k

g
, s
g
) = log
2
(1 +
rx
(k

g
, s
g
))
N
f
g
N
s
g
. (5.30)
5.2.2 Results and Discussion
For numerical analysis, bandwidth of 5 MHz with a carrier frequency of 3.6 GHz is
used. Exponential power decay prole with an average rms delay spread of 1.5 s is
considered for the channel.
Figure 5.12 shows the SINR of a standard OFDM system at a received signal
to noise ratio (SNR) of 20dB. The legend, for example 1024, denotes a standard FSB-
OFDM system with as many sub carriers. It is seen that with decreasing number of
sub carriers, i.e. increasing sub carrier bandwidth, the resilience to ICI is increased. It
may follow that the system with smallest number of sub carriers is the best, but such
c
Suvra Sekhar Das, 2007
112 Chapter 5. Adaptive Sub Carrier Bandwidth
is not observed from the capacity curves in Figure 5.13. The anomaly is understood
by considering the capacity expression (5.30) which includes the impact of guard
interval (GI). GI depends only on the maximum channel impulse response length and
should not be changed when the FFT size changes. It can be seen from Figure 5.13
that each OFDM system with a dierent sub carrier bandwidth is best in only a range
of the velocity.
Figure 5.14 shows similar curves but when the received SNR is 10dB. It can be
observed that the optimal region for a certain sub carrier bandwidth is changed com-
pared to Figure 5.13. It is clear that a particular conguration of OFDM systems (a
particular sub carrier bandwidth) is optimal only over a certain Doppler region and
a certain received SNR situation.
Next, Figure 5.15 shows the average capacity of dierent sub carrier types at
a received SNR of 20dB for the proposed VSB-OFDM system shown in Figure 5.10.
Though the new system introduces some amount of non-orthogonality, yet the
impact is not signicant as found by comparing Figure 5.13 and Figure 5.15. In a
FSB OFDM system, the interference power is almost same across all sub carrier on
an average. In case of the proposed Variable Sub Carrier Bandwidth (VSB), since
dierent sub carrier bandwidths are implemented, the interference on the sub carriers
is dierent.
At 10dB received SNR condition, the performance of the sub carriers of the
new system is found similar to that of Figure 5.14 and hence not repeated. The
structure as in Figure 5.10 provides the exibility to implement the dierent sub
carrier bandwidths in the same system simultaneously.
Dierent users having dierent SNR and Doppler conditions can be allocated
to the sub carrier type which best suits the situation, i.e. the sub carrier bandwidth
with highest capacity for the user situation can be used for the user. Such a situation
is created with two types of users, i.e. one type with very low mobility and the other
type with very high mobility. The capacity of the FSB systems is compared with that
of the proposed VSB under these situation. The comparison is given in Figure 5.16.
When the proposed VSB is used, the sub carrier bandwidth with highest capacity for
a user is selected and allocated. The fraction of system bandwidth for high mobility
users with respect to the total system bandwidth is changed from zero to unity, i.e
between the situation when only low mobility users exist and when only high mobility
users exits. The mark 0.5 denotes situation when both high and low mobility users
exists in equal proportion. For simplicity of analysis all users are supposed to be at
the same received SNR condition of 20dB. The capacity of the proposed VSB system
Center for TeleInFrastruktur (CTIF), Aalborg University
5.2 OFDMA Framework 113
is compared with that of FSB in Figure 5.16. The xaxis denotes the fraction of
system bandwidth used by high mobility users to the total available bandwdith. The
curve labeled 1024 is for FSB OFDM with 1024 sub carriers (sub carrier bandwidth
= 4.88 KHz). Similarly the curves labeled 512 and 256 represent FSB system with
as many sub carriers and hence with 9.76 KHz and 19.5 KHz of sub carrier spacing.
The curve labeled VSB is the one which can support the proposed variable sub carrier
bandwidth by using multiple sub carrier bandwidths (options used in this work are
4.88 KHz, 9.76 KHz and 19.5 KHz) simultaneously at the transmitter. In this system,
the users with smaller values of eective carrier oset use sub carriers with a smaller
sub carrier bandwidth while those with higher Doppler frequency spread use wider
subcarrier spacing. In this situation when there are users with dierent Doppler
condition, it can be seen that any one type of FSB is ecient for only a small range
of values of user distribution, while the proposed VSB has the highest capacity for all
possible distribution of user mobility conditions. It is seen here that when more low
mobility users are present, 1024 system is very good, but its eectiveness decreases
steeply with increasing portion of high mobility users. In the same way FSB with
256 sub carries is good when more high mobility users are present. When more users
are present who are almost static, the gain of the proposed VSB over FSB with 256
carriers is about 16%, and about 8% over FSB with 512 sub carriers. When most
users are highly mobile, then VSB has gain of 20% over FSB with 1024 sub carriers,
and about six percent over FSB with 512 carriers while it has similar capacity as 256
sub carrier system.
0 50 100 150 200 250
10
12
14
16
18
20
22
velocity in kmph
S
I
N
R

i
n

d
B


1024
512
256
Figure 5.12: SINR of standard OFDM systems at 20dB SNR
c
Suvra Sekhar Das, 2007
114 Chapter 5. Adaptive Sub Carrier Bandwidth
0 50 100 150 200 250
3
3.5
4
4.5
5
5.5
6
velocity in kmph
B
i
t
s
/
s
/
H
z


1024
512
256
Figure 5.13: Capacity of standard OFDM systems at 20dB
0 50 100 150 200 250
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
3
velocity in kmph
B
i
t
s
/
s
/
H
z


1024
512
256
Figure 5.14: Capacity of standard OFDM systems at 10dB
Center for TeleInFrastruktur (CTIF), Aalborg University
5.2 OFDMA Framework 115
0 50 100 150 200 250
3
3.5
4
4.5
5
5.5
6
velocity in kmph
B
i
t
s
/
s
/
H
z


1024
512
256
Figure 5.15: Capacity of VSB OFDM at 20dB
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
3
3.5
4
4.5
5
5.5
6
Fraction of system bandwidth for high mobility users
C
a
p
a
c
i
t
y

(
b
i
t
s
/
s
/
H
z
)


1024
512
256
VSB
Figure 5.16: Capacity comparison when users with dierent mobility conditions coexist
c
Suvra Sekhar Das, 2007
116 Chapter 5. Adaptive Sub Carrier Bandwidth
5.2.3 Conclusion
It has been found that a chosen sub carrier bandwidth for a FSB OFDM system is
optimum only over a small range of velocity and received signal strength conditions.
When dierent users with various mobility conditions suer from dierent amount of
ICI, FSB OFDM system is found to be optimal only for a certain range of mobility
distributions. A system which can support multiple sub carrier bandwidths simulta-
neously is proposed and analyzed in this work. Variable sub carrier bandwidths are
used in the proposed system to overcome the impact of ICI since ICI is inversely pro-
portional to the sub carrier bandwidth. This avoids complex ICI cancelation schemes
at the receiver. Depending on the condition of ICI power, which is a function of the
received signal strength and Doppler condition, users may use dierent sub carrier
bandwidths. The new scheme analyzed can vary the sub carrier bandwidth and the
number of sub carrier for each type of sub carrier bandwidth. In this way it can
optimally cater to dierent types of user conditions simultaneously. The gains of the
proposed VSB system over FSB OFDM system is in the range of eight to twenty
percent in terms of capacity. The promising results of this analysis proves the poten-
tial of the proposed VSB scheme to enhance the performance of FSB OFDM systems
though further investigations are needed to compare the gains when non ideal channel
estimation and Doppler frequency spread estimations are encountered.
5.3 Summary
In this Chapter, a new technique namely adaptive sub carrier bandwidth system
has been introduced and evaluated for its eectiveness in tolerating a variety of
Doppler conditions.
It is found to be very robust and eective over OFDM systems using xed sub
carrier bandwidths.
Two dierent architectures have been presented.
The rst one uses standard OFDM system and principles of normal OFDM
can be used in evaluating or designing the system.
The second architecture on the contrary does not maintain orthogonality
among the sub carriers but provides more exibility to the system de-
ployment where dierent FFTs with dierent sizes but having the same
Center for TeleInFrastruktur (CTIF), Aalborg University
5.3 Summary 117
sampling clock and using orthogonal frequency bands for the active regions
are used.
The rst system has the benet that it supports dierent sampling frequencies,
which in turn means it can simultaneously provide support to systems requir-
ing dierent bandwidths. This is highly important if dierent equipments with
dierent data rate capabilities are required to operate optimally in terms of
power consumption in the same system. This is because a high sampling fre-
quency drains the power, which is unnecessary for users requiring a low data
rate. Therefore users with high and low data rate can be separated and each
can optimize the power usage.
The second architecture does not allow this as the sampling frequency of all
FFTs with dierent sizes must be same. Of course it opens up the analysis
for the case where a mixed architecture may be developed. Its advantage is
that there is minimum loss in bandwidth since there is no need for guard band
between two FFTs which is needed in the previous case.
The concern for both the systems is the frame structure. It will require detailed
design and might challenge existing framework.
This kind of systems can be very useful for vehicle to vehicle communication as
well, where devices will encounter large variety of Doppler conditions frequently.
The proposed methods do not use complex receiver algorithms to cancel ICI,
instead avoids it in the an ecient way, thereby reducing the complexity of the
UE.
c
Suvra Sekhar Das, 2007
118 Chapter 5. Adaptive Sub Carrier Bandwidth
Center for TeleInFrastruktur (CTIF), Aalborg University
6
Variable Guard Interval
In the previous chapter the sub carrier bandwidth was adapted to overcome ICI and
increase the spectral eciency. However, the Guard Interval (GI) was assumed to be
xed. The focus of this chapter is to use dynamic adaptation of GI to improve the
spectral eciency further.
GI, in the form of Cyclic Prex (CP), is used in OFDM systems between
two consecutive OFDM symbols to avoid Inter Symbol Interference (ISI). The GI
duration used normally is taken to be more than the maximum excess delay of the
channel. Though the use of GI is important for a simple receiver implementation,
it reduces the spectral eciency of the system. On the other hand using a short
GI introduces ISI thereby degrading the performance. The GI required, depends on
the propagation environment. Instead of using the maximum excess channel delay
as the design factor, dynamic selection of GI can be made to decrease the loss in
spectral eciency. An algorithm for implementing dynamically varying GI length is
proposed in this chapter. The required GI is derived in this work as a function of
several important system and channel parameters which is not available in existing
literature to the best of the authors knowledge. The performance of the algorithm
is analyzed for dynamic channel conditions. Results show that the algorithm can
reduce GI signicantly and thus can improve the throughput. The proposed method
119
120 Chapter 6. Variable Guard Interval
GI
DFT Period
PreviousSymbol CurrentSymbol
Channelimpulseresponse
ISI
Figure 6.1: Eect of small GI.
performs well even in vehicular speed.
6.1 Introduction
To overcome ISI eects in the hostile multipath wireless environment, a GI is used
between consecutive OFDM symbols. This helps in using one tap equalizer per sub
carrier at the receiver. The GI is usually designed to be longer than the maximum
channel delay; its length is selected after channel measurements in the desired im-
plementation scenario. A GI shorter than the maximum excess channel delay causes
ISI, as depicted in Figure 6.1. The short GI also introduces ICI due to loss in orthog-
onality among sub carriers. It is important here to note that carrier frequency oset
and Doppler eect also cause ICI [70],[45],[74].
Cyclic prex (CP) or zero padding (ZP) [75] are usual implementations of GI.
CP causes an additional wastage of energy on top of spectral eciency loss, whereas
ZP only adds the overhead, but [76] shows that cyclic prexed OFDM is more stable
compared to other schemes. In IEEE standards 802.11a/g, 802.16a,e [11, 14, 15], as
much as 25% of the OFDM symbol duration is CP.
It is seen from several measurements [77, 78, 31] that the local rms delay spread
varies from few tens of nano seconds to few micro seconds. In such environments,
using xed GI (chosen more than the maximum delay in the environment) causes
devices which experience smaller rms delay spread to use an unnecessarily large GI.
There is a wastage of costly resource from a network point of view and power from
the user device perspective, where battery life is a major concern.
This background motivates the use of Variable Guard Interval (VGI). This
scheme is suitable when a preamble [14] is used for synchronization and channel esti-
mation. There may be a concern for the timing synchronization algorithms which use
GI for synchronization. However, using the VGI scheme is not expected to inuence
the performance of such algorithms much as it is seen that the mean square error in
Center for TeleInFrastruktur (CTIF), Aalborg University
6.1 Introduction 121
the estimation of timing and frequency oset using the GI does not improve notably
after a certain length of GI [79].
The loss due to a xed GI, can be reduced by using VGI as discussed above.
A preliminary form of this is in IEEE 802.16a,e. The Base Station (BS) can choose
several GI lengths. But once a BS chooses a GI, all User Equipment (UE), in the cell
covered by the BS, will use the same GI. Therefore, there is only one GI, which is the
maximum of the required GIs considering all users in the coverage area of the BS. In
this work, we consider the situation where, dierent GI lengths coexist simultaneously
under one BS or Access Point (AP).
A few of the relevant works in this area are as follows. [80] presents the degra-
dation when GI length is less than the channel impulse response, but its relation to
rms delay spread of the channel is not discussed. Authors in [81] present the detailed
analysis of the impact of insucient GI in OFDM systems. But the expressions are
too complex to use in a dynamic GI estimation procedure. The work in [82] presents
ISI and ICI cancelation scheme and it requires highly complex interference cancelation
scheme at the receiver. In [83], the authors propose not to use GI, but this is from
synchronization perspective and does not consider the impact of ISI. The article [84]
provides method for equalization where a lter is designed, which on convolution
with the channel yields an eective channel with a compressed impulse response.
This method is elegant but would increase the receiver complexity and signalling
overhead.The authors in [85] also advocate the use of ISI cancelation algorithm at
the receiver and the use of redundant carrier to combat ICI. Such mechanisms have
high receiver complexity which prohibit the use of low end devices and also have loss
in eciency due to use of redundant carriers.The scheme in [86] uses dierent lengths
of GI but by adjusting the sampling rate of the system. This scheme may be useful
but may not be permissible in many simple receivers. In this scheme the receiver has
to adjust the sampling rate which may create problems in synchronization of OFDM
and internal receiver components. Moreover, in this scheme the number of sub carri-
ers is dependent on the GI length needed. The number of sub carriers impact the sub
carrier bandwidth which is usually decided based on the tolerance level to Doppler
frequency spread and carrier oset. Therefore it has conicting design parameters.
In this chapter an algorithm is proposed which selects the GI dynamically. A
simplied closed form expression for maximum limit of interference and lower limit
for desired signal power is used in this chapter to arrive at the the received SINR due
to insucient GI. This is used to derive the expression for the required GI which can
be computed dynamically. The required GI is shown to be a function the rms delay
c
Suvra Sekhar Das, 2007
122 Chapter 6. Variable Guard Interval
spread of the channel, the discrete fourier transform duration, the required SNR,
which is not presented in any of the previous works by the others.
6.2 System Description
In this work, it is assumed that each transmission burst, which can be referred to as
a frame (consisting of a sequence of OFDM symbols) may have a dierent GI as per
the need of the user transmitting or receiving the frame. If multiple users share one
transmission burst then the maximum length of required GI among the users sharing
the burst is used. Higher level algorithms may be used to group users with similar
requirement of GI in such a case.
In down link, the Access Point (AP), or the BS uses a GI based on the require-
ment of the users. The GI to be used in a frame has to be estimated and this infor-
mation needs to be present at the transmitter. In Frequency Division Duplex (FDD)
mode, the intended receivers can estimate the required GI, at a previous time, and
feed back the information in the reverse link. For Time Division Duplex (TDD) mode,
channel reciprocity may be used. In such a case, the required GI can be estimated
from the reverse link channel estimate at the transmitter (AP/BS) and used in the
subsequent forward link transmission. To feed back the required GI, a device may
need up to 4 bits to indicate 16 levels of GI. This overhead is very small compared
to the frame length. In the forward link , the GI used by the AP/BS in a frame can
be signalled in the header for each frame. The required GI may be estimated every
frame, and fed back only when there is a change required thereby further reducing
the overhead. These are very much implementation specic.
6.3 Required GI
The time domain signal model for OFDM is dened in (2.19) (2.22). The received
sub carrier symbol in (3.11) can be expanded as
X
s
[k

] =
1

T
f
_
(s+1)T
s
sT
s
+T
gi
r
s
(t)e
j2
k

T
f
(tsT
s
T
gi
)
dt
=
1
T
f
_
(s+1)T
s
sT
s
+T
gi
_
_
_
_

max
0
h()
_
_
_
N
f
2
1

k=
N
f
2
X
s
[k]e
j2
k
T
f
(tsT
s
T
gi
)

T
s
(t sT
s
)
_
_
_
d + (t)
_
_
_
.e
j2t

T
f

T
f
(t sT
s
T
gi
)e
j2
k

T
f
(tsT
s
T
gi
)
dt. (6.1)
Center for TeleInFrastruktur (CTIF), Aalborg University
6.3 Required GI 123
Results exist for situations where the GI is suciently large, but here we consider
T
gi
<
max
, i.e. the channel impulse response is not completely covered by the GI, and
therefore interference from previous OFDM symbol creeps in. The OFDM symbol
number s will have inuence from the previous OFDM symbol. Figure 6.1 shows
the eect of previous OFDM symbol on the current OFDM symbol. The highlighted
portion inside the circle introduces ISI due to insucient GI, and ICI due to the loss
of orthogonality among sub carriers. Exponential power decay prole is considered for
the channel model. Channel taps are taken as Rayleigh distributed. Following such a
model,
0
=
m
=
rms
[87] ([88] also shows linear relationship between
m
and
rms
),
where
0
is the decay constant of the channel model chosen,
m
is the mean excess
delay and
rms
is the rms delay spread of the channel.
The transmit pulse for the (s 1)
th
OFDM symbol is given by

T
s
(t (s 1)T
s
) = 1,
for, (s 1)T
s
+ t < sT
s
+
= 0, otherwise. (6.2)
At the receiver, the pulse used is

T
f
(t sT
s
T
gi
) = 1, for, sT
s
+ T
gi
t < (s + 1)T
s
= 0, otherwise. (6.3)
The integral over t in (6.1) is non zero over the period
max ((s 1)T
s
+ , sT
s
+ T
gi
) t < min ((s)T
s
+ , (s + 1)T
s
)
sT
s
+ T
gi
t < sT
s
+ . (6.4)
The coherence time is assumed to be greater than at least two OFDM symbols.
Therefore (6.1) reduces to
X
s
[k

]
I
=
1

T
f
N
f
2
1

k=
N
f
2
X
s1
[k]
_

max
T
gi
h()e
j2k

T
f
e
(k,k

,s,T
f
,T
s
,T
gi
)
e
j
2
T
f
(k,k

,)(sT
s
+T
gi
)
.e
j

T
f
(T
gi
)
( T
gi
)sinc((k, k

, )
T
gi
T
f
) d, (6.5)
c
Suvra Sekhar Das, 2007
124 Chapter 6. Variable Guard Interval
where
(k, k

, s, T
f
, T
s
, T
gi
) = j2
_
T
gi
T
f
(k

k)
T
s
T
f
sk

(s 1)k
_
and
(k, k

) = k k

+. (6.6)
The power for this term can be calculated as
E
[
X
s
[k

]
I
X
s
[k

I
]
= E
|X
s1
[k

]|
2
T
f
2
N
f
2
1

k=
N
f
2
_

max
T
gi
E
|h()|
2( T
gi
)
2
sinc
2
((k, k

, )
T
gi
T
f
) d
where, E
|h()|
2 =
1

0
e

0
i.e. exponential power decay prole. (6.7)
Therefore
E
[
X
s
[k

]
I
X
s
[k

I
]
=E
|X
s1
[k

]|
2

2
1

0
N
f
2
1

k=
N
f
2
1
(k k

+ )
2
_

max
T
gi
e

0
sin
2
_
(k k

+ )
T
gi
T
f
_
.
(6.8)
Using the identity [89]
_
e
ax
sin
2
(bx)d x = e
ax
_
1
2a

1
a
2
+4b
2
_
a
2
Cos(2bx) + bSin(2bx)
_
(6.9)
and neglecting the terms with coecients e

max

0
and considering that
T
gi
T
f
to be
quite small such that the sine term can be assumed to be very small and the cosine
can be approximated to unity, the earlier expression can be reduced to
E
[
X
s
[k

]
I
X
s
[k

I
]
E
|X
s1
[k

]|
2
N
f
2
1

k=
N
f
2
2e

0

0
2
T
f
2
1 + 4

0
2
T
2
f
(k k

+ )
2
. (6.10)
Using the identity [89]
coth(x) =
1
x
+
2x

a=1
1
a
2
+ x
2
(6.11)
the interference from the previous OFDM symbol due to insucient GI can be ap-
Center for TeleInFrastruktur (CTIF), Aalborg University
6.3 Required GI 125
proximated as
E
intfI
E
|X
s1
[k]|
2

0
T
f
e
T
gi
/
0
, (6.12)
where E
|X
s1
[k]|
2 is the average power in the previous OFDM symbol. For very large
T
gi
(T
gi

0
) the term becomes negligible. It is seen that, for the interference
from the previous OFDM symbol on the current one, the impact of insucient GI
dominates that due to carrier oset. Figure 6.2 compares the expression with the
simulated interference value, marked in the gure as interference 1. It is seen that
the expression agrees very closely with the simulated value. Now we consider the
1 2 3 4 5 6 7
0
0.02
0.04
0.06
0.08
0.1
0.12
Ratio of GI over rms delay spread
I
n
t
e
r
f
e
r
e
n
c
e

P
o
w
e
r


Simulation interference 1
Theory interference 1
Simulation interference 2
Theory interference 2
aaprox interference 2
Simulation interference sum
Theory interference sum
aaprox interference sum
Figure 6.2: Interference Power due to previous OFDM symbol vs Ratio of GI over rms delay
spread.
s
th
OFDM symbol. For the region 0 < T
gi
, there is the useful part and the
interference due to ICI from carrier frequency oset. The contribution of this part of
the received signal is given by
X
s
[k

]
II
=
1

T
f
N
f
2
1

k=
N
f
2
X
s
[k]
_
T
gi
0
h()e
j2k

T
f
e
(k,k

,s,T
f
,T
s
,T
gi
)
e
j
2
T
f
(k,k

,)(sT
s
+T
gi
)
.e
j

T
f
(T
gi
)
( T
gi
)sinc((k, k

, )
T
gi
T
f
) d. (6.13)
c
Suvra Sekhar Das, 2007
126 Chapter 6. Variable Guard Interval
The power of this term is given by E
[
X
s
[k

]
II
X
s
[k

II
]
=
E
|X
s
[k

]|
2
N
f
2
1

k=
N
f
2
_
T
gi
0
E
|h()|
2sinc
2
((k, k

, )) d, (6.14)
neglecting terms with coecients e

max

0
, the component of the desired signal power
(k = k

) can be approximated using [45] as

E
usefulI
E
|X
s
[k

]|
2sinc
2
()(1 e

T
gi

0
), (6.15)
while the inter carrier interference (k ,= k

) can be approximated as
E
intfII
E
|X
s
[k

]|
2
()
2
3
(1 e

T
gi

0
). (6.16)
Next the region between T
gi
< <
max
is considered. As shown earlier, the transmit
pulse for the s
th
OFDM symbol is given by,

T
s
(t sT
s
) =1, for, sT
s
+ t < (s + 1)T
s
=0, otherwise. (6.17)
At the receiver, the portion after the GI is taken away, the pulse used is given in (6.3).
The integral over t in (6.1) is non zero over the period
max (sT
s
+ , sT
s
+ T
gi
) min ((s + 1)T
s
+, (s + 1)T
s
)
sT
s
+ T
gi
t < (s + 1)T
s
. (6.18)
Then the received sub carrier component is given by
X
s
[k

]
III
=
1
T
f
N
f
2
1

k=
N
f
2
X
s
[k]
_

max
T
gi
h()e
j2k

T
f
e
(k,k

,s,T
f
,T
s
,T
gi
)
e
j
2
T
f
(k,k

,)(sT
s
+T
gi
)
.e
j

T
f
(T
s
)
(T
s
)sinc((k, k

, )
T
s

T
f
) d. (6.19)
Center for TeleInFrastruktur (CTIF), Aalborg University
6.3 Required GI 127
The power of this component is computed as
E
[
X
s
[k

]
III
X
s
[k

III
]
=E
|X
s
[k

]|
2
T
f
2
N
f
2
1

k=
N
f
2
_

max
T
gi
E
|h()|
2(T
s
)
2
sinc
2
((k, k

, )
T
s

T
f
) d.
(6.20)
For k = k

, it is the desired sub carrier and (6.20) becomes

E
usefulII
=
E
|X
s
[k

]|
2
T
f
2
_

max
T
gi
e

0
(T
s
2

2
2T
s
)d
neglecting terms with coecients e

max

E
|X
s
[k

]|
2
T
f
2
e

T
gi

0
_
T
s
2
((T
gi
+
0
)
2
+
0
2
) 2(T
gi
+
0
)T
s
_
, (6.21)
for k ,= k

, it is the interference component

E
intfIII

E
|X
s
[k

]|
2

2
1

0
N
f
2
1

k=
N
f
2
1
k k

+
_

max
T
gi
e

0
sin
2
_
(k k

+ )
T
s

T
f
_

E
|X
s
[k

]|
2

2
1

0
N
f
2
1

k=
N
f
2
1
k k

+
_

max
T
gi
e

0
. (6.22)
Since the above leads to a loose upper bound of this component of the interference,
we use the average power of the sine function, and hence have it as
E
intfIII

E
|X
s
[k

]|
2
2
2
1

0
N
f
2
1

k=
N
f
2
1
k k

+
_

max
T
gi
e

0
, (6.23)
which can be evaluated to
E
intfIII

1
6
E
|X
s
[k

]|
2e

T
gi

0
(6.24)
but after simulation, it was found that a tighter bound is approximately given by
E
intfIII

1
8
E
|X
s
[k

]|
2e

T
gi

0
. (6.25)
Figure 6.2 compares the simulation result for the interference power for this compo-
c
Suvra Sekhar Das, 2007
128 Chapter 6. Variable Guard Interval
nent against the two approximations, marked in the Figure with interference 2. The
legend Theory is for the latter expression in (6.25) while approx is for the earlier
one in (6.24). Both theoretical expressions upper bound the result obtained from
simulations. The gap between the expression for small value of the ratio of GI over
rms delay spread is due to the approximations made but this gap becomes very small
for value of the ratio above 2. The total interference power is computed by adding
up each interference contribution, i.e.
E
intf
=E
intfI
+ E
intfII
+ E
intfIII
=E
|X
s1
[k]|
2

0
T
f
e
T
gi
/
0
+ E
|X
s
[k

]|
2
()
2
3
(1 e

T
gi

0
) +
1
8
E
|X
s
[k

]|
2e

T
gi

0
. (6.26)
Figure 6.2 shows the total interference power of simulation as compared against the
theoretical formula derived. The curves labeled as sum is the total interference
power at a received sub carrier. The theoretical formulas are indeed upper bounds
as can be seen from the gure. As indicated earlier, the legend simulation is for
the result obtained via simulation, theory is obtained from the derived expressions,
while approx is obtained from the expression using approximations as described
earlier. Similar observation is made in this case also regarding the expression and
their relation to the simulation curves as in the case of interference 2. The total
useful signal power is given by
E
useful
= E
usefulI
+ E
usefulII
E
|X
s
[k

]|
2sinc
2
()(1 e

T
gi

0
) +
E
|X
s
[k

]|
2
T
f
2
e

T
gi

0
(T
s
2
((T
gi
+
0
)
2
+
0
2
) 2(T
gi
+
0
)T
s
).
(6.27)
For large T
gi
and = 0, this term becomes E
|X
s
[k]|
2, which is the ideal case. The
expression is found to be a very close lower bound for the signal power. Since this
is cumbersome to solve, we will use a simpler expression in evaluating the desired
length of the GI. Instead of using (6.27) for E
useful
, only E
usefulI
will used for the total
useful signal power. This will of course reduce the estimate of SINR. SINR can thus
be written as
SINR (

)
E
useful
E
intfI
+ E
intfII
+ E
intfIII
+
2
[k]
, (6.28)
where
2
[k]
is the noise power per sub carrier. Figure 6.3 compares the SINR as
Center for TeleInFrastruktur (CTIF), Aalborg University
6.3 Required GI 129
1 2 3 4 5 6 7
7
8
9
10
11
12
13
14
Ratio of GI over rms delay spread
S
I
N
R

i
n

d
B


Simulation
Theory approx1
Theory approx2
Figure 6.3: SINR vs Ratio of GI over rms delay spread
.
obtained from simulations, and that obtained from the derivations given above. The
curve labeled Simulation is the SINR obtained from simulation, while the one labeled
Theory approx1, is obtained considering the tight bound, while approx2 is obtained
using the relatively relaxed bound (for simpler computation complexity). It shows
that the theoretical SINR predictions are lower bounds. Usually systems are required
to satisfy a certain BER to meet a desired quality of service. The required BER, which
has to be satised, can be mapped to a SINR value [26]. This value can be called
the required SINR (

req
). To compute the required GI, SINR (

as in (6.28)), due
to the interference, is set to be greater than the required value,

req
, i.e.

>

req
.
Upon evaluating this expression, the required GI becomes
T
gi

0
log
_
A
B
_
, (6.29)
where A =

req

0
T
f
+
1
8


2
3

2
+ sinc
2
()
and
B = sinc
2
()

req

2
3

2
+
1

,
where =

req
+

in dB scale, where

is extra SNR needed over the required

c
Suvra Sekhar Das, 2007
130 Chapter 6. Variable Guard Interval
SNR, and =
E
|X
s
[k]|
2

2
[k]
. As can be observed from the above, the required GI is thus
found to be a function of the channel rms delay spread, required signal to interference
plus noise power ratio, the signal to noise power ratio margin, and the DFT duration.
The rms delay spread of the channel has to be estimated to compute the length
of the required GI. The method in [90] may be used for such estimation. It can also
be done following the denition of rms delay spread [34]. This will require channel
coecients which can be obtained by advanced channel estimation schemes. The GI
used in the simulations is estimated following the denition, while the channel coef-
cients used in it are estimated following a Discrete Fourier Transform (DFT) based
interpolation technique [91]. The estimated GI is used in the subsequent frame. Thus
there will be a time lag between such estimation and use in a subsequent frame. In
dynamic channel conditions, the required GI may change over the time lag period.
The performance of systems using this VGI algorithm needs to be veried via simula-
tion over such channel conditions, results for which will be presented in the following
sections.
6.4 Performance and Discussion
To generate the time varying channel coecients for each tap, the exponential power
delay prole using Jakes Doppler spectrum with maximum velocity as 250 kmph is
considered. The carrier frequency is taken as 3.6 GHz. Once the GI is estimated, it is
used only in the next frame. Each frame is of 512 octets, with quadrature phase shift
keying (QPSK) modulation scheme, while using rate 1/2 convolution coding with a
constraint length of 7. The maximum tolerable relative carrier frequency oset has
been kept at 2% of the sub carrier bandwidth.
To compare the performance of VGI in dynamic environment, rst the mean
BER is veried for dierent amount of time lag between estimation of GI and its
use in subsequent OFDM symbols. Simulation conditions and results are given in
Table 6.1. The symbol T
coh0.5
implies the coherence time for 50% correlation of the
rms delay spread or mean excess delay. The term T
coh0
implies the duration when the
correlation function reaches zero for the rst time. This point was calculated from
the time correlation function of the channel. It can be seen from the table that the
proposed VGI scheme does not compromise on BER. It maintains BER which is very
close to the system using a large but xed GI. This behavior can be explained from
the condition which was used to derive the required GI. Since the GI is chosen such
that the signal to interference plus noise ratio is maintained at the receiver there is no
Center for TeleInFrastruktur (CTIF), Aalborg University
6.4 Performance and Discussion 131
Table 6.1: Bit Error Rate for SINR: 15 dB , rms delay spread: 1 s, Carrier Frequency: 3.5
GHz, Bandwidth 20 Mhz bandwidth, Number of sub carriers: 1024, Guard length for xed
GI: 128.
Lag NoDelay T
coh0.5
T
coh0
2xT
coh0
10xT
coh0
Fixed GI

0.5 5x10
5
4x10
5
3.5x10
5
4x10
5
5x10
5
5x10
5
1 3x10
5
3x10
5
2.7x10
5
3x10
5
4x10
5
3x10
5
2 1.25x10
5
1.25x10
5
1.25
5
1.25
5
1.25
5
1.25x10
5
degradation in the BER. It also shows that the algorithm is robust to time variation
of the rms delay spread and mean excess delay of the channel. Time variation of
channel coecients will give rise to time variations of the instantaneous rms delay
spread [42]. Even though there is a large gap in the estimation of the required GI
and the use in subsequent OFDM frames, there is not much degradation in the bit
error rate. This means that as long as the average rms delay spread of the channel
does not change signicantly the estimated GI can be used quite satisfactorily. The
over determination of interference and under estimation of the desired signal power
leading to under estimation of SINR as discussed in the previous Section contributes
to such phenomenon. Another issue to be considered is the discrete implementation of
the GI. During implementation, the number of samples in the GI to be used is taken
as N
gi
used
=
T
gi
T
|, where | means ceil operation and T is the sampling period.
Therefore only if the required GI changes by more than one sampling period, the
required GI will change by one sampling period.
Now the impact of the several parameters that inuence the selection of GI as
given in (6.29) will be analyzed. Since the ratio of

0
T
f
is small, the impact of variation
of the ratio of

0
T
f
on the required GI is very small. The impact of the parameters

req
and on the required GI is given in Figure 6.4. SNRgap indicates the gap
between

req
and . It can be seen that for any given scenario, for non zero value of
= 0.1 (ep2), the required GI increases compared to when = 0.02 (ep1). Without
the knowledge of allowable , the required GI estimated would be less, and therefore
there would be more interference and the required SINR will not be met. With
increasing value of required SINR, the required GI also increases. This is because
with increasing required SINR, the interference level must be kept at very low levels,
hence a larger GI is required. It is also seen that with increasing gap between

req
and
, the required GI decreases. It will be shown, that increasing this gap indenitely,
does not give any benet.
Figure 6.5 shows the bit error rate performance against increasing SNR when
c
Suvra Sekhar Das, 2007
132 Chapter 6. Variable Guard Interval
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10
6
0
1
2
3
4
5
6
RMS Delay Spread
G
I
/

R
M
S

D
e
l
a
y

S
p
r
e
a
d



SNRreq 5dB, SNR gap 0.5dB, ep 0.02
SNRreq 5dB, SNR gap 0.5dB, ep 0.1
SNRreq 5dB, SNR gap 3dB, ep 0.02
SNRreq 5dB, SNR gap 3dB, ep 0.1
SNRreq 15dB, SNR gap 0.5dB, ep 0.02
SNRreq 15dB, SNR gap 0.5dB, ep 0.1
SNRreq 15dB, SNR gap 3dB, ep 0.02
SNRreq 15dB, SNR gap 3dB, ep 0.1
Figure 6.4: Ratio of GI Vs rms delay spread. Depicting variation of T
gi
with respect to
various

.
the required SNR is kept at 15 dB; i.e. SNRgap is varied from 0.5 to 20 dB. It can
be observed that even though the SNR is increased the bit error rate for VGI reaches
an error oor. This is because the interference from the previous OFDM symbol will
also increase with increasing signal power and thus perceived SINR will not increase
and therefore BER curve will not improve beyond a certain value. It depends on
implementation, how much extra power may be permissible. It has been seen that
using 0.5 to 1 dB extra SNR may be a good tradeo.
The distribution of the required GI, when the average rms delay spread is 1s,
(instantaneous rms delay spreads may vary, as was indicated with references in the
introduction) is analyzed next. Figure 6.6 shows an empirical cumulative distribution
plot of the required GI, obtained from the simulations, when the required SNR is 15
db and the SNR gap is 0.5,1 and 2 dB. The reduction in required GI with varying
SNR gap can also be seen in these gures. In systems using xed GI, the length of
GI is usually taken about eight to ten times more than average rms delay spread(here
the xed GI was taken as 1/4 of the useful symbol duration, where the useful symbol
period was kept as 51.2s). Whereas in VGI it can be seen that in some cases as low
as 3s and maximum of about 6.2s. This is because some of the extra GI is bought
o by the additional SNR (SNR gap). Thus it can be seen that the required GI can
be reduced by a huge margin of about 60% to 20%.
Center for TeleInFrastruktur (CTIF), Aalborg University
6.4 Performance and Discussion 133
15 20 25 30 35
10
8
10
7
10
6
10
5
10
4
MeanBERVsSNRcodeduserdistTGFTheory
SNR in dB
M
e
a
n

B
E
R


VGI OFDM
Normal OFDM
Figure 6.5: Performance with increasing SNR.
3000 3500 4000 4500 5000 5500 6000 6500
0.7
0.75
0.8
0.85
0.9
0.95
1
T
gi
in nano seconds
P
r
o
b

(
T
g
i
<
A
b
s
i
s
s
a
)


15.5 dB
16 dB
17 dB
Figure 6.6: Cumulative Distribution Function of required GI.
c
Suvra Sekhar Das, 2007
134 Chapter 6. Variable Guard Interval
In absolute terms the throughput increase is about 20% as can be seen in
Figure 6.7, where VGI OFDM is for the throughput distribution of the system using
the proposed variable GI and Normal OFDM is OFDM system using xed GI. The
legends Mean marks the mean values of each throughput.
0.8 1 1.2 1.4 1.6 1.8 2
x 10
7
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Throughput
P
r
o
b

(
T
h
r
o
u
g
h
p
u
t

<

a
b
c
i
s
s
a
)


VGI OFDM
Normal OFDM
Mean VGI
Mean Normal
Figure 6.7: Gain in throughput of the proposed VGI over xed GI system for 2dB extra
SNR for 1s rms delay spread.
6.5 Conclusion
In this work, an algorithm for implementing VGI for OFDM system in presence of
carrier frequency oset in a dynamic wireless channel condition has been proposed.
The algorithm is found to reduce required GI by about 60% and it also increases the
mean throughput by about 20%, without compromising on BER but using 0.52 dB
extra SNR. The performance is evaluated at 1 micro second average rms delay spread
of the channel, at vehicular speeds of 250 kmph with 3.6 GHz of carrier frequency. The
ecient performance of the algorithm makes it a good candidate for implementation
in OFDM based wireless systems having such provisions.
Center for TeleInFrastruktur (CTIF), Aalborg University
7
Hybrid Link Adaptation
The goal of this thesis is to improve the spectral eciency for OFDM systems, with
minimum impact on User Equipment (UE) complexity. To this eect, hybrid Link
Adaptation (LA) methods are proposed and analyzed in this Chapter.
In OFDM systems, there are several degrees of freedom in time and frequency
domain, such as sub-band size, forward error control coding (FEC) rate, modulation,
power level, modulation & coding rate adaptation interval and power adaptation in-
terval. These can be exploited for Link Adaptation (LA) to achieve high spectral
eciency. The choice for adapting these parameters is dependent on dynamically
changing channel parameters such as the instantaneous channel gains, the rms delay
spread of the channel, Doppler frequency spread and average signal to noise ratio
(SNR). Optimal adaptation implies highly complex systems and high signalling over-
head. Hybrid strategies to tradeo achievable eciency and complexity are proposed
and analyzed in this Chapter. The term Hybrid is used because not only is the bit and
power is being adapted but also the adaptation framework, i.e. the adaptation window
for each of these parameters is also suggested to be adapted for achieving high spectral
eciency with optimum resolution of adaptation.
135
136 Chapter 7. Hybrid Link Adaptation
7.1 Introduction
LA techniques are known to maximize spectral eciency by exploiting the channel
variability [92],[93]. LA schemes adapt transmission parameters, such as bit and
power, according to the channel conditions so that the maximum bit rate is transmit-
ted while keeping the error rate below a target [26]. Once the values of the adaptable
parameters are selected they are kept constant over a region in time and frequency
domain where the channel is relatively at. There is advanced literature on adap-
tive modulation for single carrier systems [26]. OFDM has ne granularity of the
minimum allocation unit which is the sub carrier. The at fading experience by a
sub carrier provides inherent support needed to exploit the advantage of LA [94],[95]
in multiple dimensions. The optimal solution for achieving capacity is water lling
method [96]. Due to its high implementation complexity, suboptimal solutions such
as [94],[97], are used. There are several algorithms which address BER constraint
along with spectral eciency maximization [98] but these are mostly for uncoded
systems. When a group of sub carriers are used instead of a single sub carrier and
when forward error control coding is applied these results are no more applicable.
The algorithm used in this work has been described in detail in Appendix B. The
algorithm is however extended in this work in the sense that all references to a sub
carrier are to be replaced by the sub band (group of neighboring sub carrier forming
a data block).
The degrees of freedom that can be exploited by LA techniques increase when
they are used in the OFDM framework with the benet of improved spectral e-
ciency. However, this leads to increase in complexity of the system. LA involves
adaptation of the modulation level (M), the FEC rate (C) and the power level (P) at
the transmitter as per the channel state information fed back from the receiver. When
applied in the OFDM framework, LA additionally includes selection of adaptation in-
terval for M & C, adaptation interval for P, choice of sub-band size and bitpower
loading algorithm. Other than the fast fading (instantaneous) of the channel gains,
the dynamic variation of the channel parameters such as the root mean square (rms)
delay spread, Doppler frequency spread, average SNR condition also heavily inuence
the values to be selected for some LA parameters. The authors in [27] discuss some
of the channel eects but the work presented in this chapter provides a much more
detailed analysis, covers many more situations and gives guidelines on Hybrid LA.
When a LA scheme tries to optimally adapt so many parameters which depend
on another large set of varying channel conditions, then the system complexity is
Center for TeleInFrastruktur (CTIF), Aalborg University
7.2 System Model 137
bound to increase many folds. Therefore hybrid strategies, which limit some degrees
of freedom by slowly varying some parameters, while using fast adaptation for the
others are proposed and investigated in this work. The objective is to analyze the
tradeo between spectral eciency loss and complexity and overhead reduction that
can be obtained by the hybrid strategies.
7.2 System Model
The broad system model for LA is presented in Chapter 2. Fig. 7.1 elaborates the
diagram for OFDM systems. As mentioned, in OFDM systems, the link adaptation
is done in time and frequency domain. The Channel State Information (CSI) is fed
back from the User Equipment (UE) to the Base Station (BS) in FDD systems or is
measured at the BS in the reverse link in case of TDD systems. The CQI module
feeds back the SNR of a sub band (a group of consecutive sub carriers). It may
also feedback a change in the transmit power required to maintain the SNR. This
information is used by the Link Adaptation Control unit at the transmitter. This unit
selects the FEC coding rate, the modulation rate and the power level for each sub
band according to the bit and power loading algorithm used in the system. It might
also be used to select the sub band size, the adaptation interval, the power control
interval, and a xed coding rate when a single coding rate is to be used, details of
which are discussed in this chapter. 3GPP-LTE [2] system parameters are used in
this work and are given in Table 7.1 for reference.
Table 7.1: System Parameters
Parameter Value
Carrier Frequency 2GHz
Bandwidth 5MHz
Sampling Frequency 7.68MHz
Sub Carrier Spacing 15 KHz
Rms delay spread 0.5s2s
FFT size 512
Useful Symbol period 66.67s
Cyclic Prex 8.3s
Code rates
1
3
,
1
2
,
2
3
convolution coding
Modulation QPSK, 16-QAM, 64-QAM
c
Suvra Sekhar Das, 2007
138 Chapter 7. Hybrid Link Adaptation
BinaryInput
FEC
rate
1/3

2/3
I
n
t
e
r
l
e
a
v
e
r
S
e
r
i
a
l

t
o
P
a
r
a
l
l
e
l
Symbol
Mapping
QPSK,
16-QAM,
64-QAM
P
o
w
e
r
L
o
a
d
i
n
g
IFFT+CP
&
Parallelto
Serial
RFchainand
Antenna
ThewirelessChannel
Rx Antenna
to
Analogue
RF
to
Digital
Baseband
S
y
n
c
h
r
o
n
i
z
a
t
i
o
n
S
e
r
i
a
l

t
o
P
a
r
a
l
l
e
l
FFT and
remove
CP
Estimate
Equalize
Channel
&
CQI/channel
PowerFeedback
Link Adaptation
Control
Transmitter
Receiver
Binary
Output
D
e
c
o
d
e
r
D
e
-
I
n
t
e
r
l
e
a
v
e
r
Symbol
De
Mapping.
Parallel
to
Serial
Figure 7.1: OFDM based link adaptation transceiver architecture
Frame Structure The frame structure fundamental to LA which is being consid-
ered in this work is given in Fig. 7.2. The sub-band size is dened by the number
of consecutive sub carriers that make the block. The minimum unit over which FEC
and interleaving are applied is called a block. The Block Error Rate (BLER) is calcu-
lated with respect to this block. It is a set of consecutive sub carriers which spans a
successive sequence of OFDM symbols over a period of 0.5ms. The modulation level,
code rate and power level are applied at the block level in this work. The block size
can be made to vary in the frequency domain, by changing the sub band size. The
modulation and coding level is adapted once every adaptation interval, which can be
multiple of the block duration. Adaptation interval is the period in time over which
the modulation and coding remains unchanged and power adaptation interval is the
period between each power adaptation.
Link Adaptation Set up It is being considered that the UE receives the pilot
signals at the Downlink (DL) transmission slot, and then it measures the received
SNR. The received SNR is fed back to BS. The feed back information is sent to
BS at Uplink (UL) time slot for TDD and uplink frequency band in FDD. CQI
information conveys the channel information (SNR in this work). If no power control
is needed then only the desired bit rate (a combination of modulation and coding
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 139
0.5ms
1
18
1 6
0.5ms
1
16
Block
Number
OFDMSymbolNumber
DataBlockduration
OneDataBlockUnit
OFDMSymbolDuration
Subcarrier
Number
S
u
b
B
a
n
d
Figure 7.2: Link adaptation frame structure.
rate) may be used for CQI information, otherwise it may be quantized SNR in the
dB scale or it might be a dierence in power indicator. This depends very much
on the LA algorithm in use. A typical time sequence of the channel estimation and
feedback is in Fig. 2.22. At the beginning of each adaptation window, the resulting
CQI for all blocks are sent back to BS, which is used to decide on LA method.
7.3 Hybrid LA strategies
Hybrid LA strategies being proposed here do not exploit all degrees of freedom simul-
taneously. As mentioned, several parameters can be adapted, namely, modulation
level (M), the FEC code rate (C), power level (P), adaptation interval for M & C,
adaptation interval for P, sub-band size and choice of bitpower loading algorithm.
Dierent combination of slow and fast adaptation can be made between the available
choices so that only few parameters are adapted instantaneously using immediate
channel gains. Other parameters are adapted statistically, i.e. using average infor-
mation such as, Doppler frequency spread, rms delay spread and average SNR. The
interrelationships between the channel parameters and the link adaptation parameters
are analyzed for simplied Hybrid LA in this chapter.
c
Suvra Sekhar Das, 2007
140 Chapter 7. Hybrid Link Adaptation
7.3.1 Dierent Link Adaptation Algorithms
Here two dierent link adaptation (bit and power loading) algorithms for various
conditions of rms delay spread and Doppler conditions are considered.
Adaptive Power, Modulation and Coding (APMC): In this algorithm power,
modulation and coding rate are adapted simultaneously for each data block
to achieve the best possible spectral eciency. It uses iterative procedure to
distribute power, and nd bit loads, whose details are in [95]. The algorithm is
also described in Appendix B.
Adaptive Modulation and Coding with xed Power (AMCxP): In this algo-
rithm, transmission power for each block is assumed to be equal at the begin-
ning. Using the feedback CSI, the modulation and coding rate, which suits the
SNR level, for each block are found. After this, the power is brought down
so that the received SNR equals the threshold value. The purpose of bringing
down the power is to save power and reduce interference when deployed in a
cellular environment with aggressive frequency re-use. The dierence with the
earlier algorithm is that it avoids the iterations (of nding the best bit and
power loading combination) and hence is signicantly less complex.
The algorithms are evaluated in two dierent conditions. One has low diversity i.e.
low rms delay spread and low Doppler while the other has high diversity i.e. high
rms delay spread and high Doppler.
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
s
/
H
z
)


APMC
AMC,fixP
Figure 7.3: Spectral Eciency for SISO with Fd=50Hz,
rms
= 0.5s
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 141
5 10 15 20 25 30
10
9
8
7
6
5
4
3
2
1
0
SNR in dB
U
t
i
l
i
z
e
d

P
o
w
e
r

i
n

d
B


APMC
AMC,fixP
Figure 7.4: Power Utilization for SISO with Fd=50Hz,
rms
= 0.5s
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
s
/
H
z
)


APMC
AMC,fixP
Figure 7.5: Spectral Eciency for SISO with Fd=250Hz,
rms
= 2s
Fig. 7.3 shows the spectral eciency for the above mentioned LA algorithms
in a Single Input Single Output (SISO) system for the low diversity conditions. From
Fig. 7.3 it can be seen that APMC has quite improved performance compared to
AMCxP but at high SNR region the spectral eciency of the two algorithms comes
close to each other. This suggests that the low complexity AMCxP can be used
under the conditions of low diversity with high SNR. The behavior can be understood
c
Suvra Sekhar Das, 2007
142 Chapter 7. Hybrid Link Adaptation
5 10 15 20 25 30
25
20
15
10
5
0
SNR in dB
U
t
i
l
i
z
e
d

P
o
w
e
r

i
n

d
B


APMC
AMC,fixP
Figure 7.6: Power Utilization for SISO with Fd=250Hz,
rms
= 2s
knowing that up to 64-QAM modulation have been used, therefore above a certain
SNR condition, the spectral eciency of the best bit and power loading algorithm
cannot increase beyond certain limit. From the power utilization curves under these
conditions as in Fig. 7.4 it can be concluded that AMCxP algorithm consumes less
power and therefore may cause less interference in a cellular scenario under strong
interference.
Fig. 7.5 shows the spectral eciency and Fig. 7.6 shows the power utilization
performance for the same LA algorithms for high diversity conditions. In contrast to
the earlier condition, it can now be seen from Fig. 7.6 that the gap in performance
between the two algorithms has not decreased in the range of SNR under observation.
Therefore the APMC algorithm proves to be better compared to the other LA algo-
rithm. It can also be observed that the dierence in power utilization for SNR above
10 dB is not more than 2 dB, while in the previous case there was much larger power
saving by the AMCxP algorithm. Under high interference scenario the reduction in
interference due to use of less power might not be noticeable.
Finally, under low diversity and high SNR conditions, i.e. small cell and near
the base station it is worth using the AMCxP in order to have less complexity while
for high diversity scenarios it is suggested to use APMC algorithm. It must also be
noted that the algorithm to load bit and power can be adapted as per statistical
channel conditions at a very slow rate since the channel statistics are not expected
to change as per the coherence time. Fast bit and power loading in combination
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 143
with slow algorithm selection can be used to obtain the best performance in terms of
spectral eciency and power consumption. It is called fast adaptation here when the
adaptation interval is comparable to the coherence time of the channel coecients,
whereas, when the adaptation interval is larger than the coherence time of the channel
coecients, then it is called slow adaptation.
7.3.2 LA with Dierent Sub-channel Sizes
In this section, the inuence of rms delay spread and Doppler spread on the perfor-
mance of LA with dierent sub-channel sizes is evaluated. The APMC algorithm is
used in all analysis unless otherwise mentioned. This is because the APMC algorithm
achieves the highest spectral eciency. While the LA window in time domain is xed
at 0.5 ms, the LA in frequency domain can be performed across sub-band size of 8, 16,
32 & 64. Like the previous situation, analysis will be made in low and high diversity
channel conditions.
From Fig. 7.7 it can be seen that with small Doppler frequency and delay
spread values, a small sub-band size will lead to high spectral eciency compared to
a large sub-band size. This is because with small Doppler frequency and delay spread,
diversity is low, i.e. the channel over one data block does not vary much. So, link
adaptation gains over a at channel will dominate interleaving gain. A small sub-
channel size means less variation within the LA window, which will lead to better
performance. On the other hand, it can be seen from Figure 7.8 that with high
Doppler frequency and delay spread values, the performance for dierent sub-channel
size is quite similar. The reason is, with high Doppler frequency and delay spread
values, diversity is high even with a sub-channel size of 8. i.e. the channel is not
very at over a data block consisting of even 8 sub carrier over 6 OFDM symbols.
Moreover, a wide sub-band means higher interleaving gain, because of more variation
in the channel. Thus it can be seen that a variation in the sub-band size does not
change the spectral eciency. From Fig. 7.7 and Fig. 7.8, it can be concluded that for
low Doppler frequency and delay spread, it is better to use a small sub-channel size
to maximize system performance while for high Doppler frequency and delay spread,
it is suggested to use a large sub band size since it will use signicantly low overhead
and computation resources, while achieving the same performance.
Now it is important to recall that in Chapter 2, it is stated that some SNR
threshold was needed to select a modulation and coding which yields the best through-
put while satisfying the target error rate constraint. In the cases presented above it
c
Suvra Sekhar Das, 2007
144 Chapter 7. Hybrid Link Adaptation
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y


fd50Hz
rms
0.5s subN8
fd50Hz
rms
0.5s subN16
fd50Hz
rms
0.5s subN32
fd50Hz
rms
0.5s subN64
Figure 7.7: Spectral Eciency with Doppler 50Hz, delay spread 0.5s
can be noted that each situation, (which can be considered as a function of channel
diversity condition, the LA duration, and the sub band size) demands a dierent
switching threshold since the SNR required to meet a certain BLER is a function of
the channel condition, the data block size in time, frequency and number of bits. The
Look Up Table (LUT) used for one of the situations is presented in Table 7.2. The
threshold values give an indication of the spectral eciency performance. It can be
seen in the LUT that the starting thresholds for each modulation and coding rate are
increasing with increase in the sub band size for the 50 Hz Doppler scenario. This
means that a certain modulation and coding level (a data rate) will be selected for a
low SNR for the small sub band size as compared to a large sub band size to maintain
the target BLER while maximizing the throughput. This in turn implies that at a
lower SNR a higher data rate may be selected for the smaller sub band size, which
has already been reected in the spectral eciency plot for this scenario. The BLER
for this case on using these thresholds is in Fig. 7.9. It is shown in the gure that the
BLER is almost satised in all conditions using the LUT given above, except for the
very large sub band size, which means that the spectral eciency curves are invalid
and may shift even further to the right.
Finally it can be concluded that it is important to consider very slow rate sub
band size adaptation to minimize feedback overhead along with fast bit and power
loading to have a high spectral eciency. The sub band size needs to be small for
situations of at channel over a data block while it can be large when the diversity
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 145
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y


fd250Hz
rms
2s subN8
fd250Hz
rms
2s subN16
fd250Hz
rms
2s subN32
fd250Hz
rms
2s subN64
Figure 7.8: Spectral Eciency with Doppler 250Hz, delay spread 2s
Table 7.2: SISO,LA per 1 frame(s),BLER=0.05, tau
rms
=0.5s,fd=50Hz
SubN Mod schemes Rate2/3 Rate1/2 Rate1/3
8 4QAM 9.01dB 6.83dB 4.27dB
16QAM 15.62dB 13.43dB 10.59dB
64QAM 21.36dB 18.36dB 14.89dB
16 4QAM 10.43dB 7.83dB 4.96dB
16QAM 17.14dB 14.29dB 10.81dB
64QAM 22.98dB 20.19dB 16.09dB
32 4QAM 13.91dB 10.17dB 6.16dB
16QAM 21.29dB 16.80dB 11.83dB
64QAM 26.64dB 22.16dB 17.07dB
64 4QAM 17.38dB 11.54dB 6.60dB
16QAM 23.35dB 17.84dB 12.61dB
64QAM 29.83dB 22.83dB 17.48dB
c
Suvra Sekhar Das, 2007
146 Chapter 7. Hybrid Link Adaptation
5 10 15 20 25 30
10
3
10
2
10
1
SNR in dB
B
L
E
R


subN=8
subN=16
subN=32
subN=64
targetBLER
Figure 7.9: BLER for SISO,LA per 1 frame(s),BLER=0.05,
rms
=0.5s, fd=50Hz.
of the channel is quite high.
7.3.3 Fixed Coding Rate
In link adaptation studies [99],[27],[100],[101], it is found that adaptive bit loading
is considered without any constraint on user devices. With optimal bit loading,
there are situations where more than one coding rate is allocated to one user in
consecutive data blocks, or more than one data block at a time. Though this leads
to maximum spectrally eciency, it may not be practically feasible to use more than
once code rate for one UE. Using more than one code rate for a UE will put a
heavy signal processing burden on the UE. Therefore using only one coding rate for
a UE is advisable. In this situation it is important to identify the coding rate which
might give the best performance. The focus of this section is to nd the impact
of dierent channel conditions on the xed coding rate to be selected for a user.
Therefore dierent xed coding rates with adaptive modulation and power loading
are considered. The benchmark performance is from adaptive modulation, coding
and power control (APMC).
From Fig. 7.10 it can be seen that with small Doppler frequency and delay
spread values, Adaptive Power and Modulation with Fixed Coding rate (APMxC)
with coding rate
2
3
gives almost the same performance as APMC from 12 dB onwards;
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 147
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y


APMC
APM,FEC 2/3
APM,FEC 1/2
APM,FEC 1/3
Figure 7.10: Spectral Eciency with Doppler 50Hz, delay spread 0.5s
while for SNR less than 12 dB coding rate
1
2
has near optimal performance.
From Fig. 7.11 it can be said that rate
1
3
can be used up to a SNR of 19dB,
and rate
1
2
up to SNR of 27 dB, while beyond this rate
2
3
can be used for channel
conditions with high diversity. These SNR values are average values measured over
the whole bandwidth. By using the above thresholds, a UE can use only one coding
rate yet have performance as close to adaptive coding as possible. This would require
a much less complex system implementation while there would be negligible loss in
performance.
In Table 7.3 the average SNR values for switching from one coding rate to
another is given. The mark - indicates that the coding rate is the default rate to
start with, while the SNR values indicate the starting average SNR from where the
particular coding rate can be used and NA indicates the corresponding coding rate
not be used.
7.3.4 LA Rate
The eect of changing LA window in frequency domain have been studied in Sec-
tion 7.3.2. In this part the performance of LA with dierent adaptation rate in time
(every 1, 2, 5, 10 ms) is evaluated while using 16 sub carriers as the sub band size.
The same extreme cases as in Section 7.3.2 are considered here, i.e. Doppler frequency
c
Suvra Sekhar Das, 2007
148 Chapter 7. Hybrid Link Adaptation
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
s
/
H
z
)


APMC
APM,FEC 2/3
APM,FEC 1/2
APM,FEC 1/3
Figure 7.11: Spectral Eciency with Doppler 250Hz, delay spread 2s
50Hz, Delay spread 0.5s and Doppler frequency 250Hz, Delay spread 2s.
From Fig. 7.12 it can be seen that with small Doppler frequency and delay
spread values, the performance for LA every 0.5ms is similar to the case of LA every
1ms and 2ms. This can be understood from the coherence time which is given by,
T
c
=
9
16Fd
= 3ms. (7.1)
The T
c
is greater than 2ms. The channel is expected to be almost at in this period.
However, for LA every 5ms, there is a large performance drop, e.g. with 15dB SNR,
spectral eciency for LA every 2ms and 5ms is 2bps/Hz and 1.24bps/Hz, which is
about 40% loss in performance. With LA every 10ms, the performance is even worse,
i.e. only 1bps/Hz.
Fig. 7.13 shows the performance for the high diversity condition. It can be
observed that the dierence in performance between the highest rate of LA and
lowest rate LA is less than the earlier case. The reason is, with Doppler frequency of
250Hz, the 50% coherent time is now:
T
c
=
9
16Fd
= 0.7ms, (7.2)
which means that the coherence time is comparable to the data block duration of
0.5ms. The gap between LA every 0.5 ms and 1 ms can thus be understood. The
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 149
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y


LA per 0.5ms
LA per 1ms
LA per 2ms
LA per 5ms
LA per 10ms
Figure 7.12: Spectral Eciency with Doppler 50Hz, delay spread 0.5s
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y


LA per 0.5ms
LA per 1ms
LA per 2ms
LA per 5ms
LA per 10ms
Figure 7.13: Spectral Eciency with Doppler 250Hz, delay spread 2s
c
Suvra Sekhar Das, 2007
150 Chapter 7. Hybrid Link Adaptation
Table 7.3: SNR Threshold for Coding Rate Switching
FEC
1
3
1
2
2
3
1
3
1
2
2
3
1
3
1
2
2
3
1
3
1
2
2
3

rms
Fd subN = 8 subN = 16 subN = 32 subN = 64
50 NA NA - NA - 12 - 16 24 - 18 27
0.5 150 NA - 19 - 13 19 - 18 25 - 19 27
250 NA - 18 - 16 23 - 17 26 - 20 28
50 - 11 16 - 15 24 - 18 27 - 19 28
1 150 NA - 19 - 13 19 - 19 27 - 20 28
250 NA - 21 - 16 24 - 17 27 - 20 29
50 - 14 23 - 17 26 - 20 27 - 20 28
2 150 - 16 24 - 19 26 - 18 27 - 20 28
250 - 17 24 - 19 27 - 19 27 - 20 27
performance of the other LA rates also follow. It can be noted that the performance
of the best case is already quite low compared to the earlier case which can be noted
by comparing gures 7.12 & 7.13.
7.3.5 Dierent LA & PC Rates
In Section 7.3.4 the eect of dierent LA rates on the system performance was ana-
lyzed. The results show that if the LA time is much larger than the coherent time, the
performance will be much worse as compared to LA at a fast rate. This implies that
fast LA is needed in order to achieve a high spectral eciency but a fast LA means
high feedback overhead and more complex system as faster alteration of the system
conguration is needed. In this section, the performance for slow LA but with fast
Power Control (PC) is proposed and investigated. Fig. 7.14 shows the concept of slow
LA with fast PC. The idea is to reduce the LA rate, i.e. using LA at periods larger
than the coherence time of the channel to reduce complexity and overhead. There will
be mismatch between the true SNR and the desired SNR to maintain a target error
rate at the receiver, due to the time varying nature of the fading channel. To overcome
this but in a cheaper way (in terms of overhead and complexity) fast power control is
used between two bit loading intervals. Following the discussion in the earlier section
on using xed coding rate, the coding rate can be adapted at a much slower rate.
However, if the modulation and coding rate are both adapted simultaneously in this
analysis, then, the impact of using fast power control can be captured. Fast power
control is done at least every 0.5 ms. Other adaptation intervals can be multiple of
this. The power control uses feedback to increase, decrease the power levels. Based
on the received information, the transmitter changes the transmit power by a certain
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 151
Modulation Adaptationinterval Modulation Adaptationinterval
PowerControlInterval
Time
Figure 7.14: Combined slow LA with fast power control
percentage of the transmitted power in the previous data block in time. The details of
the algorithm are given below. The feedback overhead of power control information is
expected to be much lower than a complete SNR information feedback or rate control
feedback. The feedback overhead is expected to be signicantly reduced by engaging
a combination of slow LA with fast power control. Further fast power control requires
only stepping up or down of the transmit power and does not require re-orientation
of the modulation/demodulation and the FEC/decoder blocks which is expected to
reduce system complexity. The algorithm for the above is described below:
1. As depicted in Fig. 7.14, it can be considered that LA interval is of T
la
s, and
PC interval is of T
pc
s, where T
la
/T
pc
= K and K is an integer greater than or
equal to one.
2. Within one data block, power, modulation and coding rate are calculated using
the earlier mentioned LA algorithms. The modulation, coding rate and power
are kept xed during one data block.
3. In the beginning of the k
th
PC window ( 1 < k K), UE will compare the
instantaneous channel gain with the previous PC window. If the dierence
between the two channel gains is within a certain limit L
unchg
, then there is no
need to change the power or modulation or coding levels.
4. Base Station (BS) collects the requirements for all data blocks. The BS rst
decreases the power for all the the data blocks which requires a decrease in
power. However for those data blocks which need an increase in transmit power
(since the channel conditions has become worse), the BS needs to consider the
total power constraints. The granularity to bring down the power is always
G
down
dB, which is 20% and the maximum granularity for increasing power is
G
up
dB. The value of G
up
dB can be decided dynamically. It can begin with
c
Suvra Sekhar Das, 2007
152 Chapter 7. Hybrid Link Adaptation
20% and then it can be decreased if the step size is not able to satisfy the
requirements of the blocks requiring an increase in power level. Then the actual
increase in power level G
up,actual
can be calculated as :
G
up,actual
= G
up
min1, (P
av
+ G
down

N
down

n=1
P
k1,n
)/(G
up

N
up

n=1
P
k1,n
) (7.3)
where P
av
is the available power, N
down
is the number of blocks which need less
power and N
up
is the number of blocks which need more power. P
k1,n
is the
power assigned for the n
th
block during the k 1
th
PC window. P
av
is available
power before redistributing the power and is calculated by:
P
av
= P
T

N

n=1
P
k1,n
(7.4)
P
T
is the total power constraint which indicates the up-limit for the total trans-
mission power.
5. Step 3 is then repeated for each PC window within the same LA window. Step
2 and Step 3 are repeated for each LA window during the whole transmission
time.
In this work, G
up
= G
down
= 0.2; L
lowlim
= 0.9; L
uplim
= 1.1. Like the earlier cases
the two diversity conditions have been considered. The rst one is the low diversity
i.e.
rms
= 0.5s, SISO, fd = 50Hz as in Fig. 7.15 which shows the performance for
dierent LA & PC rates with subN = 16.
From this gure it can be seen that:
When LA rate less than or equal to 1ms, there is no noticeable dierence in
performance compared to schemes using a fast power control.
When LA is done every 2ms, a fast PC rate of every 0.5ms can increase the
spectral eciency by about 0.15 b/s/Hz, close to PC every 1ms.
When LA is done every 5ms, both PC interval of 0.5ms and 1ms gives almost
the same performance, which brings a maximum gain of about 67% in spectral
eciency over the case when only LA is done every 5ms.
When LA is done every 10ms, both PC interval of 0.5ms and 1ms gives almost
the same performance, which gives gain of about 50% improvement in spectral
eciency over the case when only LA is done every 10ms.
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 153
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y
(
b
i
t
s
/
s
/
H
z
)


LA per 0.5ms, PC per 0.5ms
LA per 1ms, PC per 0.5ms
LA per 1ms, PC per 1ms
LA per 2ms, PC per 0.5ms
LA per 2ms, PC per 1ms
LA per 2ms, PC per 2ms
LA per 5ms, PC per 0.5ms
LA per 5ms, PC per 1ms
LA per 5ms, PC per 5ms
LA per 10ms, PC per 0.5ms
LA per 10ms, PC per 1ms
LA per 10ms, PC per 2ms
LA per 10ms, PC per 5ms
LA per 10ms, PC per 10ms
Figure 7.15: Spectral Eciency for Dierent LA & PC Rates, Fd=50Hz,
rms
= 0.5s
As has been seen before the decrease in the LA rate has a severe impact on the
performance of the system. However, due to some constraints if the LA rate has to
be kept slow, then the need for fast PC is clearly evident. It can be concluded that
the combination of slow LA with fast PC appears to be very eective.
Figure 7.16 shows the performance for the high diversity case when Doppler
frequency is 250Hz and rms delay spread is 2s. It can be seen from this gure
that PC gives very limited benet as compared to the earlier case. This result is in
accordance with Section 7.3.4.
Therefore it is seen that the slow LA with fast PC is very much eective in low
diversity conditions, but when diversity of the channel increases, the gain from using
fast PC does not yield a gain as signicant as in the earlier case. The low diversity
cases occur for small cells and near BS while the high diversity is expected for large
cells and towards the cell boundary. Thus it can be said the slow LA with PC is
better suited for micro cell scenarios.
Overhead calculation The signaling overhead to implement the dierent rates
of power and modulation adaption is computed in this here. Parameters: N (total
sub-carriers), subN (No. of sub-carriers in one sub-channel) T
LA
(LA interval), T
PA
c
Suvra Sekhar Das, 2007
154 Chapter 7. Hybrid Link Adaptation
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
s
/
H
z
)


LA per 0.5ms, PC per 0.5ms
LA per 1ms, PC per 0.5ms
LA per 1ms, PC per 1ms
LA per 2ms, PC per 0.5ms
LA per 2ms, PC per 1ms
LA per 2ms, PC per 2ms
LA per 5ms, PC per 0.5ms
LA per 5ms, PC per 1ms
LA per 5ms, PC per 5ms
LA per 10ms, PC per 0.5ms
LA per 10ms, PC per 1ms
LA per 10ms, PC per 2ms
LA per 10ms, PC per 5ms
LA per 10ms, PC per 10ms
Figure 7.16: Spectral Eciency for Dierent LA & PC Rates, fd=250Hz,
rms
= 2s
(power adaptation interval), n = T
LA
/T
PA
(integer number, LA over PA rate), PA
ix
(the index of power adaptation times within one LA window).
For uplink (CQI feedback), 5 bits for one sub-band will be used. For power
adaptation only 1 bit is needed to indicate whether the power should go up or down.
Overhead
UL
= (5 + (
T
LA
T
PA
1))
N
subN
1
T
LA
=
(
T
LA
T
PA
+ 4)N
subN T
LA
For downlink, i.e. to indicate the modulation and coding rate to be used, 4
bits for each sub band will be needed to indicate up to 2
4
= 16 dierent rates. The
overhead is presented in Table 7.4.
Overhead
DL
= 4
N
subN
1
T
LA
=
4N
subN T
LA
Overhead
total
= Overhead
UL
+ Overhead
DL
It has been seen that the best performance is achieved with LA interval of 1ms
& PC interval of 1 ms i.e APMC every data block. However it is also seen in the
previous section that LA interval can be increased up to 10 ms, which is several times
more than the coherence time, without any notable performance deterioration. From
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 155
Table 7.4: Overhead in Mbps for Adapt Power LA
LAinterval 1ms 2ms 5ms 10ms
PC interval 1ms 0.5ms 1ms 0.5ms 1ms 1ms
subN=8 0.57 0.384 0.32 0.2302 0.1664 0.1152
subN=32 0.144 0.096 0.08 0.0575 0.0416 0.0288
subN=128 0.036 0.024 0.02 0.0144 0.0104 0.0072
subN=512 0.009 0.006 0.005 0.0036 0.0026 0.0018
Table 7.4, which gives an indication of the overhead needed for the dierent LA and
PC intervals for dierent sub band sizes, it can be said that the overhead for case
with 10 ms LA rate, has about ve times less overhead, i.e a reduction of about 80%
overhead is achievable.
7.3.6 Interaction between Spatial Diversity and Link Adap-
tation
The performance for APMC is evaluated for multiple antenna diversity situations. In
this analysis a xed sub-band size of 16 sub carriers is used. Doppler frequency is
50Hz and delay spread is 1s. LA interval is varied from 0.5ms, to 10ms.
5 10 15 20 25 30
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
t
r
a
l

E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
s
/
H
z
)


SISO,LA per 0.5ms
MRC1x2,LA per 0.5ms
Alt2x1,LA per 0.5ms
AS2x1,LA per 0.5ms
ALT2x2,LA per 0.5ms
AS2x2,LA per 0.5ms
Figure 7.17: Spectral Eciency for Multi-antenna Schemes, LA Per 0.5ms
From Figure 7.17 it can be seen that any form of multiple antenna tech-
nique bring in benet over SISO system under the conditions investigated. Mul-
c
Suvra Sekhar Das, 2007
156 Chapter 7. Hybrid Link Adaptation
tiple Input Single Output (MISO)-Antenna Selection (ASc) has better performance
than MISO-Alamouti among the 2x1 systems. Similarly Multiple Input Multiple
Output (MIMO)-ASc outperforms MIMO-Alamouti signicantly. This is because the
Alamouti scheme has an averaging eect while ASc takes advantage of CSI knowledge
at the transmitter in selecting the the link with the highest gain. Single Input Multi-
ple Output (SIMO)-Maximal Ratio Combining (MRC), the simplest form of multiple
antenna systems has performance close to MIMO-Alamouti. The MIMO-Alamouti
has only little gain over SIMO-MRC since the extra antenna at the transmitter is
used for diversity gain. When LA is made every 0.5 ms, as in Fig. 7.17, the following
is the order of performance for SNR of up to 23 dB. ASc 2x2, Alamouti 2x2, MRC
1x2, ASc 2x1, Alamouti 2x1, SISO. This means that when channel state information
is almost instantaneous, ASc makes the best use of this, and interestingly, using MRC
1x2, gives very close performance to Alamouti 2x2.
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
s
/
H
z
)


SISO,LA per 2ms
MRC1x2,LA per 2ms
Alt2x1,LA per 2ms
AS2x1,LA per 2ms
ALT2x2,LA per 2ms
AS2x2,LA per 2ms
Figure 7.18: Spectral Eciency for Multi-antenna Schemes, LA Per 2ms
As the LA interval is increased, ASc fails to achieve its maximum potential
due to outdated channel information (Fig. 7.18 through Fig. 7.20).
Therefore it can be concluded that when reliable CSI is available, it is better
to use ASc.
Center for TeleInFrastruktur (CTIF), Aalborg University
7.3 Hybrid LA strategies 157
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y


SISO,LA per 5ms
MRC1x2,LA per 5ms
Alt2x1,LA per 5ms
AS2x1,LA per 5ms
ALT2x2,LA per 5ms
AS2x2,LA per 5ms
Figure 7.19: Spectral Eciency for Multi-antenna Schemes, LA Per 5ms
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
SNR in dB
S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y


SISO,LA per 10ms
MRC1x2,LA per 10ms
Alt2x1,LA per 10ms
AS2x1,LA per 10ms
ALT2x2,LA per 10ms
AS2x2,LA per 10ms
Figure 7.20: Spectral Eciency for Multi-antenna Schemes, LA Per 10ms
c
Suvra Sekhar Das, 2007
158 Chapter 7. Hybrid Link Adaptation
7.4 Discussion
Some channel parameters such as average SNR, rms delay spread and Doppler fre-
quency spread can be classied as statistical measure, while channel gains for each
sub band can be considered as instantaneous measure. Adaptation of the system
parameters can be called slow (statistical) or fast (instantaneous) as per their depen-
dence on the channel parameters. It is seen that bit and power loading algorithm
selection, coding rate selection, sub band size can be made at a very slow rate using
statistical channel parameters while modulation can be adapted at a medium rate
using instantaneous channel gains if fast power control is applied in time domain.
The combinations have been investigated in this chapter with the objective to arrive
at systems with minimum complexity while having practically achievable maximum
limits. The selection of the values for statistical adaptation depend heavily on the
rms delay spread, Doppler frequency spread and average SNR values.
Some guidelines can be as follows.
Adaptive Power, Modulation and Coding (APMC) should be used at low rms
delay spread and Doppler condition, while Adaptive Modulation and Coding
with xed Power (AMCxP) is suggested for high delay spread and high Doppler
spread conditions.
Coding rate 1/3 and 1/2 are expected to be used most often while 2/3 is used
mainly for small sub band size. Coding rate selection can be based on the
average SNR condition, the rms delay spread and Doppler frequency spread.
Sub band size can be made small for low mobility and small delay spread while
large sub band size is to be used for large variation of the channel.
Power control is found to be very eective, since modulation and coding can
be adapted at a medium rate, which simplies LA implementation and reduces
the overhead needed.
Transmit diversity along with receive array gain , i.e. Alamouti-MRC com-
bination is suggested for situation of large LA interval while ASc is highly
recommended when LA interval is less than the coherence time.
Table 7.5
1
summarizes the investigations made in this Chapter.
1
The Yes entry means that the particular adaptation parameter (row) is inuenced by the
channel parameter of the column.
Center for TeleInFrastruktur (CTIF), Aalborg University
7.5 Conclusion 159
Table 7.5: Summary of Hyrbid Link Adaptation
Option Doppler
Spread
Rms
Delay
spread
Average
SNR
Instantaneous
SNR
Rate
of
adap-
tation
Bit and Power Load-
ing Algorithm
Yes Yes Yes No Slow
Sub band size Yes Yes Yes No Slow
Code rate Yes Yes Yes No Slow
MIMO mode - - Yes No Slow
Adaptation interval No Yes No No Slow
Power control inter-
val
No Yes No No Slow
Modulation selection No No No Yes Medium
Power loading No No No No Fast
7.5 Conclusion
The objective of this investigation is to arrive at the combinations of slow and fast
adaptation of dierent link parameters which can provide maximum spectral eciency
yet will have minimum implementation complexity. Accordingly several aspects of
link adaptation in OFDM systems has been analyzed. It is shown that for best
performance the bit and power loading algorithm, sub band size, LA adaptation
interval and PC interval, multiple antenna mode, FEC code rate, must be selected as
per the channel statistics such as Doppler condition, RMS delay spread, average SNR
while the selection of modulation level, and power loading, can be done as per the
instantaneous channel gains. Therefore the proposed hybrid mechanisms maximizes
the spectral eciency while having minimum complexity and feedback overhead. It
is found that using a single code rate for a user for all data blocks has very little loss
in spectral eciency compared to the algorithm which uses dierent coding rates in
dierent data blocks. The selection of the single code rate is to be based on statistical
channel parameters. Proposal to reduce the adaptation rate of the modulation level
selection even further and to compensate the outdated channel information by using a
fast power control to balance the SNR at the receiver has been made. This mechanism
reduces overhead by more than 50% while realizing near maximum spectral eciency
performance. Such techniques reduce the degrees of freedom to be exploited at a given
time but do not reduce the maximum spectral eciency. These results are signicant
to provide valuable input to the upcoming wireless OFDM standards. The results are
c
Suvra Sekhar Das, 2007
160 Chapter 7. Hybrid Link Adaptation
also expected to trigger investigation in the joint resource allocation LA algorithms
for multi user cellular environments.
An important consideration in the situations under investigation in this chap-
ter is the use of outer loop link adaptation. In such systems the BER / BLER is
tracked and if the target BER / BLER is not satised / over satised then the SNR
thresholds for adaptive modulation and coding are re-adjusted so that spectral ef-
ciency is maximized. The SNR threshold values are expected to change when the
channel statistics change. The diculty with these systems is that they are slow as
there is a convergence period over which the BLER is stabilized. In contrast, the work
described in this chapter presents the concept of using dierent LUT under dierent
channel conditions. The assumption is that, an estimate of the channel statistics is
available at the BS. With the use of the channel statistics the appropriate LUT can
be used for bit and power loading. This procedure does not involve slow adapta-
tion of the SNR thresholds but selection of a-priory SNR thresholds with the help
of information on channel statistics. Further, the work in this chapter also considers
the adaptation of modulation window, coding rate selection window and power con-
trol window in time and frequency domain based on channel conditions which is not
inherent under the framework of outer loop link adaptation.
Center for TeleInFrastruktur (CTIF), Aalborg University
8
Link Adaptation under Transceiver
Impairments
In the previous chapter various hybrid LA techniques have been discussed. However,
the analysis considered ideal transmitter and receiver operation. On the other hand
it is known that OFDM signal must go through the HPA, which has non linear
transfer function. OFDM signal has high PAPR and this drives the HPA in to the
non linear region. This chapter delves into the analysis of LA under such distortion.
Modication to the LA methods to overcome the non linear distortion problem is also
developed in this chapter. One of the pre-assumptions is that this kind of non linear
distortion is not captured in the channel quality indicator which is fed back from
the receiver. This kind of situations apply directly to WLAN type systems which
use a low PAPR preamble as training sequence for channel estimation in front of the
data symbols in a frame. Whether the channel quality indicator can carry the non
linearity information in other systems is a matter of investigation. An important issue
addressed here, which applies to all systems using link adaptation is the condition that
dierent optimal Back of (BO) in the operating point of the HPA exists for dierent
modulation and coding rates. Choosing a particular BO either causes a penalty in
terms of coverage for some modulation and coding rates while it may increase the
161
162 Chapter 8. Link Adaptation under Transceiver Impairments
BER / BLER degradation for others. This work presents the detailed analysis of such
situations as well by considering dierent BO values. Further the inuence of change
in BER / BLER target on the optimal BO is also shown.
The analysis of LA in the presence of frequency synchronization problem is
also analyzed in this chapter. It is assumed here that the residual frequency oset
and maximum Doppler spread cannot be estimated at the UE due to low complexity
implementation, but these impairments can be measured at the BS. This is in contrast
to the analysis of Chapter 5 where it is considered that the maximum value of Doppler
spread and residual carrier oset is measurable at the receiver. A method to overcome
the eect of frequency synchronization error for LA schemes is proposed here.
The Hybrid Automatic Repeat-reQuest (HARQ) mechanism is generally com-
bined with link adaptation to further improve the performance. Though the use
HARQ mechanism may be expected to give some gain for Link Adaptation systems
yet the performance of such schemes under self interference must be investigated. The
HARQ mechanisms increase received SNR by combining repeated transmission of the
same frame. It can be compared to increasing the SNR by increasing the number of
receiver antenna branches. When there is self interference, then with increase in de-
sired signal power there is a proportional increase in the interference power. Therefore
the improvement in SINR is not a direct consequence. The SINR can be improved
only if dedicated signal processing algorithms are used to cancel the interference /
extract the desired signal out of the interference components which is beyond the
scope of this work in its current state.
8.1 Inuence of Non Linear HPA
OFDM is an elegant solution to tackle the frequency selective wireless channel and
achieve high spectral eciency using LA mechanism, but, it also has its problems. The
output of the OFDM signal is the sum of a large number of sinusoids. If they combine
constructively, then a large peak is created, thus giving rise to high PAPR [1]. When
this signal goes through the HPA, before being transmitted to the air medium, the
signal suers from non linear distortion due to non ideal characteristics of the HPA.
When the signal undergoes such non linear distortion, then spectral regrowth occurs
which consists of in band intermodulation and out of band intermodulation [102].
These phenomenon aect the BLER performance and also introduce Adjacent Chan-
nel Interference (ACI) [103], [104].
Several algorithms to reduce PAPR in OFDM signals are available [105]. Some
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 163
of the schemes need side information [106], some use iterative techniques to reduce
the PAPR [107], while others need additional processing at the receiver [108]. Which
ever technique is used there is always a tradeo between the PAPR reduction and
performance degradation. Moreover some amount of PAPR always remains. To
alleviate the problem of high peaks going into the non linear region of the HPA,
usually a power Back of (BO) is used. The larger the BO, the more shifted is the
operating point of the HPA to the linear region and thus better is the reduction of the
distortion. However, BO also implies a reduction in the output power of the HPA.
This reduces the coverage area of the system. So, a minimum BO which is a tradeo
between performance degradation and coverage is used. The other alternative is to
use a very costly power amplier which has a large operating range. This would
increase the cost of the UE signicantly. There will also be a huge power wastage
due to large BO. Power wastage is a very important issue for UE where battery life
is limited.
The amount of BO needed for optimum performance varies for dierent modu-
lation and coding schemes [95]. It has been discussed in the earlier chapters, that LA
dynamically adapts the modulation level, coding rate and power as per the time vary-
ing channel gain. In case of OFDM, the adaptation is also made across the frequency
domain. The selection of the best data rate is made based on the SNR measured
at the receiver. The SNR is usually measured using TS, which is also know as pilot
symbol. The pilot symbols are designed in a way that the PAPR of the signal is
very low [109],[110],[111] so that the pilot symbols suer from very little non linear
distortion. This is done to ensure proper channel estimation and synchronization at
the receiver. Therefore, the SNR measured using the pilot symbols do not contain
information about the non linear distortion in the DATA symbols which follow the
pilot symbols. Existing works on LA do not analyze this situation.
Additionally, since LA uses varying modulation, power and coding rate for
dierent sub-bands at the same time, using a xed value of BO will not optimize the
performance. It may be suggested to use the maximum BO. Though such a method
will cause minimum distortion, it will be at the cost of coverage and power. Thus it
is important to analyze the impact of nonlinearity in OFDM system using LA.
The focus here is to analyze the impact of non linear distortion on the perfor-
mance of LA algorithms in OFDM systems and nd suitable modication to the LA
algorithm to overcome the problem. Hence PAPR reduction algorithm is not used.
It is also to be noted that if a PAPR reduction algorithm is used then the results
will be dependent on the PAPR reduction mechanism. Moreover neither the WLAN
c
Suvra Sekhar Das, 2007
164 Chapter 8. Link Adaptation under Transceiver Impairments
nor the WMAN standards specify a transmit side signal processing procedure to re-
duce PAPR and this is open for implementation. For all these reasons, only BO has
been used to reduce PAPR in the analysis in this work.
8.1.1 HPA Models
This section presents the HPA model used in the work. Its impact on the OFDM
systems is also discussed. There are several models for HPA available in the litera-
ture which are mainly based on two types of amplier. One is relatively older and is
known as Traveling Wave Tube Amplier (TWTA), which exhibits nonlinear distor-
tion in both amplitude AM/AM and phase AM/PM conversion [112], [102],[113]. The
other model is Solid State Power Amplier (SSPA), which has nonlinear distortion in
amplitude AM/AM only and is referred to as Rapps Model [114].
Available literatures suggests that the AM/PM conversion for SSPA is small
and hence is neglected in the model [1]. The Rapps HPA model for AM/AM conver-
sion is used in this work following [1]. The input output amplitude relation is given
by [1]
g(x) =
[x[
(1 +[x[
2p
)
1
2p
, (8.1)
where x is the signal amplitude and the variable p indicates the amount of nonlinearity.
A good approximation of a practical HPA can be obtained by choosing p between 2
and 3. For large values of p, the model converges to a clipping amplier, which is
perfectly linear until it reaches the maximum output power level. It is very dicult
to realize a linear HPA in practice.
Fig. 8.1 shows the input output signal amplitude relationship for the SSPA
model. From the gure it can be seen that the higher the input power, the more
is the chance that the signal will be in the nonlinear region. So, to operate in the
linear region, the signal power needs to be reduced before feeding it to the amplier
input. This power reduction is referred to as Back of (BO). The gure also shows
that when no BO is used, with p=2, for an input amplitude of 1, the output is 0.82.
For 3 dB BO, i.e. reducing the input signal by a factor of half, the output is at
0.48 when the input is 0.5, BO by 6dB means reducing the input power by a factor
of 0.25 and it can be seen that the output is very close, if not exactly same, as the
input operating point. This shows the non linear eect. Thus BO is expected to
improve BER performance but at the cost of capacity, coverage and power. To nd
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 165
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Input Amplitude
O
u
t
p
u
t

A
m
p
l
i
t
u
d
e
p=2
p=3
p=5
p=10
Ideal
3 dB BO
6 dB BO
Figure 8.1: Power BO representation in Rapps Model
the amount of distortion caused by the amplier only, the variation of power at the
input and at the output of the amplier is measured for dierent BO values and
plotted in Fig. 8.2. It can be seen from the gure that with increasing BO power, the
amount of distortion decreases, i.e. the signal operates more in the linear region.
c
Suvra Sekhar Das, 2007
166 Chapter 8. Link Adaptation under Transceiver Impairments
0 2 4 6 8 10 12
25
20
15
10
5
0
5
10
15
20
Back Off in dB
A
m
p
l
i
f
i
e
r

d
i
s
t
o
r
t
i
o
n

i
n

d
B


Amplifier Distortion Vs Back Off
Figure 8.2: Relation between Amplier Distortion and BO (in dB)
8.1.2 Eect of HPA on dierent Modulation and coding rates
Some fundamental results on PAPR issue in OFDM are briey described in the Ap-
pendix D. The eect of modulation and coding schemes on PAPR for OFDM is also
given in the Appendix D.
8.1.2.1 Eect on Power Spectrum
The frequency domain carrier arrangement for a typical OFDM system is given in
Fig. 4.1. The high frequency components are set to zero to reduce ACI. The 0
frequency, is also set to zero to avoid unnecessary power wastage by DC transmission.
The power spectrum of OFDM signal after the power amplier is shown in Fig. 8.3
for dierent values of BO. It can be seen that the out of band emission decreases with
increasing BO. With a change in signal power of 9 dB (BO of 10 dB to 1 dB) the
out of band emission increases by about 34 dB. Along with the out of band emission
it can be seen that the DC leakage increases as well and this is even slightly worse
compared to the out of band emission. Thus it is seen that small BO results in large
out of band emission which is a highly undesirable eect.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 167
256 156 56 45 145 245
30
20
10
0
10
20
30
Subcarrier Index
P
o
w
e
r
BO1
BO3
BO6
BO10
0
Figure 8.3: Spectrum plot of OFDM signal. BO indicates BO value in dB.
8.1.2.2 Signal to Distortion plus Noise Ratio (SDNR) Plot
Signal to Distortion plus Noise Ratio (SDNR) helps to understand the true usable
signal quality. SDNR can be measured both at the transmitter and at the receiver
(post equalization). The output of the HPA depends on the nonlinearity parameter
p given by the (8.1). Output of the HPA can be written as
y(t) = [x(t)[
2
e
jx
+ n(t) = f([x[
2
, ), (8.2)
where is an attenuation constant which depends on the BO value, x(t) is the input
signal while n(t) is the uncorrelated noise generated due to the amplier distortion.
Variance of the distortion caused by the amplier is given by,

2
D
= E[x(t)[
2
f([x[
2
, )] (8.3)
The transmitter side SDNR is then given by,
SDNR
Tx
=
f([x[
2
, )

2
D
+
2
w
(8.4)
where
2
w
is the variance of the white gaussian noise. When SDNR is measured at
the receiver, it will include processed noise due to equalization as well. SDNR at the
c
Suvra Sekhar Das, 2007
168 Chapter 8. Link Adaptation under Transceiver Impairments
receiver end can be expressed as,
SDNR =
1
N
N1

i=0
[X
i
[
2
[X
i


X
i
[
2
, (8.5)
where, X
i
is the transmitted constellation point at a sub carrier,

X
i
is the received
constellation point and N is the number of sub carriers. Amount of SDNR does not
depend on the modulation scheme used. However, for the same amount of SDNR,
the eect on BLER performance of dierent modulation will be dierent. Since the
decision boundaries for higher modulation in the constellation diagram are closer
compared to the low order modulation, therefore, for the same amount of distortion
the probability of correct decision for a symbol will be much lower for higher modu-
lation order than that of a lower modulation order. Fig. 8.4 shows the SDNR at the
receiver for AWGN channel, and Fig. 8.5 shows that for a fading channel for 4-QAM,
i.e. QPSK. The curves for 16-QAM and 64-QAM are shown in D.7 D.10 for AWGN
and fading channels. The impact of using a smaller BO on the BLER performance
is clearly visible from these gures. The saturation of SDNR as seen in the gure,
is where the distortion is signicantly larger than the noise power. It can be real-
ized that increasing the transmit power or reducing the noise level does not help in
improving the received signal quality.
The gures also show that though the nature of SDNR curve is the same for
both AWGN and fading channels. The amount of variation in SDNR (due to dierent
values of BO) for fading channel is much lower than that of AWGN channel. This
is because the fading channel has more inuence on the SNR degradation than the
non linear distortion. For example, in 4QAM system if the input power is 18 dB
with BO of 8 dB, then the output power is 17 dB and 9dB for AWGN and fading
channel respectively. Similarly, for modulation scheme of 16QAM with the same BO
as above, but when the input power is 24 dB then the output power is 22 dB and 15
dB for AWGN and fading channel respectively.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 169
0 2 4 6 8 10 12 14 16 18
0
2
4
6
8
10
12
14
16
18
Opearting SNR (dB)
S
D
N
R

(
d
B
)
BO=10
BO=8
BO=6
BO=5
BO=4
BO=3
BO=2
BO=1
BO=0
M=4, BO Values in dB
Figure 8.4: SDNR plot for 4QAM modulation in AWGN
Channel.
0 2 4 6 8 10 12 14 16 18
8
6
4
2
0
2
4
6
8
10
Opearting SNR (dB)
S
D
N
R

(
d
B
)


BO=10
BO=8
BO=6
BO=5
BO=4
BO=3
BO=2
BO=1
BO=0
M=4, BO Values in dB
Figure 8.5: SDNR plot for 4QAM modulation in Fading
Channel. The BO values are given in dB.
c
Suvra Sekhar Das, 2007
170 Chapter 8. Link Adaptation under Transceiver Impairments
8.1.2.3 BLER Performance of Dierent Modulation and Coding
In this section the system performance is presented for individual modulation and
coding schemes. Simulation parameters are as discussed in the last chapter (similar
to 3GPP-LTE specication [2]) except that 16 sub carriers are used as the sub band
size. Modulation schemes under investigation are 4QAM, 16QAM and 64QAM. BO is
varied from 0 dB to 10 dB. It is seen from the CDF curves presented in Section D, that
a power BO of 10 dB will accommodate almost all the high peaks and nonlinearity
eect is mostly mitigated. Therefore, performance with 10 dB BO is considered as
bench mark performance. The Doppler frequency for all the investigations was xed
at 50 Hz.
Eect of HPA and BO on all the modulation schemes are studied with FEC
coding rate of
1
3
,
1
2
and
2
3
. However, results for FEC =
1
2
is presented here while
the others are given in Appendix D. The main parameters, used to characterize the
impact of nonlinearities to in this work is BLER. Since LA concerns spectral eciency
and BLER, the main concern here is the analysis of the eect of HPA distortion on
these parameters for a given BO of the HPA
1
.
It is worth mentioning that for convenience of using results obtained here in
link adapted system, all the performance is plotted in terms of POST-SNR. POST-
SNR is dened as the SNR measured at the receiver in each sub band while PRE-SNR
is the SNR measured over the whole system bandwidth, i.e. across all sub bands. For
AWGN channel, PRE-SNR and POST-SNR are the same since there is no channel
variation. For each transmitted PRE-SNR, dierent sub carriers go through dierent
amounts of channel fade, and thus the received SNR for dierent sub band vary.
If the BLER is plotted against the PRE-SNR, then it will represent the average
performance of dierent received SNR which will not exhibit the appropriate behavior
of the system. In the simulations made, the SNR for each data block along with the
information on the correctness of the data block is collected and then the BLER vs
POST-SNR is plotted.
Fig. 8.6Fig. 8.11 show the BER vs SNR for uncoded system and BLER vs
SNR for FEC coded systems in fading channel. The results are shown for QPSK,
16-QAM and 64-QAM; For coded curves, the code rate of
1
2
is used. It can be seen
that the impact of HPA is quite dierent for coded and uncoded system. Since FEC
coding and interleaving improve the performance, coding rate in conjunction with
dierent modulation scheme has dierent impacts. The higher the modulation order,
1
The selection of BO is achieved by striking a tradeo between ACI and BLER degradation.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 171
the greater is the impact of BO values. From the gures it can also be found that
for uncoded case, 16-QAM and 64-QAM schemes show error oor quite early for a
certain power BO as compared to the coded systems. It can also be seen that low
order modulation like QPSK is minimally aected by even the smallest BO. This is
because the HPA model being used in this work has only amplitude distortion with
no phase distortion and QPSK is least aected by amplitude distortion compared to
higher order QAM modulations. As stated earlier, BO shifts the operating point of
the HPA to the linear region and thereby reduces the nonlinear distortion. It is clear
that low modulation order does not require BO at all, as only little improvement is
found between 3dB of BO applied in comparison to the case with no BO is used.
However for 16QAM modulation, a 6dB BO gives us almost the same performance
comparing to 10 BO case. This is inline with other available literatures [1], where it
is reported that increasing the BO values more than the mean PAPR does not make
much dierence as the Cumulative Distribution Function (CDF) of PAPR is quite
steep. For higher modulation, the impact of dierent BO values are more visible,
since the distortions are more detrimental to dense constellation diagrams.
Further, it is observed in the CDF curve, the median and the 90 percentile
value of the PAPR is very close. Beyond that, rest of the high peaks spans a wide
area of SNR values. It can be said that high PAPR occurs very rarely. This is also
evident from the BLER curves. For OFDM symbols with 512 subcarrier, median
value of PAPR distribution is around 8 dB and it is seen from BLER curves for all
modulation and coding rate that, there is not much performance improvement if 10
dB BO is used instead of 8dB. This means BO by median value might be good enough
to mitigate the distortion eect. Increasing back o beyond that value will be of not
much gain.
Since dierent modulation and coding rates show dierent behaviors for vari-
ation in BO power, the choices of suitable BO point for dierent modulations are
dierent. Also, it is highly complex, if not impossible, to change the BO frequently
with the change in modulation and coding rate in link adapted system. So nding
an optimum BO point considering modulation and coding rate becomes necessary for
optimum performance. In case of OFDM systems using LA, where more than one
modulation order is expected to be used across the frequency domain in one OFDM
symbol, it might not be possible to use a dynamic BO. For a proper analysis of the
above scenario i.e. balancing the BO and BLER degradation can be made using
Total Degradation (TDEG), which is dened for a certain BLER threshold as the
amount of BO plus the SNR degradation in BLER performance as compared to the
c
Suvra Sekhar Das, 2007
172 Chapter 8. Link Adaptation under Transceiver Impairments
0 5 10 15 20 25
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
E
R
)
,

U
n
c
o
d
e
d


M=4, BO=10
M=4, BO=8
M=4, BO=6
M=4, BO=5
M=4, BO=4
M=4, BO=3
M=4, BO=2
M=4, BO=1
Figure 8.6: BER vs SNR curve for uncoded and M=4 in
fading channel
0 5 10 15 20
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)


M=4, BO=10
M=4, BO=8
M=4, BO=6
M=4, BO=5
M=4, BO=4
M=4, BO=3
M=4, BO=2
M=4, BO=1
Figure 8.7: BLER vs SNR curve for C =
1
2
and M=4 in
fading channel
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 173
0 5 10 15 20 25 30 35
5.5
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
POSTSNR (dB)
l
o
g
1
0
(
B
E
R
)
,

U
n
c
o
d
e
d


M=16, BO=10
M=16, BO=8
M=16, BO=6
M=16, BO=5
M=16, BO=4
M=16, BO=3
M=16, BO=2
M=16, BO=1
Figure 8.8: BER vs SNR curve for uncoded and M=16 in
Fading channel
5 10 15 20 25 30
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)


M=16, BO=10
M=16, BO=8
M=16, BO=6
M=16, BO=5
M=16, BO=4
M=16, BO=3
M=16, BO=2
M=16, BO=1
Figure 8.9: BLER vs SNR curve for C =
1
2
and M=16 in
fading channel
c
Suvra Sekhar Das, 2007
174 Chapter 8. Link Adaptation under Transceiver Impairments
0 5 10 15 20 25 30 35 40 45
4
3.5
3
2.5
2
1.5
1
0.5
POSTSNR (dB)
l
o
g
1
0
(
B
E
R
)
,

U
n
c
o
d
e
d


M=64, BO=10
M=64, BO=8
M=64, BO=6
M=64, BO=5
M=64, BO=4
M=64, BO=3
M=64, BO=2
M=64, BO=1
Figure 8.10: BER vs SNR curve for uncoded and M=64 in
Fading channel
10 15 20 25 30 35 40
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)
Fading Channel, FEC=1by2, BO in dB


M=64, BO=10
M=64, BO=8
M=64, BO=6
M=64, BO=5
M=64, BO=4
M=64, BO=3
M=64, BO=2
M=64, BO=1
Figure 8.11: BLER vs SNR curve for C =
1
2
and M=64 in
fading channel
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 175
performance in the system without any distortion [102].
Table 8.1 explains how a TDEG can be found. This is for fading channel
Table 8.1: Table for Calculation of Total Degradation in dB
BO Values Ideal(10dB) 8 6 5 4 3 2
Measured SNR 14.6 14.7 15.6 16.3 17.2 18.9 23.1
Degradation - 0.1 1.0 1.7 2.6 4.3 8.5
TDEG - 8.1 7.0 6.7 6.6 7.3 10.5
4 5 6 7 8 9 10
5
6
7
8
9
10
11
12
13
14
15
Power BO (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)


M=64, C=1by2
M=16, C=1by2
M=4, C=1by2
Figure 8.12: TD plot for FEC =
1
2
with BLER Threshold= 0.1 in Fading Channel
with BLER threshold of 0.1, with 16-QAM and FEC coding rate of 0.5. In this table
the reference performance is taken from that with 10 dB BO. For a certain value
of BLER, the corresponding SNR without HPA system (or HPA with a BO of 10
dB or more) is used as the basic reference point. Then for that BLER threshold,
the corresponding SNR values for dierent BO values for a particular modulation
and coding rate is noted. The dierence between this value and the reference value
represents the degradation. The TDEG as dened earlier is the sum of the degradation
and the corresponding power BO.
Fig. 8.12 and Fig. 8.13 show the TDEG curve for two dierent threshold of
BLER 0.1 and 0.05 respectively. The optimum BO can be dierent for dierent
BLER threshold. From the gures, it is seen that for 16-QAM coding rate 1/2, the
minimum BO is 4dB and 5dB for BLER threshold of 0.1 and 0.05 respectively. So,
c
Suvra Sekhar Das, 2007
176 Chapter 8. Link Adaptation under Transceiver Impairments
4 5 6 7 8 9 10
5
6
7
8
9
10
11
12
13
14
Power BO in dB
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

i
n

d
B


M=64, C=1by2
M=16, C=1by2
M=4, C=1by2
Figure 8.13: TD plot for FEC =
1
2
with BLER Threshold= 0.05 in Fading Channel
it becomes important to look at the TDEG curves for dierent BLER threshold to
satisfy dierent QoS criteria.
Several set of curves for dierent modulation and coding rates in AWGN and
fading channel are given in Section D.
8.1.3 Link Adaptation under HPA Impairments
The LA algorithm considered in this chapter is simplied since the focus of investiga-
tion is the impact on non linear distortion on LA system and not the LA algorithm
itself. Accordingly transmit power is brought down to maintain the SNR at the re-
ceiver for a chosen modulation level (identied by using a xed transmit power in the
PREAMBLE) while a coding rate of 1/2 is used, and hence it is mainly the modu-
lation that is adapted as per the channel condition. The ow chart of the algorithm
is given in Fig. 8.14. It is shown in [26] that the SNR threshold for switching the
modulation level as per the channel gain is a function of the received SNR.
It is shown in (8.4) that the received SNR (actually SDNR) during the data
symbol of an OFDM system includes distortion due to the non linear HPA and is a
function of the BO. This implies that the thresholds are expected to be a function
of the BO. This means that the Look Up Table (LUT), which contains the threshold
values for dierent modulation and coding rates at a certain BLER, needs to be
updated.
For uncoded system, it is possible to obtain analytical expression for revised
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 177
Initialize
Parameters
CalculatetheSNR,Initialmodulation
andpowerdistribution
P ornomore
SubChannel
L T
=P
Distributethe
bitloading
ReducethePowertothethreshold
level
Calculatethebitsloadedineach
subchannelandSTOP
Figure 8.14: Flow chart of bit loading algorithm used for analyzing the inuence of HPA in
LA OFDM systems
threshold but, this will not be applicable to coded system and in those situations
where a number of sub carriers form a sub band and when the system has to operate
in varying channel conditions of rms delay spread and Doppler frequency spread. Since
most practical systems such as 3GPP-LTE, WiMAX and IEEE 802.11g WLAN, use
FEC coding, analytical expression obtained for uncoded system will not be usable.
Therefore, computer based simulations are used in this work in order to nd the
additional margin needed to update LUT in order to maintain the QoS constraint.
8.1.3.1 Performance of LA based OFDM System
Performance of LA system is investigated for both AWGN and fading channel. From
the basic BLER vs SNR curve for dierent BO values, the nature of performance
degradation can be found. The reference LUT is prepared from the BLER vs SNR
curves with 10 dB power BO since that does not introduce any non linear distortion.
Results for LA based system presented here are for FEC=
1
2
. The required target
BLER is set as 0.1.
c
Suvra Sekhar Das, 2007
178 Chapter 8. Link Adaptation under Transceiver Impairments
Table 8.2: Variation in number of sub carriers Vs SNR.
SNR (dB) -23 -21 -19 -17 -15 -10 -5 0 10 20
N 1.18 2.03 3.41 5.7 8.9 25.41 61.32 136.3 365.98 490.17
PAPR 3.89 4.91 5.67 7.31 8.03 8.22 8.25 8.26
8.1.3.2 CDF of PAPR for LA System
From the expression of PAPR and also the Fig. D.1 it is clear that PAPR depends
on the number for subcarrier used in a system. Though LA algorithm used in this
work is a simplied version of the APMC, it is important to consider the class of
algorithm such as APMC since they load dierent number of sub carriers as per the
SNR condition. The link adaptation algorithm described in Appendix B uses power
redistribution to the best sub carriers. This is because when the SNR is very low,
the unused power from bad sub carriers is distributed to the better sub carriers to
achieve a higher throughput. The relation between SNR and the number of loaded
subcarrier is shown as in Table 8.2, where N denotes the average number of sub
carrier activated, and PAPR denotes the average PAPR in dB. The IFFT size has
been 512. It can be seen that when the average SNR is very low, only few sub carriers
are selected. This happens because the total power when distributed over these small
number of carrier can cross the threshold for loading non zero number of bits. In these
conditions, the mean PAPR is found to be quite low. The CDF of PAPR is shown in
Fig. 8.15. There is hardly any change in PAPR for LA when the average SNR varies
between -5 and 20 dB. The reason might be that high PAPR occurs rarely. When the
number of sub carriers loaded exceed a certain threshold, there is not much inuence
on further increase in the number of sub carriers up to size of the IFFT. From this
one may be led to conclude that since there is no change in the CDF of PAPR, there
should not be any inuence of the PAPR in the performance of LA based OFDM
systems. However, the reverse has been partly observed in the earlier sections in this
chapter and will also be observed in the upcoming sections as well.
8.1.3.3 Performance in AWGN Channel
Now, the performance of LA systems in OFDM will be analyzed with AWGN channel
condition which will be followed by the more practical fading channel condition. The
rst row of the Table 8.3 shows the threshold for dierent modulation level when
BLER constraint was set to 0.1. The Fig. 8.16 shows the performance with the ideal
scenario, i.e. when HPA distortion is not taken into account and the rst row the
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 179
2 3 4 5 6 7 8 9 10 11 12
10
4
10
3
10
2
10
1
10
0
PAPR (dB)
P
r
o
b
a
b
i
l
i
t
y

(
p
a
p
r
<
a
b
c
i
s
s
a
)


23 dB , N
sc
=1.184,P
uitl
=89.9315%
19 dB , N
sc
=3.412,P
uitl
=91.8207%
17 dB , N
sc
=5.696,P
uitl
=92.2655%
15 dB , N
sc
=8.94,P
uitl
=92.5447%
10 dB , N
sc
=25.416,P
uitl
=92.8215%
5 dB , N
sc
=61.324,P
uitl
=92.9102%
20 dB , N
sc
=490.176,P
uitl
=92.9814%
Figure 8.15: PAPR distribution for LA based OFDM system.
Table 8.3: LUT with basic and updated values for system with FEC=
1
2
in AWGN Chan-
nel(Values in dB)
Power BO None QPSK 16QAM 64QAM
Ideal 0 5.1 11.6 17.3
6 0 5.1+0.5 11.6+0.9 17.3+2.9
5 0 5.1+0.8 11.6+1.5 17.3+7.1
table is used. It satises the BLER constraint. Fig. 8.19 gives the corresponding
SE. However when the power amplier distortion sets in, due to small BO, it clearly
fails to satisfy the performance as is seen from Fig. 8.17. In this case the reference
LUT is used while power BO less than 10 dB is considered. It can be seen that the
performance of the system with 5 dB back o is worse than that of with 6dB back
o power. This is expected as it is already found that the smaller the BO, larger the
distortion and therefore, the greater the margin by which the system fails to meet
the target BLER threshold. The Fig. 8.19 shows the severe impact on SE when not
considering the impact of PAPR into LA systems. If however a power BO of 10 dB
is used then the performance is same as reference system as shown in Fig. 8.16. But,
this large amount of power back o drastically reduce the transmit power thereby
reducing coverage which is not desired.
Therefore it is important to investigate methods which can compensate for the
c
Suvra Sekhar Das, 2007
180 Chapter 8. Link Adaptation under Transceiver Impairments
4 6 8 10 12 14 16 18
10
2
10
1
10
0
POSTSNR (dB)
B
L
E
R
HPA0Basic LUT
HPA1BO10Basic LUT
Figure 8.16: Performance of LA system with basic LUT when no power amplier is applied.
HPA-0 implies no HPA situation while HPA-1 implies that HPA is used in the simulation.
degradation. Modication of the LUT as discussed previously by adding extra margin
may compensate for the BLER degradation. The margin will be dierent for dierent
BO as it is already seen that the SDNR is a function of the power BO parameter. The
less is the BO, the higher is the distortion which results in more BLER performance
degradation and hence larger increment in threshold may be required. The Table 8.3
shows the ideal scenario as well as the added margin for non ideal cases with dierent
BO and the above discussion is also reected in the table. The Fig. 8.18 shows that
the system satises the QoS constraint after adding the margin while Fig. 8.19 shows
that spectral eciency performance is restored to nearly the performance of the basic
system.
The zigzag nature of the curves in Fig. 8.17 and in Fig. 8.18 when HPA is used
is due to the power adaptation algorithm used in the system. For each sub channel,
the measured SNR is used to switch the modulation level and the power for that sub
channel is reduced to the threshold level which is just enough to maintain the SNR
at the receiver. However, reducing power of the sub band is similar to BO by that
amount which leads the sub band power to be closer to the linear region of the HPA
and this improves the performance. Now for the same modulation level, there can
be two dierent transmit power situation, one near the next lower threshold and the
other near the next higher modulation threshold, depending on the channel condition.
In the situation when the transmit power is near the next lower threshold, then the
power BO is more compared to the other case, and hence the former is expected to
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 181
4 6 8 10 12 14 16 18
10
2
10
1
10
0
POSTSNR (dB)
B
L
E
R
HPA1BO6Basic LUT
HPA1BO5 Basic LUT
Figure 8.17: Performance of LA system with basic LUT when power amplier is used
have a better BLER performance than the latter. This is observed in the nature of
the curves shown in Fig. 8.17 and Fig. 8.18.
8.1.3.4 Performance in Fading Channel
The previous Section was dedicated to AWGN channel to show the inuence of the
nonlinear HPA. In this section, the fading channel is considered to get practical results.
The LA issues for OFDM in fading channel are not similar to that of the AWGN
channel. For AWGN channel, the PRE and POST-SNR values are identical. So the
adaptation is based on the PRE-SNR, but for fading channel the PRE and POST-
SNR are dierent quantities. While the PRE-SNR is the measure of the average
SNR across the whole bandwidth the POST-SNR is the measured SNR across each
sub band. At a certain PRESNR value, dierent sub channels can be loaded with
dierent modulations and the satisfaction of the BLER criteria on the PRE-SNR level
does not indicate the same for POST SNR values of the sub bands. In other words,
for a given PRE-SNR, say 15 dB, there may be sub bands with POST-SNR varying
between 5 to 20 dB (the range may be even wider). Therefore these sub bands will be
loaded with dierent modulations. If the BLER is measured over the whole bandwidth
and plotted against the PRE-SNR it will not give indication on the performance of
each modulation level, i.e. it may happen that a particular modulation order does not
meet the required BLER but the average BLER over the whole bandwidth does satisfy
the target BLER, and this may not be desired. Therefore the analysis in this part
c
Suvra Sekhar Das, 2007
182 Chapter 8. Link Adaptation under Transceiver Impairments
5 10 15 20 25
10
2
10
1
POSTSNR (dB)
B
L
E
R
HPA1BO6Revised LUT
HPA1BO5Revised LUT
Figure 8.18: Performance of LA system with revised LUT when power amplier is used
shows the BLER performance for dierent POST-SNR values. In the simulations,
three dierent PRESNR values of 10 dB, 20 dB and 25 dB are chosen and the
BLER performance vs the POST-SNR values is presented for each.
Table 8.4 shows the threshold of the reference system. System performance
Table 8.4: LUT with reference values for system with FEC =
1
2
in Fading Channel(Values
in dB)
Power BO None QPSK 16QAM 64QAM
Ideal 0 8.46 14.76 20.58
using the basic thresholds are shown in Figure 8.20 and 8.21 when 6 dB and 4 dB of
BO is considered respectively. In both the cases system fails when 20 dB and 25 dB
of transmit SNR is used, while the target BLER is satised when 10 dB PRE-SNR
is used.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 183
5 10 15 20
0
0.5
1
1.5
2
2.5
3
POSTSNR
S
E
(
b
/
s
/
H
z
)


Basic LUT
BO3Basic LUT
BO6Basic LUT
BO6Revisd LUT
BO5Basic LUT
BO5Revised LUT
Figure 8.19: Spectral Eciency comparison for LA system with and without PAPR con-
sideration
6 8 10 12 14 16 18 20
10
2
10
1
10
0
POSTSNR (dB)
B
L
E
R


HPA 1BO 6 no margin SNR10
HPA 1BO 6 no margin SNR20
HPA 1BO 6 no margin SNR25
Figure 8.20: Performance of LA system with basic LUT for 6 dB of BO power
c
Suvra Sekhar Das, 2007
184 Chapter 8. Link Adaptation under Transceiver Impairments
5 10 15 20
10
2
10
1
10
0
POSTSNR (dB)
B
L
E
R


HPA 1BO 4 no margin SNR10
HPA 1BO 4 no margin SNR20
HPA 1BO 4 no margin SNR25
Figure 8.21: Performance of LA system with basic LUT for 4 dB of BO power
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 185
Then, extra margins are added to each threshold to satisfy the BLER con-
straint. The Table 8.5, 8.6 and 8.7 give the new LUT with additional margin required
to satisfy the BLER constraint for 10, 20 and 25 dB of PRESNR.
Table 8.5: LUT with updated values for system with FEC =
1
2
and PRESNR of 10 dB in
Fading Channel(Values in dB)
Power BO None QPSK 16QAM 64QAM
10 - 8.46+0 14.76+0 20.58+0
Table 8.6: LUT with updated values for system with FEC =
1
2
and PRESNR of 20 dB in
Fading Channel(Values in dB)
Power BO None QPSK 16QAM 64QAM
6 - 8.46++1.75 14.76+0.3 20.58+0
4 - 8.46+2.5 14.76+0.5 20.58+0
Table 8.7: LUT with updated values for system with FEC =
1
2
and PRESNR of 25 dB in
Fading Channel(Values in dB)
Power BO None QPSK 16QAM 64QAM
6 - 8.46+2.5 14.76+1.65 20.58+0
4 - 8.46+4.0 14.76+1.75 20.58+0.5
The performances using the revised LUT are shown in Figure 8.22 and 8.23.
When PRESNR is 10 dB, which is slightly higher than the switching threshold
for QPSK, most of the sub carriers are expected to be loaded with QPSK and few
of the subcarrier may experience very good channel and are loaded with 16QAM.
As seen from the performance presented earlier in the chapter, QPSK is not much
aected by HPA non linearities. This indicates that the target BLER level is satised
for this modulation. The very few channels which are loaded with 16QAM might be
experiencing a good average SNR over the sub channel. So they also satisfy the target
BLER and hence no additional margin is required. This may be because of the nature
of frequency domain fading of the channel. The fading characteristics may aect the
performance of the FEC decoder. Since the interleaver distributes the bits within
one data block, the gain from interleaving is also limited to the channel properties
of a data block consisting of 16 consecutive sub carriers in the frequency domain
and 6 OFDM symbols in the time domain. This can be substantiated by referring
c
Suvra Sekhar Das, 2007
186 Chapter 8. Link Adaptation under Transceiver Impairments
6 8 10 12 14 16 18 20
10
2
10
1
10
0
POSTSNR (dB)
B
L
E
R


HPA 1BO 6 use margin SNR20
HPA 1BO 6 use margin SNR10
HPA 1BO 6 use margin SNR25
Figure 8.22: Performance of LA system with revised LUT for 6 dB of BO power
6 8 10 12 14 16 18 20 22
10
2
10
1
10
0
POSTSNR
B
L
E
R


HPA 1BO 4 use margin SNR10
HPA 1BO 4 use margin SNR20
HPA 1BO 4 use margin SNR25
Figure 8.23: Performance of LA system with revised LUT for 4 dB of BO power
to the work [38], in Chapter 2 Section 3, which discusses the average bandwidth of
fade vs, ratio of channel gain to the RMS channel gain. It is shown in the work
that with increase in the channel gain relative to the RMS channel gain, the average
bandwidth of fade increases. In other words, the work suggests that the good sub
carriers will experience more at channel in the frequency domain than the deeply
faded sub carriers. This explains the result obtained in the LA. Interestingly it is
worth comparing the additional margins being used for the AWGN situation and
the fading channel situation. In the AWGN channel, the additional margin is solely
because of the non linear distortions, whereas in fading channel the eect of frequency
selectivity is more prominent as understood by the discussion above.
For PRESNR of 20 dB, most of the sub channels are likely to be loaded
Center for TeleInFrastruktur (CTIF), Aalborg University
8.1 Inuence of Non Linear HPA 187
with 16QAM and those experiencing high channel gains are loaded with 64 QAM.
The sub channels which experience deep fades may get loaded with 4QAM. Even
though they were loaded with 4QAM, it may be due to the extremely poor channel
condition of some sub carriers that they fail to satisfy the BLER constraint. Hence,
some additional margin is needed. This additional margin in turn implies that a
particular modulation order is selected only if the average SNR over the sub channel
exceeds an additional threshold thereby ensuring better channel condition suitable
for the interleaver and FEC coder to satisfy the BLER target.
It can be seen from the Figure 8.22 and 8.23 that with additional margin the
BLER requirement is successfully met. It can also be seen from the tables that higher
the BO, lower is the additional margin required and this is accordance to the earlier
discussion.
At a PRE-SNR of 25 dB more margin is needed for dierent modulation scheme
as compared to that of 20 dB PRESNR. It can be noted here that threshold for
switching of LA system changes depending on the PRESNR condition. This kind of
revelation is not possible in the analysis made for single carrier systems [26]. Thus for
implementing LA in OFDM based system, along with the primary distortion, it is also
very important to consider the PRE-SNR condition for updating the SNR thresholds
for switching the modulation and coding rate, otherwise though the average BLER
might appear to be satised, the system may pay a penalty by not being able to
transmit a particular modulation successfully and this especially aects the low order
modulations.
8.1.4 Conclusion
It is seen that the distribution of PAPR in case of LA algorithms, which loads dierent
number of sub carrier at dierent SNRs, is independent of SNR, when the average
SNR is above 0 dB, i.e. PAPR of LA algorithms does not depend on how many sub
carriers are loaded but the PAPR is similar to the case when all sub carriers of the
IFFT are used. It has been found that high PAPR has severe impact on the LA
based OFDM systems. It has been shown that LA-OFDM systems designed without
considering the very important non linear distortion eects of the HPA will fail to meet
a target BER. Additional margin must be used with the SNR switching thresholds for
adaptive modulation and coding to meet the target BLER. It is also found that along
with the eects of the distortion it is also vital to take into account the PRE-SNR
condition while nding the SNR margins. In this section only FEC=
1
2
is presented
c
Suvra Sekhar Das, 2007
188 Chapter 8. Link Adaptation under Transceiver Impairments
which is representative of the kind of problem to be faced in these conditions. Results
for other coding rates can be obtained following the discussion made in this work.
With the modied algorithm for LA, that uses additional margin and updates the
LUT (based on the BO information) to make the system maintain the BLER target,
several LUT for dierent combination of back o power and coding rate may be
needed which best ts the situation.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.2 LA under ICI 189
8.2 LA under ICI
In the previous section of this chapter the impairment at the transmitter end is
presented. In this section the focus is on the impairment at the receiver end. The
impact of ICI due to frequency synchronization is discussed in this section. A method
to overcome the eect of ICI was presented in Chapter 5 of this thesis, where it was
assumed that the maximum Doppler spread and residual frequency oset is estimated
at the User Equipment (UE) and that the sub carrier bandwidth can be adapted to
mitigate ICI. However in this section the inuence on LA when the receivers do
not have the capability to measure the frequency osets and use xed sub carrier
bandwidth is analyzed. If the measured channel quality of the sub carriers does not
include the ICI factor then there might be severe impact on the LA system which
heavily depends on the feedback of measured signal strength. It is therefore necessary
to analyze the performance of LA systems under these conditions. However, it is
assumed that maximum Doppler spread and residual frequency oset are measurable
at the BS. This information is used to make adjustments to LA.
8.2.1 Introduction
ICI in OFDM systems is caused by carrier oset due to frequency synchronization
error [68],[45] and due to Doppler frequency spread [69],[70]. The analytical expres-
sions for ICI are given in Chapter 5. It is seen that the amount of ICI is dependent
on the ratio of eective carrier oset (residual carrier frequency oset and maximum
doppler spread) to the sub carrier spacing and it also proportional to the received
signal strength.
Figures 8.248.29 show the impact of the eective oset for dierent modula-
tions for dierent SNR conditions for uncoded and rate half FEC coded system, in
fading channel. In the analysis, 512 sub carriers in 5MHz bandwidth at 2 GHz carrrier
frequency is considered. In all of the above cases the adaptation rate is considered
to be 0.5ms. The channel rms delay spread of 2s is used. It can be realized that
for un coded system a target BER of 10
3
can hardly be maintained at high mobility
conditions for high order modulations. For coded case the metric is BLER. In this
case the error oor is not easily seen except for 64-QAM. But the impact of high
Doppler frequency spread is evident. For QPSK, the performance degradation from
50Hz to 400Hz is about 5 dB at 10
1
BLER level. At the same BLER threshold, the
performance degradation for 16-QAM is about 6dB and that for 64-QAM is about 17
dB. At lower BLER thresholds the performance degradation is even worse.
c
Suvra Sekhar Das, 2007
190 Chapter 8. Link Adaptation under Transceiver Impairments
0 5 10 15 20 25 30
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
POSTSNR
1
0
*
L
o
g
1
0
(
B
E
R
)


50 Hz
150 Hz
225 Hz
300 Hz
400 Hz
Figure 8.24: Impact of frequency oset on 4-QAM in fading channel.
0 5 10 15 20 25 30
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR
1
0
*
L
o
g
1
0
(
B
L
E
R
)


50 Hz
150 Hz
225 Hz
300 Hz
400 Hz
Figure 8.25: Impact of frequency oset on 4-QAM, FEC rate 1/2, in fading channel.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.2 LA under ICI 191
0 5 10 15 20 25 30 35 40
6
5
4
3
2
1
0
POSTSNR
1
0
*
L
o
g
1
0
(
B
E
R
)


50 Hz
150 Hz
225 Hz
300 Hz
400 Hz
Figure 8.26: Impact of frequency oset on 16-QAM in fading channel.
0 5 10 15 20 25 30 35 40
3
2.5
2
1.5
1
0.5
0
POSTSNR
1
0
*
L
o
g
1
0
(
B
L
E
R
)


50 Hz
150 Hz
225 Hz
300 Hz
400 Hz
Figure 8.27: Impact of frequency oset on 16-QAM, FEC rate 1/2, in fading channel.
c
Suvra Sekhar Das, 2007
192 Chapter 8. Link Adaptation under Transceiver Impairments
0 5 10 15 20 25 30 35 40 45 50
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR
1
0
*
L
o
g
1
0
(
B
E
R
)


50 Hz
150 Hz
225 Hz
300 Hz
400 Hz
Figure 8.28: Impact of frequency oset on 64-QAM in fading channel.
0 5 10 15 20 25 30 35 40 45 50
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR
1
0
*
L
o
g
1
0
(
B
L
E
R
)


50 Hz
150 Hz
225 Hz
300 Hz
400 Hz
Figure 8.29: Impact of frequency oset on 64-QAM, FEC rate 1/2, in fading channel.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.2 LA under ICI 193
Table 8.8: LUT for dierent Doppler spread conditions for code rate of 1/2 at BLER
threshold of 10
1
Modulation QPSK 16-QAM 64-QAM
SNR threshold in dB at 50 Hz 8.52 14.67 20.46
SNR threshold in dB at 150 Hz 10.52 16.67 22.46
SNR threshold in dB at 255 Hz 12.27 18.42 24.12
SNR threshold in dB at 300 Hz 13.52 19.67 25.46
SNR threshold in dB at 400 Hz 17.22 13.37 29.16
8.2.2 LA under undetected ICI
If the LUT for LA is generated based only one Doppler condition, say the 50Hz case,
then the BLER performance for other Doppler conditions can be seen from Fig. 8.30.
It can be seen that higher the Doppler frequency spread, the larger is the deviation
of the BLER performance from the target BLER of 10
1
. This can be understood
in the light of the increasing ICI with increase in Doppler frequency spread. It has
been described before that the threshold to switch from one modulation to another
or to select a modulation for a given coding rate depends on the SINR. In the above
situation, it is assumed that the true Signal to Interference plus Noise Ratio (SINR)
is not measured in the receiver. i.e. the ICI component has not been captured at the
receiver in the SINR meaurement, i.e.
SINR =
E[[X[
2
][H[
2
sinc
2
()

2
(8.6)
where
2
is the noise power and is the relative frequency oset. Here the deterio-
ration in SINR due to ICI is not considered and hence the wrong decision is made in
selecting the modulation level. If receivers do not have the capability to estimate the
residual carrier oset or the maximum Doppler spread then they are expected to be
in this situation.
Following the above formula, the estimate of SINR is much more optimistic
than the actual SINR. However as mentioned earlier it is assumed that the maximum
Doppler oset is measurable at the Base Station (BS). Then to make the system
meet the target BLER, the SNR thresholds must be modied. The updated LUT
is given in Table 8.8. Though the UE may not have the capability to measure the
residual oset, the BS my be able to do so and indicate to the UE to use a dierent
SNR threshold corresponding to the Doppler condition.
The BLER of the system using the new threshold as dened in Table 8.8 is
c
Suvra Sekhar Das, 2007
194 Chapter 8. Link Adaptation under Transceiver Impairments
given in Fig. 8.31. It can be seen that the system now satises the target BLER
threshold at all Doppler conditions. The spectral eciency curves for this system
are in Fig. 8.32. At 20dB, there is about 17% decrease in spectral eciency when
the eective Doppler increases from 50 to 150 Hz, and the decrease is about 30%
when the Doppler is about 225 Hz and when the Doppler is 400 Hz, the decrease in
spectral eciency is about 70%. The decrease in performance is due to the increased
threshold which can be easily seen from the Table 8.8.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.2 LA under ICI 195
6 8 10 12 14 16 18 20 22 24
10
2
10
1
10
0
PRESNR
B
L
E
R


Fd 50Hz
Fd 150Hz
Fd 225Hz
Fd 300Hz
Fd 400Hz
Figure 8.30: Bler performance without additional margin
for coding rate 1/2
5 10 15 20 25 30 35
10
2
10
1
10
0
PRESNR
B
L
E
R


Fd 150Hz
Fd 225Hz
Fd 300Hz
Fd 400Hz
Figure 8.31: Bler performance with additional margin for
coding rate 1/2
c
Suvra Sekhar Das, 2007
196 Chapter 8. Link Adaptation under Transceiver Impairments
5 10 15 20 25 30 35
0
0.5
1
1.5
2
2.5
3
PRESNR
S
E
(
b
/
s
/
H
z
)


Fd 50Hz
Fd 150Hz
Fd 225Hz
Fd 300Hz
Fd 400Hz
Figure 8.32: Spectral eciency performance with additional margin for coding rate 1/2
8.2.3 Conclusion
In this section, it is shown that if the user equipment which feeds back the SNR to
the BS in a LA system, cannot estimate the Doppler frequency spread, then the LA
fails to meet the target error rate criteria. However, it is also shown that if the BS
has the required capability to measure the Doppler frequency spread of a user, then
a suitable adjustment to the SNR switching threshold for the adaptive modulation
and coding can be made. This would allow the LA system to satisfy the error rate
constraint while maximizing the spectral eciency. Even though the LA system with
the new thresholds can meet the required error rate there is a severe impact on the
spectral eciency of the systems. It is seen that when the Doppler frequency spread
increases from 50Hz to 150Hz, the spectral eciency drops by about 17%, while if
the Doppler is about 400Hz, the performance drops by about 70%.
8.3 Summary
In this chapter, the impairments at the transmitter and at the receiver ends
have been considered for evaluating the performance of LA systems in OFDM
framework
At the transmitter side the inuence of non linear power amplier is analyzed.
Center for TeleInFrastruktur (CTIF), Aalborg University
8.3 Summary 197
It is seen that non linear operation of the power amplier has a detrimental
eect on the LA systems. The target error rate criteria is not met by the LA
when the HPA eects are not measured in the fed back SNR.
The HPA introduces uncorrelated Gaussian noise, and the power of this noise
increases with reducing power back o.
The distortion introduced by the HPA is not measurable by a TS sequence since
the TS design ensures that the time domain signal does not have a high PAPR.
Therefore, the SNR fed back by the UE does not include the measure of the
distortion although the eect is present in the data symbols which follow the
TS.
Since the BS has knowledge about the power BO and the size of FFT which gives
an indication of the PAPR to be expected in a data symbol, it can leverage this
information to adjust the SNR switching thresholds for the adaptive modulation
and coding schemes.
It is also shown that the optimum power BO obtained from the Total Degradation
(TDEG) curves is dierent for each modulation and coding and that it varies
with the required BLER.
It is seen that higher order modulations require a higher SNR margin with
respect to the reference SNR threshold for the same amount of power BO in
the HPA. This is true because the higher order modulations have higher er-
ror probability for the same distortion noise power compared to a low order
modulation.
SNR margins required for fading channel condition when compared against the
margins required for AWGN show quite distinct behavior. This is because the
frequency selective fading nature of the channel has a very strong inuence on
the thresholds. In this case it is found that the SNR thresholds for the low
order modulation need to be increased.
Finally it is shown that by adjusting the SNR thresholds the LA schemes can
maintain the required BLER and also maximize the throughput.
In the second part of the chapter the inuence of ICI due to Doppler frequency
spread is analyzed.
c
Suvra Sekhar Das, 2007
198 Chapter 8. Link Adaptation under Transceiver Impairments
In contrary to an earlier chapter, it is assumed here that the maximum Doppler
is not measurable at the UE. Therefore the SNR fed back does not include the
eect of ICI.
It is found that under such circumstances the LA systems fail to meet the BLER
target.
However if it is assumed that the BS has the signal processing capability to
estimate the Doppler spread, then the SNR switching thresholds can be adjusted
and the LA system can again be made to satisfy the BLER constraint.
It is seen that even though the LA system can be made to satisfy the target
BLER using the modied LUT, yet, there is a signicant loss in spectral e-
ciency of the system. However, it is understood that with the modication in
the LUT the system can sustain communication while meeting the target BLER
even though the throughput is low, which is not possible without the mentioned
modication.
Center for TeleInFrastruktur (CTIF), Aalborg University
9
Conclusions and Future Work
The main conclusions of this work and possible areas of further investigation are
discussed in this chapter.
9.1 Conclusions
The focus of this work is to enhance the spectral eciency of OFDM based wireless
systems. Multi carrier spread spectrum based transmission schemes obtain frequency
diversity due to spread spectrum properties and are robust against multi path fad-
ing due to the multi carrier architecture. However, for indoor conditions where the
channel varies slowly with time and only a sub set of all sub carriers are allocated
to one user, the outage performance of such systems is very low. It is proposed in
Chapter 3, to use sub carrier hopping in such systems to overcome the low outage.
The proposed sub carrier hopping with multi carrier spread spectrum is shown to im-
prove the outage by 15%. It is also found that using interleaved sub carrier allocation
is better than using consecutive sub carrier allocation for one user. Additionally it
has been found that successive interference cancelation algorithms at the receiver are
useful in only few scenarios. Several such guidelines are presented in this work.
Multi carrier spread spectrum techniques need additional signal processing
199
200 Chapter 9. Conclusions and Future Work
complexity over OFDM systems therefore the performance of such systems is com-
pared against OFDM to nd the better scheme. It is shown that, OFDM has better
robustness to channel estimation errors than multi carrier spread spectrum systems
when the systems are fully loaded. In partial load conditions, multi carrier spread
spectrum schemes can provide a better BER performance than OFDM systems. Since,
operating the system in full load gives higher data rate, and as it is found that OFDM
is robust to receiver impairments under full load, therefore techniques to improve
spectral eciency of multi carrier based systems are investigated for OFDM systems.
Though OFDM is relatively more robust compared to multi carrier spread
spectrum schemes, yet, it is not able to sustain the performance once the receiver
impairments of channel estimation error and residual phase error exceed a certain
amount. Therefore compensating for the receiver impairments is necessary to achieve
a high performance. To track the residual phase error, pilot sub carriers are embedded
among the data sub carriers. The pilot sub carriers are pre dened signals which are
thus necessary overhead. It is found that the pilots enjoy a extra SNR. Therefore it
is proposed in Chapter 4 to transmit additional information bits on pilot sub carriers
to reduce the pilot overhead. Several combinations of modulation order to be loaded
on pilot sub carriers and data sub carriers have been shown. The proposed scheme
reduces the pilot overhead without performance degradation. It is seen that this
scheme provides a gain between 5% and 15% depending on the scenario.
In outdoor condition a major contributor to the frequency synchronization
errors is the Doppler frequency spread, which causes ICI. The ICI is dependent on
the received signal strength and velocity of user and scatters in the environment.
There is a large variation in the range of values of ICI. A novel technique of using
adaptive sub carrier bandwidth to overcome inter carrier interference without using
complex interference cancelation algorithms at the receiver is suggested in Chapter 5.
Two dierent transmitter architectures are presented which provide a throughput gain
in the range of 10% and 30% over OFDM system using xed sub carrier bandwidth.
The performance of OFDM is also limited by the overhead due to GI, which
is necessary to overcome the eect of ISI. An algorithm for selecting the GI duration
dynamically, in OFDM systems, which is a function of the target SNR, the OFDM
symbol duration, the maximum allowable carrier oset or Doppler frequency spread
and the rms delay spread of the channel, is presented in Chapter 6. The scheme of
using adaptive GI duration is seen to provide a throughput gain of up to 20% by
reducing the GI overhead by as much as 60%.
To enhance the spectral eciency of OFDM systems while maintaining a target
Center for TeleInFrastruktur (CTIF), Aalborg University
9.1 Conclusions 201
error rate link adaptation methods using adaptive modulation, coding, and power are
under investigation. OFDM provides an ideal framework for implementation of LA
schemes, but as the degrees of freedom increase when LA is combined with OFDM, the
implementation complexity also increases. Therefore several schemes to adapt some
parameters slowly while other parameters at a fast rate so as to limit the system
complexity is proposed in Chapter 7. Among other useful results in this chapter,
it is shown that by using fast power control along with slow adaptive modulation
along with even slower coding rate adaptation the overhead can be reduced by more
than 50% while maintaining the spectral eciency nearly as much as the optimum
performance. The need to adapt the sub band size and the suggestion for using only
one FEC rate at a time for a user is also made in this chapter. Several valuable
guidelines and important results have been obtained which can provide signicant
inputs to the upcoming OFDM based wireless communication standard.
Though the LA schemes in combination with OFDM are seen to provide very
high spectral eciency, the investigation of their performance under transmitter and
receiver impairments is very important in order to understand the true potential of
the schemes. Accordingly Chapter 8 presents the analysis of the LA - OFDM system
under the inuence of non linear HPA and the eect of ICI. The TS, which precedes
the data symbols, is designed to have a much lower PAPR than that of the data
symbols. Therefore the non linear distortion due to the HPA aects the TS much less
than the data symbols. The received SNR measured using the TS will therefore not
capture the non linear distortion due to the HPA. The erroneous SNR fed back by the
UE causes the BS to select a modulation order and coding rate which cannot meet
the target BLER. However, it is found that if the transmitter has a knowledge of
the amount of non linear distortion, then suitable adjustments to the SNR switching
thresholds can be made, which can ensure a high spectral eciency while meeting
the target BLER. In case of the ICI impairment, it is considered that the maximum
Doppler spread or the residual carrier oset is not measurable by the user equipment
due to limited complexity. However, it is considered that the base station has the
capability to estimate these parameters. Under such situations a similar observation
is made, i.e. the SNR switching thresholds need to be adjusted in order to meet the
target BLER and maintain a high spectral eciency.
Finally it can be concluded that using the transmission techniques proposed
in this thesis, signicant increase in spectral eciency can be achieved for OFDM
based wireless systems. Notably the gains of each of the proposed techniques are non
overlapping. Therefore if these techniques are used simultaneously in future wireless
c
Suvra Sekhar Das, 2007
202 Chapter 9. Conclusions and Future Work
systems, then the benet from each scheme will add up to provide a signicant total
improvement in spectral eciency.
9.2 Future Works
Several methods to enhance the spectral eciency of OFDM based wireless systems
are proposed in this thesis. However there are many possible areas which can be
explored further. It is found in this work that OFDM performs better than multi
carrier spread spectrum under full load conditions where as the reverse is true for low
load situations. Therefore hybrid adaptive multiplexing techniques which combine
these two schemes suitably can be a possible issue for further investigation.
The use of semi blind pilot has been recommended in this work, which increases
the spectral eciency of OFDM systems. An extension of the analysis can be to devise
systems which can dynamically adapt the pilot sub carrier density to further enhance
the performance.
To overcome varying ICI conditions the use of variable sub carrier bandwidth
is suggested here. The analysis presented assumes perfect estimation of maximum
Doppler oset and SNR. Investigation on the eect of Doppler and SNR estima-
tion error and algorithms for estimation of these parameters can be areas of further
research.
To reduce the GI overhead of OFDM systems, an algorithm to dynamically
choose the GI duration is given in this thesis, which considers ideal estimation of
channel rms delay spread and SNR. Evaluation of the system under practical con-
straints of imperfect estimation of rms delay spread and SNR are possible issues of
future investigation.
Algorithms for simplied link adaptation in OFDM framework have been pre-
sented in this work, but they are analyzed for single user scenario. As these algorithms
are required to operate in multi user situations, joint link adaptation - resource alloca-
tion in a OFDMA framework for multicellular environments is foreseen as a natural
extension of the work. It is found in this work that the non linear distortion and
inter carrier interference have signicant impact on the performance of LA - OFDM
schemes. To tolerate the impairments, a large number of SNR look up tables are
needed which increase the complexity of the system. Therefore reducing the com-
plexity in this perspective can be investigated further.
Center for TeleInFrastruktur (CTIF), Aalborg University
A
Selected Publications Related to
the thesis
A.1 Publications
A.1.1 IPR
1. Enhancing spectral eciency of OFDM systems by Data Transmission over
Pilot Tones, September, 2004, Patent Application No 963/MUM/2004.
2. A novel multirate orthogonal frequency division multiplexing system proposal
to reduce intercarrier interference, September, 2004, Patent Application No
964/MUM/2004.
A.1.2 Journal
1. Suvra Sekhar Das, E. De Carvalho, and R. Prasad, Variable sub-carrier band-
width in OFDM framework IEE Electronic Letters Vol 43 , Issue 1, Jan. 2006
pp. 46-48.
203
204 Chapter A. Selected Publications Related to the thesis
2. Suvra Sekhar Das, E. De. Carvalho, and R. Prasad, Performance Analysis
of OFDM systems with Adaptive Sub Carrier Bandwidth, accepted, IEEE
Transactions in Wireless Communications.
3. Suvra Sekhar Das, E. De. Carvalho, and R. Prasad, Dynamically Varying
Guard Interval Length for OFDM Systems, to be submitted with revisions,
IEEE Transactions in Wireless Communications.
4. Suvra Sekhar Das, E. De. Carvalho, and R. Prasad, Hybrid Link Adaptation
Strategies for OFDM systems, to be submitted.
5. Suvra Sekhar Das, E. De. Carvalho, and R. Prasad, Adaptive Bit and Power
Loading with PAPR and ICI Considerations for OFDM Based Systems, to be
submitted.
A.1.3 Conference
1. Suvra S Das, Muhhamad Imadur Rahman, Nidcha Pongsuwanich, Fleming B.
Fredericksen, Ramjee Prasad.Inuence of PAPR on Link Adaptation Algo-
rithms in OFDM Systems, IEEE VTC Spring 2007.
2. Suvra S Das, Muhhamad Imadur Rahman, Nidcha Pongsuwanich, Fleming B.
Fredericksen, Elisabeth De Carvalho, Ramjee Prasad.Adaptive Modulation
and Power Loading with PAPR Considerations for OFDM Based Systems,
Accepted IEEE VTC Fall 2007.
3. Suvra Sekhar Das, Muhammad Imadur Rahman, Yuyane Wang, Fleming B.
Fredericksen, Elisabeth De Carvalho, Ramjee Prasad. Hybrid Link Adaptation
Strategies for OFDM based systems. Accepted for publication in IEEE VTC
Fall 2007.
4. Muhammad Imadur Rahman, Suvra Sekhar Das, Yuyane Wang, Fleming B.
Fredericksen, Elisabeth De Carvalho, Ramjee Prasad. Link Adaptation Strate-
gies for Multi-Antenna Assisted WiMAX-like System, Accepted IST mobile
summit 2007.
5. Suvra S Das, Muhammad Imadur Rahman, Yuyane Wang, Fleming B. Freder-
icksen, Ramjee Prasad, Inuence of Inter Carrier Interference on Link Adap-
tation Algorithms in OFDM Systems, Accepted IST mobile Summit 2007.
Center for TeleInFrastruktur (CTIF), Aalborg University
A.1 Publications 205
6. Suvra S Das, Elisabeth De Carvalho, Ramjee Prasad, VGI OFDM in dynamic
channel conditions, IEEE PIMRC 2006
7. Suvra Sekhar Das, Ramjee Prasad, et. al. Variable Guard Interval OFDM in
presence of carrier frequency oset, IEEE Globecom, November, 2005.
8. Suvra S Das, Elisabeth De Carvalho, Ramjee Prasad, Time Correlation func-
tion for RMS delay spread of a channel model, IST Mobile Summit 2006,
Myconos, Greece, June, 2006
9. Suvra Sekhar Das, Elisabeth De Carvalho, Ramjee Prasad, et. al. Variable
SubCarrier Bandwidths in OFDM(A) Framework, IEEE VTC Spring 2007,
Dublin, Ireland
10. Suvra S Das, Elisabeth De Carvalho, Ramjee Prasad, et. al . Dynamically
Adaptive Bandwidth for Sub Carriers in OFDM based Wireless Systems, IEEE
WCNC 2007, Hongkong.
11. Das, S.S.; Rahman, M.I.; Fitzek, F.H.P.; Multi rate orthogonal frequency
division multiplexing,2005 IEEE International Conference on Communications
, 2005. ICC 2005. Volume 4, 16-20 May 2005 Page(s):2588 - 2592
12. Suvra S Das, Rajeshwar Rao, Ramjee Prasad, Performance of MultiCarrier
Access Schemes with Receiver Impairments in Down LinkIndoor Environment,
VTC Spring 2006, Melbourne, Australia.
13. Suvra S Das, Muhammad Imadur Rahman, Frank H.P. Fitzek, Ramjee Prasad
On Performance of SCH-OFDMA-CDM in Frequency Selective Indoor Envi-
ronment, IEEE Vehicular Technology Conference Spring 2005 Wireless Access
Stockholm, Sweden, May, 2005, conference proceedings.
14. Suvra S Das, et. al. Using Pilots to Transmit Data Information in OFDM
based WLANs, Wreless Personla Multimedia Communication WPMC 2004,
September, 2004, Abano Terma Italy.
15. Suvra S Das, Ratnam V. Rajakumar, Muhammad I.Rahman, Arpan Pal,Frank
H.P.Fitzek, Ole Olsen,Ramjee Prasad, Low Complexity Residual Phase Track-
ing Algorithm for OFDM-based WLAN Systems, CSNDSP Symposium 2004,
Fourth International Symposium 20-22 July 2004, Newcastle, UK, Juli, 2004,
pp. 128-131
c
Suvra Sekhar Das, 2007
206 Chapter A. Selected Publications Related to the thesis
16. Suvra Sekhar Das, Muhammad Imadur Rahman, Frank H.P. Fitzek, Ramjee
Prasad, et. al. Low Complexity Semi Blind Residual Phase Tracking for
OFDM based WLANs, WPMC 2005, Aalborg, Denmark, September, 2005,
A.2 Chapter wise Publications
Table A.1: Chapter wise Publications
Chapter Number Publication
3 Conference: 12,13
4 Conference: 14,15,16. IPR: 1
5 Conference: 9,10,11. Journal: 1,2. IPR: 2
6 Conference: 6, 7, 8. Journal: 3
7 Conference: 3,4. Journal: 4
8 Conference: 1,2,5. Journal: 5
Center for TeleInFrastruktur (CTIF), Aalborg University
B
Link Adaptation
B.1 Bit and Power Loading Algorithm
An algorithm is presented here which has the highest possible throughput of the
two algorithms referred [115, 94], while the complexity is between the two. The
algorithm is referred to as simple adaptive modulation and power distribution algo-
rithm(SAMPDA). It is designed based on the two algorithms, simple rate adaptation
algorithm (SRA) [115] and adaptive power distribution (APDA) [94]. In the algo-
rithm, greedy approach is used, but unlike the APDA, instead of starting from 0
power and 0 bits as the beginning, the algorithm is initiated with equal power for
all sub-carriers. Then by comparing received signal to noise ratio (SNR) with the
SNR-lookup table, loaded bits for each subcarrier can be found, and power required
for each subcarrier is recalculated. The ow chart for the algorithm is in Figure B.1.
The parameters used in describing the algorithm are given in Table B.1
The SAMPDA algorithm works as follows:
Step 1: Initialization Equal power per sub carriers for SNR measurement at
the receiver
P
n
= P
T
/N
207
208 Chapter B. Link Adaptation
Feedback measured SINR on each
sub carrierto the transmitter
Transmit same power on all sub carriers
Choose allowable modulation
for the received SINR
Find the best sub carrier and re calculated
the power and modulation
Exclude
impossible
modulation
modes
Caclculate number
of bits loaded
for each subcarrier
and stop
Check if total loaded
power exceeds total available
transmit power
Check if total allocated
power is greater than total allowed
transmit power
Updateparameters, bit and power are
distributed to the best sub carrier
No
No
Figure B.1: Flow diagram of the SAMPDA algorithm
Table B.1: Parameter description
P
T
Transmit power threshold
P
L
Loaded power
N Number of sub-carriers
F The highest modulation level

mod
=
[0, 1, ..., F]
Usable modulation set
P
1N
Vector of power for each subcarrier
k
1N
Vector of loaded bits for each subcarrier
M
1N
Vector of Modulation scheme for each subcarrier
n Sequence number of the subcarrier
g
n
Channel gain at the n
th
subcarrier
P
k 1N
Incremental power per incremental bit

n
Signal to noise ratio in each subcarrier

2
n
Noise power in each sub-carrier
SNR
f
Required SNR to maintain the target BER for the f
th
modulation level
NaN Not a number
n

Sequence number of the BEST subcarrier which has the

minimum value of
P
k
.
Center for TeleInFrastruktur (CTIF), Aalborg University
B.1 Bit and Power Loading Algorithm 209
Step 2: Initial modulation scheme and power calculation Find initial
modulation scheme based on the feedback of SNR from the receiver.

n
=
P
n
g
2
n

2
n
M
n
= f
n
where SNR
f
n

n
< SNR
f
n
+1
P
n
=
SNR
M
n

2
n
g
2
n
; P
L
=
N

n=1
P
n
P
k n
=
(SNR
M
n
+1
SNR
M
n
)
2
n
2g
2
n
if M
n
,= F
P
k n
= NaN if M
n
== F
Step 3: Check the Termination Condition: If P
L
= P
T
or min(M) = F,
go to step 6, else continue;
Step 4: Iteration starts: Find the BEST subcarrier:
n

= argmin
n
P
k
Recalculate power and modulation scheme for the n

th subcarrier:
M
n
= M
n
+ 1; P
n
=
SNR
M
n

2
n
g
2
n

Step 5: Check Whether the Distributed Power Overows: if

N
n=1
P
n

P
T
, exclude the infeasible Modulations:
P
k n

= NaN; M
n
= M
n
1
P
n
=
SNR
M
n

2
n
g
2
n

go to step 3, else update the parameters:

P
L
=
N

n=1
P
n
P
k n

=
(SNR
M
n
+1
SNR
M
n

)
2
n
2g
2
n

if M
n
,= F
c
Suvra Sekhar Das, 2007
210 Chapter B. Link Adaptation
P
k n

= NaN if M
n
== F
go to step 3;
Step 6: End: Calculate bits loaded for each subcarrier:
k
n
= log2(2
2M
n
) = 2 M
n
and stop.
After these six steps, bit and power for each subcarrier are stored in the two
N length vectors k
1N
and P
1N
, which will be used for the transmission.
Analysis and Simulation Number of sub-carriers is taken as N = 512, system
bandwidth, B = 5MHz, target bit error rate (BER) is taken as 10
2
. Baseband
modulation schemes such as QPSK, 16-QAM, 64-QAM are considered switchable
options. Number of symbols per frame is 10.
Analysis in terms of Spectral Eciency From the spectral eciency curves for
the three algorithms in Figure B.2 it is seen that the SAMPDA algorithm achieves
the same spectral eciency as the APDA algorithm, which is better than the SRA
by more than 30% between 10db and 20dB and by more than 15% between 20 and
30dB.
Complexity Analysis in terms of number of Iterations For the SNR region
less than 10 dB, the algorithm uses the least number of iterations as shown by Fig-
ure B.3. Between 10dB and 30 dB SNR, the algorithm uses signicantly less number
of iterations than the APDA algorithm but slightly higher number compared to the
SRA. Beyond 40dB the SAMPDA algorithm the number of iterations for the algo-
rithms is very close to that of SRA. It is interesting to observe that where as the
SAMPDA needs increasing number of iterations with increasing SNR, that for the
algorithm is almost independent of the SNR until 40 dB.
Center for TeleInFrastruktur (CTIF), Aalborg University
B.1 Bit and Power Loading Algorithm 211
Figure B.2: Spectral eciency achievement of the adaptation algorithms
Figure B.3: Number of iterations required by dierent adaptation algorithms
c
Suvra Sekhar Das, 2007
212 Chapter B. Link Adaptation
Center for TeleInFrastruktur (CTIF), Aalborg University
C
Hybrid Link Adaptation
In this part the results for Hybrid Link Adaptation using WiMAX parameters are
presented.
Bit and Power Loading Algorithms There can be some variation on the bit
and power loading algorithm used for LA as discussed below. Adaptive Power,
Modulation and Coding (APMC): APMC will adapt power, modulation and
coding rate all together [115, 94]. It uses iterative procedure to distribute power, and
nd bit loads, whose details are in [95]. AMCFP: This case considers xed power,
i.e the iterations for bit and power loading are not used, which makes it very simple
and fast.
AMC adapt P: In the AMCFP algorithm above, since power is xed, one can
expect some power to be wasted since signal to noise ratio (SNR) threshold required
for the M& C rate selected for a block must be less than the available SNR. Therefore
to save this power, the transmit power level can be adjusted so that the received SNR
at the block is maintained just at the threshold required to maintain the target error
rate.
213
214 Chapter C. Hybrid Link Adaptation
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
x 10
7


T
h
r
o
u
g
h
p
u
t

(
b
i
t
s
/
s
e
c
)
SNR in dB
APMC, rms d.s. 0.5 us, 20kmph
AMC fixP, rms d.s. 0.5, 20kmph
AMC adaptP, rms d.s. 0.5, 20kmph
APMC,rms d.s. 2 us, 70kmph
AMC fixP,d.s. 2 us, 70kmph
AMC adaptP,d.s. 2 us, 70kmph
Figure C.1: Throughput comparison of dierent Link adaptation algorithms at 0.5 s and
2.0 s rms delay spread and Doppler condition for sub-band size of 8 sub carriers
Performance Comparison The throughput comparison of the three algorithms
mentioned above is shown in Fig. C.1. The legend u is used to represent micro
seconds which refers to the rms delay spread of the channel. In all cases the APMC
has the best throughput. It is followed by AMCFP, while AMC adapt P has the
worst performance. AMCFP has almost similar performance to APMC, when there
is high diversity in the channel condition, i.e. large Doppler and rms delay spread,
but it has notable performance loss in case of low Doppler and low rms delay spread
condition. Therefore it can be suggested that for low Doppler and low rms delay
spread condition, APMC be used, whereas when the diversity in the channel increases,
it is better to use AMCFP. Finally it can be suggested to combine selection of bit
and power loading algorithm at a very slow rate (based on statistical measure) along
with fast (instantaneous) adaptation of modulation and coding rate.
Power Utilization Fig. C.2 shows the power utilization of the dierent algorithms.
A low power utilization means low power transmission. This in turn means low in-
terference condition in multi cellular scenario, where aggressive frequency re-use is
followed. In such a scenario, the algorithm which has the lowest power utilization, may
be the best one use. In this viewpoint the algorithm which brings down the transmit
power to meet the threshold of the received SNR, but avoids iterative power distri-
bution seems to have the best performance under all channel conditions. Considering
the above results we will used xed power in most of the analysis unless mentioned.
Center for TeleInFrastruktur (CTIF), Aalborg University
215
5 10 15 20 25 30
9
8
7
6
5
4
3
2
1
0
S
a
v
e
d

p
o
w
e
r

i
n

d
B
SNR in dB


APMC,2 us, 200 kmph,subN=8
AMC fixP,2 us, 200 kmph, subN=8
AMC adaptP,2 us, 200kmph, subN=8
AMC fixP,0.5 us, 20 kmph, subN=8
AMC adaptP,0.5 us, 20 kmph, subN=8
APMC,0.5 us, 20 kmph, subN=32
AMC fixP,0.5 us, 20 kmph, subN=32
AMC adaptP,0.5 us, 20 kmph, subN=32
Figure C.2: Power utilization comparison of dierent Link adaptation algorithms at dierent
rms delay spread and Doppler condition for sub-band size of 8 and 32 sub carriers
Sub band Size In this part the inuence of rms delay spread and the Doppler ve-
locity on dierent sub-band sizes (8,32,128 and 512 carriers) are investigated. Fig. C.3
shows the throughput for LA system when modulation (M) & coding rate (C) are
adapted every 2ms while keeping P xed. It can be seen from the gure that, when
rms delay spread is small i.e. coherence bandwidth is large, and the velocity is also
low, i.e. coherence time is large, then the sub-band size of 8 sub carriers has the
highest throughput. It is followed by sub-band size of 32 which is followed by 512.
Interestingly at high velocity and high rms delay spread i.e. small coherence band-
width and small coherence time, the sub-band size of 8 sub carriers has very similar
performance as that of sub-band with 128 sub carriers. Therefore it can be concluded
that for very high velocity and high rms delay spread condition, it is better to use a
large sub-band size since it will use signicantly low overhead, whereas when velocity
and rms delay spread are low, it is suggested to use small sub-band size. It must
be noted that sub band size selection can be a statistical adaptation in combination
with instantaneous adaption of modulation and coding rate.
c
Suvra Sekhar Das, 2007
216 Chapter C. Hybrid Link Adaptation
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
x 10
7


SNR in dB
T
h
r
o
u
g
h
p
u
t
trms=half,velocity=20kmph,subN=8
trms=half,velocity=20kmph,subN=32
trms=half,velocity=20kmph,subN=128
trms=half,velocity=20kmph,subN=512
trms=2us,velocity=200kmph,subN=8
trms=2us,velocity=200kmph,subN=128
Figure C.3: Throughput performance of dierent sub-band sizes for dierent rms delay
spread, Doppler velocity.
Fixed Coding rate In this part comparison of the scheme which adapts both
modulation and coding every each frame, is made against the scheme which uses a
single coding rate with adaptive modulation. Using multiple coding rates simulta-
neously for user means that the user equipment needs multiple FECs and decoders,
which has prohibitive increase in implementation complexity of the transmitter and
receiver. Therefore using only a single FEC coder (i.e only one FEC rate) for one
user is highly desired. Then, selection of the FEC code rate becomes very important
which is discussed in this section. Fig. C.4 shows the throughput comparison for
sub-band with 8 sub carriers. In the gure RateAdapt means adaptive modulation
and coding simultaneously however as mentioned earlier, the power is kept constant.
It can be seen that the performance of xed coding rate is not far from the optimal
adaptive modulation coding scheme. The channel statistics such as rms delay spread,
Doppler velocity and average SNR are needed to choose the threshold for switching
from one coding to another so that performance is close to being optimal. Similar
behaviour has been found for dierent sub band size. In Table C.1 the average SNR
values for switching from one coding rate to another is given. The mark - indicates
that that coding rate is the default coding rate to start with, while the SNR values
indicate the starting average SNR from where the particular coding rate can be used
and NA indicates the corresponding coding rate not be used. It can be concluded
that if small sub-band size is used then all coding rates are important, but when large
sub-band size is selected then coding rates 1/3 and 1/2 are enough.
Center for TeleInFrastruktur (CTIF), Aalborg University
217
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
x 10
7


T
h
r
o
u
g
h
p
u
t

(
b
i
t
s
/
s
e
c
)
SNR in dB
RateAdapt,trms=half,velocity=20kmph,subN=8
BitAdapt,trms=half,velocity=20kmph,subN=8,rate=2by3
BitAdapt,trms=half,velocity=20kmph,subN=8,rate=1by3
RateAdapt,trms=2us,velocity=20kmph,subN=8
BitAdapt,trms=2us,velocity=20kmph,subN=8,rate=2by3
BitAdapt,trms=2us,velocity=20kmph,subN=8,rate=1by2
BitAdapt,trms=2us,velocity=20kmph,subN=8,rate=1by3
RateAdapt,trms=2us,velocity=200kmph,subN=8
BitAdapt,trms=2us,velocity=200kmph,subN=8,rate=1by2
BitAdapt,trms=2us,velocity=200kmph,subN=8,rate=1by3
Figure C.4: Throughput performance comparison for xed coding with adaptive modulation
Vs adaptive modulation and coding for sub-band size of 8 sub carriers.
Table C.1: Average SNR thresholds (in dB) for switching coding rate for dierent rms delay
spread and Doppler condition
0.5s, 20kmph 2s, 20 kmph 2s, 200 kmph
subN
Code rate
1
3
1
2
2
3
1
3
1
2
2
3
1
3
1
2
2
3
8 11.5 15.5 14.5 24 19.5 30
32 13.5 21.5 20.5 N.A. 20 N.A.
128 21.5 N.A. 21.5 N.A. 20 N.A.
512 23.5 N.A. 22 N.A. 22.5 N.A.
c
Suvra Sekhar Das, 2007
218 Chapter C. Hybrid Link Adaptation
M & C Adaptation Rate Fig. C.5 shows the impact of dierent adaptation
interval for rms delay spread of 0.5 s and velocity 20 kmph. It can be seen that
decrease in the adaptation rate has severe impact on the achievable spectral eciency
using adaptive modulation and coding when the size of sub band is up to 32 sub
carriers. When the sub band size is larger, i.e. for subN=512 there is very little
impact of decrease rate of adaptation. This happens because when a large sub band
size is used, there is a large diversity gain and hence there will be little variation
in the average channel gain and therefore even a large adaptation interval can be
used easily.When sub band size is made large there is little impact on the adaptation
time interval; i.e. short term adaptation in time domain is not necessary when the
adaptation window is large in frequency domain even in low mobility conditions.
Center for TeleInFrastruktur (CTIF), Aalborg University
219
5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
x 10
7


T
h
r
o
u
g
h
p
u
t
SNR in dB
LA per 0.5ms, subN=8
LA per 2ms, subN=8
LA per 5ms, subN=8
LA per 0.5ms, subN=512
LA per 10ms, subN=512
LA per 0.5ms, subN=32
LA per 2ms, subN=32
LA per 5ms, subN=32
Figure C.5: Throughput comparison for dierent adaptation rates, for rms delay spread of
0.5s at 20 kmph
c
Suvra Sekhar Das, 2007
220 Chapter C. Hybrid Link Adaptation
Center for TeleInFrastruktur (CTIF), Aalborg University
D
LA in OFDM systems under HPA
PAPR in OFDM
OFDM symbol is generated by superimposing several carriers. These carriers may add
up constructively which results in high amplitude and when it comes to power consid-
eration, the problem is even worse. This problem is widely knows as Peak to Average
Power Ratio (PAPR) problem. Large PAPR of a system makes the implementation of
Digital to Analogue Converter (DAC) and Analogue to Digital Converter (ADC) to
be extremely dicult. The design of RF amplier also becomes increasingly dicult
as the PAPR increases. Since the source of the problem is constructive addition of
subcarrier, with the increase of subcarrier number in an OFDM symbol, PAPR will
also increase. PAPR can be dened mathematically as:
PAPR =
max [x(n)[
2
[x(n)[
2
(D.1)
Several methods have been proposed to reduce the nonlinear eect. Some tech-
nique directly deals with nonlinearity of the amplier, such as pre-distortion, negative
feedback, linear amplication with nonlinear component (LINC), feed forward etc.
Other techniques reduce the PAPR of the OFDM signal so that operating point does
221
222 Chapter D. LA in OFDM systems under HPA
not uctuate over a large range. There are several techniques to reduce PAPR. They
are divided mainly in three category, namely Signal Distortion Techniques, Coding
Techniques and nally the Scrambling Technique. A very good comparison of these
methods can be found in [116].
CDF of PAPR
OFDM signal is generated using (2.19). There X
s
[k] when multiplied by the complex
exponential forms a complex number, the real and imaginary parts are zero mean
random numbers with uniform distribution. Due to the operation of the

operator
which adds a large number of such terms the output of the IFFT can be approximated
to be Gaussian distributed, following the central limit theorum. As the output of the
IFFT is a complex sequence, the amplitude of the OFDM signal can be considered
Rayleigh distributed, while the power distribution is central chi squared distributed
with two degrees of freedom. The cumulative distribution function of the power of
the OFDM signal is therefore given by:
F(z) =
_
z
0
1
2
2
e

u
2
2
du = 1 e
z
(D.2)
where is the variance of the real or imaginary part. Assuming that the samples
z to be mutually uncorrelated and the cumulative distribution function for the peak
power per OFDM symbol is given by [117]:
P(PAPR z) = F(z)
N
= (1 e
z
)
N
(D.3)
where N is the number of subcarrier.
PAPR and Number of Subcarrier
The Figure D.1 shows the theoretical and simulated CDF of PAPR for dierent
number of subcarrier. Theoretical curve is generated using the (D.3). It is seen that
there is no dierence between theoretical and simulation values except for the very
low CDF. This because the probability of such situations are very low and it requires
a very large number of simulations to get the exact curves. Here CDF for 128, 512
and 1024 subcarrier are shown and it shows that PAPR is proportional to the number
of subcarrier.
The CDF of PAPR reaches 90 percentile value very quickly and rest of the
values span over wide range of SNR.
Center for TeleInFrastruktur (CTIF), Aalborg University
223
3 4 5 6 7 8 9 10 11 12
10
4
10
3
10
2
10
1
10
0
PAPR in dB
C
D
F

o
f

P
A
P
R
Theory, N=128
Simulation N=128
Theory, N=512
Simulation N=512
Thory, N=1024
Simulation N=1024
Figure D.1: Comparison of theoretical and simulated CDF of PAPR
Eect on Constellation points
Fig. D shows the performance for two dierent BO values, 3 dB and 6 dB. This can
be compared with Fig. D.4 which shows the received constellation points for 16QAM
modulation scheme when no power amplier was used. It can be seen that the received
constellation points become aected and scattered due to the eect of the nonlinear
distortion of the HPA.
4 2 0 2 4
4
3
2
1
0
1
2
3
4
Real Axis
I
a
m
g
i
n
a
r
y

A
x
i
s
16QAM, HPA=1, BO=6dB
Figure D.2: Eect of BO of 6 dB on 16QAM
constellation points
4 2 0 2 4
4
3
2
1
0
1
2
3
4
Real Axis
I
m
a
g
i
n
a
r
y

A
x
i
s
16QAM, HPA=1, BO=3dB
Figure D.3: Eect of BO on 16QAM con-
stellation points
Eect Modulation and Coding Schemes on PAPR
The Fig. D.5 shows the eect of dierent modulation schemes for a rate
1
2
coding rate
when the IFFT size if 512. All curves overlap, which implies that PAPR for OFDM
c
Suvra Sekhar Das, 2007
224 Chapter D. LA in OFDM systems under HPA
4 2 0 2 4
4
3
2
1
0
1
2
3
4
Real Axis
I
m
a
g
i
n
a
r
y

A
x
i
s
16QAM, HPA=0
Figure D.4: 16QAM basic constellation points
is independent of modulation scheme.
4 5 6 7 8 9 10 11 12
10
3
10
2
10
1
10
0
PAPR indB
C
D
F

o
f

P
A
P
R
M4c1by2N512
M16c1by2N512
M64c1by2N512
Figure D.5: Eect of dierent modulation
scheme on CDF of PAPR when FFT size is
512. M indicates the modulation level, and
C the coding rate.
4 5 6 7 8 9 10 11 12
10
3
10
2
10
1
10
0
PAPR in dB
C
D
F

o
f

P
A
P
R
M16c1by3N128
M16c1by2N128
M16c2by3N128
Figure D.6: Eect of dierent coding rate
on the CDF of PAPR when FFT size is 128.
M indicates the modulation level, and C the
coding rate.
Fig. D.6 shows that PAPR is not aected by the coding rate. Here modulation
and number of subcarrier were xed to 16QAM and 128 respectively while coding
rate was varied. And the result is same as mentioned above. i.e. there is no impact of
coding rate on PAPR distribution. So, PAPR is an inherent property of the subcarrier
of the symbol itself which is not inuenced by the parameters like modulation, FEC
coding etc.
Center for TeleInFrastruktur (CTIF), Aalborg University
225
SDNR Plots
5 10 15 20 25 30
0
5
10
15
20
25
30
Opearting SNR (dB)
S
D
N
R

(
d
B
)
BO=10
BO=8
BO=6
BO=5
BO=4
BO=3
BO=2
BO=1
BO=0
M=16, BO Values in dB
Figure D.7: SDNR plot for 16QAM modu-
lation in AWGN Channel
10 15 20 25 30
0
5
10
15
20
Opearting SNR(dB)
S
D
N
R
(
d
B
)


BO=10
BO=8
BO=6
BO=5
BO=4
BO=3
BO=2
BO=1
BO=0
M=16, BO Values in dB
Figure D.8: SDNR plot for 16QAM modu-
lation in Fading Channel
10 15 20 25 30 35 40
10
15
20
25
30
35
Opearting SNR (dB)
S
D
N
R

(
d
B
)
BO=10
BO=8
BO=6
BO=5
BO=4
BO=3
BO=2
BO=1
BO=0
M=64, BO Values in dB
Figure D.9: SDNR plot for 64QAM modu-
lation in AWGN Channel
15 20 25 30 35 40
5
10
15
20
25
30
Opearting SNR (dB)
S
D
N
R
(
d
B
)


BO=10
BO=8
BO=6
BO=5
BO=4
BO=3
BO=2
BO=1
BO=0
M=64, BO Values in dB
Figure D.10: SDNR plot for 64QAM mod-
ulation in Fading Channel
Performance of Dierent Modulation and Coding
c
Suvra Sekhar Das, 2007
226 Chapter D. LA in OFDM systems under HPA
0 2 4 6 8 10 12 14 16 18
6
5.5
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
POSTSNR (dB)
l
o
g
1
0
(
B
E
R
)
,

U
n
c
o
d
e
d


M=4, BO=10
M=4, BO=8
M=4, BO=6
M=4, BO=5
M=4, BO=4
M=4, BO=3
M=4, BO=2
M=4, BO=1
M=4, BO=0
Figure D.11: BER vs SNR curve for un-
coded and M=4 in AWGN channel
0 5 10 15
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR in dB
l
o
g
1
0
(
B
L
E
R
)


M=4, BO=10
M=4, BO=8
M=4, BO=6
M=4, BO=5
M=4, BO=4
M=4, BO=3
M=4, BO=2
M=4, BO=1
M=4, BO=0
Figure D.12: BLER vs SNR curve for
C =
1
2
and M=4 in AWGN channel
10 15 20 25 30
6
5
4
3
2
1
POSTSNR (dB)
l
o
g
1
0
(
B
E
R
)
,

U
n
c
o
d
e
d


M=16, BO=10
M=16, BO=8
M=16, BO=6
M=16, BO=5
M=16, BO=4
M=16, BO=3
M=16, BO=2
M=16, BO=1
M=16, BO=0
Figure D.13: BER vs SNR curve for un-
coded and M=16 in AWGN channel
5 10 15 20 25 30
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)


M=16, BO=10
M=16, BO=8
M=16, BO=6
M=16, BO=5
M=16, BO=4
M=16, BO=3
M=16, BO=2
M=16, BO=1
M=16, BO=0
Figure D.14: BLER vs SNR curve for
C =
1
2
and M=16 in AWGN channel
15 20 25 30 35 40
6
5
4
3
2
1
POSTSNR (dB)
l
o
g
1
0
(
B
E
R
)
,

U
n
c
o
d
e
d


M=64, BO=10
M=64, BO=8
M=64, BO=6
M=64, BO=5
M=64, BO=4
M=64, BO=3
M=64, BO=2
M=64, BO=1
M=64, BO=0
Figure D.15: BER vs SNR curve for un-
coded and M=64 in AWGN channel
10 15 20 25 30 35 40
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)


M=64, BO=10
M=64, BO=8
M=64, BO=6
M=64, BO=5
M=64, BO=4
M=64, BO=3
M=64, BO=2
M=64, BO=1
M=64, BO=0
Figure D.16: BLER vs SNR curve for
C =
1
2
and M=64 in AWGN channel
Center for TeleInFrastruktur (CTIF), Aalborg University
227
0 2 4 6 8 10
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)
M=4, BO=10
M=4, BO=9
M=4, BO=8
M=4, BO=7
M=4, BO=6
M=4, BO=5
M=4, BO=4
M=4, BO=3
M=4, BO=2
Figure D.17: BLER vs SNR curve for C =
1
3
and M=4 in AWGN channel
0 5 10 15
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)M=4, BO=10
M=4, BO=9
M=4, BO=8
M=4, BO=7
M=4, BO=6
M=4, BO=5
M=4, BO=4
M=4, BO=3
M=4, BO=2
Figure D.18: BLER vs SNR curve for C =
2
3
and M=4 in AWGN channel
5 10 15 20 25 30
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)
M=16, BO=10
M=16, BO=9
M=16, BO=8
M=16, BO=7
M=16, BO=6
M=16, BO=5
M=16, BO=4
M=16, BO=3
M=16, BO=2
Figure D.19: BLER vs SNR curve for C =
1
3
and M=16 in AWGN channel
5 10 15 20 25 30
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)
M=16, BO=10
M=16, BO=9
M=16, BO=8
M=16, BO=7
M=16, BO=6
M=16, BO=5
M=16, BO=4
M=16, BO=3
M=16, BO=2
Figure D.20: BLER vs SNR curve for C =
2
3
and M=16 in AWGN channel
10 15 20 25 30 35 40
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)M=64, BO=10
M=64, BO=9
M=64, BO=8
M=64, BO=7
M=64, BO=6
M=64, BO=5
M=64, BO=4
M=64, BO=3
M=64, BO=2
Figure D.21: BLER vs SNR curve for C =
1
3
and M=64 in AWGN channel
10 15 20 25 30 35 40
4
3.5
3
2.5
2
1.5
1
0.5
0
POSTSNR (dB)
l
o
g
1
0
(
B
L
E
R
)
M=64, BO=10
M=64, BO=9
M=64, BO=8
M=64, BO=7
M=64, BO=6
M=64, BO=5
M=64, BO=4
M=64, BO=3
M=64, BO=2
Figure D.22: BLER vs SNR curve for
C =
2
3
and M=64 in AWGN channel
c
Suvra Sekhar Das, 2007
228 Chapter D. LA in OFDM systems under HPA
0 2 4 6 8 10
2
4
6
8
10
12
14
Power Bo (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)
M=4, C=1by2
M=16, C=1by2
M=64, C=1by2
Figure D.23: TD plot for FEC =
1
2
with
BLER Threshold= 0.1 in AWGN
0 2 4 6 8 10
2
4
6
8
10
12
14
16
Power BO (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)
M=4, C=1by2
M=16, C=1by2
M=64, C=1by2
Figure D.24: TD plot for FEC =
1
2
with
BLER Threshold= 0.05 in AWGN
0 2 4 6 8 10
4
5
6
7
8
9
10
Power BO (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)

M=4,C=1by3
M=16,C=1by3
M=64,C=1by3
Figure D.25: TD plot for FEC =
1
3
with
BLER Threshold= 0.1 in AWGN
0 2 4 6 8 10
2
4
6
8
10
12
14
Power BO (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)
M=4, C=1by3
M=16, C=1by3
M=64, C=1by3
Figure D.26: TD plot for FEC =
1
3
with
BLER Threshold= 0.05 in AWGN
0 2 4 6 8 10
3
4
5
6
7
8
9
10
Power Bo (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)
M=4, C=1by3
M=16, C=1by3
M=64, C=1by3
Figure D.27: TD plot for FEC =
1
3
with
BLER Threshold= 0.1 in Fading Channel
0 2 4 6 8 10
2
4
6
8
10
12
14
Power Bo (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)
M=4, C=1by3
M=16, C=1by3
M=64, C=1by3
Figure D.28: TD plot for FEC =
1
3
with
BLER Threshold= 0.05 in Fading Channel
Center for TeleInFrastruktur (CTIF), Aalborg University
229
0 2 4 6 8 10
3
4
5
6
7
8
9
10
11
12
Power Bo (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)
M=4, C=2by3
M=16, C=2by3
M=64, C=2by3
Figure D.29: TD plot for FEC =
2
3
with
BLER Threshold= 0.1 in AWGN
0 2 4 6 8 10
4
6
8
10
12
14
16
Power Bo (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)
M=4, C=2by3
M=16, C=2by3
M=64, C=2by3
Figure D.30: TD plot for FEC =
2
3
with
BLER Threshold= 0.05 in AWGN
0 2 4 6 8 10
3
4
5
6
7
8
9
10
11
12
Power Bo (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)


M=4,C=2by3
M=16,C=2by3
M=64,C=2by3
Figure D.31: TD plot for FEC =
2
3
with
BLER Threshold= 0.1 in Fading Channel
0 2 4 6 8 10
2
4
6
8
10
12
14
Power Bo (dB)
T
o
t
a
l

D
e
g
r
a
d
a
t
i
o
n

(
d
B
)


M=4, C=2by3
M=16, C=2by3
M=64, C=2by3
Figure D.32: TD plot for FEC =
2
3
with
BLER Threshold= 0.05 in Fading Channel
c
Suvra Sekhar Das, 2007
230 Chapter D. LA in OFDM systems under HPA
Center for TeleInFrastruktur (CTIF), Aalborg University
List of Abbreviations
ABL Adaptive Bit Loading
ACI Adjacent Channel Interference
ADC Analogue to Digital Converter
ADSL Asynchronous Digital Subscriber Line
APMC Adaptive Power, Modulation and Coding
APMxC Adaptive Power and Modulation with Fixed Coding rate
AMCxP Adaptive Modulation and Coding with xed Power
AP Access Point
ASB Adaptive Sub Carrier Bandwidth
ASc Antenna Selection
AWGN Additive White Gaussian Noise
BDM Band Division Multiplexing
BER Bit Error Rate
BLER Block Error Rate
BO Back of
BS Base Station
CDF Cumulative Distribution Function
CDMA Code Division Multiple Access
231
232 Chapter D. LA in OFDM systems under HPA
CP Cyclic Prex
CSI Channel State Information
CQI Channel Quality Index
DAB Digital Audio Broadcasting
DAC Digital to Analogue Converter
DFT Discrete Fourier Transform
DL Downlink
DVB Digital Video Broadcasting
EGC Equal Gain Combining
FDM Frequency Division Multiplexing
FDD Frequency Division Duplex
FEC Forward Error Correction
FER Frame Error Rate
FFT Fast Fourier Transform
FSB Fixed Sub Carrier Bandwidth
HARQ Hybrid Automatic Repeat-reQuest
HDSL High bit-rate Digital Subcarrier Lines
HPA High Power Amplier
GI Guard Interval
ICI Inter Carrier Interference
IFFT Inverse Fast Fourier Transform
IO Interfering Objects
ISI Inter Symbol Interference
LA Link Adaptation
Center for TeleInFrastruktur (CTIF), Aalborg University
233
LUT Look Up Table
MC-SS Multi Carrier Spread Spectrum
MC-SS-MA Multi Carrier Spread Spectrum Multiple Access
MIMO Multiple Input Multiple Output
MISO Multiple Input Single Output
MMSE Minimum Mean Square Error
MRC Maximal Ratio Combining
OFDM Orthogonal Frequency Division Multiplexing
OFDM-BDM Orthogonal Frequency Division Multiplexing - Band Division
Multiplexing
OFDMA-CDM Orthogonal Frequency Division Multiple Acess with Code Division
Multiplexing
OFDM-TDM Orthogonal Frequency Division Multiplexing - Time Division
Multiplexing
OFDM-TDMA Orthogonal Frequency Division Multiplexing - Time Division
Multiple Access
OFDMA Orthogonal Frequency Division Multiple Access
OFDMA-TDMA Orthogonal Frequency Division Multiple Access - Time Division
Multiple Access
PAM Pulse Amplitude Modulation
PAPR Peak to Average Power Ratio
PC Power Control
PCs Personal Computers
PER Packet Error Rate
PSK Phase Shift Keying
c
Suvra Sekhar Das, 2007
234 Chapter D. LA in OFDM systems under HPA
QAM Quadrature Amplitude Modulation
QoS Quality of Service
QPSK Quadrature Phase Shift Keying
RF Radio Frequency
RMS Root Mean Square
SCH Subcarrier hopping
SCH-OFDMA-CDM Sub-Carrier Hopped Orthogonal Frequency Division Multiple
Access with Code Division Multiplexing
SCH-MC-SS Sub-Carrier Hopped Multi Carrier Spread Spectrum
SDNR Signal to Distortion plus Noise Ratio
SE Spectral Eciency
SIC Successive Interference Cancelation
SIMO Single Input Multiple Output
SINR Signal to Interference plus Noise Ratio
SISO Single Input Single Output
SNR Signal to Noise Ratio
SSPA Solid State Power Amplier
TDD Time Division Duplex
TDEG Total Degradation
TDM Time Division Multiplexing
TS Training Sequence
TWTA Traveling Wave Tube Amplier
UE User Equipment
UL Uplink
Center for TeleInFrastruktur (CTIF), Aalborg University
235
VDSL Very High speed Digital Subscriber Line
VGI Variable Guard Interval
VSB Variable Sub Carrier Bandwidth
WiMAX Worldwide Interoperability for Microwave Access
WLAN Wireless Local Area Network
WMAN Wireless Metropolitan Area Network
WPAN Wireless Personal Area Network
c
Suvra Sekhar Das, 2007
236 Chapter D. LA in OFDM systems under HPA
Center for TeleInFrastruktur (CTIF), Aalborg University
Bibliography
[1] Prasad R., OFDM for Wireless Communications. Artech House Publishers,
2004.
[2] Technical specication group radio access network; physical layer aspects for
evolved universal terrestrial radio access (utra): Release 7, 3rd Generation
Partnership Project, Draft Standard 3GPP TR 25.814 V7.1.0, 2006-09.
[3] Nee R. V. & Prasad R., OFDM for Wireless Multimedia Communications.
Artech House Publishers, 2000.
[4] Hanzo L., M unster M., Choi B.J., Keller T., OFDM and MC-CDMA for Borad-
band Multi-User Communications, WLANs, and Broadcasting. Wiley, 2003.
[5] Kuipers B. W. M., MULTICARRIER - ORTHOGONAL FREQUENCY DI-
VISION MULTIPLE ACCESS, Ph.D. dissertation, Aalborg University, 2006.
[6] Weinstein S.B., Ebert P.M., Data Transmission by Frequency Division Mul-
tiplexing Using the Discrete Fourier Transform, IEEE Transactions on Com-
munications, vol. 19, no. 5, Oct. 1971.
[7] Chow P.S. , Tu J.C., Cio J. M., A Discrete Multitone Transceiver System
for HDSL Applications, IEEE Journal on Selected Areas in Communications,
vol. 9, no. 6, Sep. 1991.
[8] Chow P.S., Tu J.C. and Cio J.M., Performance Evaluation of a Multichannel
Transceiver System for ADSL and VHDSL services, IEEE Journal on Selected
Areas in Communications, vol. 9, no. 6, Aug. 1991.
[9] ETSI, Radio broadcasting systems; Digital Audio Broadcasting (DAB) to
mobile, portable and xed receivers, European Telecommunication Standard,
Standard EN-300-401, May 1997.
237
238 BIBLIOGRAPHY
[10] Hiperlan2, Broadband Radio Access Networks (BRAN),HIPERLAN Type
2;Physical (PHY) layer, ETSI, Tech. Rep., 1999.
[11] Part 11: Wireless lan medium access control (mac) and physical layer (phy)
specications amendment 4: Further higher data rate extension in the 2.4 ghz
band, IEEE, Standard IEEE Std 802.11g.-2003, June 2003.
[12] ETSI, Digital Video Broadcasting (DVB);Framing structure, channel coding
and modulation for digital terrestrial television, European Telecommunication
Standard, Standard EN-300-744, 2004-2006.
[13] , Digital Video Broadcasting (DVB); Transmission System for Handheld
Terminals (DVB-H), European Telecommunication Standard, Standard EN-
302-304, 2004.
[14] Part 16: Air Interface for Fixed Broadband Wireless Access SystemsAmend-
ment 2: Medium Access Control Modications and Additional Physical Layer
Specications for 2-11 GHz, IEEE, Standard IEEE Std 802.16a-2003, 2003.
[15] Part 16: Air interface for xed broadband wireless access systems. amendment
for physical and medium access control layers for combined xed and mobile
operation in licensed bands, IEEE, Standard IEEE Std 802.16e/D12,October
2005, 2005.
[16] Cherry, S, Edholms law of bandwidth, IEEE Spectrum, vol. 41, no. 7, pp.
5860, July 2004.
[17] Kaiser S., Fazel K., A exible spread-spectrum multi-carrier multiple-access
system for multi-media applications, in The 8th IEEE International Sympo-
sium on Personal, Indoor and Mobile Radio Communications, vol. 1, Sep. 1997,
pp. 100104.
[18] Narasimhan, R., Performance of diversity schemes for OFDM systems with fre-
quency oset, phase noise, and channel estimation errors, IEEE Transactions
on Communications, vol. 50, no. 10, pp. 15611565, Oct. 2002.
[19] Zhao Y., Haggman, S. G., Intercarrier interference self-cancellation scheme for
OFDM mobile communication systems, IEEE Transactions on Communica-
tions, vol. 49, no. 7, pp. 1185 1191, July 2001.
[20] Stuber G.L., Principles of mobile communication. Kluwer Academic, 2001.
Center for TeleInFrastruktur (CTIF), Aalborg University
BIBLIOGRAPHY 239
[21] Hou W, Chen B., ICI cancellation for OFDM communication systems in time-
varying multipath fading channels, IEEE Transactions on Communications,
vol. 4, no. 5, Sep. 2005.
[22] Tomasin S, et al., Iterative interference cancellation and channel estimation
for mobile OFDM, IEEE Transactions on Wireless Communications, vol. 4,
no. 1, pp. 238 245, 2005.
[23] Moose P.H., A Technique for Orthogonal Frequency Division Multiplexing
Frequency Oset Correction, IEEE Transactions on Communications, vol. 42,
no. 10, pp. 29082914, Oct. 1994.
[24] Fletcher, P.N., Iterative decoding for reducing cyclic prex requirement in
OFDM modulation, IEE Electronic Letters, vol. 39, no. 6, pp. 539541, Mar
2003.
[25] Chen S.; Zhu C., ICI and ISI analysis and mitigation for OFDM systems with
insucient cyclic prex in time-varying channels, vol. 50, no. 1, pp. 7883,
Feb. 2004.
[26] Chung S. T., Goldsmith A. J., Degrees of freedom in adaptive modulation:
a unied view, IEEE Transactions on Communications, vol. 49, no. 1, pp.
15611571, Sep. 2001.
[27] Ahn, C.-J.; Sasase, I, The eects of modulation combination, target BER,
Doppler frequency, and adaptation interval on the performance of adaptive
OFDM in broadband mobile channel, IEEE Transactions on Consumer Elec-
tronics, vol. 48, no. 1, pp. 167 174, 2002.
[28] Goldsmith A. J. and Greenstein L., Eect of average power estimation error
on adaptiveMQAM modulation, in IEEE ICC, June 1997, pp. 11051109.
[29] Medbo J., Andersson H., Schramm P., Asplund H., Berg J.-E. , Channel mod-
els for HIPERLAN/2 in dierent indoor scenarios. COST259, EURO-COST,
Report, 1998.
[30] Erceg V., et al., An Empirically Based Path Loss Model for Wireless Channels
in Suburban Environments, IEEE Journal on Selected Areas in Communica-
tions, vol. 17, no. 7, pp. 1205 1211, July 1999.
c
Suvra Sekhar Das, 2007
240 BIBLIOGRAPHY
[31] , A Model for the Multipath Delay Prole of Fixed Wireless Channels,
IEEE Journal on Selected Areas in Communications, vol. 17, no. 3, pp. 399410,
March 1999.
[32] , Channel model for xed wireless application, IEEE 802.16 working
group, Tech. Rep., 2001.
[33] Emerson D.T., THE WORK OF JAGADIS CHANDRA BOSE: 100 YEARS
OF MM-WAVE RESEARCH, in International Microwave Symposium Digest,
1997., IEEE MTT-S, vol. 2. IEEE,, June 1997, pp. 553556.
[34] Rappaport T. S., Wireless Communications Principles and Practice. Prentice
Hall Inc., 1996.
[35] Proakis J.G., Digital Communications, 3rd ed. Mc Graw Hill, 1995.
[36] A. F. Molisch, Wireless Communications, 1st ed. John Wiley & Sons, 2005.
[37] Lawton, M.C.; Davies, R.L.; McGeehan, J.P., An analytical model for indoor
multipath propagation in the picocellular environment, in Sixth International
Conference on Mobile Radio and Personal Communications, Dec. 1991, pp. 18.
[38] Witrisal K., OFDM Air Interface Design for Multimedia Communications,
Ph.D. dissertation, Delft University of Technology, The Netherlands, April 2002.
[39] Paetzold M., Mobile Fading Channels: Modelling, Analysis, & Simulation, 1st
Edition. John Wiley & Sons, 2002.
[40] Greenstein L.J., et. al., A New Path-Gain/Delay-Spread Propagation Model
for Digital Cellular Channels, IEEE Transactions on Vehicular Technology,
vol. 46, no. 2, pp. 477485, May 1997.
[41] Cox D., Leck R., Correlation Bandwidth and Delay Spread Multipath Propaga-
tion Statistics for 910-MHz Urban Mobile Radio Channels, IEEE Transactions
on Communications, vol. 23, no. 11, pp. 12711280, 1975.
[42] Das S.S., Prasad R.,, Time Correlation function for RMS delay spread of a
channel model, in IST Mobile Summit, June 2006.
[43] X. Fuqin, Digital Modulation Techniques. Artech House Publishers, 2000.
[44] Standard
Center for TeleInFrastruktur (CTIF), Aalborg University
BIBLIOGRAPHY 241
[45] Speth M., Fechtel S.A., Fock G., Meyr H., Optimum Receiver Design for Wire-
less Broad-Band Systems Using OFDMPart I, IEEE Transactions on Commu-
nications, vol. 47, no. 11, pp. 1668 1677, Nov. 1999.
[46] Das S. S., et al., Multi Rate Orthogonal Frequency Division Multiplexing, in
IEEE ICC 2005 , vol. 4, May 2005, pp. 2588 2592.
[47] N. Yee, J.P. Linnarz, G. Fettweis, Multi-Carrier CDMA in indoor wireless
radio networks, in IEEE PIMRC 93, Sep. 1993.
[48] A. Chouly, et al., Orthogonal multicarrier technique applied to direct se-
quence spread spectrum CDMA systems, in IEEE Telecommunication con-
ference (GLOBECOM93), Nov.-Dec. 1993.
[49] Fazel K., Performance of CDMA/OFDM for mobile communication sys-
tem, in IEEE international conference on Universal Personal Communications
(ICUPC93), Oct. 1993, pp. 975979.
[50] QingXin C., Elvano S. S., Pasupathy S., Multicarrier CDMA with Adaptive
Frequency Hopping for Mobile Radio Systems, IEEE Journal on Selected Areas
in Communications, vol. 14, no. 9, Dec. 1996.
[51] Yang L-L, Hanzo L.,, Performance of Generalized Multicarrier DS-CDMA
Over Nakagami-m Fading Channels, IEEE Transactions on Communications,
vol. 50, no. 6, pp. 956966, June 2002.
[52] Pongsuwanich N., PAPR issues in Link Adaptation for WiMAX like OFDM
Systems, Masters Thesis, Aalborg University, Sep. 2006.
[53] Heiskala J.& Terry J., OFDM Wireless LANs: A Theoritical and Practical
Guide, 2nd ed. Sams Publishing, July 2001.
[54] Prasad R., Hara S., An Overview of Multi-Carrier CDMA, in 4th IEEE
International Symposium on Spread Spectrum Tehcniques and Applications,
ISSSTA96, Sep. 1996,, p. 107114.
[55] Paulraj A., et al., Introduction to Space Time Wireless Communiactions.
CAMBRIDGE, 2003.
[56] Steendam H., Moeneclaey M., The Eect of Carrier Phase Jitter on MC-
CDMA Performance, IEEE Transactions on Communications, vol. 47, no. 2,
pp. 195198, Feb. 1999.
c
Suvra Sekhar Das, 2007
242 BIBLIOGRAPHY
[57] Legouable R, et al., Performance Comparison between OFDM/TDMA vs MC-
CDMA and DS-CDMA vs Techniques in HIPERLAN/2, in IST Mobile and
Wireless Telecommunications Summit, Aveiro, Portugal, 15-18th June 2003,
Aveiro, Portu, June 2003.
[58] Hanzo, L.; Lie-Liang Yang; Manster, M.; Byoung-Jo Choi;, Recital on
multicarrier communications: space-time coded versus adaptive OFDM/MC-
CDMA, in IST Workshop on Mobile Future, 2004 and the Symposium on
Trends in Communications., 24-26 Oct. 2004, pp. x xxviii.
[59] Tsumura S., Mino R., Hara S., Hara Y., Performance comparison of OFDM-
FH and MC-CDM in single- and multi-cell environments, in 61st IEEE Vehic-
ular Technology Conference, VTC Spring, vol. 3, June 2005, pp. 17301734.
[60] M. Tanda, Blind symbol-timing and frequency-oset estimation in OFDM sys-
tems with real data symbols, IEEE Transactions on Communications, vol. 52,
no. 10, pp. 16091612, Oct. 2004.
[61] Ma X. , Giannakis G.B: , Barbarossa S., Non-data-aided frequency-oset and
channel estimation in OFDM and related block transmissions, in IEEE ICC ,
2001, pp. 18661870.
[62] M. Speth, et al., Optimum Receiver Design for Wireless Broad-Band Systems
Using OFDM - Part II: A case study, IEEE Transactions on Communications,
vol. 49, no. 4, April 2001.
[63] Yang B., Letaief K.B., Cheng R.S. & Cao Z. ,, Timing Recovery for OFDM
Transmission, IEEE Journal on Selected Areas in Communications, vol. 18,
no. 11, Nov. 2000.
[64] Yang B., Ma Z., Cao Z., ML-oriented DA sampling clock synchronization for
OFDM systems, WCC - ICCT 2000, vol. 1, pp. 781 784, Aug. 2000.
[65] Abhayawardhana,V.S., Wassell, I.J.,, Residual frequency oset correction for
coherently modulated OFDM systems in wireless communication, in IEEE
Vehicular Technology Conference Spring, vol. 2, 2002, pp. 777781.
[66] Miaoudakis,et al., An all-digital feed-forward CFO cancellation scheme for
HIPERLAN/2 in multipath environment, in The 13th IEEE International
Symposium on Personal, Indoor and Mobile Radio Communications, 2002, 4,
Ed., Sep. 2002, pp. 1518.
Center for TeleInFrastruktur (CTIF), Aalborg University
BIBLIOGRAPHY 243
[67] Das S. S., et al., Low Complexity Residual Phase Tracking Algorithm for
OFDM-based WLAN Systems, in Proceedings of CSNDSP 2004, July 2004.
[68] Pollet T., Bladel V. M., Moeneclaey M., BER sensitivity of OFDM systems
to carrier frequency oset and Wiener phase noise, IEEE Transactions on
Communications, vol. 43, no. 234, pp. 191 193, Feb.-March-April 1995.
[69] Robertson P., Kaiser S.,, The eects of Doppler spreads in OFDM(A) mobile
radio systems, in IEEE VTC 50th Vehicular Technology Conference, vol. 1,
Sep. 1999, pp. 1922.
[70] Wang T., et al., Performance Degradation of OFDM Systems Due to Doppler
Spreading, IEEE Transactions on Wireless Communications, vol. 5, no. 6, pp.
14221432, June 2006.
[71] Steendam H., Moeneclaey M., Analysis and Optimization of the Performance
of OFDM on Frequency-Selective Time-Selective Fading Channels, IEEE
Transactions on Communications, vol. 47, no. 12, pp. 18111819, Dec. 1999.
[72] Hung C. P., Chen S. G., Chen K. L.,, Design of an ecient variable-length
FFT processor, in ISCAS, vol. 2, May 2004, pp. 833836.
[73] Cai J., Song W., Li Z., Doppler spread estimation for mobile OFDM systems
in Rayleigh fading channels, IEEE Transactions on Consumer Electronics,
vol. 49, no. 4, pp. 973977, Nov. 2003.
[74] Zhao Y., Haggman S. -G., Sensitivity to Doppler shift and carrier frequency
errors in OFDM systems-the consequences and solutions, in IEEE VTC 1996,
vol. 3, pp. 1564 1568.
[75] Muquet B., et al., Cyclic prexing or zero padding for wireless multicarrier
transmissions? IEEE Transactions on Communications, vol. 50, no. 12, Dec.
2002.
[76] Manton J. H., Dissecting OFDM: The Independent Roles of the Cyclic Prex
and the IDFT Operation, IEEE Communications Letters, vol. 5, no. 12, pp.
474476, Dec. 2001.
[77] Rappaport, T.S., Seidel, S.Y., Takamizawa K.,, Statistical channel impulse
response models for factory and open plan building radio communicate system
c
Suvra Sekhar Das, 2007
244 BIBLIOGRAPHY
design, IEEE Transactions on Communications, vol. 39, no. 5, pp. 794 807,
May 1991.
[78] Mohr, W., Radio propagation for local loop applications at 2 GHz, in Third
Annual International Universal Personal Communications, Oct. 1994, pp. 119
123.
[79] Beek V. D., et al., ML estimation of time and frequency oset in OFDM
systems, IEEE Transactions on Signal Processing, vol. 45, no. 7, pp. 1800
1805, July 1997.
[80] Seoane J.L. et al., Analysis of intertone and interblock interference in OFDM
when the length of the cyclic prex is shorter than the length of the impulse
response of the channel, vol. 1. IEEE Globecom97.
[81] Pollet T., et al., Performance Degradation of MultiCarrier Systems Caused
by an Insucient Guard Interval, in International Workshop on Coppler Wire
Access Systems Bridging the last copper drop, Oct. 1997.
[82] Toeltsch M., Molisch, A.F., Ecient OFDM transmission without cyclic prex
over frequency-selective channels, in IEEE PIMRC 2000, vol. 2, Sep.
[83] Bakker J.D., Eliminating the OFDM cyclic prex, in IEEE PIMRC, vol. 2,
Sep. 2002, pp. 834837.
[84] Schur R., Impulse compression for OFDM transmission over time-varying mul-
tipath channels, in IEEE 56th VTC 2002-Fall, vol. 2, Sep.
[85] Park C.J., Im G-H. , Ecient DMT/OFDM transmission with insucient
cyclic prex, IEEE Communications Letters, vol. 8, no. 9, pp. 576578, Sep.
2004.
[86] Lim C., et al., Novel OFDM Transmission Scheme to Overcome ISI Caused by
Multipath Delay Longer than Cyclic Prex, in IEEE VTC 2005-Spring, vol. 3,
June.
[87] Halford K., Webster M., Multipath Measurements in Wireless LANs, intersil,
Application Note AN9895.1, 2001.
[88] Rappaport T. S., Characterization of UHF Multipath Radio Channels in Fac-
tory Buildings, IEEE Transactions on Antennas and Propagation, vol. 37,
no. 8, Aug. 1989.
Center for TeleInFrastruktur (CTIF), Aalborg University
BIBLIOGRAPHY 245
[89] Gradshteyn I.S., Ryzhik I.M., Table of Integral, Series and Products. Academic
Press Inc., 1980.
[90] Tang X., Bar-Ness Y., Maximum Likelihood Channel Delay Estimation with
Limited Number of Samples Using Diagonal Matrix Approximation, Wireless
Personal Communication, vol. 27, no. 3, pp. 183194, 2003.
[91] Garcia F. -G., et al., DFT-based channel estimation in 2D-pilot-symbol-aided
OFDM wireless systems, in IEEE VTC 2001 Spring, vol. 2.
[92] Hayes J., Adaptive Feedback Communications, IEEE Transactions on Com-
munications, vol. 16, no. 1, pp. 2934, Feb. 1968.
[93] Cavers, J, Variable-Rate Transmission for Rayleigh Fading Channels, IEEE
Transactions on Communications, vol. 20, no. 1, pp. 1522, Feb. 1972.
[94] Toyserkani A. T. et al., Sub-carrier based Adaptive Modulation in HIPER-
LAN/2 System, in IEEE International Conference on Communications, 2004
, June.
[95] Das S.S., M.I. Rahman et al., Inuence of PAPR on Link Adaptation Algo-
rithms in OFDM Systems, in IEEE VTC Spring07, Dublin, Ireland, 22-25
April 2007.
[96] Viswanath P., Asymptotically optimal water-lling in vector multipleaccess
channels, IEEE Transactions on Information Theory, vol. 47, pp. 241267,
Jan. 2001.
[97] Chow P.S., Cio J.M., Bingham J.A.C., A practical discrete multitone
transceiver loading algorithm for data transmission over spectrally shaped chan-
nels, IEEE Transactions on Communications, vol. 43, pp. 773775, Feb.-March
1995.
[98] Wyglinski A.M. et. al, Bit Loading With BER-Constraint for Multicarrier
Systems, IEEE Transactions on Wireless Communications, vol. 4, no. 4, pp.
13831387, July 2005.
[99] Wong C. Y., Cheng R.S., Lataief K.B., Murch R.D.,, Multiuser OFDM with
adaptive subcarrier, bit, and power allocation, IEEE Journal on Selected Areas
in Communications, vol. 17, no. 10, pp. 17471758, Oct. 1999.
c
Suvra Sekhar Das, 2007
246 BIBLIOGRAPHY
[100] Zhang Y. J.; Letaief, K.B., Multiuser adaptive subcarrier-and-bit allocation
with adaptive cell selection for OFDM systems, IEEE Transactions on Wire-
less Communications, vol. 3, no. 5, pp. 1566 1575, Sep. 2004.
[101] Kulkarni, G.; Adlakha, S.; Srivastava, M., Subcarrier Allocation and Bit Load-
ing Algorithms for OFDMA-Based Wireless Networks, IEEE Transactions on
Mobile Computing, vol. 4, no. 6, pp. 652 662, Nov.-Dec. 2005.
[102] Paolo Banelli and Giuseppe Barua and Saverio Cacopardi, Eect of HPA Non
Linearity on Frequency Multiplexed OFDM Signals, in IEEE Transaction on
Broadcasting, vol. 47, no. 2, June 2001, pp. 123136.
[103] Ramjee Prasad , OFDM for Wireless Communication. Artech House Publish-
ers, 2004.
[104] A. Molisch, Wireless Communication. IEEE Press, 2005.
[105] Han S. H., and Lee J.H. , An Overview of Peak-to-Average Power Ratio Reduc-
tion Techniques for Multicarrier Transmission, IEEE Transactions on Wireless
Communications, April 2005.
[106] Alsusa, E.; Yang, L., Low-complexity post-IFFT PAPR reduction technique
for OFDM systems, IEE Electronic Letters, vol. 42, no. 19, pp. 11231124,
Sep. 2006.
[107] Nikookar, H.; Lidsheim, K.S., Random phase updating algorithm for OFDM
transmission with low PAPR, IEEE Transactions on Broadcasting, vol. 48,
no. 2, pp. 123128, June 2002.
[108] Park M.; et al., PAPR reduction in OFDM transmission using Hadamard
transform, in IEEE International Conference on Communications ICC, vol. 1,
June 2000, pp. 430433.
[109] You Y.-H.; et al., Training sequence design and channel estimation of OFDM-
CDMA broadband wireless access networks with diversity techniques, IEEE
Transactions on Broadcasting, vol. 49, no. 4, pp. 354361, Dec. 2003.
[110] Chong E. T., Wassell, I.J., Near-optimum training sequences for OFDM sys-
tems, in The 9th Asia-Pacic Conference on Communications, APCC, vol. 1,
Sep. 2003, pp. 119123.
Center for TeleInFrastruktur (CTIF), Aalborg University
BIBLIOGRAPHY 247
[111] Drieberg, M.; Yew Kuan Min; Jeoti, V.;, Low PAPR preamble for IEEE802.16a
OFDM system, in 13th IEEE International Conference on Networks, Jointly
held with the IEEE 7th Malaysia International Conference on Communication.,
vol. 2, Nov. 2005.
[112] Adel A. M. Saleh, Frequency-Independent and Frequency-Dependent Nonlin-
ear Models of TWT Ampliers, in IEEE Transaction on Communication, vol.
COM-29, no. 11, November 1981, pp. 17151720.
[113] G Santella and F Mazzenga, A hybrid analytical-simulation procedure for per-
formance evaluationin M-QAM-OFDM schemes in presence of nonlinear distor-
tions, in IEEE Transactions on Vehicular Technology, vol. 47, no. 1, February
1998, pp. 142151.
[114] C. Rapp, Eects of HPA-nonlinearity on a 4-DPSK/OFDM-signal for a digital
sound broadcasting signal, in ESA, Second European Conference on Satellite
Communications (ECSC-2) p 179-184 (SEE N92-15210 06-32), Oct. 1991, pp.
179184.
[115] Lei M., et al., An Adaptive Power Distribution Algorithm for Improving Spec-
tral Eciency in OFDM, IEEE Transactions on Broadcasting, vol. 50, no. 3,
pp. 347 351, Sep. 2004.
[116] S. H. Han and J. H. Lee, An overview of peak-to-average power ratio reduction
techniques for multicarrier transmission, in IEEE Wireless Communications,
vol. 12, no. 2, April 2005, pp. 56 65.
[117] Richard Van Nee and Ramjee Prasad , OFDM Wireless Multimedia Communi-
cation. Artech House Publishers, 2000.
c
Suvra Sekhar Das, 2007
248 BIBLIOGRAPHY
Center for TeleInFrastruktur (CTIF), Aalborg University
Vita
Suvra Sekhar Das was born in 1977 in a small town Naihati, in the state of West
Bengal in India. He did his schooling from St. Patricks H.S. school in Asansol, West
Bengal. He obtained his B.Eng. degree in Electronics and Communication Engineer-
ing from Birla Institute of Technology (Deemed University), Mesra, Ranchi, India
in the year 2000 with a First Class with Distinction. He joined Tata Consultancy
Services (TCS), India an Assistant Systems Engineer (Trainee) in June 2000. He was
involved with the multi media group where he was involved with building simula-
tors for interactive video televisions. He joined the Embedded systems group in TCS
Kolkata in the year 2001. For the next two years he was engaged in developing several
digital signal processing applications and wireless LAN IEEE 802.11a receiver front
end algorithms. Since September 2003, he has been posted at the Wireless Network-
ing (WING) Group at Aalborg University (AAU) as a Researcher from TCS where
he later started pursuing his PhD studies.
His current research interest is in optimization for wireless systems at the
physical and medium access control layer for multi user cellular networks. He has
been an active IEEE volunteer and has taken part in numerous IEEE activities, most
notably as the TPC chair of the 1
st
AAU IEEE Student Paper Conference, AISPC07,
held in Aalborg in January 2007.
Permanent Address: Center for TeleInFrastruktur (CTiF)
Aalborg University
Niels Jernes Vej 12
9220 Aalborg Ost
Denmark
e-mail: [email protected]
249