OFDM Thesis Lawrey

Adaptive Techniques for Multiuser OFDM
Thesis submitted by
Eric Phillip LAWREY BE (Hons)
in December 2001
for the degree of Doctor of Philosophy

in Electrical and Computer Engineering
School of Engineering
James Cook University
Access to Thesis
I, the undersigned, the author of this thesis, understand that James Cook University
will make it available for use within the University Library and, by microfilm or
other photographic means, allow access to users in other approved libraries. All users
consulting this thesis will have to sign the following statement:
"In consulting this thesis I agree not to copy or closely paraphrase it in whole or in
part without the written consent of the author; and to make proper written
acknowledgement for any assistance which I have obtained from it."
Beyond this, I do not wish to place any restriction on access to this thesis.
…………………………………. ……………..
(signature) (date)
ii
Declaration
I declare that this thesis is my own work and has not been submitted in any form for
another degree or diploma at any university or other institution of tertiary education.
Information derived from the published work of others has been acknowledged in the
text and a list of references is given.
…………………………………. ……………..
(signature) (date)
iii
Acknowledgements
I wish to convey warmest thanks to my parents; Cheryl and Michael, who have given
me endless support, and provided me with the opportunity to reach this far with my
studies.
I wish to thank my supervisor A/Prof Cornelis Jan Kikkert for his support and
encouragement.
I also thank A/Prof Greg Allen, and Dr. Graham Woods of the Department of
Electronic and Computer Engineering at James Cook University for their expert help
and recommendations.
I would like to thank Craig Blackburn for the many hours of white board discussion,
and support.
I would also like to thank both Noel Lovisa and Noel Lovisa (senior) for their
support as friends and for sharing my excitement of discoveries made during this
thesis.
I would particularly like to thank Dr. Nicole Kessissoglou for her motivational
discussions about research and publications, without which, I would probably still be
not finished.
I also would like to thank all staff of the School of Engineering at James Cook
University for their kind help.
iv
ABSTRACT
The demand for high-speed mobile wireless communications is rapidly growing.

OFDM technology promises to be a key technique for achieving the high data
capacity and spectral efficiency requirements for wireless communication systems of
the near future.
This thesis presents an investigation into methods for maximising the spectral
efficiency of Orthogonal Frequency Division Multiplexing (OFDM) systems. As part
of this, an investigation of detrimental effects on OFDM is presented, showing the
effect of: band pass filtering, the use of a raised cosine guard period, clipping
distortion, Additive White Gaussian Noise (AWGN) on modulation BER rate, time
synchronisation error, and frequency offset errors.
An investigation of two adaptive techniques is also presented. These techniques

utilise knowledge obtained by dynamically tracking the radio channel response, to
optimise the user frequency, and subcarrier modulation. Adaptive modulation
independently optimises the modulation scheme applied to each subcarrier so that the
spectral efficiency is maximised, while maintaining a target Bit Error Rate (BER).
For a fading channel, adaptive modulation results in an improvement of 12 - 16 dB in
the Signal to Noise Ratio (SNR) required to maintain a given BER, as compared with
fixed modulation. Adaptive user allocation exploits the difference in frequency
selective fading between users, to optimise user subcarrier allocation. In a multipath
environment the fading experienced on each subcarrier varies from user to user, thus
by utilising user/subcarrier combinations that suffer the least fading, the overall
performance is maximised. Adaptive user allocation results in an additional average
signal power improvement of 3 - 5 dB.
The simulated performance of the adaptive techniques is presented using a set of

measured wide bandwidth (70 MHz) frequency response measurements taken at 1
v
GHz. These measurements show the changes in the frequency selective fading with
small increments in space, allowing the effects of different tracking rates for the
adaptive allocations schemes to be investigated.
This thesis also presents a method for maximising the signal strength within
buildings, by using transmission repeaters. This is similar to the Single Frequency
Networks used in DAB and DVB systems, except applied to small-scale bi-
directional communications. Using multiple repeaters causes multipath problems in
most conventional systems, however OFDM has a sufficiently high multipath
tolerance to combine the multipath energy. It was found to decrease the path loss by
7 dB for an indoor system with two repeaters, and up to 20 dB for eight repeaters.
In addition, two techniques are presented for reducing the Crest Factor (peak to
average power ratio of the RF signal envelope) of OFDM signals. The first technique
is a phasing scheme for OFDM pilot symbols, which uses genetic algorithms to
optimise the phase angle of each subcarrier. This technique achieves a lower CF than
any previously published techniques, obtaining a CF as low as 0.65 dB, which is 2
dB lower than commonly used techniques. The second technique reduces the CF of
data carrying symbols, by including additional subcarriers that are optimised in
amplitude and phase to cancel out the peaks in the overall OFDM symbol. This was
found to produce a net improvement of 4 dB to the worst-case symbol CF.
vi
TABLE OF CONTENTS
Abstract....................................................................................................................... v
Table of Contents ..................................................................................................... vii
List of Tables............................................................................................................ xiii
List of Figures........................................................................................................... xv
Glossary .................................................................................................................. xxiv
Simulation Files ................................................................................................... xxviii
Chapter 1 Introduction.............................................................................................. 1
1.1 Third Generation Wireless Systems ............................................................ 2
1.2 4th Generation Systems and beyond ............................................................. 3
1.3 Orthogonal Frequency Division Multiplexing............................................. 4
1.3.1 Digital Audio Broadcasting.......................................................... 5
1.3.2 Digital Video Broadcasting.......................................................... 7
1.3.3 Multiuser OFDM.......................................................................... 9
1.3.4 HiperLAN2 and IEEE802.11a ................................................... 10
1.4 Thesis outline and contributions to the field .............................................. 11
1.4.1 Chapter 2 Basic Principles of OFDM ........................................ 11
1.4.2 Chapter 3 Radio Propagation ..................................................... 12
1.4.3 Chapter 4 Multiuser OFDM ....................................................... 13
1.4.4 Chapter 5 Access Point Repeaters.............................................. 14
1.4.5 Chapter 6 Generation of low Crest Factor Multi- Tone Signals . 14
1.4.6 Chapter 7 Crest Factor Minimisation of OFDM Data Symbols . 15
1.4.7 Chapter 8 Summary.................................................................... 15
vii
Chapter 2 Basic Princlipes of OFDM..................................................................... 16
2.1.1 Orthogonality ............................................................................. 18
2.1.2 Frequency Domain Orthogonality.............................................. 20
2.2 OFDM Generation and Reception............................................................. 21
2.2.1 Serial to Parallel Conversion...................................................... 22
2.2.2 Subcarrier Modulation................................................................ 23
2.2.3 Frequency to Time Domain Conversion.................................... 25
2.2.4 RF Modulation ........................................................................... 26
2.2.5 Real verses Complex OFDM generation ................................... 27
2.3 Guard Period .............................................................................................. 30
2.3.1 Protection against Time Offset................................................... 32
2.3.2 Protection against ISI ................................................................. 32
2.3.3 Guard Period Overhead and Subcarrier Spacing........................ 36
2.4 Bandlimiting of OFDM and windowing.................................................... 37
2.4.1 Band Pass Filtering..................................................................... 39
2.4.2 Computational Complexity of FIR band pass filtering .............. 42
2.4.3 Effect of band pass filtering on OFDM performance ................ 43
2.5 Raised Cosine Guard Period ...................................................................... 48
2.6 Effect of Additive White Gaussian Noise on OFDM ................................ 51
2.6.1 Modulation schemes................................................................... 52
2.6.2 OFDM verses Single Carrier Transmission............................... 53
2.6.3 Modulation Limitations of systems............................................ 55
2.6.4 Simulation Set up ....................................................................... 56
2.6.5 Gray Coding ............................................................................... 57
2.6.6 Coherent Modulation.................................................................. 62
2.6.7 Differential Phase Modulation................................................... 62
2.6.8 Differential QAM ....................................................................... 64
2.6.9 Simulation Results...................................................................... 67
2.7 Effect of pilot symbol channel equalisation in coherent modulation........ 69
2.8 Effect of Distortion on OFDM .................................................................. 72
2.8.1 Distortion Modelling .................................................................. 74
2.8.2 Spectral SpreadinG from Distortion Clipping............................ 77
2.8.3 Effective SNR from Clipping Distortion ................................... 79
viii
2.8.4 Effect of pilot symbol type on the BER with distortion............. 80
2.9 Reduction of spectral spreading by Pre- filter Clipping ............................. 82
2.10 Effect of Time Synchronisation Error ....................................................... 86
2.11 Effect of Frequency Synchronisation Error............................................... 88
Chapter 3 Radio Propagation ................................................................................. 91

3.1 Path loss and Attenuation .......................................................................... 91
3.2 Shadowing and slow fading....................................................................... 93
3.3 Multipath Effects ....................................................................................... 95
3.4 Delay Spread.............................................................................................. 97
3.5 Short term fading ....................................................................................... 99
3.6 Frequency selective fading ...................................................................... 100
3.7 Frequency-Space Fading ......................................................................... 102
3.7.1 Equalisation.............................................................................. 103
3.8 Frequency Space Fading measurements .................................................. 104
3.8.1 Measurement Locations ........................................................... 104
3.8.2 Measurement Results ............................................................... 107
3.9 Rate of change of radio channels ............................................................. 109
3.9.1 Experimental Set up ................................................................. 110
3.9.2 Scope and Scaling of the Results ............................................. 112
3.9.3 Measurement Location............................................................. 113
3.9.4 Effect of walking toward the transmitter channel response ..... 114
3.9.5 Effect of typing on a keyboard near the transmitter on the
channel response ...................................................................................... 117
3.9.6 Effect of a moving person on the channel response ................. 119
Chapter 4 Multiuser OFDM.................................................................................. 122

4.1 Network Structure.................................................................................... 123
4.1.1 Peer-to-peer Networking .......................................................... 123
4.1.2 Single Cell Base Station........................................................... 123
4.1.3 Forward Channel...................................................................... 124
4.1.4 Reverse Channel....................................................................... 125
4.2 Adaptive modulation ............................................................................... 127
ix
4.2.1 Limitations of Adaptive Modulation........................................ 130
4.2.2 Effective SNR measurement .................................................... 131
4.2.3 Single User Adaptive Modulation............................................ 133
4.3 Multiuser Subcarrier Allocation .............................................................. 138
4.3.1 Fixed frequency grouped subcarriers ....................................... 140
4.3.2 Random frequency hopping ..................................................... 143
4.3.3 Time Division Multiple Access Allocation.............................. 146
4.3.4 Comb Spread Subcarriers ......................................................... 149
4.3.5 Adaptive User Allocation and Adaptive Bandwidth................ 151
4.3.6 Summary of user allocation scheme performance ................... 158
4.4 Tracking Rate Required for Adaptive Modulation and Adaptive User
Allocation............................................................................................................. 159
4.4.1 Effect of delay on the performance of adaptive modulation.... 163
Chapter 5 Access Point Repeaters ........................................................................ 165

5.1 Spatial Diversity by multiple repeat transmissions ................................. 167
5.1.1 APR spacing for HiperLAN2 system....................................... 170
5.2 Implementation of an Access Point Repeater.......................................... 173
5.3 Experimental Set up ................................................................................. 174
5.4 Discussion of measured results................................................................ 177
5.5 Reduction in cellular interference............................................................ 179
5.6 Shadow Modelling ................................................................................... 179
5.6.1 Intercept calculations ................................................................ 181
5.6.2 Spacial Anti-Aliasing ............................................................... 183
5.6.3 Model Verification................................................................... 183
5.6.4 Simulated Results ..................................................................... 184
5.7 Access Point Repeater Performance with a moving interferer................ 188
5.7.1 SNR calculation........................................................................ 188
5.7.2 Results ...................................................................................... 189
5.8 Conclusion............................................................................................... 193
Chapter 6 Generation of low Crest Factor Multi-Tone Signals......................... 194

6.1 Crest Factor Measurement ....................................................................... 196
x
6.1.1 Random Phase DMT symbols.................................................. 197
6.1.2 Measurement of Crest Factor ................................................... 199
6.1.3 Estimation of error in CF measurements.................................. 200
6.1.4 Simulated Results ..................................................................... 203
6.1.5 CF Errors in Practice................................................................ 204
6.2 Crest Factor Minimisation using Genetic Algorithms ............................. 204
6.2.1 Shapiro-Rudin Phases .............................................................. 205
6.2.2 Newmann Phases...................................................................... 206
6.2.3 S. Narahashi and T. Nojima phasing scheme........................... 208
6.2.4 Genetic Algorithm Phase Optimisation.................................... 210
6.3 Phasing Schemes for Null Symbol Synchronisation............................... 219
6.3.1 Spectral Flatness....................................................................... 221
6.3.2 Experimental Verification........................................................ 224
Chapter 7 Crest Factor Minimisation of OFDM Data Symbols ........................ 228

7.1 Background .............................................................................................. 228
7.2 Peak Reduction Carriers .......................................................................... 229
7.3 Results...................................................................................................... 230
7.4 Effect of PRC Position............................................................................. 232
7.5 Grouped PRCs ......................................................................................... 234
7.6 Spread PRCs ............................................................................................ 234
7.7 Conclusion............................................................................................... 237
Chapter 8 Summary............................................................................................... 238

8.1 Future Work ............................................................................................. 238
8.2 Conclusion............................................................................................... 241
References............................................................................................................... 244
Publications resulting from this PhD ................................................................... 256
Appendix A Window functions ....................................................................... 257

A.1 Window function transition width and side- lobe power.......................... 259
A.2 Window function performance when used for FIR filters ....................... 260
A.3 Window Loss ........................................................................................... 263
xi
A.4 Lawrey Window Function....................................................................... 264
Appendix B BER verses SNR for OFDM in AWGN .................................... 266

B.1 Coherent Quadrature Amplitude Modulation.......................................... 266
B.2 Coherent Phase Shift Keying................................................................... 270
B.3 Differential QAM .................................................................................... 272
B.4 Differential Phase Shift Keying............................................................... 274
Appendix C RF Measurements ...................................................................... 277

C.1 Diversity .................................................................................................. 283
Appendix D Phase tables for evolved low Crest Factor Discrete Multi-Tone
signals ...................................................................................................... 285
D.1 Phase difference between evolved low CF DMT symbols and Newmann’s
phasing scheme .................................................................................................... 292
xii
LIST OF TABLES
Table 1-1, DAB Transmission parameters for each transmission mode [20] ........................ 6
Table 1-2, DVB transmission parameters [18]................................................................... 8
Table 1-3, SNR required and net bit rate for a selection of the coding and modulation
combinations for DVB ................................................................................. 9
Table 1-4, Summary of characteristics of IEEE802.11b, IEEE802.11a and

HiperLAN2. Derived from [97] ...................................................................11
Table 1-5, Physical Layer for HiperLAN2 and IEEE802.11a. Derived from [99] ................11
Table 2-1, 4 bit Gray coding in binary.............................................................................58
Table 2-2, IQ diagrams for modulations used in the OFDM simulations. (script
s0045) ........................................................................................................59
Table 2-3, Phase mapping for differential QPSK..............................................................63
Table 2-4, SNR degradation in dB due to imperfect channel estimation. ............................71
Table 3-1, Table of typical path loss exponent for different environments..........................95
Table 3-2, Measured delay spread for a variety of environments. ......................................99
Table 4-1, Required SNR to maintain a BER below a given threshold. ............................132
Table 4-2, Summary of system performance when using different user allocation
schemes. Note: all schemes used adaptive modulation. ................................158
Table 5-1, Material properties of the internal walls.........................................................184
Table 6-1, Shapiro-Rudin phase sequences for generating low CF DMT signals ...............205
Table 7-1, Optimised positions found for 10 BPSK data subcarriers................................237
xiii
Table A - 1, Transition width and side-lobe power of different window functions.
(script s0039) ............................................................................................260
Table A - 2, Side-lobe power of different window functions when used with the
window method of FIR filter design. (script s0071) .....................................261
Table A - 3, Transition of FIR filters using different window functions. (script s0071) .....262
Table A - 4, Window loss for different window functions (script s0075) ..........................264
Table A - 5, Coefficients for Lawrey5 and Lawrey6 window functions, and other
common window functions........................................................................265
xiv
LIST OF FIGURES
Figure 1-1, Current and future mobile systems. The general trend will be to provide
higher data rates and greater mobility. Derived from [12] and [13]. ................. 4
Figure 2-1, Time domain construction of an OFDM signal. ..............................................19
Figure 2-2, Frequency response of the subcarriers in a 5 tone OFDM signal. (script
0006) .........................................................................................................21
Figure 2-3, Block diagram showing a basic OFDM transceiver. ........................................22
Figure 2-4, Example IQ modulation constellation. 16-QAM, with gray coding of the
data to each location. (script s0045)..............................................................24
Figure 2-5, IQ plot for 16-QAM data with added noise. (script s0083)...............................25
Figure 2-6, OFDM generation, IFFT stage.......................................................................26
Figure 2-7, RF modulation of complex base band OFDM signal, using analog
techniques. .................................................................................................26
Figure 2-8, RF modulation of complex base band OFDM signal, using digital
techniques. (DDS = Direct Digital Synthesis) ...............................................27
Figure 2-9, DC offset OFDM signal, W - bandwidth, f off - frequency offset from DC, f c
centre frequency. ........................................................................................28
Figure 2-10, Construction of the subcarriers for generating a real output time domain
waveform. ..................................................................................................29
Figure 2-11, Construction of the subcarriers for complex signal representation of

OFDM signals ............................................................................................30
Figure 2-12, Addition of a guard period to an OFDM signal .............................................31
Figure 2-13, Function of the guard period for protecting against ISI. .................................33
Figure 2-14, Effectiveness of adding a guard period for removal of ISI. (script s0082) .......36
Figure 2-15, Time waveform of a single carrier OFDM signal, showing 3 symbols. ...........37
xv
Figure 2-16, Spectrum of a 52 subcarrier OFDM signal with no band-limiting. (script
s0050) ........................................................................................................38
Figure 2-17, Spectrum of a 1536 subcarrier OFDM signal with no band-limiting.

(script s0050)..............................................................................................38
Figure 2-18, Spectrum of a 20 subcarrier OFDM, with and without band pass
filtering. Centre subcarrier was not used. (script s0050) .................................41
Figure 2-19, Effective SNR as a function of the time offset for a band pass filtered 52
subcarrier OFDM signal. (script s0057) ........................................................44
Figure 2-20, Section of the waveform that the receiver FFT is taken from depending
on the time offset........................................................................................44
Figure 2-21, Effective SNR for each subcarrier as a result of band pass filtering.
(script s0057) ..............................................................................................45
Figure 2-22, Subcarrier effective SNR, due to band pass filtering. (script s0085) ................47
Figure 2-23, Construction of a RC guard period. ..............................................................48
Figure 2-24, Envelope of OFDM symbols with a flat guard period and an overlapping
raised cosine guard period. ..........................................................................48
Figure 2-25, Side-lobe power for an OFDM signal with 20 subcarriers as the length of
the RC guard period is varied. (script s0050) ................................................50
Figure 2-26, Side-lobe power for an OFDM signal with 100 subcarriers as the length
of the RC guard period is varied. (script s0050) ............................................50
Figure 2-27, Side-lobe power for an OFDM signal with 4000 subcarriers as the length
of the RC guard period is varied. (script s0050) ............................................51
Figure 2-28, IQ plot of 16-PSK using gray coding. Note that each successive IQ
location changes by only a single bit. ...........................................................58
Figure 2-29, IQ data mapping for differential 16-QAM. ...................................................64
Figure 2-30, Bit Error Rate verses the Energy per Bit to Noise Ratio for a selection of
OFDM modulation schemes. (script s0041) ..................................................67
Figure 2-31, SNR required to maintain a BER of < 1x10-5 , for all the simulated
modulation schemes. (script s0046) ..............................................................69
Figure 2-32, Effect of distortion on a 2 tone signal, showing harmonics and IMD.
(script s0049) ..............................................................................................72
xvi
Figure 2-33, Measured Input verses Output for an MAR6 amplifier, showing
distortion. ...................................................................................................75
Figure 2-34, Input verses Output for a power limited linear amplifier. ...............................76
Figure 2-35, Spectrum of OFDM signal with clipping distortion. (script s0048) .................78
Figure 2-36, Strength of the spectral spreading as a function of the Output Power
Backoff. (script s0048) ................................................................................78
Figure 2-37, Crest Factor distribution verses number of subcarriers for an OFDM
signal. (script s0032) ...................................................................................79
Figure 2-38, Effective SNR of an OFDM transmission verses the Output Power
Backoff for 8, 64 and 512 transmission subcarriers. (script s0052) .................79
Figure 2-39, Effective SNR verses output power backoff using different pilot
symbols. (script s0054) ................................................................................81
Figure 2-40, Pre-filter clipping is to reduce spectral spreading due to clipping in the
power amplifier. .........................................................................................82
Figure 2-41, Spectrum of a 64 subcarrier OFDM signal with pre-filter clipping. (script
s0063) ........................................................................................................85
Figure 2-42, Side-lobe power caused by IMD as a result of clipping in the power
amplifier, for different levels of pre-filter clipping. (script s0063) ..................85
Figure 2-43, Effective SNR of the OFDM transmission as a function of the OBO in
the Power Amplifier, and the Pre-filter clipping used. (script s0062) ..............86
Figure 2-44, Effective SNR of an OFDM signal with a time offset error, using a flat
guard period of 40 samples. (script s0053) ....................................................87
Figure 2-45, Effective SNR of an OFDM signal with a time offset error, using a
raised cosine guard period. (script s0069) .....................................................88
Figure 2-46, Effective SNR for Coherent QAM with frequency offset error. (script
s0068) ........................................................................................................90
Figure 3-1, Expanding RF pulse from a central transmitter. (script s0029) .........................92
Figure 3-2, Effect of receiver position on the relative strength of the received direct,
reflected and diffracted signals.....................................................................94
Figure 3-3, Impulse response and phasor plot for multipath channel. .................................96
Figure 3-4, Example channel impulse response for a 2.4 km transmission at 910 MHz
on Manhattan Island, New York City [71]. ...................................................97
xvii
Figure 3-5, Plot of fast fading with distance. This is measured data for a short indoor
link. (script s0030) ....................................................................................100
Figure 3-6, Two path transmission to demonstrate frequency selective fading. (see
text) .........................................................................................................101
Figure 3-7, Frequency selective fading for a short indoor link. (script s0028) ...................102
Figure 3-8, Short term fading, showing fading dependence on distance and frequency.
(script s0030) ............................................................................................103
Figure 3-9, Plan of measurement locations for frequency space fading measurements.......105
Figure 3-10, Set up of equipment for frequency space fading measurements. ...................105
Figure 3-11, Experimental set up for measuring the radio channel response as a
function of distance...................................................................................106
Figure 3-12, Measured frequency space measurement for a 14.9 m non-LOS indoor
link. (script s0030) ....................................................................................107
Figure 3-13, Path loss of the measurements taken. (script s0030) ....................................108
Figure 3-14, Fading probability distribution for each of the measurement locations.
(script s0030) ............................................................................................109
Figure 3-15, Experimental set up. The Network Analyser was used to measure the
amplitude and phase of the channel. The Digital Oscilloscope was used
to capture the results for further processing. ................................................112
Figure 3-16, Photo of the room where the propagation measurements were taken.............113
Figure 3-17, Plan view of the test room, showing the layout of the benches, and the
position of the transmitter and receiver. Note: All lengths are in metres........113
Figure 3-18, Movement of the person in the test room during the measurement period .....114
Figure 3-19, Radio channel response variation due to a person walking toward the
transmitter. (At 940 MHz) (script s0028) ....................................................116
Figure 3-20, Fading causes by a person pretending to type on a keyboard 15 cm from

the transmitter antenna. (script s0028) ........................................................118
Figure 3-21, Movement of the person relative to the transmitter and receiver...................120
Figure 3-22, Fading caused by a person walking in the same room as the transmitter
and receiver. (script s0028)........................................................................121
Figure 4-1, Peer-to-peer network. Users communicate directly to each other, with no
centralised control. ....................................................................................123
xviii
Figure 4-2, Base Station connected system. All data passes through the base station. ........124
Figure 4-3, Forward Link in a single cell system. ...........................................................125
Figure 4-4, Reverse link of an OFDM system using a single cell network and FDM
for user allocation. ....................................................................................126
Figure 4-5, Adaptive Modulation. The modulation scheme is set based on the SNR of
the channel. ..............................................................................................129
Figure 4-6, Algorithm for estimating the signal plus noise, to noise ratio ((S+N)/N)
for a subcarrier. This approximates the SNR when the noise is
significantly smaller than the signal power..................................................131
Figure 4-7, Signal to Noise Ratio of wide band radio channel used for single user
adaptive modulation simulation. (script s0059) ...........................................134
Figure 4-8, Modulation scheme allocated to each subcarrier. (script s0059) .....................135
Figure 4-9, Location of the bit errors for the adaptive modulation simulation. (script
s0059) ......................................................................................................136
Figure 4-10, Location of bit errors for fixed modulation of QPSK, for the same
channel as used for the adaptive modulation simulation (script s0059)..........137
Figure 4-11, Frequency allocations for the fixed frequency group of subcarriers
simulation. (script s0059) ..........................................................................141
Figure 4-12, Simulated performance from allocating a fixed frequency group of

subcarriers to each user. (script s0059) ......................................................142
Figure 4-13, Frequency allocations for each user with random frequency hopping.
(script s0059) ............................................................................................144
Figure 4-14, Simulated performance for randomly hopped groups of subcarriers.

(script s0059) ............................................................................................145
Figure 4-15, Frequency allocations for each user with TDMA. (script s0059) ..................147
Figure 4-16, Simulated performance for TDMA allocation. (script s0059) .......................148
Figure 4-17, Frequency allocations for each user with a comb frequency pattern.
(script s0059) ............................................................................................149
Figure 4-18, System performance when using adaptive modulation and comb
frequency allocation. (script s0059) ............................................................150
Figure 4-19, Simulated adaptive user allocation, utilising adaptive bandwidth,

adaptive frequency, and adaptive modulation. (script s0059) ........................156
xix
Figure 4-20, Continuation of results from Figure 4-19 ....................................................157
Figure 4-21, BER as a function of the distance between tracking points for a multiuser
system using adaptive modulation and adaptive user allocation. (script
s0061) ......................................................................................................160
Figure 4-22, Calculated BER for adaptive modulation. (script s0065) ..............................163
Figure 4-23, BER as a function of the delay between channel measurement and
adaptive modulation allocation (script s0085) .............................................164
Figure 4-24, Location of the bit errors in a single user adaptive modulation scheme at
a SNR of 42 dB. Channel measurement Figure A - 12 was used for this
simulation. (script s0059) ..........................................................................164
Figure 5-1, Methods for obtaining spatial diversity. ........................................................166
Figure 5-2, Pictorial demonstration showing the difference in shadowing when using
a single transmitter (a), and 4 transmitters (b).............................................167
Figure 5-3, Increase in the multipath delay spread as a result of using Access Point
Repeaters. ................................................................................................170
Figure 5-4, Possible implementation of an Access Point Repeater. ..................................173
Figure 5-5, Path loss measurement locations in the ECE building for APR testing. ...........175
Figure 5-6, Photo of the mobile receiver trolley, showing the rotating antenna on top,
spectrum analyser in the middle, and the battery for powering the pre-
amp..........................................................................................................176
Figure 5-7, Transmitter set up simulating Access Point Repeaters. The total
transmitter power from Tx2a and Tx2b were matched to the Tx1 power.......176
Figure 5-8, The receiver antenna was swept over a circle to average out effect of
fading. ......................................................................................................177
Figure 5-9, Measured path loss with one transmitter over the area of the ECE building
at 990 MHz. (script s0011) ........................................................................177
Figure 5-10, Measured path loss for two transmitters over the area of the ECE
building at 990 MHz. (script s0011) ...........................................................178
Figure 5-11, Probability of path loss over the measured area of the building. (script
s0011) ......................................................................................................179
Figure 5-12, Attenuation calculations for a ray passing through two objects.....................180
xx
Figure 5-13, Diagram showing the method for find the intercepts between a ray and
an object...................................................................................................182
Figure 5-14, Polar to rectangular conversion used for plotting of simulated path loss........182
Figure 5-15, Example of spatial aliasing due to finite number of rays used.......................183
Figure 5-16, Cumulative Distribution Function of the path loss within the measured
and simulated building. (script s0011) ........................................................184
Figure 5-17, Simulated path loss for one transmitter at 990 MHz. (script s0011) ..............185
Figure 5-18, Simulated path loss for two transmitters at 990 MHz. (script s0011) .............186
Figure 5-19, Simulated path loss for four transmitters for 990 MHz. (script s0011) ..........186
Figure 5-20, Simulated path loss for 8 transmitters for 990 MHz. (script s0011) ...............187
Figure 5-21, Cumulative Probability Distribution for simulated results, showing the
result for 1, 2, 4 and 8 transmitters. (script s0011) .......................................187
Figure 5-22, Received signal power for interferer and wanted signal. (script s0010) .........191
Figure 5-23, Average spectral efficiency over the area of the top building, as a
function of the movement of the interferer. (script s0009) ............................192
Figure 5-24, Quality of service provided in the top building in the simulation. (script
s0009) ......................................................................................................192
Figure 6-1, The magnitude of the complex base band DMT signal (shown in red)
represents the envelope. (script s0001) .......................................................195
Figure 6-2, Envelope for 26 tone DMT symbol for different phasing schemes. (script
s0005) ......................................................................................................197
Figure 6-3, Crest Factor probability distribution function (split into 40 bins) for
simulated random phase DMT signals, with 8, 32 and 128 tones (script
s0032) ......................................................................................................198
Figure 6-4, CF distribution of phase modulated OFDM signals verses the number of
tones. (script s0032) ..................................................................................199
Figure 6-5, Accurate measurement of CF requires accurate measurement of the peaks

in the waveform, which can be found using interpolation. (script s0004) .......200
Figure 6-6, Worst case peak estimation sampling. ..........................................................200
Figure 6-7, Maximum error in CF measurements as a function of over sampling of the

DMT signals. (script s0002).......................................................................203
xxi
Figure 6-8, Crest Factor of a multitone signal when using Shaprio-Rudin's phasing
scheme. (script s0067) ...............................................................................205
Figure 6-9, Time envelope and spectrogram for DMT based Shaprio-Rudin's phasing
scheme. (script s0055) ...............................................................................206
Figure 6-10, Crest Factor for Newmann's Phasing scheme. (script s0067) ........................207
Figure 6-11, Time envelope and spectrogram for DMT based on Newmann’s
quadratic phasing scheme. (script s0055) ....................................................207
Figure 6-12, Crest Factor as a function of the number of tones for S. Narahashi and T.
Nojima phasing scheme [106] (script s0067) ...............................................208
Figure 6-13, Time envelope and spectrogram for DMT based on S. Narahashi and T.
Nojima’s quadratic phasing scheme. (script s0055) .....................................209
Figure 6-14, Crest Factor for the steepest descent optimised phasing scheme proposed
in [107]. (script s0034) ..............................................................................209
Figure 6-15, Genetic Algorithm used to minimise the CF of the DMT waveforms............213
Figure 6-16, Best crest factor found using GA verses the number of tones in the
signal. (script s0020) .................................................................................214
Figure 6-17, Power distribution of the time waveform for DMT signals generated
using a range of phasing schemes. (script s0055) .........................................214
Figure 6-18, Crest factor distribution as a function of the simulated generations.

(script s0021) ............................................................................................215
Figure 6-19, Effect of population size for genetic algorithm optimisation of CF, for 26
tone signal. (script s0021) ..........................................................................216
Figure 6-20, Effect of percentage population used to breed the next generation for
genetic algorithm optimisation of CF. (population = 100). (script s0021) ......217
Figure 6-21, Effect of changing the amplitude of the mutation noise for genetic
algorithm optimisation of CF. (script s0021) ...............................................217
Figure 6-22, Effect of initial phase sequence used for genetic algorithm optimisation
of the CF. (script s0021) ...........................................................................218
Figure 6-23, (a) Crest factor and (b) time spectral flatness as a function of the phase
equation coefficient A. (script s0056) .........................................................222
Figure 6-24, Fitness of the phase scheme for null symbol time synchronisation as a
function of the phase gain coefficient A. (script s0056) ................................223
xxii
Figure 6-25, Crest Factor and Time spectral flatness as a function of the number of
tones. (script s0056) ..................................................................................224
Figure 6-26, Time waveform and spectrogram of a 224 tone OFDM signal using the
time spectral flatness optimised phasing scheme. (script s0055) ...................224
Figure 6-27, Experimental set up for investigating the performance of the time
synchronisation method. ............................................................................225
Figure 6-28, Measured time synchronisation accuracy for different reference symbols. ....227
Figure 7-1, CF verses number of edge grouped PRCs (8 BPSK data subcarriers)..............230
Figure 7-2, Maximum CF verses number of data subcarriers and edge grouped PRCs,
where M is the number of PRCs.................................................................231
Figure 7-3, Net CF improvement verses number of edge grouped PRCs. (10 BPSK
data subcarriers)........................................................................................232
Figure 7-4, PRC position combinations .........................................................................232
Figure 7-5, CF verses position of 2 grouped PRCs (10 BPSK data subcarriers) ................233
Figure 7-6, CF verses position of 4 grouped PRCs (10 BPSK data subcarriers) ................233
Figure 7-7, CF verses the number of spread PRCs (10 BPSK data subcarriers).................235
Figure 7-8, Net improvement in CF, position optimised PRCs (10 BPSK data
subcarriers)...............................................................................................236
xxiii
GLOSSARY
2G Second Generation mobile phone system (GSM, IS-95)

3G Third Generation mobile phone system (systems using WCDMA)
4G Fourth Generation mobile phone system
AC Alternating Current (Frequency above 0 Hz)
ADSL Asymmetric Digital Subscriber Line
alpha Path loss exponent (rate of path loss change with distance)
AM Amplitude Modulation
APR Access Point Repeater
ASK Amplitude Shift Keying
AWGN Additive White Gaussian Noise
b/s/Hz Bits per second per hertz (unit of spectral efficiency)
BER Bit Error Rate
bps Bits per second
BPSK Binary Phase Shift Keying
BS Base Station
CD Compact Disc
CDMA Code Division Multiple Access
CF Crest Factor (peak to average power ratio of the RF envelope)
DAB Digital Audio Broadcasting
dB Decibel (ratio in log scale)
dBc Decibel relative to main signal power
dBm Decibel relative to 1 milliwatt
DC Direct Current (0 Hz)
DDS Direct Digital Synthesiser
DFT Discrete Fourier Transform
xxiv
DMT Discrete Multi-Tone
DPSK Differential Phase Shift Keying
DSBSC Double Side Band Suppressed Carrier
DS-CDMA Direct Sequence Code Division Multiple Access
DSP Digital Signal Processing
DSSS Direct Sequence Spread Spectrum
DVB Digital Video Broadcasting
DVB-C Digital Video Broadcasting – Cable
DVB-S Digital Video Broadcasting – Satellite
DVB-T Digital Video Broadcasting - Terrestrial
EBNR Energy per Bit to Noise Ratio
FDM Frequency Division Multiplexing
FEC Forward Error Correction
FFT Fast Fourier Transform
FIR Finite Impulse Response (digital filter)
FM Frequency Modulation
Fs Sample Frequency
FSK Frequency Shift Keying
GA Genetic Algorithm
GHz Gigahertz - 109 Hz
GMSK Gaussian Minimum Shift Keying
GSM Global System for Mobile communications
HDTV High Definition Television
HiperLAN2 HIgh PErformance Radio Local Area Network, WLAN standard
(Europe) based on OFDM, with a maximum data rate of 54 Mbps.
Similar to IEEE802.11a
Hz Hertz (cycles per second)
ICI Inter-Carrier Interference
IEEE802.11a WLAN standard (U.S.) based on OFDM, with a maximum data
rate of 54 Mbps. Similar to HiperLAN2
xxv
IEEE802.11b WLAN standard (U.S.) based on DSSS, with a maximum data rate
of 11 Mbps
IF Intermediate Frequency
IFFT Inverse Fast Fourier Transform
IMD Inter-Modulation Distortion
IQ Inphase Quadrature
ISI Inter-Symbol Interference
ISM Industrial Scientific Medical
IS-95 Mobile phone standard using CDMA transmission method.
JCU James Cook University
K Kelvin
kbps Kilo bits per second (103 bps)
kHz Kilohertz - 103 Hz
km Kilometer (103 m)
λ Lambda - RF wavelength
LO Local Oscillator
LOS Line Of Sight
m Metre
Mbps Mega bits per second (106 bps)
MHz Megahertz - 106 Hz
MPEG Moving Picture Experts Group (Video compression standard)
NF Receiver Noise Figure
OBO Output power BackOff
OFDM Orthogonal Frequency Division Multiplexing
PAPR Peak to Average Power Ratio
PC Personal Computer
PDA Personal Digital Assistant
ð Pi (3.14159265 ….)
PLL Phase Locked Loop
PM Phase Modulation
PRC Peak Reduction Carriers
xxvi
PRS Pseudo Random Sequence
PSK Phase Shift Keying
QAM Quadrature Amplitude Modulation
QOS Quality Of Service
QPSK Quadrature Phase Shift Keying
RAKE Multiple tap multipath equalisation for CDMA
RC Raised Cosine (Guard Period)
RF Radio Frequency
RMS Root Mean Squared
SFN Single Frequency Network
SHARC Super Harvard ARChitecture, Digital Signal Processor by Analog
Devices)
SIR Signal to Interference Ratio
SNR Signal to Noise Ratio
SSB Single Side Band
SSPA Solid State Power Amplifier
TDM Time Division Multiplexing
TDMA Time Division Multiple Access
TWTA Travelling Wave Tube Amplifier
µm Micrometre (10-6 m)
UMTS Universal Mobile Telecommunications System
µs Microsecond (10-6 s)
VSB Vestigal Side Band
W Watt (energy per unit time, one joule per second)
W-CDMA Wide-band Code Division Multiple Access
WLAN Wireless Local Area Network
WLL Wireless Local Loop
xxvii
SIMULATION FILES
Each of the simulation plots shown in this thesis were produced using a Matlab
simulation script. Each of these simulation files was assigned a unique script number.
The start of the script file name is shown on each of the plots, for example: script
s0059 corresponds to a complete filename of s0059_aofdm.m. The simulation files
for this thesis are included in electronic format on CD. This CD is available from the
Department of Electrical and Computer Engineering at James Cook University,
Australia. The script corresponding to each plot can be found by using a file search
tool.
Most of the simulation files generate several plots from a single simulation; see the
comments in each simulation file for more details. Where possible the simulation
scripts were written to produce the plots shown in this thesis. There are however
several scripts that were modified to produce a series of figures. The settings in these
scripts must be set to the required application to produce the required plot.
xxviii
Chapter 1 Introduction
CHAPTER 1 INTRODUCTION
Wireless communications is an emerging field, which has seen enormous growth in

the last several years. The huge uptake rate of mobile phone technology, Wireless
Local Area Networks (WLAN) and the exponential growth of the Internet have
resulted in an increased demand for new methods of obtaining high capacity wireless
networks.
Most WLAN systems currently use the IEEE802.11b standard, which provides a
maximum data rate of 11 Mbps [96]. Newer WLAN standards such as IEEE802.11a
[100] and HiperLAN2 [98], [99] are based on OFDM technology and provide a
much higher data rate of 54 Mbps. However systems of the near future will require
WLANs with data rates of greater than 100 Mbps, and so there is a need to further
improve the spectral efficiency and data capacity of OFDM systems in WLAN
applications.
For cellular mobile applications, we will see in the near future a complete
convergence of mobile phone technology, computing, Internet access, and potentially
many multimedia applications such as video and high quality audio. In fact, some
may argue that this convergence has already largely occurred, with the advent of
being able to send and receive data using a notebook computer and a mobile phone.
Although this is possible with current 2G (2nd Generation) Mobile phones, the data
rates provided are very low (9.6 kbps – 14.4 kbps) and the cost is high (typically
$0.20 - $1.30 AUD per minute) [24], [25], limiting the usefulness of such a service.
The goal of third and fourth generation mobile networks is to provide users with a
high data rate, and to provide a wider range of services, such as voice
communications, videophones, and high speed Internet access. The higher data rate
of future mobile networks will be achieved by increasing the amount of spectrum
1
allocated to the service and by improvements in the spectral efficiency. OFDM is a

potential candidate for the physical layer of fourth generation mobile systems. This
thesis presents techniques for improving the spectral efficiency of OFDM systems
applied in WLAN and mobile networks.
1.1 THIRD GENERATION WIRELESS SYSTEMS
Third generation mobile systems such as the Universal Mobile Telecommunications

System (UMTS) [1], [2], [3], [4] and CDMA2000 [6] will be introduced over the
next 1-5 years (2002 onwards) [5]. These systems are striving to provide higher data
rates than current 2G systems such as the Global System for Mobile communications
(GSM) [22], [23] and IS-95. Second generation systems are mainly targeted at
providing voice services, while 3rd generation systems will shift to more data
oriented services such as Internet access.
Third generation systems use Wide-band Code Division Multiple Access (W-
CDMA) as the carrier modulation scheme [10]. This modulation scheme has a high
multipath tolerance, flexible data rate, and allows a greater cellular spectral
efficiency than 2G systems. Third generation systems will provide a significantly
higher data rate (64 kbps – 2 Mbps) [1] than second-generation systems (9.6 – 14.4
kbps). The higher data rate of 3G systems will be able to support a wide range of
applications including Internet access, voice communications and mobile
videophones. In addition to this, a large number of new applications will emerge to
utilise the permanent network connectivity, such as wireless appliances, notebooks
with built in mobile phones, remote logging, wireless web cameras, car navigation
systems, and so forth. In fact most of these applications will not be limited by the
data rate provided by 3G systems, but by the cost of the service.
The demand for use of the radio spectrum is very high, with terrestrial mobile phone
systems being just one of many applications vying for suitable bandwidth. These
applications require the system to operate reliably in non-line-of-sight environments
with a propagation distance of 0.5 - 30 km, and at velocities up to 100 km/hr or
higher. This operating environment limits the maximum RF frequency to 5 GHz, as
operating above this frequency results in excessive channel path loss, and excessive
2
Doppler spread at high velocity. This limits the spectrum available for mobile
applications, making the value of the radio spectrum extremely high.
In Europe auctions of 3G licenses of the radio spectrum began in 1999. In the United
Kingdom, 90 MHz of bandwidth [8] was auctioned off for £22.5 billion [9]. In
Germany the result was similar, with 100 MHz of bandwidth raising $46 billion (US)
[7]. This represents a value of around $450 Million (US) per MHz. The length of
these license agreements is 20 years [8] and so to obtain a reasonable rate of return of
8% on investment, $105 Million (US) per MHz must be raised per year. It is
therefore vitally important that the spectral efficiency of the communication system
is maximised, as this is one of the main limitations to providing a low cost high data
rate service.
1.2 4TH GENERATION SYSTEMS AND BEYOND
Research has just recently begun on the development of 4th generation (4G) mobile
communication systems. The commercial rollout of these systems is likely to begin
around 2008 - 2012, and will replace 3rd generation technology. Few of the aims of
4G networks have yet been published, however it is likely that they will be to extend
the capabilities of 3G networks, allowing a greater range of applications, and
improved universal access. Ultimately 4G networks should encompass broadband
wireless services, such as High Definition Television (HDTV) (4 - 20 Mbps) and
computer network applications (1 - 100 Mbps). This will allow 4G networks to
replace many of the functions of WLAN systems. However, to cover this application,
cost of service must be reduced significantly from 3G networks. The spectral
efficiency of 3G networks is too low to support high data rate services at low cost.
As a consequence one of the main focuses of 4G systems will be to significantly
improve the spectral efficiency.
In addition to high data rates, future systems must support a higher Quality Of
Service (QOS) than current cellular systems, which are designed to achieve 90 - 95%
coverage [11], i.e. network connection can be obtained over 90 - 95% of the area of
the cell. This will become inadequate as more systems become dependent on wireless
networking. As a result 4G systems are likely to require a QOS closer to 98 - 99.5%.
3
In order to achieve this level of QOS it will require the communication system to be
more flexible and adaptive. In many applications it is more important to maintain
network connectivity than the actual data rate achieved. If the transmission path is
very poor, e.g. in a building basement, then the data rate has to drop to maintain the
link. Thus the data rate might vary from as low as 1 kbps in extreme conditions, to as
high as 20 Mbps for a good transmission path. Alternatively, for applications
requiring a fixed data rate, the QOS can be improved by allocating additional
resources to users with a poor transmission path.
A significant improvement in spectral efficiency will be required in order for 4G

systems to provide true broadband access. This will only be achieved by significant
advances in multiple aspects of cellular network systems, such as network structure,
network management, smart antennas, RF modulation, user allocation, and general
resource allocation. This thesis will focus on several aspects, primarily RF
modulation, user allocation and network structure.
Vehicular
4G systems 5G systems
Mobility
(2012) (2020)
3G systems
(2003)
Pedestrian
Stationary 2G systems Hiperlan/2

(2002)
0.03 0.1 0.3 1 3 10 30 100
Data Rate (Mbps)
Figure 1-1, Current and future mobile systems. The general trend will be to
provide higher data rates and greater mobility. Derived from [12] and [13].
1.3 ORTHOGONAL FREQUENCY DIVISION

MULTIPLEXING
Orthogonal Frequency Division Multiplexing (OFDM) is an alternative wireless

modulation technology to CDMA. OFDM has the potential to surpass the capacity of
CDMA systems and provide the wireless access method for 4G systems.
4
OFDM is a modulation scheme that allows digital data to be efficiently and reliably
transmitted over a radio channel, even in multipath environments. OFDM transmits
data by using a large number of narrow bandwidth carriers. These carriers are
regularly spaced in frequency, forming a block of spectrum. The frequency spacing
and time synchronisation of the carriers is chosen in such a way that the carriers are
orthogonal, meaning that they do not cause interference to each other. This is despite
the carriers overlapping each other in the frequency domain. The name ‘OFDM’ is
derived from the fact that the digital data is sent using many carriers, each of a
different frequency (Frequency Division Multiplexing) and these carriers are
orthogonal to each other, hence Orthogonal Frequency Division Multiplexing.
The origins of OFDM development started in the late 1950’s [30] with the
introduction of Frequency Division Multiplexing (FDM) for data communications. In
1966 Chang patented the structure of OFDM [32] and published [31] the concept of
using orthogonal overlapping multi-tone signals for data communications. In 1971
Weinstein [33] introduced the idea of using a Discrete Fourier Transform (DFT) for
implementation of the generation and reception of OFDM signals, eliminating the
requirement for banks of analog subcarrier oscillators. This presented an opportunity
for an easy implementation of OFDM, especially with the use of Fast Fourier
Transforms (FFT), which are an efficient implementation of the DFT. This suggested
that the easiest implementation of OFDM is with the use of Digital Signal Processing
(DSP), which can implement FFT algorithms. It is only recently that the advances in
integrated circuit technology have made the implementation of OFDM cost effective.
The reliance on DSP prevented the wide spread use of OFDM during the early
development of OFDM. It wasn’t until the late 1980’s that work began on the
development of OFDM for commercial use, with the introduction of the Digital
Audio Broadcasting (DAB) system.
1.3.1 DIGITAL AUDIO BROADCASTING

DAB was the first commercial use of OFDM technology [19], [20]. Development of
DAB started in 1987 and services began in U.K and Sweden in 1995. DAB is a
replacement for FM audio broadcasting, by providing high quality digital audio and
information services. OFDM was used for DAB due to its multipath tolerance.
5
Broadcast systems operate with potentially very long transmission distances (20 -
100 km). As a result, multipath is a major problem as it causes extensive ghosting of
the transmission. This ghosting causes Inter-Symbol Interference (ISI), blurring the
time domain signal.
For single carrier transmissions the effects of ISI are normally mitigated using
adaptive equalisation. This process uses adaptive filtering to approximate the impulse
response of the radio channel. An inverse channel response filter is then used to
recombine the blurred copies of the symbol bits. This process is however complex
and slow due to the locking time of the adaptive equaliser. Additionally it becomes
increasing difficult to equalise signals that suffer ISI of more than a couple of symbol
periods.
OFDM overcomes the effects of multipath by breaking the signal into many narrow
bandwidth carriers. This results in a low symbol rate reducing the amount of ISI. In
addition to this, a guard period is added to the start of each symbol, removing the
effects of ISI for multipath signals delayed less than the guard period (see section 2.3
for more detail). The high tolerance to multipath makes OFDM more suited to high
data transmissions in terrestrial environments than single carrier transmissions.
Parameter Transmission Mode

I II III IV
Bandwidth 1.536 MHz 1.536 MHz 1.536 MHz 1.536 MHz
Modulation DQPSK DQPSK DQPSK DQPSK
Frequency Range ≤ 375 MHz ≤ 1.5 GHz ≤ 3 GHz ≤ 1.5 GHz
(Mobile reception)
Number of subcarriers 1536 384 192 768
Symbol Duration 1000 µs 250 µs 125 µs 500 µs
Guard Duration 246 µs 62 µs 31 µs 123 µs
Total Symbol Duration 1246 µs 312 µs 156 µs 623 µs
Maximum Transmitter 96 km 24 km 12 km 48 km
Separation for SFN
Table 1-1, DAB Transmission parameters for each transmission mode [20]
Table 1-1 shows the system parameters for DAB. DAB has four transmission modes.
The transmission frequency, receiver velocity and required multipath tolerance all
determine the most suitable transmission mode to use.
Doppler spread is caused by rapid changes in the channel response due to movement
of the receiver through a multipath environment. It results in random frequency
6
modulation of the OFDM subcarriers, leading to signal degradation. The amount of

Doppler spread is proportional to the transmission frequency and the velocity of
movement. The closer the subcarriers are spaced together, the more susceptible the
OFDM signal is to Doppler spread, and so the different transmission modes in DAB
allow trade off between the amount of multipath protection (length of the guard
period) and the Doppler spread tolerance.
The high multipath tolerance of OFDM allows the use of a Single Frequency
Network (SFN), which uses transmission repeaters to provide improved coverage,
and spectral efficiency. For traditional FM broadcasting, neighbouring cities must
use different RF frequencies even for the same radio station, to prevent multipath
causes by rebroadcasting at the same frequency. However, with DAB it is possible
for the same signal to be broadcast from every area requiring coverage, eliminating
the need for different frequencies to be used in neighbouring areas.
The data throughput of DAB varies from 0.6 - 1.8 Mbps depending on the amount of
Forward Error Correction (FEC) applied. This data payload allows multiple channels
to be broadcast as part of the one transmission ensemble. The number of audio
channels is variable depending on the quality of the audio and the amount of FEC
used to protect the signal. For telephone quality audio (24 kbps) up to 64 audio
channels can be provided, while for CD quality audio (256 kb/s), with maximum
protection, three channels are available.
More information on DAB can be found in [20] and [21].
1.3.2 DIGITAL VIDEO BROADCASTING

The development of the Digital Video Broadcasting (DVB) standards was started in
1993 [14]. DVB is a transmission scheme based on the MPEG-2 standard, as a
method for point to multipoint delivery of high quality compressed digital audio and
video. It is an enhanced replacement of the analogue television broadcast standard, as
DVB provides a flexible transmission medium for delivery of video, audio and data
services [17]. The DVB standards specify the delivery mechanism for a wide range
of applications, including satellite TV (DVB-S), cable systems (DVB-C) and
terrestrial transmissions (DVB-T) [15]. The physical layer of each of these standards
7
is optimised for the transmission channel being used. Satellite broadcasts use a single
carrier transmission, with QPSK modulation, which is optimised for this application
as a single carrier allows for large Doppler shifts, and QPSK allows for maximum
energy efficiency [16]. This transmission method is however unsuitable for terrestrial
transmissions as multipath severely degrades the performance of high-speed single
carrier transmissions. For this reason, OFDM was used for the terrestrial
transmission standard for DVB. The physical layer of the DVB-T transmission is
similar to DAB, in that the OFDM transmission uses a large number of subcarriers to
mitigate the effects of multipath. DVB-T allows for two transmission modes
depending on the number of subcarriers used [18]. Table 1-2 shows the basic
transmission parameters for these two modes. The major difference between DAB
and DVB-T is the larger bandwidth used and the use of higher modulation schemes
to achieve a higher data throughput. The DVB-T allows for three subcarrier
modulation schemes: QPSK, 16-QAM (Quadrature Amplitude Modulation) and 64-
QAM; and a range of guard period lengths and coding rates. This allows the
robustness of the transmission link to be traded at the expense of link capacity. Table
1-3 shows the data throughput and required SNR for some of the transmission
combinations.
DVB-T is a uni-directional link due to its broadcast nature. Thus any choice in data
rate verses robustness affects all receivers. If the system goal is to achieve high
reliability, the data rate must be lowered to meet the conditions of the worst receiver.
This effect limits the usefulness of the flexible nature of the standard. However if
these same principles of a flexible transmission rate are used in bi-directional
communications, the data rate can be maximised based on the current radio
conditions. Additionally for multiuser applications, it can be optimised for individual
remote transceivers.
Parameter 2k Mode 8k Mode

Number subcarriers 1705 6817
Useful Symbol Duration (Tu) 896 µs 224 µs
Carrier Spacing (1/ Tu) 1116 Hz 4464 Hz
Bandwidth 7.61 MHz 7.61 MHz
Table 1-2, DVB transmission parameters [18].
8
Subcarrier Code Rate SNR for BER = 2 × 10 -4 after Bit rate (Mbps)
Modulation Viterbi (dB) Guard Period (Fraction of
Useful symbol duration)
Gaussian Rayleigh
Channel Channel 1/4 1/32
QPSK ½ 3.1 5.4 4.98 6.03
QPSK 7/8 7.7 16.3 8.71 10.56
16-QAM ½ 8.8 11.2 9.95 12.06
16-QAM 7/8 13.9 22.8 17.42 21.11
64-QAM ½ 14.4 16.0 14.93 18.10
64-QAM 7/8 20.1 27.9 26.13 31.67
Table 1-3, SNR required and net bit rate for a selection of the coding and
modulation combinations for DVB
Note: Code rate can be any of the following values: 1/2, 2/3, 3/4, 5/6, 7/8. The
Guard Period duration can be any following values: 1/4, 1/8, 1/16, 1/32.
Derived from [18].
1.3.3 MULTIUSER OFDM

DAB and DVB systems are only uni-directional from the base station to the users.
Not much work has been done on using OFDM for two-way communications or for
multiuser applications. These applications include wireless modems, Wireless Local
Area Networks (WLAN's), Wireless Local Loop (WLL), mobile phones, and mobile
high speed internet. This thesis aims to look at applying OFDM to such applications,
and to look at the resulting advantages and problems. This thesis also presents some
new techniques that can be used to improve broadcast and multiuser OFDM systems.
This thesis presents the performance of adaptive modulation and adaptive user
allocation schemes in a multiuser OFDM system. These techniques improve the
spectral efficiency and QOS.
Fattouche [81] patented a method for implementing a wireless multiuser OFDM

system in 1992, predating any published research in this field. This system used half
duplex Time Division Multiplexing (TDM) to allow multiuser access, with the base
stations and portable units taking turns to transmit. Carrier modulation was fixed and
used D8-PSK (Differential 8 Phase Shift Keying). The system was bandwidth limited
by using a raised cosine guard period. Fattouche is the founder WiLan Inc., which is
one of the few companies currently producing multiuser OFDM modems.
Williams and Prodan [82], patented the use of multiuser OFDM in cable applications
in 1995. This introduced the use of a hybrid user allocation, using Frequency
9
Division Multiplexing (FDM) and TDM. In this system the users were allocated time
and frequency slots depending on the data demand. This patent however, fails to
address problem of obtaining and maintaining accurate time and frequency
synchronisation between users, which is critical for maintaining orthogonality
between users.
Cimini, Chuang, Sollenberger [83] outlined an Advanced Cellular Internet Service

using multiuser OFDM. The aim of this system was to provide Internet access at a
data rate of 1 – 2 Mbps. This system uses time synchronised base stations, which are
allocated time slots in a self-organising fashion. These base station time slots are
then broken down in to time slots for users. In addition to TDM, users are allocated
subcarriers dynamically based on the channel Signal to Interference Ratio (SIR), to
allow minimisation of inter-cellular interference.
Wahlqvist [84] described one possible implementation of multiuser OFDM in a

wireless environment, outlining a user allocation scheme where users were allocated
small blocks of time and frequency. In this scheme, each transmission block consists
of a small group of subcarriers, (5 - 10) and a small number of symbols, about 11 in
length. The aim of this structure is to allocate time and frequency slots to utilise the
high correlation between neighbouring subcarriers, and the small channel variation
between a small group of symbols. This allows the block to be characterised with a
simple pilot tone structure.
1.3.4 HIPERLAN2 AND IEEE802.11A

Development of the European Hiperlan standard was started in 1995, with the final
standard of HiperLAN2 being defined in June 1999. HiperLAN2 pushes the
performance of WLAN systems, allowing a data rate of up to 54 Mbps [97].
HiperLAN2 uses 48 data and 4 pilot subcarriers in a 16 MHz channel, with 2 MHz
on either side of the signal to allow out of band roll off. User allocation is achieved
by using TDM, and subcarriers are allocated using a range of modulation schemes,
from BPSK up to 64-QAM, depending on the link quality. Forward Error Correction
is used to compensate for frequency selective fading.
IEEE802.11a has the same physical layer as HiperLAN2 with the main difference
10
between the standard corresponding to the higher-level network protocols used.
HiperLAN2 is used extensively as an example OFDM system in this thesis. Since the
physical layer of HiperLAN2 is very similar to the IEEE802.11a standard these
examples are applicable to both standards.
Standard 802.11b 802.11a HiperLAN2

Spectrum 2.4 GHz 5.2 GHz 5.2 GHz
Modulation Technique DSSS OFDM OFDM
~ Max physical rate 11 Mbps 54 Mbps 54 Mbps
~ Max data rate, layer 3 5 Mbps 32 Mbps 32 Mbps
Medium access control CSMA/CA TDMA/TDD
Connectivity Conn. less Conn. less Conn. orientated
Table 1-4, Summary of characteristics of IEEE802.11b, IEEE802.11a and

HiperLAN2. Derived from [97]
Parameter Value
Channel Spacing 20 MHz
IFFT used for 20 MSPS 64
Data Subcarriers 48
Pilot Subcarriers 4
Carrier Spacing (Fc) 312.5 kHz (=20 MHz/64)
Nominal Bandwidth 16.25 MHz (=312.5 kHz × 52)
Useful Symbol Period 3.2 µsec (=1/ Fc)
Guard Period 0.8 µsec
Modulation Schemes BPSK, QPSK, 16-QAM, 64-QAM
Coding Rate 1/2, 2/3, 3/4
Table 1-5, Physical Layer for HiperLAN2 and IEEE802.11a. Derived from [99]
1.4 THESIS OUTLINE AND CONTRIBUTIONS TO THE

FIELD
1.4.1 CHAPTER 2 BASIC PRINCIPLES OF OFDM

Chapter 2 provides an introduction to OFDM in general and outlines some of the
problems associated with it. This chapter describes what OFDM is, and how it can be
generated and received. It also looks at why OFDM is a robust modulation scheme
and some of its advantages over single carrier modulation schemes. Detrimental
effects such as noise, band pass filtering, distortion, time synchronisation error and
frequency synchronisation error are described. In addition to this, an investigation
into the effectiveness of using raised cosine guard periods for side-lobe suppression
11
is presented. This chapter establishes the performance of OFDM over a wide range
of modulation schemes from BPSK through to 4096-QAM and over a range of
detrimental effects. This work forms the foundation for the development of adaptive
modulation schemes that are presented in Chapter 4.
1.4.2 CHAPTER 3 RADIO PROPAGATION

This chapter presents an introduction to radio propagation effects such as path loss,
frequency selective fading, Doppler spread and multipath delay spread. The
performance of any wireless communication systems is highly dependent on the
propagation channel, and so a detailed knowledge of radio propagation is important
for optimisation of wireless communications. Chapter 4 presents a method for
exploiting the effects of multipath propagation in multiuser OFDM applications in
order to improve the spectral efficiency of the system and so an understanding of
radio propagation characteristics is needed before different forms of multiuser
OFDM can be discussed. This chapter provides a review of well-known propagation
effects and extends these to look at the effects on wide bandwidth transmissions.
Most radio fading models in literature are based on statistical results, making them of
little use for investigating the performance of the adaptive modulation scheme
presented in Chapter 4. The performance of this technique is dependent on the
correlation of the frequency selective fading and how fast the fading changes with
position of the transceiver. To overcome this problem this chapter includes an
experimental investigation of small-scale frequency selective fading of multipath
environments, which is a measure of the variation in the multipath fading with small
changes in position. This work is used in Chapter 4 to simulate the performance of
adaptive modulation.
Also presented is an experiment measuring the rate of change of the radio channel
with movement of nearby people. This also looks at the error in tracking the channel
under these conditions, giving a guide to how fast adaptive schemes must actively
track the radio channel characteristics.
12
1.4.3 CHAPTER 4 MULTIUSER OFDM

This chapter looks at methods for applying OFDM in multiuser applications. It
provides a study into user allocation schemes, and the application of adaptive
modulation mobile RF environments.
Adaptive modulation is a method for obtaining a high spectral efficiency in a fading

environment. The HiperLAN2 and IEEE802.11a standard use system wide
modulation tracking, allowing the modulation of all subcarriers to be changed to
match the channel SNR. The modulation used in such systems is based on the
average subcarrier SNR, and so the effects of fading still cause large increases in the
error rate.
This thesis proposes that the adaptive modulation is applied on a subcarrier by

subcarrier basis, as a method for combating the effects of frequency selective fading.
Each subcarrier in the OFDM system transmits a different amount of data based on
the Signal to Noise Ratio (SNR) for that subcarrier. The channel performance is
tracked on a regular basis using the bi-directional nature of the link to ensure that
both the transmitter and receiver know what modulation scheme is currently being
used on each subcarrier.
This technique has been applied to wire line (telephone cable) OFDM systems such
as xDSL (x – Asymmetric or Symmetric, Digital Subscriber Loop) [87]. Such a
technique is relatively easy to apply to copper cables, due to the static nature of the
channel response. In a mobile environment the frequency response of the radio
channel changes as a result of the relative movement of the transmitter, receiver and
the surrounding objects. This response changes significantly for movements as small
as a fraction of a wavelength, which is typically 10-30 cm. This dynamic nature of
the channel makes it difficult to track its response, especially when mobile users are
travelling faster than walking speed.
Work has been done studying the use of adaptive modulation in single carrier
systems [90] - [93], however not much work has been published on use of adaptive
modulation in OFDM systems.
This chapter outlines applying adaptive modulation to OFDM systems in a multiuser

mobile environment. Additionally an adaptive frequency allocation scheme is
13
presented for optimisation of user subcarrier frequency. This chapter investigates the
feasibility of such techniques and the possible advantages such as increased capacity,
improved quality of service, and a significant reduction in required forward error
correction.
1.4.4 CHAPTER 5 ACCESS POINT REPEATERS

OFDM can withstand high levels of multipath delay spread. This property can be
exploited to reduce the path loss in WLAN systems. For example the signal from
single base station signal can be simultaneously transmitted from multiple locations
within a cell. This simple technique reduces shadowing caused by walls and
obstructions, improving the coverage of WLAN systems. The signals from the
multiple transmitters appear as multipath at the receiver, but can be compensated for
by the high multipath tolerance of OFDM. These multiple transmitters are referred to
as Access Point Repeaters (APR) because the process is reversible in that the base
station transmits and receives from these APRs. This technique is similar to Single
Frequency Networks used in DAB [14], and DVB systems [20]. The main difference
is this technique is proposed for bi-directional communications and for small-scale
systems. Using APRs coupled with adaptive modulation allows WLAN OFDM
systems to operate at very high spectral efficiencies while maintaining a high Quality
Of Service (QOS), and so this technique provides a significant advance.
1.4.5 CHAPTER 6 GENERATION OF LOW CREST FACTOR

MULTI-TONE SIGNALS
OFDM signals require highly linear amplification to prevent Inter-Carrier
Interference (ICI) and spectral spreading due to Inter-Modulation Distortion (IMD)
[54]. Linear amplifiers have a low energy efficiency making them difficult to use in
battery-operated equipment. Additionally OFDM signals have a high Crest Factor
(CF) (peak to average power ratio of the envelope of the RF signal) due to the multi-
carrier modulation. As a result of the high CF, the output power of the amplifier must
be kept low to prevent clipping and distortion.
This chapter describes how the CF of a sampled signal can be measured accurately,
and outlines a new phasing technique for generating OFDM pilot reference symbols
14
with a very low CF.
The new phasing scheme uses genetic algorithms to optimise the phase of the
subcarriers to minimise the CF of the waveform. This form of optimisation has never
been applied to this application. The result of this process is that symbols with a CF
as low as 0.65 dB are possible. This is 0.7 dB lower than any previously known
algorithm. The very low CF of the OFDM signals makes them useful for pilot
symbols in coherent modulation or for adaptive modulation to allow tracking of the
radio channel. Because of their ultra low CF the average power of these symbols can
be boosted significantly (~6 dB) while still maintaining a lower peak power than data
symbols. This improves the SNR of the channel estimation, reducing the number of
pilot symbols needed, and potentially the channel estimation speed.
1.4.6 CHAPTER 7 CREST FACTOR MINIMISATION OF OFDM

DATA SYMBOLS
This chapter describes a technique for a better than 5 dB reduction in the CF for data
carrying OFDM signals. This improvement is achieved by inserting additional
subcarriers into the signal referred to as peak reduction carriers (PRC). These PRC
are set in phase and amplitude, using a codebook, to minimise the overall symbol CF.
The codebook is obtained using a search of all possible signal combinations. This
technique is optimised for signals using a low number of subcarriers, and low
modulation schemes such as BPSK and QPSK, due to the brute force searching
technique employed.
1.4.7 CHAPTER 8 SUMMARY

A conclusion is provided, which summarises the major results obtained this thesis,
and outlines possible future work in this field.
15
Chapter 2 Basic Principles of OFDM
CHAPTER 2 BASIC PRINCLIPES OF

OFDM
Orthogonal Frequency Division Multiplexing (OFDM) is very similar to the well-

known and used technique of Frequency Division Multiplexing (FDM). OFDM uses
the principles of FDM to allow multiple messages to be sent over a single radio
channel. It is however in a much more controlled manner, allowing an improved
spectral efficiency.
A simple example of FDM is the use of different frequencies for each FM

(Frequency Modulation) radio stations. All stations transmit at the same time but do
not interfere with each other because they transmit using different carrier
frequencies. Additionally they are bandwidth limited and are spaced sufficiently far
apart in frequency so that their transmitted signals do not overlap in the frequency
domain. At the receiver, each signal is individually received by using a frequency
tuneable band pass filter to selectively remove all the signals except for the station of
interest. This filtered signal can then be demodulated to recover the original
transmitted information.
OFDM is different from FDM in several ways. In conventional broadcasting each

radio station transmits on a different frequency, effectively using FDM to maintain a
separation between the stations. There is however no coordination or synchronisation
between each of these stations. With an OFDM transmission such as DAB, the
information signals from multiple stations is combined into a single multiplexed
stream of data. This data is then transmitted using an OFDM ensemble that is made
up from a dense packing of many subcarriers. All the subcarriers within the OFDM
signal are time and frequency synchronised to each other, allowing the interference
between subcarriers to be carefully controlled. These multiple subcarriers overlap in
the frequency domain, but do not cause Inter-Carrier Interference (ICI) due to the
16
orthogonal nature of the modulation. Typically with FDM the transmission signals
need to have a large frequency guard-band between channels to prevent interference.
This lowers the overall spectral efficiency. However with OFDM the orthogonal
packing of the subcarriers greatly reduces this guard band, improving the spectral
efficiency.
All wireless communication systems use a modulation scheme to map the

information signal to a form that can be effectively transmitted over the
communications channel. A wide range of modulation schemes has been developed,
with the most suitable one, depending on whether the information signal is an
analogue waveform or a digital signal. Some of the common analogue modulation
schemes include Frequency Modulation (FM), Amplitude Modulation (AM), Phase
Modulation (PM), Single Side Band (SSB), Vestigial Side Band (VSB), Double Side
Band Suppressed Carrier (DSBSC) [121], [122]. Common single carrier modulation
schemes for digital communications include, Amplitude Shift Keying (ASK),
Frequency Shift Keying (FSK), Phase Shift Keying (PSK) and Quadrature
Amplitude Modulation (QAM) [121] - [123].
Each of the carriers in a FDM transmission can use an analogue or digital modulation
scheme. There is no synchronisation between the transmission and so one station
could transmit using FM and another in digital using FSK. In a single OFDM
transmission all the subcarriers are synchronised to each other, restricting the
transmission to digital modulation schemes. OFDM is symbol based, and can be
thought of as a large number of low bit rate carriers transmitting in parallel. All these
carriers transmit in unison using synchronised time and frequency, forming a single
block of spectrum. This is to ensure that the orthogonal nature of the structure is
maintained. Since these multiple carriers form a single OFDM transmission, they are
commonly referred to as ‘subcarriers’, with the term of ‘carrier’ reserved for
describing the RF carrier mixing the signal from base band. There are several ways
of looking at what make the subcarriers in an OFDM signal orthogonal and why this
prevents interference between them.
17
2.1.1 ORTHOGONALITY
Signals are orthogonal if they are mutually independent of each other. Orthogonality
is a property that allows multiple information signals to be transmitted perfectly over
a common channel and detected, without interference. Loss of orthogonality results
in blurring between these information signals and degradation in communications.
Many common multiplexing schemes are inherently orthogonal. Time Division
Multiplexing (TDM) allows transmission of multiple information signals over a
single channel by assigning unique time slots to each separate information signal.
During each time slot only the signal from a single source is transmitted preventing
any interference between the multiple information sources. Because of this TDM is
orthogonal in nature. In the frequency domain most FDM systems are orthogonal as
each of the separate transmission signals are well spaced out in frequency preventing
interference. Although these methods are orthogonal the term OFDM has been
reserved for a special form of FDM. The subcarriers in an OFDM signal are spaced
as close as is theoretically possible while maintain orthogonality between them.
OFDM achieves orthogonality in the frequency domain by allocating each of the

separate information signals onto different subcarriers. OFDM signals are made up
from a sum of sinusoids, with each corresponding to a subcarrier. The baseband
frequency of each subcarrier is chosen to be an integer multiple of the inverse of the
symbol time, resulting in all subcarriers having an integer number of cycles per
symbol. As a consequence the subcarriers are orthogonal to each other. Figure 2-1
shows the construction of an OFDM signal with four subcarriers.
18
Figure 2-1, Time domain construction of an OFDM signal.
(1a), (2a), (3a) and (4a) show individual subcarriers, with 1, 2, 3, and 4 cycles
per symbol respectively. The phase on all these subcarriers is zero. Note, that
each subcarrier has an integer number of cycles per symbol, making them
cyclic. Adding a copy of the symbol to the end would result in a smooth join
between symbols. (1b), (2b), (3b) and (4b) show the FFT of the time
waveforms in (1a), (2a), (3a) and (4a) respectively. (4a) and (4b) shows the
result for the summation of the 4 subcarriers.
Sets of functions are orthogonal to each other if they match the conditions in
equation (2-1). If any two different functions within the set are multiplied, and
integrated over a symbol period, the result is zero, for orthogonal functions. Another
way of thinking of this is that if we look at a matched receiver for one of the
orthogonal functions, a subcarrier in the case of OFDM, then the receiver will only
see the result for that function. The results from all other functions in the set integrate
to zero, and thus have no effect.
C i= j
∫ s (t )s (t )dt =  0
T
(2-1)
i≠ j
i j
0
Equation (2-2) shows a set of orthogonal sinusoids, which represent the subcarriers
for an unmodulated real OFDM signal.
sin (2πkf o t ) 0 < t < T k = 1, 2, K M

s k (t ) =  (2-2)
 0 otherwise
19
where f o is the carrier spacing, M is the number of carriers, T is the symbol period.
Since the highest frequency component is Mf o the transmission bandwidth is
also Mf o .
These subcarriers are orthogonal to each other because when we multiply the
waveforms of any two subcarriers and integrate over the symbol period the result is
zero. Multiplying the two sine waves together is the same as mixing these
subcarriers. This results in sum and difference frequency components, which will
always be integer subcarrier frequencies, as the frequency of the two mixing
subcarriers has integer number of cycles. Since the system is linear we can integrate
the result by taking the integral of each frequency component separately then
combining the results by adding the two sub-integrals. The two frequency
components after the mixing have an integer number of cycles over the period and so
the sub-integral of each component will be zero, as the integral of a sinusoid over an
entire period is zero. Both the sub-integrals are zeros and so the resulting addition of
the two will also be zero, thus we have established that the frequency components are
orthogonal to each other.
2.1.2 FREQUENCY DOMAIN ORTHOGONALITY

Another way to view the orthogonality property of OFDM signals is to look at its
spectrum. In the frequency domain each OFDM subcarrier has a sinc, sin(x)/x,
frequency response, as shown in Figure 2-2. This is a result of the symbol time
corresponding to the inverse of the carrier spacing. As far as the receiver is
concerned each OFDM symbol transmitted for a fixed time (TFFT) with no tapering at
the ends of the symbol. This symbol time corresponds to the inverse of the subcarrier
spacing of 1/T FFT Hz 1. This rectangular, boxcar, waveform in the time domain
results in a sinc frequency response in the frequency domain. The sinc shape has a
narrow main lobe, with many side-lobes that decay slowly with the magnitude of the
frequency difference away from the centre. Each carrier has a peak at the centre
frequency and nulls evenly spaced with a frequency gap equal to the carrier spacing.
1 OFDM signals typically use a time domain guard period. This guard period extends the length of the
transmitted OFDM symbol, but is removed at the receiver (see section 2.3). As a result we can ignore its effect on
the spectrum seen by the receiver.
20
The orthogonal nature of the transmission is a result of the peak of each subcarrier
corresponding to the nulls of all other subcarriers. When this signal is detected using
a Discrete Fourier Transform (DFT) the spectrum is not continuous as shown in
Figure 2-2 (a), but has discrete samples. The sampled spectrum are shown as ‘o’s in
the figure. If the DFT is time synchronised, the frequency samples of the DFT
correspond to just the peaks of the subcarriers, thus the overlapping frequency region
between subcarriers does not affect the receiver. The measured peaks correspond to
the nulls for all other subcarriers, resulting in orthogonality between the subcarriers.
1 1
0.8 0.8
Tx Power
Tx Power
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
Frequency (Carrier spacings) Frequency (Carrier spacings)
(a) (b)
Figure 2-2, Frequency response of the subcarriers in a 5 tone OFDM signal.
(script 0006)
(a) shows the spectrum of each carrier, and the discrete frequency samples seen
by an OFDM receiver. Note, each carrier is sinc, sin(x)/x, in shape. (b) Shows
the overall combined response of the 5 subcarriers (thick black line).
2.2 OFDM GENERATION AND RECEPTION
OFDM signals are typically generated digitally due to the difficulty in creating large
banks of phase lock oscillators and receivers in the analog domain. Figure 2-3 shows
the block diagram of a typical OFDM transceiver. The transmitter section converts
digital data to be transmitted, into a mapping of subcarrier amplitude and phase. It
then transforms this spectral representation of the data into the time domain using an
Inverse Discrete Fourier Transform (IDFT). The Inverse Fast Fourier Transform
(IFFT) performs the same operations as an IDFT, except that it is much more
computationally efficiency, and so is used in all practical systems. In order to
transmit the OFDM signal the calculated time domain signal is then mixed up to the
required frequency.
21
The receiver performs the reverse operation of the transmitter, mixing the RF signal
to base band for processing, then using a Fast Fourier Transform (FFT) to analyse the
signal in the frequency domain. The amplitude and phase of the subcarriers is then
picked out and converted back to digital data.
The IFFT and the FFT are complementary function and the most appropriate term
depends on whether the signal is being received or generated. In cases where the
signal is independent of this distinction then the term FFT and IFFT is used
interchangeably.
Carrier
Phase I I I
Serial Guard Frame RF
Modulation
to Mapping IFFT Q Period Q Sync Q Modulator
Serial Insertion Insertion Amplifier
Para.
TX data
Transmitter Carrier Time Waveform
Amplitude
Frequency Correction LO
Carrier
Phase
I I
Para. Modulation Guard RF
to Slicer FFT Q Period Q Amplifier
Serial
Serial Removal Demod
RX data
Carrier
Amplitude
Time Sync.
Frame
Receiver Detection
Figure 2-3, Block diagram showing a basic OFDM transceiver.
2.2.1 SERIAL TO PARALLEL CONVERSION

Data to be transmitted is typically in the form of a serial data stream. In OFDM, each
symbol typically transmits 40 - 4000 bits, and so a serial to parallel conversion stage
is needed to convert the input serial bit stream to the data to be transmitted in each
OFDM symbol. The data allocated to each symbol depends on the modulation
scheme used and the number of subcarriers. For example, for a subcarrier modulation
of 16-QAM each subcarrier carries 4 bits of data, and so for a transmission using 100
subcarriers the number of bits per symbol would be 400.
For adaptive modulation schemes such as described in section 4.2, the modulation
scheme used on each subcarrier can vary and so the number of bits per subcarrier
22
also varies. As a result the serial to parallel conversion stage involves filling the data
payload for each subcarrier. At the receiver the reverse process takes place, with the
data from the subcarriers being converted back to the original serial data stream.
When an OFDM transmission occurs in a multipath radio environment, frequency

selective fading can result in groups of subcarriers being heavily attenuated, which in
turn can result in bit errors. These nulls in the frequency response of the channel can
cause the information sent in neighbouring carriers to be destroyed, resulting in a
clustering of the bit errors in each symbol. Most Forward Error Correction (FEC)
schemes tend to work more effectively if the errors are spread evenly, rather than in
large clusters, and so to improve the performance most systems employ data
scrambling as part of the serial to parallel conversion stage. This is implemented by
randomising the subcarrier allocation of each sequential data bit. At the receiver the
reverse scrambling is used to decode the signal. This restores the original sequencing
of the data bits, but spreads clusters of bit errors so that they are approximately
uniformly distributed in time. This randomisation of the location of the bit errors
improves the performance of the FEC and the system as a whole.
2.2.2 SUBCARRIER MODULATION

Once each subcarrier has been allocated bits for transmission, they are mapped using
a modulation scheme to a subcarrier amplitude and phase, which is represented by a
complex In-phase and Quadrature-phase (IQ) vector. Figure 2-4 shows an example
of subcarrier modulation mapping. This example shows 16-QAM, which maps 4 bits
for each symbol. Each combination of the 4 bits of data corresponds to a unique IQ
vector, shown as a dot on the figure. A large number of modulation schemes are
available allowing the number of bits transmitted per carrier per symbol to be varied.
An analysis of the performance of a range of commonly used modulation schemes is
presented in section 2.6.
23
I-Q diagram for 16QAM

1
1000 1001 1011 1010
0.5
1100 1101 1111 1110
Imaginary
0
0100 0101 0111 0110
-0.5
0000 0001 0011 0010

-1
-1 -0.5 0 0.5 1
Real
Figure 2-4, Example IQ modulation constellation. 16-QAM, with gray coding

of the data to each location. (script s0045)
Subcarrier modulation can be implemented using a lookup table, making it very

efficient to implement.
In the receiver, mapping the received IQ vector back to the data word performs
subcarrier demodulation. During transmission, noise and distortion becomes added to
the signal due to thermal noise, signal power reduction and imperfect channel
equalisation. Figure 2-5 shows an example of a received 16-QAM signal with a SNR
of 18 dB. Each of the IQ points is blurred in location due to the channel noise. For
each received IQ vector the receiver has to estimate the most likely original
transmission vector. This is achieved by finding the transmission vector that is
closest to the received vector. Errors occur when the noise exceeds half the spacing
between the transmission IQ points, making it cross over a decision boundary.
24
Figure 2-5, IQ plot for 16-QAM data with added noise. (script s0083)
2.2.3 FREQUENCY TO TIME DOMAIN CONVERSION

After the subcarrier modulation stage each of the data subcarriers is set to an
amplitude and phase based on the data being sent and the modulation scheme; all
unused subcarriers are set to zero. This sets up the OFDM signal in the frequency
domain. An IFFT is then used to convert this signal to the time domain, allowing it to
be transmitted. Figure 2-6 shows the IFFT section of the OFDM transmitter. In the
frequency domain, before applying the IFFT, each of the discrete samples of the
IFFT corresponds to an individual subcarrier. Most of the subcarriers are modulated
with data. The outer subcarriers are unmodulated and set to zero amplitude. These
zero subcarriers provide a frequency guard band before the nyquist frequency and
effectively act as an interpolation of the signal and allows for a realistic roll off in the
analog anti-aliasing reconstruction filters.
25
End Time
Frequency
Parallel to serial
Modulation
Subcarrier
IQ vector
I
Data >
IFFT
Q
Output
Base
Input
Band
Data
OFDM
Zeros Guard signal
Period
Symbol
Start
Figure 2-6, OFDM generation, IFFT stage
2.2.4 RF MODULATION
The output of the OFDM modulator generates a base band signal, which must be
mixed up to the required transmission frequency. This can be implemented using
analog techniques as shown in Figure 2-7 or using a Digital Up Converter as shown
in Figure 2-8. Both techniques perform the same operation, however the performance
of the digital modulation will tend to be more accurate due to improved matching
between the processing of the I and Q channels, and the phase accuracy of the digital
IQ modulator.
Anti-aliasing IQ modulator
Low pass
I
DAC
cos
Complex RF
90°
OFDM RF carrier output
Base Band
sin
Q
DAC
Digital Analog
Figure 2-7, RF modulation of complex base band OFDM signal, using analog
techniques.
26
Digital Up Converter
I Interpolation
Filter
cos RF
Complex output
90 °
OFDM DDS
Base Band DAC
sin
Q Interpolation
Filter
Digital Analog
Figure 2-8, RF modulation of complex base band OFDM signal, using digital
techniques. (DDS = Direct Digital Synthesis)
2.2.5 REAL VERSES COMPLEX OFDM GENERATION

For most wireless applications the OFDM signal is generated at base band using
complex samples, then modulated up to the required frequency using an IQ
modulator, as shown in Figure 2-7 and Figure 2-8. The IQ modulator frequency
shifts the OFDM signal from DC to the required RF frequency, and converts the
complex signal into a real signal. A transmitted RF signal is always a real signal as it
is just a variation in field intensity. It is however possible to directly generate a real
OFDM signal. This is useful in wired applications, such as ADSL. In these
applications the transmitted signal is generally from just above DC to an upper limit
determined by the required signal bandwidth. The required transmission signal is a
real signal as only a single cable is used. If a complex signal were used then two
wires would be needed, one for the real signal and one for the imaginary signal.
A real signal is equivalent to a complex base band signal, centred on DC, mixed to
the new centre frequency using an IQ modulator:
W
fc = +f off (2-3)
2
where f c is the frequency translation required to shift the complex base band signal to
form the real OFDM signal, W is the signal bandwidth and f off is the offset from DC,
also see Figure 2-9. In wired applications such as ASDL, the lower most subcarrier
is offset from DC by a small amount compared with the signal bandwidth. This
27
means that the real signal can be generated directly using the IFFT stage instead of
requiring the use of an IQ modulator for frequency translation.
W
Power
OFDM
signal
Frequency
foff
fc
Figure 2-9, DC offset OFDM signal, W - bandwidth, f off - frequency offset from
DC, f c centre frequency.
Figure 2-10 shows the set up of the OFDM signal in the frequency domain for the
generation of a real waveform. With a real waveform the useable bandwidth of the
signal is only half the sampling frequency, and so to generate a real OFDM signal
only one half of the available subcarriers can be used for data modulation. To create
a real waveform the upper frequency bins of the IFFT must be set to the complex
conjugate of the mirror of the lower half.
This can be contrasted with the construction of a complex base band OFDM signal as
shown in Figure 2-11. In this case all of the frequency bins can be used for subcarrier
modulation, with the main limitation being that the outer bins must be kept as zero to
allow reconstruction of the analog signal, without aliasing occurring. In most
applications the subcarrier corresponding to DC is not used. Its removal simplifies
the implementation hardware. Most OFDM system currently using an analog base
band the same as shown in Figure 2-7. In order for the DC subcarrier to be used it
requires that the IQ outputs are DC coupled to the IQ mixer. This is difficult to
achieve in hardware as offset errors result in large errors in the generated IQ vector.
Using AC coupling reduces the complexity of the implementation and so the DC
subcarrier is usually not used. If digital modulation is used as shown in Figure 2-8
then the DC subcarrier can be used.
28
Figure 2-10, Construction of the subcarriers for generating a real output time
domain waveform.
(The slope on the top of the signal block is to show spectral reversing of the
signal)
29
Figure 2-11, Construction of the subcarriers for complex signal representation

of OFDM signals
2.3 GUARD PERIOD
For a given system bandwidth the symbol rate for an OFDM signal is much lower
than a single carrier transmission scheme. For example for a single carrier BPSK
modulation, the symbol rate corresponds to the bit rate of the transmission. However
for OFDM the system bandwidth is broken up into Nc subcarriers, resulting in a
symbol rate that is Nc times lower than the single carrier transmission. This low
symbol rate makes OFDM naturally resistant to effects of Inter-Symbol Interference
(ISI) caused by multipath propagation.
Multipath propagation is caused by the radio transmission signal reflecting off

30
objects in the propagation environment, such as walls, buildings, mountains, etc.

These multiple signals arrive at the receiver at different times due to the transmission
distances being different. This spreads the symbol boundaries causing energy leakage
between them. Multipath propagation is discussed in more detail in Chapter 3.
The effect of ISI on an OFDM signal can be further improved by the addition of a
guard period to the start of each symbol. This guard period is a cyclic copy that
extends the length of the symbol waveform. Each subcarrier, in the data section of
the symbol, (i.e. the OFDM symbol with no guard period added, which is equal to
the length of the IFFT size used to generate the signal) has an integer number of
cycles. Because of this, placing copies of the symbol end-to-end results in a
continuous signal, with no discontinuities at the joins. Thus by copying the end of a
symbol and appending this to the start results in a longer symbol time. Figure 2-12
shows the insertion of a guard period.
Copy
Guard Guard
IFFT IFFT output IFFT
Period Period
Time
TG TFFT
Symbol N -1 Ts Symbol N +1
Symbol N
Figure 2-12, Addition of a guard period to an OFDM signal
The total length of the symbol is Ts=TG + TFFT, where Ts is the total length of the
symbol in samples, TG is the length of the guard period in samples, and TFFT is the
size of the IFFT used to generate the OFDM signal.
In addition to protecting the OFDM from ISI, the guard period also provides
protection against time-offset errors in the receiver.
31
2.3.1 PROTECTION AGAINST TIME OFFSET

To decode the OFDM signal the receiver has to take the FFT of each received
symbol, to work out the phase and amplitude of the subcarriers. For an OFDM
system that has the same sample rate for both the transmitter and receiver, it must use
the same FFT size at both the receiver and transmitted signal in order to maintain
subcarrier orthogonality. Each received symbol has TG + TFFT samples due to the
added guard period. The receiver only needs TFFT samples of the received symbol to
decode the signal. The remaining TG samples are redundant and are not needed. For
an ideal channel with no delay spread (see section 3.4 for a description of delay
spread) the receiver can pick any time offset, up to the length of the guard period,
and still get the correct number of samples, without crossing a symbol boundary.
Because of the cyclic nature of the guard period changing the time offset simply
results in a phase rotation of all the subcarriers in the signal. The amount of this
phase rotation is proportional to the subcarrier frequency, with a subcarrier at the
nyquist frequency changing by 180° for each sample time offset. Provided the time
offset is held constant from symbol to symbol, the phase rotation due to a time offset
can be removed out as part of the channel equalisation. In multipath environments
ISI reduces the effective length of the guard period leading to a corresponding
reduction in the allowable time offset error.
2.3.2 PROTECTION AGAINST ISI

In an OFDM signal the amplitude and phase of the subcarrier must remain constant
over the period of the symbol in order for the subcarriers to maintain orthogonality.
If they are not constant it means that the spectral shape of the subcarriers will not
have the correct sinc shape, and thus the nulls will not be at the correct frequencies,
resulting in Inter-Carrier Interference. At the symbol boundary the amplitude and
phase change suddenly to the new value required for the next data symbol. In
multipath environments ISI causes spreading of the energy between the symbols,
resulting in transient changes in the amplitude and phase of the subcarrier at the start
of the symbol. The length of these transient effects corresponds to the delay spread of
the radio channel. The transient signal is a result of each multipath component
32
arriving at slightly different times, changing the received subcarrier vector. Figure
2-13 shows this effect. Adding a guard period allows time for the transient part of the
signal to decay, so that the FFT is taken from a steady state portion of the symbol.
This eliminates the effect of ISI provided that the guard period is longer than the
delay spread of the radio channel. The remaining effects caused by the multipath,
such as amplitude scaling and phase rotation are corrected for by channel
equalisation.
Figure 2-13, Function of the guard period for protecting against ISI.
The guard period protects against transient effects due to multipath, removing
the effects of ISI, provided it is longer than the channel delay spread. This
example shows the instantaneous phase of a single carrier for 3 symbols.
The addition of guard period removes most of the effects of ISI; however in practice,
multipath components tend to decay slowly with time, resulting in some ISI even
when a relatively long guard period is used.
Figure 2-14 shows the simulated performance of an OFDM system in the presence of
static multipath. In this case the multipath impulse response (see section 3.3 for more
detail) followed an exponential decay with a time constant of 8 samples, resulting in
an RMS delay spread of 3.5 samples. Each sample in the impulse response was
complex and Gaussian distributed. The RMS delay spread is a common parameter to
33
estimate the spread of the multipath energy in time, and used to estimate the level of
ISI in single carrier communications. Section 3.4 provides a more detailed
description of RMS delay spread with typical values for a range of environments. A
more appropriate measure is the time over which 99% of the total accumulated
impulse energy arrived, which in this simulation was 16 samples.
The results shown in Figure 2-14 plot the effective SNR of the demodulated OFDM
signal as a function of the channel SNR. Effective SNR is used extensively though
out this thesis as a measure of the performance of the communications link. It is a
measure of the signal to noise ratio as seen by the OFDM receiver after
demodulation, where the signal power is the magnitude of the wanted signal, and the
noise is the combined error in the received signal due to all the detrimental effects in
the system including channel noise, IMD, filtering, ISI, ICI, frequency errors, time
offset errors, channel equalisation errors, etc. The effective SNR provides a measure
of the OFDM performance, independent of the modulation scheme. Traditionally the
BER is used to measure the performance of a link, however in this thesis OFDM is
considered the work with a large number of modulation schemes making BER a poor
method of measurement. The BER of any particular modulation scheme can be
estimated from the effective SNR by finding the BER of the modulation scheme in
an AWGN channel with a SNR equal to the effective SNR (see section 2.6.9
Figure 2-14 shows the effect of multipath on the OFDM transmission. Ideally the
effective SNR should follow the channel SNR, however detrimental effects such as
ISI lead to degraded performance. We can see from the results that as the length of
the guard period is increased the maximum effective SNR improves. For example,
the effective SNR of the OFDM signal only reaches a maximum of 15 dB when the
guard period length is 4 samples in length, but reaches 25 dB when a guard period of
16 samples is used. This is a result of more of the ISI energy being removed by the
guard period. This shows that having a guard period (16 samples) that is more than
four times the multipath RMS delay spread (3.5 samples) still results in significant
ISI.
The low effective SNR for when the guard period was a similar length to the channel
RMS delay spread is fine for robust modulation schemes such as BPSK and QPSK,
but is insufficient for higher spectral efficiency modulation schemes such as 64-
34
QAM and 256-QAM. Traditionally the RMS delay spread has been used as a
measure of ISI and the allowable symbol rate in a multipath environment [68], [69],
[70]. However if a higher spectral efficiency is required a more appropriate measure
is needed. To achieve very high spectral efficiencies an effective SNR of greater than
35 dB must be able to be reached. In this case it required a guard period of at least 64
samples in length. This length of the guard period corresponds to the time it took for
the impulse energy to decay to –35 dBc. Thus if we require a SNR of 25 dB then we
have a guard period that is at least long enough to remove all impulse reflections that
are stronger than –25 dBc.
The last two results in the simulation show the performance when using a guard
period of 64 samples, with an IFFT size of 128, and 512. In the 128-point IFFT
simulation, 80 subcarriers were used while in the 512-point simulation, 320
subcarriers were used, making the bandwidth of both systems the same. In order for
the OFDM carriers to remain orthogonal to each other, the channel response must be
approximately flat over the bandwidth of each subcarrier (see section 3.7.1 ). The
simulation using 320 subcarriers divides the channel response using finer subcarriers,
and hence the variation of the channel fading over their bandwidth of each subcarrier
is more constant, improving the performance. The effective SNR for the 128 IFFT
size is not limited by the guard period, but instead by poor channel equalisation
caused by an insufficient number of subcarriers. For OFDM to operate effectively,
the frequency response must be approximately flat over the bandwidth of a
subcarrier. If insufficient subcarriers are used then the frequency response changes
too rapidly, leading to degraded performance.
35
Exp. Decay T=8 samp, RMS del.=3.5 samp, 99% Energy=16 samp
55
Effective SNR of demod. OFDM (dB)

50 Guard=4, IFFT=128
Guard=8, IFFT=128
45
Guard=16, IFFT=128
40 Guard=32, IFFT=128
Guard=64, IFFT=128
35
Guard=64, IFFT=512
30
25
20
15
10
0
5 10 15 20 25 30 35 40 45 50 55 60
Channel SNR (dB)
Figure 2-14, Effectiveness of adding a guard period for removal of ISI. (script
s0082)
2.3.3 GUARD PERIOD OVERHEAD AND SUBCARRIER

SPACING
Adding a guard period lowers the symbol rate, however it does not affect the
subcarrier spacing seen by the receiver. The subcarrier spacing is determined by the
sample rate and the FFT size used to analyse the received signal.
Fs
∆f = (2-4)
N FFT
In Equation (2-4), Äf is the subcarrier spacing in Hz, Fs is the sample rate in Hz, and
NFFT is the size of the FFT. The guard period adds time overhead, decreasing the
overall spectral efficiency of the system.
36
2.4 BANDLIMITING OF OFDM AND WINDOWING
OFDM in the time domain is equivalent to a sum of modulated sinusoidal carriers

that are each windowed in time with a rectangular window function, also known as a
boxcar window function (see Appendix A for more details). This window defines the
boundary of each OFDM symbol, and determines the frequency response of the
generated OFDM signal. Figure 2-15 shows an example time waveform for a single
carrier OFDM transmission using Phase Shift Keying (PSK). The amplitude of the
subcarrier is fixed and the phase is varied from symbol to symbol to transmit the data
information. The subcarrier phase is constant for the entire symbol, resulting in a step
in phase between symbols. These sharp transitions between symbols result in
spreading in the frequency domain. Figure 2-16 shows the spectrum of a 52
subcarrier OFDM signal (same as HiperLAN2, or IEEE802.11a) with no band-pass
limiting. The out of band components only fall off slowly due to the sinc roll off of
each subcarrier. Figure 2-17 shows the spectrum of a 1536 subcarrier OFDM signal
(same as Type I DAB). The side-lobes roll off faster than the 52-subcarrier case, as a
fraction of the system bandwidth. However the side-lobes are still significant (> -40
dBc) even far away from the edge of the OFDM main signal block. These side-lobes
increase the effective bandwidth of the OFDM signal, degrading the spectral
efficiency. There are two common techniques for reducing the level of the side-lobes
to acceptable limits; these are band pass filtering the signal, or adding a RC guard
period.
0.5
-0.5
-1
0 Symbol 1 1 Symbol 2 2 Symbol 3 3
Figure 2-15, Time waveform of a single carrier OFDM signal, showing 3

symbols.
Note: Each symbol has an integer number of cycles per symbol, which is
required for orthogonality. Also there are sharp phase transients between
symbols. (script s0037)
37
OFDM Spectrum, 52 subcarriers, No filtering
-10
Power (dB)
-20
-30
-40
-50
-60 -40 -20 0 20 40 60
Frequency (Normalised to Subcarrier Spacings)
Figure 2-16, Spectrum of a 52 subcarrier OFDM signal with no band-limiting.

(script s0050)
This is the same frequency response as an un-filtered HiperLAN2 signal. For

HiperLAN2 the subcarrier spacing corresponds to 312.5 kHz. The DC
subcarrier has not been used, making the signal symmetrical around DC.
OFDM Spectrum, 1536 subcarriers, No filtering
-10
Power (dB)
-20
-30
-40
-50
-2000 -1500 -1000 -500 0 500 1000 1500 2000
Figure 2-17, Spectrum of a 1536 subcarrier OFDM signal with no band-

limiting. (script s0050)
This is the same frequency response as an un-filtered Type I DAB signal. For
DAB the subcarrier spacing corresponds to 1 kHz.
38
2.4.1 BAND PASS FILTERING

Whenever signals are converted from the digital domain to an analog waveform for
transmission, filtering is used to prevent aliasing occurring [120]. This effectively
band pass filters the signal, removing some of the OFDM side-lobes. The amount of
side-lobe removal depends on the sharpness of the filters used. In general digital
filtering provides a much greater flexibility, accuracy and cut off rate than analog
filters, making them especially useful for band limiting of an OFDM signal.
Figure 2-18 (a) shows the frequency response of the OFDM signal with no filtering.
Figure 2-18 (b) - (e) shows examples of a band pass filtered OFDM signal. These
signals have been filtered with a Finite Impulse Response (FIR) filter [120]
developed using the windowing method (see Appendix A). A low number of
subcarriers were used in these plots so that the roll off of the FIR filtering could be
seen. The filtering removes virtually all of the side lobes, but does so at the cost of
the computational expense of implementing the FIR filtering, and it reduces the
effective SNR of the OFDM channel (see section 2.6 for calculating the BER for a
given modulation scheme and effective SNR). The act of filtering the OFDM signal,
chops off significant energy from the outer subcarriers, distorting their shape and
causing ICI.
No publications to date could be found studying the effects of band pass filtering of
OFDM signals, looking at the reduction in the effective SNR caused by the filtering.
Very sharp cut off filters allow separate blocks of OFDM signals to be packed very
closely in the frequency domain, improving the spectral efficiency. But this tight
filtering can result in a degraded effective SNR, and so its effects must be taken into
consideration when designing a system.
39
OFDM Spectrum, 20 subcarriers, no filtering

10
-10
-30
Power (dB)
-50
(a) -70
-90
-110
-130
- 3 2 - 2 8 - 2 4 - 2 0 - 1 6 - 1 2 -8 -4 0 4 8 12 16 20 24 28 32
OFDM Spectrum, Filter Cutoff: 8 subcarriers, Window Func: 3

10
8 sub-
-10 carriers
-30
Power (dB)
-50
(b) -70
-90
-110
-130
- 3 2 - 2 8 - 2 4 - 2 0 - 1 6 - 1 2 -8 -4 0 4 8 12 16 20 24 28 32
OFDM Spectrum, Filter Cutoff: 2 subcarriers, Window Func: 3

10
-10
-30
Power (dB)
-50
(c) -70
-90
-110
-130
- 3 2 - 2 8 - 2 4 - 2 0 - 1 6 - 1 2 -8 -4 0 4 8 12 16 20 24 28 32
Figure 2-18 continued next page
40
OFDM Spectrum, Filter Cutoff: 8 subcarriers, Window Func: 1.5

10
-10
-30
Power (dB)
-50
(d) -70
-90
-110
-130
- 3 2 - 2 8 - 2 4 - 2 0 - 1 6 - 1 2 -8 -4 0 4 8 12 16 20 24 28 32
OFDM Spectrum, Filter Cutoff: 2 subcarriers, Window Func: 1.5

10
-10
-30
Power (dB)
-50
(e)
-70
-90
-110
-130
- 3 2 - 2 8 - 2 4 - 2 0 - 1 6 - 1 2 -8 -4 0 4 8 12 16 20 24 28 32
Figure 2-18, Spectrum of a 20 subcarrier OFDM, with and without band pass
filtering. Centre subcarrier was not used. (script s0050)
(a) OFDM spectrum with no band pass filtering.
Filtered results were band pass filtered using an FIR filter, which was developed
using the windowing method with a Kaiser window function. See Appendix A.
(script s0050)
(b) Kaiser window width of 3 (Side lobe attenuation of 89 dB). The transition
width of the filter was 8 subcarrier spacings (24 tap FIR filter)
(c) Kaiser window width of 3 (Side lobe attenuation of 89 dB). The transition
(d) Kaiser window width of 1.5 (Side lobe attenuation of 40 dB). The transition
(e) Kaiser window width of 1.5 (Side lobe attenuation of 40 dB). The transition
41
2.4.2 COMPUTATIONAL COMPLEXITY OF FIR BAND PASS

FILTERING
Using digital band pass filtering is a very successful method for removing the side-
lobes generated by OFDM signals. The problem with digital filtering is the high
computational overhead. To implement an FIR band pass filter the number of taps
required corresponds to:
 W ⋅ IFFT 
Ntaps = ceil  T  (2-5)
 FT 
where Ntaps is the number of taps in the FIR filter, WT is the transition width of the
window function used to generate the FIR filter (see section A.2 in the appendix for a
description of window functions), IFFT is the size of the IFFT used to generate the
signal, and FT is the transition width of the filter normalised for subcarrier spacings.
Ceil is a rounding toward infinity, i.e. ceil(1.1) = 2.
For example, to generate the signal shown in Figure 2-18 (b) requires filtering with a
24 tap FIR filter. This can be calculated from the signal specifications. The signal
was generated using an IFFT size of 64 and so IFFT = 64. A Kaiser window function
with a transition width of 3 was used. This results in a stop band attenuation of 89 dB
(see Figure A - 5 in the appendix). The side-lobe power of the unfiltered OFDM
signal is –20 dBc and so after filtering it should be –109 dBc. This matches the
results shown in Figure 2-18 (b). The transition width of the window function used
was 3.0 and so the number of taps would be:
3.0 × 64 
N taps = ceil   = 24 (2-6)
 8 
Each tap of the FIR filter requires two Multiply and ACcumulate operations (MAC)
as a result of the complex samples, and so for a sample rate of 20 MHz the number of
calculations would be 20×106 ×24×2 = 960 MMAC (Million MACs).
In applications where the required number of taps in the filter is high (> 100), it is
probably more efficient to implement it using an FFT implementation of an FIR
filter. Another method for reducing the number of calculations is to implement the
filtering using an IIR filter, however a review of the amount of ISI caused by the
42
non-linear phase of the filter would need further investigation.
2.4.3 EFFECT OF BAND PASS FILTERING ON OFDM

PERFORMANCE
In the time domain, an OFDM symbol is rectangular in shape, which corresponds to
a sinc decay in the frequency domain, as shown in Figure 2-16. If we were to band
pass filter an OFDM signal with a brick wall filter then the signal would become
rectangular in the frequency domain, causing the time domain waveform to have a
sinc decay between symbols. This in turn results in ISI degrading the performance.
The ISI caused by the filtering can be removed by using a guard period of sufficient
length, and by choosing the time offset to synchronise in the middle of the guard
period, so that most of the ISI energy is removed.
Figure 2-19 shows the simulated performance of a band pass filtered OFDM signal,
with different transition widths for the filter, in a channel with no channel noise. This
plot shows the performance of the OFDM transmission when the time
synchronisation offset was varied. The guard period used in this simulation was of
the same length as the IFFT section of the symbol. This very long guard period was
used so that the effect of the time offset could be varied over a large range, while still
maintaining a time offset within the guard period. The effective SNR was calculated
by averaging effective SNR over all the subcarriers in the transmission. When the
time offset is 0 this corresponds to the receiver taking the FFT of the IFFT section of
the transmitted signal. When the time offset is negative this corresponds to the
receiver taking the FFT over the IFFT section and part of the symbol guard period
(see Figure 2-20).
The lowest ISI is achieved when the time offset is negative and half the guard period
length. The sharper the filter cuts off the signal (in the figure the sharpest filter
removes the side-lobes down to below –100 dBc within 2 carrier spacings), the
longer the ISI. The residual capping of the SNR at 85 dB is caused by ICI, and
distortion of the subcarriers at the edges of the signal.
43
Filter transition in carrier spacing (Window Function Width:3)

90
80
Effective SNR (dB)

70
60
50
40
2 carr. cutoff
30 4 carr. cutoff
8 carr. cutoff
20
-0.5 -0.45 -0.4 -0.35 -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0
Time Offset (Fraction of Symbol)
Figure 2-19, Effective SNR as a function of the time offset for a band pass
filtered 52 subcarrier OFDM signal. (script s0057)
The guard period in this test was 50% of total symbol time, thus guard period
length = useful symbol time. Carr. cutoff corresponds to the transition width of
the filter in subcarrier spacings.
Symbol N Symbol N+1
Guard Guard
IFFT Period IFFT output Period IFFT
Neg. time offset
Receiver FFT
Time offset:0
Receiver FFT
Positive time offset

Receiver FFT
Figure 2-20, Section of the waveform that the receiver FFT is taken from
depending on the time offset.
The effective SNR of a band pass filtered OFDM signal depends on the effects of
both ISI and ICI. Figure 2-21 shows the performance of an OFDM system matching
the specifications of HiperLAN2 or IEE802.11a. In this case 52 carriers are used, and
the guard period is 20% of the total symbol time. The effective SNR varies with
subcarrier number as the filter distorts the response of the outer subcarriers the most.
The highest modulation scheme used in HiperLAN2 and IEEE802.11a systems is 64-
44
QAM, which requires an effective SNR of greater than 26 dB. We can see from the
results in Figure 2-21 that the effective SNR exceeds 26 dB for all carriers even
when using a very sharp band pass filter that cuts off within one half of a subcarrier
spacing.
53 subcarriers, Guard 20%, Wind. Width 1.5, Filter Atten: 40 dB

100
95
90
85
80
Effective SNR (dB)
75
70
65
60
(a) 55
50
45 0.5 carr. cutoff
40 3 carr. cutoff
35 4 carr. cutoff
30 5 carr. cutoff
25 6 carr. cutoff
20
5 10 15 20 25 30 35 40 45 50
Subcarrier number from edge of signal
53 subcarriers, Guard 20%, Wind. Width 3, Filter Atten: 89 dB

85
80
75
70
Effective SNR (dB)
65
60
55
50
(b) 45
40
35
0.5 carr. cutoff
30 4 carr. cutoff
25 6 carr. cutoff
20 8 carr. cutoff
15 10 carr. cutoff
10
5 10 15 20 25 30 35 40 45 50
Figure 2-21, Effective SNR for each subcarrier as a result of band pass filtering.
(script s0057)
An odd number of subcarriers (53) were simulated to make the signal
symmetrical about DC. The band pass filtering was implemented using an FIR
filter based one the windowing method using a Kaiser window. Each result
shows five different sharpness of the filter (carr. cutoff).
(a) Kaiser window width of 1.5 (Side lobe attenuation of 40 dB).

(b) Kaiser window width of 3 (Side lobe attenuation of 89 dB).
45
Signals with a larger number of carriers than the one shown in Figure 2-21, still have
the effective SNR limited by the band pass filtering. Figure 2-22 shows the effective
SNR for a signal with many subcarriers. Only the lower section is shown, as the SNR
is symmetrical about the centre subcarrier. These results can be used to estimate the
subcarrier effective SNR for any sized OFDM signal. In this simulation the only
source of noise is the ISI caused by the band pass filtering of the signal.
For example: let us assume that we wanted to know the effective SNR of the 26th
subcarrier from the edge of a HiperLAN2 signal when it is band pass filtered using a
transition width of 6 subcarriers and we use a Kaiser window with a window width
of 3, and a guard period of 20% is used. From Figure 2-22 (g) we can estimate the
effective SNR to be 64 dB. In a HiperLAN2 signal the 26th carrier corresponds to the
middle of the signal bandwidth and so it suffers from degradation from the filtering
of both edges of the signal in the frequency domain. This degrades the effective SNR
of the 26th subcarrier by 3 dB, thus the resulting in an effective SNR of 61 dB. This
matches the direct simulated results shown in Figure 2-21 (b).
We can see that for each halving of the guard period length the transition width of
the band pass filter must be approximately doubled in order to maintain the same
effective SNR. Subcarriers more than 10 subcarrier spacings from the signal
boundary have an effective SNR of greater than 30 dB regardless of the sharpness of
the filtering provided some guard period is present. This effective SNR is sufficiently
high to support a modulation scheme up to 128-QAM with a low error rate (see
Figure 2-31), providing that there are no other detrimental effects. For subcarriers on
the edge of the OFDM signal, their effective SNR can be as low as 20 dB, which
may pose a small problem for modulation schemes above 32-QAM.
Band pass filtering of OFDM signals allows the side-lobes to be removed from an
OFDM signal, effectively reducing its bandwidth, and improving the spectral
efficiency. It does this with little or no extra overheads in the time domain, as the
guard period is needed for protection from multipath delay spread.
46
Guard 0%, Wind. Width 1.5, Filter Atten: 40 dB
90
1 carr. cutoff
80 24 carr. cutoff
32 carr. cutoff
70
Effective SNR (dB) 48 carr. cutoff
60 64 carr. cutoff
50
40
30
20
10
0
0 1 2
10 10 10
(a) (b)
90
80
70
Effective SNR (dB)
60
50
40
30 1 carr. cutoff
12 carr. cutoff
20
16 carr. cutoff
10 24 carr. cutoff
32 carr. cutoff
0
0 1 2
10 10 10
(c) (d)
90
80
70
Effective SNR (dB)
60
50
40
30 1 carr. cutoff
6 carr. cutoff
20
8 carr. cutoff
10 12 carr. cutoff
16 carr. cutoff
0
0 1 2
10 10 10
(e) (f)
90
80
70
Effective SNR (dB)
60
50
40
30 1 carr. cutoff
3 carr. cutoff
20
4 carr. cutoff
10 6 carr. cutoff
8 carr. cutoff
0
0 1 2
10 10 10
(g) (h)
Figure 2-22, Subcarrier effective SNR, due to band pass filtering. (script s0085)
An OFDM signal with a large number of subcarriers (701) was simulated and
only the lower 100 subcarriers is show. The guard period is specified as a
percentage of the total symbol time (IFFT and the guard period)
47
2.5 RAISED COSINE GUARD PERIOD
One of the simplest methods for suppressing the side-lobes of an OFDM signal is to
round the guard period of the OFDM signal, tapering it smoothly to zero before the
next symbol. This tapering smooths the transition between symbols, resulting in
reduced side-lobe power. Figure 2-23 shows the make up of a single OFDM symbol
with a Raised Cosine (RC) guard period. This section of the guard period is
windowed with a squared cosine shape (cos(θ)2 ), hence the name raised cosine.
Flat Copy and window

Raised Guard
Raised
Cosine Period
Cosine
Guard
Guard
Period
FFT section Period
Copy and window
Figure 2-23, Construction of a RC guard period.
The raised cosine section of a guard period can be overlapped with the previous and
next symbol as this section of the guard period only provides minimal protection
against multipath and timing errors, and is ignored at the receiver. Because this
section tapers to zero it results in minimal additional ISI. The main advantage of
overlapping is that the length of the raise cosine section can be made double in length
without incurring additional time overhead. Figure 2-24 shows a diagram of
overlapping symbols.
Raised Cosine Guard Period can

overlap with next symbol
Symbol X Symbol X+1
TGF TFFT TGRC
Figure 2-24, Envelope of OFDM symbols with a flat guard period and an
overlapping raised cosine guard period.
48
The effect of adding a RC guard period to an OFDM signal was simulated to

determine the level of the out of band side-lobes. Figure 2-25, Figure 2-26 and
Figure 2-27 shows the spectrum of OFDM signals with an RC guard period. The
spectrum shown has been shifted to the left so that just the upper edge of the OFDM
signal is shown. The results were presented in this manner to normalise them so that
the side-lobes could be compared regardless of the number of subcarriers used in the
signal. In Figure 2-25 the lower edge of the OFDM signal can also be seen. The
frequency axis has been normalised to units of subcarrier spacings.
The spectrum of an OFDM signal only varies slightly as the number of subcarriers is
increased. The main difference occurs for the side-lobe level when little or no RC
guard period is used. It can be seen in the simulations that as the number of
subcarriers is increased from 20 (Figure 2-25) to 4000 (Figure 2-27) the level of the
side-lobes, 200 subcarrier spacings from the edge of the signal, increases by about 8
dB. The increase is a result of the combined effect of all the subcarriers in the signal.
In Figure 2-25 through to Figure 2-27 the RC guard period length has been specified
as a percentage of the flat section of the OFDM symbol, that is:
TGRC
RC = 100 ⋅ % (2-7)
TFFT + TGF
Where RC is the raised cosine percentage, TGRC is the length of the RC guard period,
TFFT is the length of the FFT section of the symbol and TGF is the length of the flat
guard period, see Figure 2-24. For example: The IEEE802.11a standard recommends
a RC guard period of 100 ns. The useful symbol period (TFFT) is 3.2 ìs and the total
guard period is 800 ns. The flat section of the guard period is the total guard period
(800 ns) minus the RC section of the symbol (100 ns), thus TGF = 700 ns. The RC
section is thus:
100
RC = 100 ⋅ = 2.56% (2-8)
700 + 3200
From Figure 2-25 and Figure 2-26 we can estimate that the side-lobes will be –40
dBc approximately 30 subcarriers from the edge of the signal. Since HiperLAN2
uses 52 signal subcarriers (spaced at intervals of 312.5 kHz) the total system
bandwidth using a threshold of –40 dBc is (2×30+52)×312.5 kHz = 35 MHz. The
49
spacing of the HiperLAN2 channels is 20 MHz, and so a bandwidth of 35 MHz is too

high. This indicates that the addition of the RC guard period is insufficient to reduce
the side-lobes sufficiently, thus additional band pass filtering is required.
OFDM Spectrum, 20 subcarriers
0 No RC
RC: 0.2%
-10
RC: 0.4%
-20 RC: 0.8%
RC: 1.6%
-30
RC: 3.1%
-40 RC: 6.3%
RC: 12.5%
Power (dB)
-50
RC: 25.0%
-60
-70
-80
-90
-100
-110
-120
-130
0 40 80 120 160 200 240 280 320
Frequency wrt. edge subcarrier (Subcarrier Spacings)
Figure 2-25, Side-lobe power for an OFDM signal with 20 subcarriers as the
length of the RC guard period is varied. (script s0050)
0 No RC
RC: 0.2%
-10
RC: 0.4%
-20 RC: 0.8%
RC: 1.6%
-30
RC: 3.1%
-40 RC: 6.3%
RC: 12.5%
Power (dB)
-50
RC: 25.0%
-60
-70
-80
-90
-100
-110
-120
-130
0 40 80 120 160 200 240 280 320
50
0 No RC
RC: 0.2%
-10
RC: 0.4%
-20 RC: 0.8%
RC: 1.6%
-30
RC: 3.1%
-40 RC: 6.3%
RC: 12.5%
Power (dB)
-50
RC: 25.0%
-60
-70
-80
-90
-100
-110
-120
-130
0 40 80 120 160 200 240 280 320
2.6 EFFECT OF ADDITIVE WHITE GAUSSIAN NOISE

ON OFDM
Noise exists in all communications systems operating over an analog physical

channel, such as radio. The main sources are thermal background noise, electrical
noise in the receiver amplifiers, and inter-cellular interference. In addition to this
noise can also be generated internally to the communications system as a result of
Inter-Symbol Interference (ISI), Inter-Carrier Interference (ICI), and Inter-
Modulation Distortion (IMD). These sources of noise decrease the Signal to Noise
Ratio (SNR), ultimately limiting the spectral efficiency of the system. Noise, in all its
forms, is the main detrimental effect in most radio communication systems. It is
therefore important to study the effects of noise on the communications error rate and
some of the trade offs that exists between the level of noise and system spectral
efficiency.
Most types of noise present in radio communication systems can be modelled

accurately using Additive White Gaussian Noise (AWGN). This noise has a uniform
spectral density (making it white), and a Gaussian distribution in amplitude (this is
also referred to as a normal distribution or bell curve). Thermal and electrical noise
51
from amplification, primarily have white Gaussian noise properties, allowing them to
be modelled accurately with AWGN. Also most other noise sources have AWGN
properties due to the transmission being OFDM. OFDM signals have a flat spectral
density and a Gaussian amplitude distribution provided that the number of carriers is
large (greater than about 20 subcarriers), because of this the inter-cellular
interference from other OFDM systems have AWGN properties. For the same reason
ICI, ISI, and IMD also have AWGN properties for OFDM signals.
2.6.1 MODULATION SCHEMES

Digital data is transferred in an OFDM link by using a modulation scheme on each
subcarrier. A modulation scheme is a mapping of data words to a real (In phase) and
imaginary (Quadrature) constellation, also known as an IQ constellation. For
example 256-QAM (Quadrature Amplitude Modulation) has 256 IQ points in the
constellation (see Table 2-2(h)), constructed in a square with 16 evenly spaced
columns in the real axis and 16 rows in the imaginary axis. The number of bits that
can be transferred using a single symbol corresponds to log2 (M), where M is the
number of points in the constellation, thus 256-QAM transfers 8 bits per symbol.
Each data word is mapped to one unique IQ location in the constellation. The
resulting complex vector I + j ⋅ Q , corresponds to an amplitude of I 2 + Q 2 and a
phase of ∠( I + j ⋅ Q) where j = − 1 .
Increasing the number of points in the constellation does not change the bandwidth
of the transmission, thus using a modulation scheme with a large number of
constellation points, allows for improved spectral efficiency. For example 256-QAM
has a spectral efficiency of 8 b/s/Hz, compared with only 1 b/s/Hz for BPSK.
However, the greater the number of points in the modulation constellation, the harder
they are to resolve at the receiver. As the IQ locations become spaced closer
together, it only requires a small amount of noise to cause errors in the transmission.
This results in a direct trade off between noise tolerance and the spectral efficiency
of the modulation scheme and was summarised by Shannon's Information Theory
[53], which states that the maximum capacity of a channel of bandwidth W, with a
signal power of S, perturbed by white noise of average power N, is given by
52
C = W log 2 1 + 
S
(2-9)
 N
The spectral efficiency of a channel is a measure of the number of bits transferred per
second for each Hz of bandwidth and thus the spectral efficiency SE is given by
= log 2 1 + 
C S
SE = (2-10)
W  N
where both the signal and noise is linear scale, and the spectral efficiency is
measured in b/s/Hz. If the SNR is significantly higher than one then each doubling of
the signal power (a 3 dB increase) the ideal spectral efficiency increases by 1 b/s/Hz
2.6.2 OFDM VERSES SINGLE CARRIER TRANSMISSION

The BER of an OFDM system is dependent on several factors, such as the
modulation scheme used, the amount of multipath, and the level of noise in the
signal. However if we look at the performance of OFDM with just AWGN then the
performance of OFDM is exactly the same as that of a single carrier coherent
transmission using the same modulation scheme.
If we look at just a single OFDM subcarrier (since the subcarriers are orthogonal to
each other, this does not effect the performance in any way) then this is exactly the
same as a single carrier transmission that is quadrature modulated with no band pass
filtering. The transmitted amplitude and phase is held constant over the period of the
symbol and is set based on the modulation scheme and the transmitted data. This
transmitted vector is then updated at the start of each symbol. This results in a sinc
frequency response, which is the required response for OFDM.
The optimal receiver for such a single carrier transmission is to use a coherent
matched receiver, which can be implemented by mixing the signal to DC using an IQ
mixer. This results in an IQ output that describes the amplitude and phase of the
received modulated carrier. The amplitude and phase of the transmitted signal is
constant over the symbol period, and so the optimal method of removing the most
noise from the signal is to use an integrate-and-dump filter. This filter averages the
received IQ vector over the entire symbol, then performs IQ demodulation on the
average.
53
The demodulation of an OFDM signal is performed in exactly the same manner. In

the receiver a FFT is used to estimate the amplitude and phase of each subcarrier.
The FFT operation is exactly equivalent to IQ mixing each of the subcarriers to DC
then applying an integrate-and-dump over the number of samples in the FFT. From
this we can see that the FFT performs the same operation as the matched receiver for
the single carrier transmission, except now for a bank of subcarriers.
From this we can conclude that in AWGN, OFDM will have the same performance
as a single carrier transmission with no band limiting.
However, most propagation environments suffer from the effects of multipath

propagation. For a given fixed transmission bandwidth, the symbol rate for a single
carrier transmission is very high, where as for an OFDM signal it is N times lower,
where N is the number of subcarriers used. This lower symbol rate results in a
lowering of the ISI. In addition to lowering of the symbol rate, OFDM systems can
also use a guard period at the start of each symbol. This guard period removes any
ISI shorter than its length. If the guard period is sufficiently long, then all the ISI can
be removed.
Multipath propagation results in frequency selective fading (see section 3.6 for more
details) that leads to fading of individual subcarriers. Most OFDM systems use
Forward Error Correction to compensate for the subcarriers that suffer from severe
fading. The adaptive modulation scheme proposed in section 4.2 matches the
modulation scheme of each subcarrier to its SNR. The additional spectral efficiency
of those subcarriers that have a SNR greater than the average (due to constructive
interference) tends to compensate for subcarriers that are subjected to fading
(destructive interference). As a result of this the performance of such an OFDM
system in a multipath environment is similar to its performance in an AWGN
channel. The performance of the OFDM system will be primarily determined by the
noise seen at the receiver. However, the performance of a single carrier transmission
will degrade rapidly in the presence of multipath.
54
2.6.3 MODULATION LIMITATIONS OF SYSTEMS

Most current mobile communication systems, specifically GSM, IS-95, and 3rd
Generation Systems, only use modulation schemes with a high noise tolerance, such
as BPSK, QPSK or similar. This results in a low spectral efficiency, but gives
improved robustness. These systems use fixed modulation schemes due to the
problems with obtaining a high SNR.
The symbol rate of single carrier systems has to be high if they are to obtain a high
bit rate, and as a result, systems such as GSM require complex equalisation (up to 4
symbol periods) to cope with multipath propagation. GSM systems are designed to
cope with a maximum delay spread of 15 µs, which corresponds to the typical delay
spread experienced at a transmission distance of 30 - 35 km. The symbol rate for
GSM is 270 kHz corresponding to a symbol period of 3.7 µs, thus ISI caused by the
multipath spans over 4 symbol periods. This would normally completely destroy the
transmitted information, but is recovered in practice by using complex adaptive
equalisation. Although this works for robust modulation schemes such as Gaussian
Minimum Shift Keying (GMSK) [22], [23] as used in the GSM system, it is difficult
to successfully apply to higher modulation schemes, as the residual errors in the
equalisation will cause a high error rate.
In DS-CDMA systems the problem is not primarily limited by multipath, but instead
inter-user interference. DS-CDMA systems utilise the fact that by spreading the user
information over a wide bandwidth it allows multiple users to transmit at the same
frequency [26], [27]. Each of these users spread the information signal by
multiplying it by a unique higher speed Pseudo Random Sequence (PRS). At the
receiver the signal from each user is extracted by multiply the by the same PRS and
integrating over the period of an information symbol. This process is however non-
orthogonal in the reverse link, resulting in users appearing as noise to each other. The
system capacity is maximised when the number of users is maximised, resulting in
very high levels of noise. This results in the system typically operating at an Energy
per Bit to Noise Ratio (EBNR) of around 5 - 8 dB after demodulation. This rules out
the use of high spectral efficiency modulation schemes since the SNR is too low.
55
OFDM on the other hand, minimises both of these effects. Multipath is minimised by
using a low symbol rate and the use of a guard period. Equalisation of the channel
can be easily achieved through the use of pilot symbols and or pilot tones. This type
of equalisation is accurate and results in minimal residual error, thus allowing a high
average SNR. Additionally, users in OFDM are kept orthogonal to each other, by use
of time division multiplexing or synchronised frequency division multiplexing,
minimising inter-user interference. Both these advantages mean that a high effective
channel SNR can be maintained even in a multiuser, multipath environment. This
potential for a high SNR means that high modulation schemes can be used in OFDM
systems, allowing for improved system spectral efficiency.
Additionally each subcarrier can be allocated a different modulation scheme based

on the measured channel conditions. These measurements can be easily obtained as
part of the channel equalisation step, allowing subcarriers to be dynamically
allocated modulation schemes based on the SNR of each subcarrier. These variations
in SNR arise due to interference, transmission distance, frequency selective fading,
etc. This technique is known as adaptive modulation and is presented in section 4.2.
Those subcarriers with a low SNR can be allocated to use BPSK (1 b/s/Hz) or to
transmit no data at all. Subcarriers with a high SNR can transmit higher modulation
schemes such as 256-QAM (8 b/s/Hz) allowing a higher system throughput. The
modulation allocation is flexible in OFDM systems allowing them to be optimised to
local current conditions, rather than having to always use a low modulation scheme
just to ensure the system operates during worst-case conditions.
2.6.4 SIMULATION SET UP

The effect of AWGN on OFDM was simulated for a wide range of subcarrier
modulation schemes. The IQ diagrams are shown in Table 2-2. The lower density
modulation schemes show the data mapping used.
The results presented show the BER performance as a function of the channel SNR.
Other simulations in this thesis, present the performance of OFDM under a range of
detrimental effects. These simulations measure the performance by finding the
effective SNR of the channel instead of the BER. These simulations show the
communication performance that corresponds to an equivalent AWGN SNR. The
56
results presented in this section can then be used to predict the BER for the particular
modulation scheme used.
The symbol error rate of most common modulation schemes has been derived in
algebraic form [117]. However, the derivation of the BER is difficult due to
possibility of multiple bit errors per symbol. Additionally the BER for only square
QAM modulation schemes, such as 16-QAM, 64-QAM, 256-QAM, etc, can be
calculated directly. To overcome this problem the BER performance of OFDM was
obtained by using simulation. There are four main categories of modulation
presented, which are, coherent QAM, coherent PSK, differential QAM, and
differential PSK.
2.6.5 GRAY CODING

The IQ plot for a modulation scheme shows the transmitted vector for all data word
combinations. Each data word combination must be allocated a unique IQ vector.
Gray coding is a method for this allocation so that neighbouring points in the
constellation only differ by a single bit. This coding helps to minimise the overall bit
error rate as it reduces the chance of multiple bit errors occurring from a single
symbol error. Figure 2-28 shows an example of gray coding for 16-PSK. Equation
(2-11) shows the sequencing for gray coding in decimal format.
Gray coding can be used for all PSK modulation schemes (QPSK, 8-PSK, 16-PSK,
etc), and square QAM (16-QAM, 64-QAM, 256-QAM, etc). For square QAM each
axis is mapped separately using gray coding.
G1 = {0, 1}
G2 = {0, 1, 3, 2}
G3 = {0, 1, 3, 2, 6, 7, 5 4} (2-11)
M
{ ( ) ( )
GN +1 = GN 1, 2, 3, K , 2 N , GN 2 N , 2 N − 1, 2 N − 2, K, 1 + 2 N }
57
Decimal Gray Coding Decimal Gray Coding

0 0, 0, 0, 0 8 1, 1, 0, 0
1 0, 0, 0, 1 9 1, 1, 0, 1
2 0, 0, 1, 1 10 1, 1, 1, 1
3 0, 0, 1, 0 11 1, 1, 1, 0
4 0, 1, 1, 0 12 1, 0, 1, 0
5 0, 1, 1, 1 13 1, 0, 1, 1
6 0, 1, 0, 1 14 1, 0, 0, 1
7 0, 1, 0, 0 15 1, 0, 0, 0
Table 2-1, 4 bit Gray coding in binary
0,1,1,0 Imaginary
0,1,1,1 0,0,1,0
0,1,0,1 0,0,1,1
0,1,0,0 0,0,0,1
1,1,0,0 0,0,0,0
Real
1,1,0,1
1,0,0,0
1,1,1,1 1,0,0,1
1,1,1,0 1,0,1,1
1,0,1,0
Figure 2-28, IQ plot of 16-PSK using gray coding. Note that each successive IQ
location changes by only a single bit.
58
I-Q diagram for BPSK I-Q diagram for QPSK
1 1 01
0.5 0.5
Imaginary
Imaginary
0 1 0 0 11 00
-0.5 -0.5
-1 -1 10
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1

Real Real
(a) (b)
I-Q diagram for 8QAM I-Q diagram for 16QAM
1
101 1000 1001 1011 1010
0.5
0.5 011 001 1100 1101 1111 1110
Imaginary
Imaginary
0 111 100 0
010 000 0100 0101 0111 0110

-0.5
-0.5
-1
110 0000 0001 0011 0010

-1
-1 -0.5 0 0.5 1 1.5 -1 -0.5 0 0.5 1
Real Real
(c) (d)
11100 11101 11110 11111 1

1
10110 10111 11000 11001 11010 11011

0.5
0.5
Imaginary
Imaginary
10000 10001 10010 10011 10100 10101
0 0
01010 01011 01100 01101 01110 01111
-0.5
-0.5
00100 00101 00110 00111 01000 01001
-1
00000 00001 00010 00011 -1
-1 -0.5 0 0.5 1
-1 -0.5 0 0.5 1
Real
Real
(e) (f)
Table 2-2, IQ diagrams for modulations used in the OFDM simulations. (script
s0045)
59
1
1
0.5 0.5
Imaginary
Imaginary
0 0
-0.5 -0.5
-1 -1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1

Real Real
(g) (h)
1
1
0.5 0.5
Imaginary
Imaginary
0 0
-0.5 -0.5
-1 -1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1

Real Real
(i) (k)
(l) (m)
Table 2-2, continued
60
I-Q diagram for 8PSK I-Q diagram for 16PSK
1 011 1 0110
0111 0010
010 001 0101 0011
0.5 0.5
0100 0001
Imaginary
Imaginary
0 110 000 0 1100 0000
1101 1000
-0.5 -0.5
111 100 1111 1001
1110 1011
-1 101 -1 1010
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1

Real Real
(n) (o)
I-Q diagram for 64PSK
I-Q diagram for 32PSK
1
1 01101 01100 00100
01111 00101
01110 00111
01010 00110
01011 00010 0.5

0.5
01001 00011
Imaginary
Imaginary
01000 00001
0 11000 00000 0
11001 10000
11011 10001
-0.5 11010 10011 -0.5

11110 10010
11111 10110
11101 10111
-1 11100 10100 10101
-1
-1 -0.5 0 0.5 1
-1 -0.5 0 0.5 1
Real
Real
(p) (q)
I-Q diagram for 128PSK I-Q diagram for 256PSK
1 1
0.5 0.5
Imaginary
Imaginary
0 0
-0.5 -0.5
-1 -1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
Real Real
(r) (s)
Table 2-2, continued.
61
2.6.6 COHERENT MODULATION

Coherent modulation is achieved by transmitting the IQ constellation data vectors
with absolute phase angles, i.e. if BPSK was used then 0° or 180° would be
transmitted. At the receiver it would compare the received phase to 0° or 180°. Phase
rotations and amplitude scaling (important for QAM), greatly increase the error rate,
or completely destroy all communications. This problem is however overcome by
using channel equalisation to remove this scaling of the channel before
demodulation. The phase rotation of the channel and the amplitude scaling is
measured using pilot symbols and pilot tones, which contain a known IQ
transmission vector. In a static channel with no movement, the response of the
channel will be constant, and thus once measured and corrected for, data can be sent
reliably. However in most applications radio channels are non-static. Frequency
selective fading cause complete fades in the spectrum approximately once every
wavelength of movement, causing the response of the channel to change rapidly
during movement. Tracking of the channel requires continual updates in the channel
equalisation, thus regular pilot symbols/tones must be inserted into the transmission.
The greater the number of pilot signals the faster the channel tracking rate, however
this also causes significant overhead. A more detailed analysis of the effectiveness of
pilot tones in channels effected by AWGN can be found in section 2.7, page 69.
2.6.7 DIFFERENTIAL PHASE MODULATION

Another common method for subcarrier modulation is to send the data differentially.
Instead of each symbol being independent of each other, the transmitted information
is sent as a difference between symbols vectors. Differential Phase Shift Keying
(DPSK) is the most common method of sending differential information. Instead of
mapping data to an absolute phase angle, as in the case of coherent modulation,
DPSK maps the data to a phase difference between symbols. The transmitted phase
corresponds to the cumulative sum of the phase differences. For example, for
differential QPSK each symbol transmits 2 bits of information, corresponding to 4
different phase differences. Table 2-2(b) shows the IQ diagram for coherent QPSK.
62
D-QPSK has the same IQ diagram except that each data combination corresponds to
a phase difference. The most obvious method for allocating word combinations to
phase differences, is to linearly map the binary word combinations to a linear phase
difference, as shown in Table 2-3. For example, if the data to be transmitted is {1,0
1,1 0,0 0,1} then the differential phase would be {180°, 270°, 0°, 90°}, thus if the
starting phase is 0° then the transmitted phase would be {180°, 180°+270° = 90°,
(90°+0°) = 90°, (90°+90°) = 180°}.
In a noisy channel phase errors can result in the received phase being closer to the
next or previous phase difference combinations, causing a symbol error. The number
of bits in error depends on the data word mapping. Linear mapping is not optimal as
a wrap around error from 270° to 0° causes a double bit error (1,1) to (0,0). By using
gray coding, the number of bit errors can be reduced by ensuring that the phase-
difference combinations that are closest to each other, only differ by a single bit in
the data word.
Data Word Phase Difference Phase Difference

(linear mapping) (gray coding)
0,0 0° 0°
0,1 90° 90°
1,0 180° 270°
1,1 270° 180°
Table 2-3, Phase mapping for differential QPSK.
This shows two options for allocating data word combinations to the transmitted
phase difference.
Differential modulation has the advantage of cancelling out channel phase rotations,
eliminating the need for additional channel equalisation. Additionally the phase
tracking of the channel is effectively updated at the symbol rate, thus tracking the
channel very quickly. Differential modulation is thus highly suited to mobile
communication. The disadvantage of differential modulation is the limited range of
modulation schemes, and that it requires about 3 dB higher SNR than coherent
modulation. The output symbol phase corresponds to the phase difference between
the present and previous symbols, and as a result the symbol noise is doubled
(degrading the performance by 3 dB) compared with the phase noise of a single
symbol (as used in coherent modulation).
63
2.6.8 DIFFERENTIAL QAM

Differential mapping can be applied to QAM modulation, with some limitations. For
differential modulation to work, the mapping of the data to modulation domain must
wrap. For example with differential PSK the transmitted phase is found by mapping
each data word to a phase from 0 to 2π, then integrating this mapping from symbol to
symbol. The phase wraps in a circular fashion as it is constrained to the range 0 to
2π. The receiver decodes the phase by taking the phase difference between symbols.
To achieve this type of mapping with a QAM scheme the input data can be split into
two data streams of N/2 bits each per symbol, where N is the number of bits per
symbol. The number of bits per symbol must be even and so this mapping can only
be done for square QAM mapping such as 16-QAM, 64-QAM, etc. Each N/2 bit data
stream is mapped to the (I) real and (Q) imaginary axes to form the resulting
transmitted vector. Because the signal is differential modulated each axis is modulo
integrated from symbol to symbol. Figure 2-29 shows an example of differential
QAM. If we examine just one axis, then if the data to be transmitted was: {1, 2, 0, 3,
1}, and we started with a reference of 0, then the differentially encoded signal can be
found by taking the cumulative sum of the data words {1, 3, 3, 6, 7} then modulo
wrapping it to the bounds 0 to 3, resulting in {1, 3, 3, 2, 3}. The receiver decodes this
by taking the difference, {1, 2, 0, -1, 1}, remembering that the reference at the start
was 0, then wrapping from 0 to 3, resulting in {1, 2, 0, 3, 1}.
Imaginary
0, 3 1, 3 2, 3 3, 3
0, 2 1, 2 2, 2 3, 2
Real
0, 1 1, 1 2, 1 3, 1
0, 0 1, 0 2, 0 3, 0
Figure 2-29, IQ data mapping for differential 16-QAM.
Data is split in two equal numbers of bits, then modulo integrated from symbol
to symbol. Each number represents 2 bits (0 - 3) with each pair being 4 bits.
64
Coherent Quadrature Amplitude Modulation
-1
10
-2
10
Bit Error Rate
-3
10
BPSK
10
-4
QPSK 8 16 32 64
-5
10
-6
10
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Energy per Bit to Noise Ratio (dB)
(a) BPSK, QPSK, 8-QAM, 16-QAM, 32-QAM, 64-QAM

Note: BPSK and QPSK have the same EBNR performance.
Coherent Quadrature Amplitude Modulation
-1
10
-2
10
Bit Error Rate
-3
10
-4 128 256 512 1024 2048 4096

10
-5
10
-6
10
10 12 14 16 18 20 22 24 26 28 30 32 34
(b) 128-QAM, 256-QAM, 512-QAM, 1024-QAM, 2048-QAM, 4096-QAM
Figure 2-30 part 1 of 3. Note: Modulation schemes with an order number of bits
per symbol (8-QAM, 32-QAM, 128-QAM, etc) don’t use gray coding, resulting
in a higher BER at a low EBNR.
65
Coherent Phase Shift Keyiing
-1
10
-2
10
Bit Error Rate
-3
10
10
-4
8 16 32 64 128 256
-5
10
-6
10
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
(c) 8-PSK, 16-PSK, 32-PSK, 64-PSK, 128-PSK, 256-PSK
Differential QAM
-1
10
-2
10
Bit Error Rate
-3 QPSK
10
-4 BPSK 16 64 256 1024 4096

10
-5
10
-6
10
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36
(d) D-BPSK, D-QPSK, D-16QAM, D-64QAM, D-256QAM, D-1024QAM, D-4096QAM

Figure 2-30 part 2 of 3
66
Differential Phase Shift Keyiing
-1
10
-2
10
Bit Error Rate
-3
10
-4 8 16 32 64 128 256
10
-5
10
-6
10
4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42
(f) D-8PSK, D-16 PSK, D-32 PSK, D-64 PSK, D-128 PSK, D-256 PSK
Figure 2-30, Bit Error Rate verses the Energy per Bit to Noise Ratio for a
selection of OFDM modulation schemes. (script s0041)
The number matching each line on the plot corresponds to the modulation
scheme. For example, for part (e) the numbers 8, 16, 32, etc correspond to D-
8PSK, D-16PSK, and D-32PSK, where D means differential.
2.6.9 SIMULATION RESULTS

The simulated performance for the modulation schemes tested is shown in Figure
2-30. These show the BER as a function of the Energy per Bit to Noise Ratio
(EBNR). This is a measure of the energy efficiency of a modulation scheme. If a
higher EBNR is needed to transfer data for a given modulation scheme, then it means
that more energy is required for each bit transfer. Low spectral efficiency modulation
schemes, such as BPSK and QPSK, require a lower EBNR, and hence are more
energy efficient. For a power limited system, with unbounded bandwidth, the
maximum data rate could be achieved using BPSK or QPSK. However, in most
applications the available bandwidth is the limiting factor and so the data rate is
maximised by using a more spectrally efficient modulation schemes such as 256-
67
QAM. The BER verses the SNR can be calculated from EBNR shown on the plots in
Figure 2-30. The SNR for each modulation takes into account the number of bits per
symbol, and so the signal power corresponds to the energy per bit times the number
of bits per symbol. In log scale the SNR for a given EBNR can be found with:
SNR = 10 ⋅ log 10 ( Nb ) + EBNRdB
where SNR is in dB, Nb is the number of bits per symbol for the modulation scheme
and EBNRdB is the EBNR in dB. For example, for 256-PSK the number of bits
transferred per symbol is 8 bits/symbol and thus the SNR is 10.log10 (8)+EBNR, thus
for an EBNR of 40 dB, the SNR is 49 dB.
Figure 2-31 shows a comparison between all of the modulation schemes simulated. It
shows the required SNR for a fixed BER of 1×10-5 . Coherent QAM performs best
requiring the least SNR, while differential PSK is the worst. Also shown is
Shannon’s limit, which represents the lowest possible SNR for a given spectral
efficiency over which zero error communications can occur. In order for a
communication system to approach Shannon’s limit powerful forward error
correction coding techniques must be used. For the BER shown coherent QAM is
approximately 7.5 dB worse than Shannon’s limit. For QAM the required SNR for a
fixed BER increases by approximately 3 dB for each additional 1 b/s/Hz in spectral
efficiency, which matches the same slope as Shannon’s limit. In comparison, for
PSK the required SNR increases by 6 dB for each addition 1 b/s/Hz resulting in the
capacity of PSK modulation techniques being approximately half that of QAM for
the same SNR. The low efficiency of PSK is a result of under utilisation of the IQ
vector space. PSK only uses the phase angle to convey information, with amplitude
being ignored. QAM uses both amplitude and phase for information transfer and so
is more efficient.
68
50
SNR required for BER < 1x10-5 ( d B )

45
40
35
30
25
20
15 Coherent QAM
Coherent PSK
10
Diff QAM
5 Diff PSK
Shannon's Limit
0
1 2 3 4 5 6 7 8 9 10 11 12
Modulation Spectral Efficiency (bits/Hz/sec)
Figure 2-31, SNR required to maintain a BER of < 1x10-5 , for all the simulated
modulation schemes. (script s0046)
Also shown is Shannon’s Limit. Note that the slope for QAM is 3 dB per
bit/Hz/sec, whereas for PSK it is 6 dB per bit/Hz/sec.
2.7 EFFECT OF PILOT SYMBOL CHANNEL

EQUALISATION IN COHERENT MODULATION
Coherent modulation schemes rely on channel equalisation to compensate for phase

rotations and amplitude scaling. Without equalisation, the error rate is excessively
large making communications impossible. The effectiveness of the channel
equalisation depends on the accuracy of the channel characterisation measurements.
However, these measurements are subjected to the same channel noise as the data
transmission, resulting in measurement noise. This noise can be minimised by
allocating more energy to the pilot tones / symbols, which can be achieved by
increasing the relative power of the pilot tones / symbols, or by averaging multiple
pilot symbols.
Averaging multiple pilot symbols allows the noise on the channel measurements to
be made arbitrarily small, by increasing the number of symbols averaged over.
However the greater the amount of averaging the larger the overhead used by the
channel characterisation. This limits the allowable characterisation rate for a fixed
69
amount of overhead, effectively determining the maximum velocity of movement

that the system can handle.
A better method for minimising the noise on the channel characterisation is to

increase the transmitted power for the pilot symbols. However care must be taken
when increasing the power of the pilot symbols, as it can result in excessive
distortion. This will lead to inter-modulation distortion products limiting the effective
SNR, defeating the original purpose. This problem can be minimised by using pilot
symbols that have a low Crest Factor (CF). See section 6.1 for information about the
CF of signals. The CF of random data words is very high, typically 8 - 10 dB, with a
worst case of 10log10 (N) where N is the number of subcarriers (see Figure 6-4).
Because of this the average transmitted power is usually 8 - 10 dB lower than the
peak output power of the transmitter to prevent clipping (see section 2.8). This can
be exploited as it allows a low CF pilot symbol to be boosted in power without
causing any clipping. Chapter 6 outlines several methods for generating low CF pilot
symbols, and shows that Genetic Algorithms (GA) can be used to produce pilot
symbols with a CF of less than 1 dB. This allows the pilot symbol to be boosted in
power by 3 - 6 dB, compared with data symbols. This will give the same
performance as sending 2 - 4 pilot symbols, with a symbol power equal to the data
symbols.
Table 2-4 shows the detrimental effects of channel measurement noise on the BER
for different modulation schemes. The noise in the channel characterisation means
that to achieve the same BER performance as with perfect characterisation, the SNR
must be higher. When only a single pilot symbol is used the noise performance is
about 3 dB worse, which is almost exactly the same as the performance loss when
using differential modulation. The SNR performance loss reduces almost
proportionally with the number of pilot tones, thus the loss when using 4 pilot tones
is 4 times lower (in dB). For example from Table 2-4(b) at a BER of 1x10-4 the
performance loss when using only a single pilot symbol is 2.4 dB, when four
symbols are used the loss reduces to 0.6 dB.
From the simulated results a single low CF pilot symbol with a boosted power of 6
dB would result in the performance loss due to imperfect channel characterisation of
0.1 dB to 0.9 dB depending on the modulation scheme.
70
Number of BER BER BER BER

Pilot Symbols 1 × 10 -2 1 × 10 -3 1 × 10 -4 1 × 10 -5
1 1.55 1.2 1 0.8
2 0.6 0.35 0.2 0.15
3 0.3 0.2 0.15 0.15
4 0.2 0.15 0.1 0.1
(a) BPSK

Pilot Symbols 1 × 10 -2 1 × 10 -3 1 × 10 -4 1 × 10 -5
1 2.5 2.4 2.4 2.35
2 1.3 1.3 1.2 1.2
3 0.85 0.8 0.8 0.7
4 0.65 0.6 0.6 0.45
8 0.35 0.3 0.3 0.2
(b) QPSK

Pilot Symbols 1 × 10 -2 1 × 10 -3 1 × 10 -4 1 × 10 -5
1 2.7 2.95 3.05 3.2
2 1.5 1.65 1.7 1.75
3 1.05 1.1 1.15 1.15
4 0.8 0.85 0.85 0.85
8 0.4 0.45 0.4 0.4
(c) 16-QAM

Pilot Symbols 1 × 10 -2 1 × 10 -3 1 × 10 -4 1 × 10 -5
1 2.85 3.15 3.4 3.6
2 1.6 1.7 1.9 2.05
3 1.1 1.2 1.3 1.4
4 0.85 0.9 0.95 1
8 0.45 0.45 0.5 0.45
(d) 64-QAM

Pilot Symbols 1 × 10 -2 1 × 10 -3 1 × 10 -4 1 × 10 -5
1 2.95 3.3 3.6 3.75
2 1.7 1.85 2.05 2.25
3 1.25 1.3 1.35 1.5
4 0.9 0.95 1.05 1.1
8 0.45 0.5 0.5 0.55
(e) 256-QAM
Table 2-4, SNR degradation in dB due to imperfect channel estimation.
Pilot symbols were used to perform channel estimation. These pilot symbols are
subject to channel noise and thus result in imperfect channel estimation and an
increase in the BER. The tables show degradation in SNR (i.e. need a higher
SNR for the same error rate) as compared with perfect channel estimation.
71
These results were obtained by simulating the OFDM transmission using

different number of reference symbols. The total number of bits simulated was
7×109 . (BER simulations were performed using script s0041_ofdmber, and the
tables generated using s0042_ref_degradation)
2.8 EFFECT OF DISTORTION ON OFDM
One of the problems with OFDM is that the signal has a high peak power compared
with its average power. When an RF carrier is modulated with an OFDM signal it
results in a similar variation in power of the carrier envelope. This results in the
requirement that the signal is amplified and transmitted in a linear way. It is very
difficult to maintain a high degree of linearity at high power levels, and so most of
the distortion in a radio transmission usually occurs in the power amplifier of the
transmitter. Some additional distortion can occur in the receiver if it is not designed
properly, but in general it is relatively easy to keep the level of distortion in the
receiver significantly lower than the transmitter. Distortion in the transmitter causes
the most problems in the transmission chain, as it can result in spectral spreading,
which can cause interference to neighbouring systems in RF frequency. For this
reason we will focus on the effects of distortion in the transmitter.
30
Fundamental, 2 tone signal

20
Power Spectrum Magnitude (dB)
10
IMD
2nd Harmonic
-10
IMD 3rd Harmonic
4th Harmonic
-20
-30
-40
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency
Figure 2-32, Effect of distortion on a 2 tone signal, showing harmonics and

IMD. (script s0049)
72
Non-linearities in the transmission result in two main distortion products, Inter-

Modulation Distortion (IMD) and harmonics. Figure 2-32 shows the effect of
clipping distortion on a 2-tone signal. OFDM signals are made up from a large
number of subcarriers resulting in many distortion products.
Harmonics result in frequency components at X times the RF carrier frequency,

where X is an integer. Thus if we have a 900 MHz RF carrier the harmonics will
occur at 1800 MHz, 2.7 GHz, etc. Harmonics can easily be removed using a
relatively simple low pass filter on the output of the transmitter.
IMD is much more of a problem as it results in distortion components, which are in-
band and out of band but close to the main transmission. These components are a
result of mixing between each of the harmonics of the system, and subsequent
mixing between the IMD products. In-band components result in added noise to the
OFDM signal at the receiver, effectively limiting the SNR of the system, even in the
absence of other sources of noise. Out of band components spread the signal in
bandwidth, resulting in potential interference with other radio communications in
neighbouring frequency bands. Even if the signal is perfectly band-limited before
going to the transmitter power amplifier, spectral spreading will occur if the power
amplifier is non-linear. Spectral spreading can be slightly reduced by using analog
band pass filters after the power amplifier, however these will generally not have a
sharp roll off because they are operating at the RF frequency.
The effect of distortion on OFDM transmissions has previously been studied by [54]-
[56]. In [54] the effect of three different amplifier distortion models on the
performance of DAB transmission is compared. The amplifier models compared are
a Solid State Power Amplifier (SSPA), a Travelling Wave Tube Amplifier (TWTA)
and a perfectly linearised amplifier. The results show that the optimal Output power
Back Off (OBO) (see next section for explanation of OBO) for DAB is
approximately 2 - 3 dB, with only a small difference of 0.6 dB due to the different
amplifier models. These results are however limited to OFDM transmissions using
QPSK, which is a very robust modulation scheme, and hence robust against effects
of distortion. Modulation schemes that have a higher spectral efficiency (such as 16-
QAM, 256-QAM, etc) are more susceptible to the effects of distortion due to the
requirement of a higher effective SNR. In [55] the performance of both a QPSK and
73
a 16-QAM OFDM system was investigated. The optimal OBO for QPSK
transmissions were found to be 3 dB in this study, which compare well with results
presented in [54]. For 16-QAM the optimal OBO was found to be higher,
approximately 6 dB. The performance of a 64-QAM modulated OFDM signal was
investigated in [56], which showed that the optimal OBO is 6 dB for clipping
distortion, and closer to 10 dB for a smooth limiter. Each of these papers show the
performance of a fixed system, limiting the usefulness of the results. A more general
study is presented showing the performance of an OFDM system as a function of
clipping distortion. The effect of distortion causes in-band noise due to IMD,
resulting in a lowering of the effective SNR of the channel. The results are presented
as an effective SNR of the OFDM channel, and so are independent of any particular
modulation scheme. The BER for a particular modulation scheme can be found from
the effective SNR by using the results shown in section 2.6.
2.8.1 DISTORTION MODELLING

Distortion results from any non-linearity in the system. It is also dependent on the
actual signal waveform and its power level. A signal with a high peak to average
power ratio will tend to have peaks which operate much closer to the maximum
range of the amplifier than say a sine wave, for the same average power. The amount
of distortion is dependent on the signal envelope distribution as well as shape of the
non-linearity of the amplifier. It is therefore difficult to compare the distortion
analysis of one signal type to another.
Figure 2-33 shows the input verses output for a real amplifier. This result shows that
the amplifier is not perfectly linear up to the clipping power, but tends to roll off as it
becomes saturated. Additionally it shows that the response is not symmetrical in that
the positive half of the output clips at a lower level than the negative side.
74
Figure 2-33, Measured Input verses Output for an MAR6 amplifie r, showing
distortion.
This amplifier is a class A MMIC RF amplifier. This is an inverting amplifier

resulting in the linear slope being negative. Also the signals are AC coupled,
resulting in the plot not passing through 0.
One way to reduce the effects of non-linearities in the transmitter power amplifier is
the use of predistortion [57]-[61]. This involves predistorting the signal before the
power amplifier in such a way as to cancel the distortion caused by the power
amplifier. This predistortion is typically done at base band by changing the amplitude
and phase of the time waveform. The most general form of predistortion uses feed-
back from the power amplifier output to achieve an accurate linearisation.
Predistortion can reduce distortion from non-linearities of the transmitter power

amplifiers. It however cannot prevent distortion from clipping of the signal at high
power levels, due to the limited peak power of any real amplifier.
Figure 2-34 shows the input verses output of the distortion model used for the
simulations. In fact this is a model of a perfectly linear, but power limited amplifier,
which is the best result that could be obtained. Distortion is a result of clipping of the
signal power.
75
Output
Input
Figure 2-34, Input verses Output for a power limited linear amplifier.
This is the distortion model used for the simulations.
One method for estimating the efficient use of the output power amplifier is the
Output power Back Off (OBO), which is defined as
Psat
OBO = 10 ⋅ log 10 (2-12)
Po
where Psat is the saturation power of the amplifier and Po is the average output signal
power. It is desirable to operate at a low OBO to allow high utilisation of the
available power. This however tends to result in excessive distortion degrading the
signal, and generating out of band spectral spreading.
The OBO is measured based on the average power of the signal after clipping,
making it difficult to directly determine the required amount of clipping to be applied
to the input signal. The amount of clipping required varies depending on the signal
amplitude distribution. To overcome this difficulty, an iterative solution was used to
determine this level to meet the required OBO.
The simulation results show the effect of clipping distortion on the performance of
OFDM. For real amplifiers the onset of distortion is not as abrupt as clipping, and so
the performance of the system will be degraded even before the signal power reaches
the clipping power. This must be taken into account when studying the results of the
simulations.
76
2.8.2 SPECTRAL SPREADING FROM DISTORTION CLIPPING

Figure 2-35 shows the spectrum of an OFDM signal that is clipped in power. The
OFDM signal was band pass limited using an FIR filter, before being subjected to
clipping distortion. This was done to remove the OFDM side lobes so that the
spectral spreading caused by the clipping could be easily seen. The out of band
interference increases as the OBO is reduced, resulting in significant out of band
noise for OBOs below 8 dB. It can also be seen that the out of band interference
slowly decays with frequency. Figure 2-36 shows the strength of the spectral
spreading as a function of the OBO for an OFDM signal with 100 subcarriers. This
shows that for an OBO of greater than 10 dB there is virtually no spectral spreading.
This plot shows two results, the first being the level of out of band noise 100
subcarrier-spacings away from the signal centre (100% of system bandwidth), and
the other 200 subcarriers away from the centre (200% of the system bandwidth).
Several other tests were done to find the effect of changing the number of
subcarriers, however it was found that the results were almost independent of the
number of system subcarriers.
The worst case CF for an OFDM signal is 10×log10 (N) where N is the number of
subcarriers, and so for a 100 subcarrier signal the worst case CF is 20 dB. This would
tend to suggest that significant distortion should occur for an OBO all the way up to
20 dB. This is not the case due to the CF distribution of OFDM signals. Even though
the worst case CF is very high, the probability of it occurring is incredibly small.
Figure 2-37 shows the CF distribution for OFDM signals. This shows that for signals
with 100 subcarriers, only 1% of the symbols have a CF greater than 10 dB. Even
these symbols will only have a very small fraction of the total symbol energy above
10 dB, and so clipping at 10 dB OBO has little effect.
77
OFDM spectrum verses Clipping Distortion
-10
-20
OB: 4dB
Power (dB)
-30
-40
OB: 6dB
-50
-60 OB: 8dB
-70
No Distortion
-80
-200 -100 0 100 200
Frequency (Carrier Spacings)
Figure 2-35, Spectrum of OFDM signal with clipping distortion. (script s0048)
This is for a 100 subcarrier OFDM signal, which has been frequency band
limited using an FIR (see section 2.4.1 ) before the distortion was applied.
-10
-20
-30
Side Lobe Power (dBc)
-40
-50
-60
-70
-80
55% BW from center
-90 100% BW from centre
200% BW from centre
-100
1 2 3 4 5 6 7 8 9 10 11
Output Power Backoff (dB)
Figure 2-36, Strength of the spectral spreading as a function of the Output

Power Backoff. (script s0048)
The distortion is caused by clipping of the signal. The spectrum of the OFDM
signal goes up to 50% of the signal bandwidth (BW) from centre and so the
55% BW from the centre result corresponds to the out of band energy just
outside the pass band.
78
(a) (b)
Figure 2-37, Crest Factor distribution verses number of subcarriers for an
OFDM signal. (script s0032)
(a) Random phase modulation. (b) Random QAM modulation.
2.8.3 EFFECTIVE SNR FROM CLIPPING DISTORTION

80
8 carriers
64 carriers
70
512 carriers
60
Effective SNR (dB)
50
40
30
20
10
0
0 1 2 3 4 5 6 7 8 9
Figure 2-38, Effective SNR of an OFDM transmission verses the Output Power
Backoff for 8, 64 and 512 transmission subcarriers. (script s0052)
Figure 2-38 shows the effective SNR of an OFDM transmission subjected to clipping
distortion. The number of carriers has relatively little effect on the results. The CF of
most OFDM signals is around 8 - 12 dB, and so we would expect clipping distortion
to start to have a detrimental effect on the signal for an OBO of 8 - 12 dB. Figure
2-37 shows the distribution of the CF for random OFDM signals as a function of the
number of subcarriers. This shows that the CF levels out at about 10 - 12 dB even for
79
large numbers of subcarriers. In addition to this, even if the CF is greater than the
OBO, the peaks in the signal are only a small fraction of the total symbol period and
thus clipping them result in only minor distortion to the overall symbol. This means
that the effect of clipping distortion is almost independent of the number of
subcarriers in the OFDM signal.
Figure 2-31 shows that for QPSK an effective SNR of 12 dB is needed to maintain a
BER lower than 1×10-5 . From Figure 2-38 an OBO of as low as 2.5 dB can be used
while still maintaining a sufficient effective SNR for QPSK. This matches closely
with published results in [54] and [55], which showed an optimal OBO for QPSK of
2 - 3 dB. For 64-QAM, Figure 2-31 shows that an effective SNR of 26 dB is needed,
which corresponds to a minimum OBO of 5.5 dB from Figure 2-38. This is slightly
lower than the result published in [56] of 6 dB. This slight difference is due to the
lower BER used and the increased number of subcarriers used in the simulations in
[56].
2.8.4 EFFECT OF PILOT SYMBOL TYPE ON THE BER WITH

DISTORTION
Coherent carrier modulation requires channel equalisation in order to work correctly.
If the pilot symbols used for this are effected by distortion or noise then the overall
system BER greatly increases. Pilot symbols consist of an OFDM symbol with a
known amplitude and phase, allowing channel characteristics to be calculated.
Preferably the phasing of the subcarriers should be chosen in such a way as to
minimise the pilot symbol CF. This will help to prevent distortion from effecting the
symbol. Additionally if a low CF symbol is used, its power can be boosted compared
with the data symbols without causing clipping. This helps to improve the accuracy
of the channel equalisation, minimising the BER in the presence of noise. (see
section 2.7).
80
AWGN SNR: 25 dB, 64 carriers
24
22
20
Effective SNR (dB)

18
16
14
12
10
8
6 (a) Boost: 0dB, CF:0.65dB
4 (b) Boost: 6dB, CF:2.7dB
(c) Boost: 6dB, CF:0.65dB
2
(d) Boost: 6dB, CF:8.2dB
0
1 2 3 4 5 6 7 8 9 10 11 12
Figure 2-39, Effective SNR verses output power backoff using different pilot
symbols. (script s0054)
(a) Genetic Algorithm Optimised phasing scheme (average power equal to data
power), (b) Quadratic Phasing Scheme [106], boosted by 6 dB in power, (c) GA
Optimised phasing scheme boosted by 6dB, (d) random phase angle symbol
boosted by 6 dB.
Figure 2-39 shows the effective SNR verses OBO for four different pilot symbol
schemes. As well as distortion, AWGN was added to assess the effectiveness of the
pilot symbol for estimating the channel response. The SNR of the simulation was set
to 25 dB. As the OBO is increased the effective SNR of the OFDM channel also
increases, approaching the upper limit of 25 dB. Figure 2-39 (a) shows the result for
a single pilot symbol that has a CF of 0.65 dB (see section 6.2 for more detail on the
phasing scheme used). From an OBO of 1 to 4 dB the effective SNR follows the
same performance as that simulated in Figure 2-38. Even though the CF of the pilot
symbol is 0.65 dB the data symbols are not, resulting in the degradation shown in the
simulation. Above 4 dB the effect of the AWGN begins to limit the effective SNR.
Above an OBO of 7 dB the effective SNR is capped at 22.5 dB. This is a degradation
of 2.5 dB compared with the ideal performance of 25 dB, and is a result of noise on
the channel characterisation, which is estimated from the pilot symbol.
Figure 2-39 (c) shows the performance with the same pilot symbol boosted by 6 dB.
The resulting cap on effective SNR is 24.1 dB, reducing the degradation to only 0.9
81
dB. This improvement in the degradation is due to the improved channel estimation
from the boosted pilot symbol power. The disadvantage of this boosted power is that
clipping of the pilot symbol occurs when the OBO is below 6 dB. This results in
rapid degradation of the effective SNR as the channel estimation is degraded.
Figure 2-39 (b) shows a similar result to (c) except that a different low CF pilot
symbol was used. See section 6.2.3 for details on generation of this pilot symbol.
The CF of this pilot symbol was 2.7 dB, making it 2.05 dB higher than the ultra low
CF used in (c). Despite the higher CF, the performance was only slightly worse by
approximately 0.8 dB, which is due to the small percentage of symbol power in the
peaks.
Figure 2-39 (d) shows the performance when a boosted random pilot symbol was
used. The CF of this symbol was 8.2 dB and so its performance was significantly
worse than the other pilot symbols used.
This simulation has shown that the pilot symbol power can be boosted above the
average power, provided that it has a low CF. This boosting of the power increases
the effectiveness of the pilot symbol, improving channel characterisation.
2.9 REDUCTION OF SPECTRAL SPREADING BY PRE-

FILTER CLIPPING
Clip to Bandpass Clip to specified

specifi ed Pre- filter user Output Power
From filter OBO FIR Amplifier OBO Simulated
OFDM Transmitted
modulator signal
Figure 2-40, Pre-filter clipping is to reduce spectral spreading due to clipping in

the power amplifier.
OFDM without any output filtering has a broad spectrum, as discussed in Section 2.4
page 37. One method for reducing this problem is to filter the output of the OFDM
transmission. This filtering can also be used to reduce the amount of output spectral
spreading caused by clipping in the output power amplifier.
Applying clipping to the OFDM signal in a digital form before the power amplifier,
82
allows the CF to be reduced, minimising the distortion and spectral spreading.

However, clipping the signal in the digital domain also causes spectral spreading,
eliminating the effectiveness of such a technique. However if the clipped signal is
band pass filtered before being sent to the power amplifier, then this will remove the
spectral spreading from the first clipping stage. The reduction in the CF reduces the
amount of subsequent clipping in the power amplifier, and consequently the spectral
spreading. This is a very simple method for reducing the CF of OFDM signals. The
main problem with this technique is that IMD caused by the clipping limits the
effective SNR, reducing the capacity of the system.
Numerous other techniques for crest factor minimisation have been published [108] -
[113]. Section Chapter 7 provides a description of these techniques and presents a
new method for CF minimisation of OFDM signals. This investigation into pre-filter
clipping is to show the effectiveness of such a simple technique for spectral
spreading minimisation.
If we were designing an OFDM system with 64 subcarriers and had decided that we
needed an effective SNR of 35 dB, then by looking at Figure 2-38 we can see that the
lowest OBO that can be used is 6.5 dB. At this OBO the spectral spreading due to
IMD, 64 subcarriers from the system centre frequency, is –36 dBc (see Figure 2-36,
100% system BW at OBO of 6.5 dB). However if we know that the power amplifier
in the transmitter is going to clip the waveform, then why not clip the signal before
sending the signal to the power amplifier. Most of the spectral spreading can then be
removed by digitally band-pass filtering the signal before transmission. This reduces
the CF of the signal, reducing the amount of subsequent clipping in the PA, and
spectral spreading in the broadcast signal. This eases the requirements on the analog
band-pass filter after the PA.
Figure 2-41 shows the effect of pre-filter clipping on the output spectrum. This
spectrum also shows the filtering of the OFDM signal. In this simulation the IFFT
size used to generate the signal was 128 points, the transition width of the filter was 8
subcarrier spacings and the window function used to generate the filter had a window
width of 2 (similar to hamming window function, see Appendix A). The resulting
filter required 128/8×2 = 32 taps. A guard period of 16 samples was used to remove
the ISI caused by the filtering.
83
When no pre-filter clipping is applied, the out of band side-lobes are exactly as
predicted by Figure 2-36. However as pre-filter clipping is applied to the signal the
side-lobe energy is reduced, with a 10 dB improvement for pre-filter clipping OBO
of 6 dB, and a 34 dB improvement with a pre-filter clipping OBO of 5 dB. Clearly by
clipping and filtering the signal before the PA can significantly reduce the broadcast
spectral spreading. The question is how much has the pre-filter clipping reduced the
effective SNR of the OFDM transmission.
Pre-filter clipping reduces the output spectral spreading, however it does not remove
the in-band IMD products. Additionally for the pre-filter clipping to be effective the
OBO used must be lower than the OBO used at the power amplifier. Typically it
must be 0.5 - 1.5 dB lower to have a significant effect. This consequently increases
the in-band distortion products reducing the effective SNR. Figure 2-43 shows the
resulting effective SNR of the OFDM transmission for different levels of pre-filter
clipping, and the OBO used in the power amplifier. We can see that for our example
system earlier, that if we used a per-filter clipping of 6 dB, and an OBO in the power
amplifier of 6.5 dB, that effective SNR drops to 31 dB. Thus we gain 10 dB side-lobe
suppression for a loss in effective SNR of 4 dB (35 dB to 31 dB). Alternatively we
can increase the pre-filter clipping OBO to 7 dB and the output power amplifier
OBO to 8 dB, the effective SNR increases by 3 dB to 38 dB, and the side-lobe
suppression increased by 37 dB to –77 dBc, all for the cost of only 1.5 dB reduced
output power. To achieve this level of out of band suppression without using pre-
filter clipping the power amplifier OBO would have needed to be increased to 9.7
dB, compared with 8 dB with the pre-filter clipping.
Pre-filter clipping can be used to trade off between the effective SNR of the
transmission and the out of band suppression. Its use in practice will depend on the
amount of distortion the system can tolerate. Also it can only be used in a system that
uses digital filtering before the power amplifier. In a multiuser OFDM system,
filtering removes the orthogonal nature between users which are using neighbouring
subcarriers in the spectrum, thus limiting its used to system wide filtering.
84
Figure 2-41, Spectrum of a 64 subcarrier OFDM signal with pre-filter clipping.

(script s0063)
The signal was clipped, digitally band pass filtered then re-clipped in the PA to
an OBO of 6.5 dB.
Figure 2-42, Side-lobe power caused by IMD as a result of clipping in the

power amplifier, for different levels of pre-filter clipping. (script s0063)
85
64 Carriers, Guard 1/8 IFFT size, Filt Width: 8, Wind Width: 2

55
Pre-Filter OBO: 4dB
50 Pre-Filter OBO: 5dB
Pre-Filter OBO: 6dB
45
Pre-Filter OBO: 7dB
Effective SNR (dB)

40 No Pre-Filtering
35
30
25
20
15
10
0 1 2 3 4 5 6 7 8 9 10 11
Figure 2-43, Effective SNR of the OFDM transmission as a function of the

OBO in the Power Amplifier, and the Pre-filter clipping used. (script s0062)
2.10 EFFECT OF TIME SYNCHRONISATION ERROR
OFDM is relatively tolerant against timing errors, due to the inclusion of the guard
period between symbols. For a channel with no multipath delay spread, the time
offset error can be as much as the length of the guard period with no loss of
orthogonality results, only a phase rotation in the subcarriers. The phase rotation is
corrected as part of the channel characterisation, hence resulting in no degradation in
performance. Time offset errors greater than the guard period result in a rapid loss in
performance, as the section of the symbol that the FFT is applied to will contain
some of the neighbouring symbol, leading to Inter-Symbol Interference.
Figure 2-44 shows the effective SNR of an OFDM transmission as a function of the
time offset error. The zero point in time is taken with respect to the FFT section of
the symbol. A positive time offset results in some of the next symbol being captured
in the FFT, thus lowering the effective SNR. A negative time offset results in part of
the guard period being used in the FFT. Since this is a cyclic extension of the symbol
no ISI occurs. In a dispersive channel the effective length of the guard period is
reduced by the delay spread of the channel resulting in a corresponding reduction in
the allowable time offset error.
Figure 2-45 shows the effect of using an overlapping raised cosine guard period (see
86
Figure 2-24) on the timing error tolerance of the system. In this simulation the raise
cosine section of the guard period was 50% of the total guard period time. This
shows that the raised cosine section of the guard period only provides partial
protection against time offsets, compared with a flat guard period. The effective
length of the raised cosine guard period is shorter than its actual length. For a cut off
of 30 dB effective SNR, the raised cosine guard period provides time offset tolerance
equivalent to one quarter (5 samples) its actual length (20 samples).
Guard Period: 40 samples, IFFT size:128

60
50
Effective SNR (dB)
40
30
20
10
-60 -40 -20 0 20

Time Offset Error (samples)
flat guard period of 40 samples. (script s0053)
The time origin is with respect to the start of the IFFT section of the symbol,
just after the guard period. A positive time error means the FFT in the receiver
is capturing some of the next symbol; a negative time error results in the
receiver getting the guard period.
87
Flat GP: 20, Raised Cos GP: 20, Overlap: 20, IFFT size:128
60
50
Effective SNR (dB)

40
30
20
10
-60 -50 -40 -30 -20 -10 0 10 20 30

Time Offset Error (samples)
raised cosine guard period. (script s0069)
The length of the flat guard period is 20 samples, with an overlapping raised
cosine guard period of 20 samples.
2.11 EFFECT OF FREQUENCY SYNCHRONISATION

ERROR
One significant problem with OFDM is its sensitivity to frequency offsets affecting
the performance. The demodulation of an OFDM signal with an offset in the
frequency can lead to a high bit error rate. This is caused by the loss of orthogonality
between the subcarriers resulting in inter-carrier interference (ICI), and a lack of
correction for phase rotation of the received data vectors.
Frequency errors will tend to occur from two main sources. These are local oscillator
errors and Doppler spread. Any difference between transmitter and receiver local
oscillators will result in a frequency offset. This offset is usually compensated for by
using frequency tracking, however any residual errors result in a degraded system
performance.
Movement of the transmitter or receiver results in Doppler shift in the signal. This
appears as a frequency offset for free space propagation. This offset is usually
corrected for as part of the local oscillator compensation. A much more serious
88
problem is that of Doppler spread, which is caused by movement of the transmitter or

receiver in a multipath environment. Doppler spread is caused by the different
relative velocity of each of the reflected multipath components, resulting in the signal
being Frequency Modulated. This FM modulation on the subcarriers tends to be
random due to the large number of multipath reflections that occur in typical
environments. This Doppler spread is typically poorly compensated for and results in
degradation of the signal.
Figure 2-46 shows the effect of frequency errors on the effective SNR of an OFDM
transmission using coherent QAM. Any frequency offset results in a continual phase
rotation of all the received subcarrier vectors. The greater the frequency offset, the
greater the phase rotation. If the channel characterisation is only performed at the
start of each frame then the unresolved frequency errors will lead to degradation in
the performance with time. The first symbol after the channel compensation will
have the maximum effective SNR, which will degrade to a low effective SNR by the
end of the frame. Figure 2-46 shows the effective SNR of the 1st , 4th , 16th , and 64th
symbol in an OFDM transmission with only channel compensation at the start of a
frame. The spikes in the response for a frequency offset error > 10-2 , are caused by
full rotations of the received vector.
The performance of differential phase modulation will be similar to that of the SNR
of the first symbol after characterisation, as phase rotations will be corrected for
every symbol.
89
Number symbols after pilot reference (Coherent QAM)

80
1
70 4
16
60
64
Effective SNR (dB)

50
40
30
20
10
-10
-5 -4 -3 -2 -1
10 10 10 10 10
Frequency Offset Error (carrier spacings)
Figure 2-46, Effective SNR for Coherent QAM with frequency offset error.
(script s0068)
The effective SNR for the 1st , 4th , 16th , and 64th , symbol in an OFDM frame with
channel equalisation, only at the start of the frame.
There have been numerous techniques developed to measure and track the frequency
offset [35] - [52]. Also it has been stated that the frequency accuracy must be
maintained to within 2 - 4% to prevent significant performance loss [35] - [37]. In a
mobile multi-user environment the problem is worse, as the transmission from each
user can have a different frequency offset. Even if each user is synchronized to the
base station perfectly, there will still be significant different frequency offsets for
each user due to Doppler shift. Frequency offset in a single user OFDM link isn't a
significant problem as it can be compensated for with minimal increased receiver
complexity. However in a multi-user case there is no easy way of correcting the
frequency errors.
90
Chapter 3 Radio Propagation
CHAPTER 3 RADIO PROPAGATION
This chapter provides a brief review of the major characteristics of radio propagation,
and the problems these cause in transmitting digital information.
Radio propagation effects such as path loss, frequency selective fading, Doppler
spread and multipath delay-spread limit the effectiveness of wireless
communications. An understanding of radio propagation is needed before different
forms of multiuser OFDM can be discussed. This chapter provides a review of well-
known propagation effects and extends these to look at the effects on wide
bandwidth transmissions. This chapter includes an experimental investigation of
small-scale frequency selective fading of multipath environments. It looks at the
variation of the multipath fading with small changes in space. This work was needed
to allow different user allocation schemes, such as adaptive subcarrier modulation, to
be investigated. This work provides insight into how fast radio channels change with
space and time. Although there has been extensive work done on narrow band radio
modelling, not much work has been done on wide band modelling (10 - 100 MHz
bandwidth). Most models in literature are based on statistical results, making them of
little use for investigating user allocation schemes in an OFDM system. For this the
clustering of the frequency fading and change with space is needed to decide how
subcarrier hopping systems will work. This work is very important for OFDM
systems development because it provides a deterministic relationship between
frequency selective fading and space (movement).
3.1 PATH LOSS AND ATTENUATION
During propagation, radio signals weaken with distance. This is due to the wave front
of the radio signal expanding and thus reducing in power density. In free space, the
propagating wave expands as a sphere and thus the power density reduces in
91
proportion to the surface area of this sphere. If the signal is transmitted using a
directional antenna, the signal still expands as a sphere, except that the energy
density is concentrated to one or more areas (see Figure 3-1). If we transmitted the
same energy from an omnidirectional as a direction antenna, the integrated energy
over the surface area of the RF sphere, the energy would be the same. Figure 3-1
shows an expanding RF pulse, if we were to imagine a sinusoidal transmission
(single frequency) it would be continuous stream of expanding spheres, with the
power of these following a sinusoid waveform.
(a) (b)
Figure 3-1, Expanding RF pulse from a central transmitter. (script s0029)
The transmitter is shown as an × in the cut away section. (a) For an

omnidirectional antenna, the energy density is even in all directions. (b) For a
high gain antenna, the energy is concentrated in one direction; it still however
expands as a sphere.
The area of a sphere is proportional to the radius squared, and thus in free space the
RF field strength reduces proportionally with distance squared. Equation (3-1)
calculates the received power over transmission in free space.
λ 
2
PR = PT GT G R   (3-1)
 4πR 
Where PR is the received signal power (watts), PT is the transmitted power (watts),
GR gain of the receive antenna with respect to an isotropic antenna, GT gain of the
92
transmitter antenna, λ wavelength of the RF carrier (m), and R is the transmission

distance in metres.
Free space propagation is very predictable, and can be used to accurately model
satellite communications and directional links with no obstructions, such as short-
range microwave directional point-to-point links. However for most terrestrial
communications such as mobile phones and wireless LAN systems, the environment
is much more complex making propagation modelling much more difficult.
3.2 SHADOWING AND SLOW FADING
In most mobile radio applications, the environment through which the

communication must occur is cluttered and has many obstructions such as buildings,
hills, trees, walls, etc. These objects cause reflections off their surface and
attenuation of signals passing through them, resulting in shadowing. These shadows
can result in large areas with high path loss, causing problems with communication.
The amount of shadowing is dependent on the size of the obstructing objects, the
structure of the material, and the frequency of the RF signal. Most materials are
highly transparency at RF frequencies compared with visible light, making non-Line
Of Sight (LOS) propagation possible. Although many materials are transparent,
metal objects act much like mirrors due to being good reflectors, making them
opaque, resulting in wiring within walls, foil insulation, metal roofing, reinforcing in
concrete, etc being opaque. In addition, large objects such as buildings and hills
absorb much of the RF energy passing through, resulting in deep shadowing behind
them. Under these conditions most of the received energy usually comes from
reflections and diffraction around the object, rather than from the direct path.
Diffraction occurs at edges of obstructing objects in the transmission path. At the

diffraction edge, the signal reradiates as a spherical wavefront originating from the
diffracting edge. This allows it to bend partially around the object. Diffraction does
not have any significant effect on shadowing for visible light due to the small
wavelength of the light (0.4 - 0.7 µm) as compared with most object sizes (0.1 - 10
m). However, at microwave frequencies, the wavelength is relatively large (0.03 –
0.3 m) and thus diffraction significantly reduces shadowing.
93
The received signal is made up from a combination of direct path transmission,

reflected signals and diffracted signals. The amount of received power is the
resulting sum of transmission through these different signal paths. Moving the
receiver, transmitter, or objects in the environment will result in a change in the path
loss due to the transmission path changing, such as moving from a direct line of sight
signal to behind a building (see Figure 3-2). This variation in the path loss occurs
over large distances (typically 10 - 100 wavelengths) and is dependent on the size of
the objects causing the shadowing, rather than the wavelength of the RF signal. Due
to the slow changing nature of this variation, it is commonly referred to as slow
fading.
P
Receiver
Position 1
P Position 2
R D
Transmitter
P - Direct Path
R
D - Diffraction
Building R - Reflection
Figure 3-2, Effect of receiver position on the relative strength of the received
direct, reflected and diffracted signals.
The path loss measured in terrestrial non-LOS links is higher than the predicted value
when using the free space path loss equation (3-1). The signal power in cluttered
environments typically falls off faster than free space. One common method to model
this increased rate of decay is to change the distance dependent exponent from 2
(path loss is proportional to the distance squared) to a higher value. An adjusted path
loss equation is shown in (3-2).
α
λ 
PR = PT GT GR   (3-2)
 4πR 
Where α is the path loss exponent, other terms are defined after Equation (3-1).
94
Environment Frequency Path loss σ (dB)
(MHz) α)
Exponent (α
Retail stores 914 2.2 8.7
Grocery store 914 1.8 5.2
Office, hard partition 1500 3.0 7.0
Office, soft partition 900 2.4 9.6
Office, soft partition 1900 2.6 14.1
Indoor to street (Suburban 900 3.0 7.0
Home)
Factory
Textile/chemical (LOS) 1300 2.0 3.0
Textile/chemical (LOS) 4000 2.1 7.0
Textile/chemical (OBS) 4000 2.1 9.7
Metalworking (LOS) 1300 1.6 5.8
Metalworking (OBS) 1300 3.3 6.8
Paper/cereals 1300 1.8 6.0
Outdoor to indoor 5850 3.4 8.0

penetration of home *
Table 3-1, Table of typical path loss exponent for different environments.
Also specified is the standard deviation (σ) measurements compared with the
straight line predicted from the path loss exponent. LOS – Line Of Sight path,
OSB – Obstructed path. Results obtained from [63] and * [64]
Using (3-2) only gives a rough estimate for the path loss, with no consideration of
the actual environment and fine scale effects such a slow and fast fading.
3.3 MULTIPATH EFFECTS
In a radio link, the RF signal from the transmitter may be reflected off objects such
as hills, buildings, vehicles, walls, etc. Some of these reflections will arrive at the
receiver, effectively creating multiple transmission paths, commonly referred to as a
multipath environment. The radio signal travels over a different distance for each of
these paths, and thus takes a different amount of propagation time. If we were to
transmit an RF pulse in a multipath environment we would receive a signal like the
one shown in Figure 3-3 (a). Each impulse corresponds to one path, with the strength
of each impulse dependent on the path loss for that path. For a fixed frequency
signal, (i.e. a sine wave) the propagation delay results in a phase rotation of the
signal. The amount of phase rotation corresponds to 360° for each wavelength of
path length travelled. Each of the multipath signals will have a different propagation
distance and thus a different phase rotation. These signals add at the receiver,
95
resulting in constructive or destructive interference. Each of the multipath signals can

be represented as a phasor, which has vector length corresponding to signal power
and the angle corresponding to the phase. The received signal corresponds to the
vector sum of the multipath phasors (see Figure 3-3 (b)). Destructive interference
occurs when the vector sum adds to zero. This is also referred to as a ‘null’.
Constructive interference occurs when all the signals have a similar phase,
reinforcing each other.
Impulse Response Phasor Plot

3
Imag
1 2
4
Power
4 5
5
2
Resultant 1
3
Real
Propagation Time
(a) (b)
Figure 3-3, Impulse response and phasor plot for multipath channel.
(a) Impulse response of a radio channel with 5 significant paths. (b) Phasor plot
of the same channel for a particular RF frequency. The length of each vector
corresponds to the power and the angle of the vector corresponds to the phase of
the RF signal at the receiver for that path. The receiver sees the vector addition
of all the multipath signals.
Figure 3-4 shows an example of a channel impulse response. This is for an urban link
over 2.4 km. This signal has a strong direct path signal (0 µs) and two large
reflections at 4.5 µs and 7.5 µs.
96
Figure 3-4, Example channel impulse response for a 2.4 km transmission at 910
MHz on Manhattan Island, New York City [71].
Transmitter height of 120 m, receiver height 2 m. ‘d’ is the average excess delay
(centre of gravity with respect to first arrival delay), ‘s’ is the RMS delay
spread, and ‘o’ is the absolute transmission delay.
3.4 DELAY SPREAD
Delay spread is a measure of the spread in the time over which the multipath signals
arrive. It is a measure of the time dispersion of a channel, and is very important in
determining how fast the symbol rate can be in digital communications.
A symbol is a period over which one or more groups of bits of information are sent.
For a single carrier transmission, using Binary Phase Shift Keying (BPSK) as the
modulation scheme, each symbol carries one bit of information. The symbol
corresponds to the period required to send the phase information as 0° or 180°, which
corresponds to the digital information of zero or one respectively. The faster the
phase is varied the faster the symbol rate and the higher the data rate and bandwidth.
For OFDM transmission, each symbol corresponds to a parallel transmission of
many low bandwidth carriers. The symbol time in this case corresponds to the period
over which the amplitude and phase of the data carriers is remained fixed
corresponding to one data vector.
97
The impulse response of a multipath channel can be modelled by
m
H (t ) = ∑ Ak ⋅ δ (t − Tk ) (3-3)
k =1
where Ak is the complex amplitude and Tk is the time delay of each impulse; m is the
number of received multipath components, and is δ the dirac delta function.
The RMS delay spread is one of the most widely used measurements for
characterising the delay spread of a multipath channel and was originally used by
Cox [72].
0 .5
 m 
 ∑ (Tk − D ) ⋅ Pk
2

Trms =  k =1 m  (3-4)
 
 ∑ Pk 
k =1
where Trms is the rms delay spread, and D is the average excess time delay, which is
equivalent to the centre of energy for the impulse response.
m
∑T k ⋅P k
D= k =1
m
(3-5)
∑P k
k =1
where P is the power of each impulse and is the magnitude of the complex impulse
squared:
Pk = Ak
2
(3-6)
Delay spread results in time blurring, where energy from previous data symbols
becomes mixed in with current symbols. This causes interference, known as Inter-
Symbol Interference (ISI), because previous symbols are uncorrelated, effectively
adding noise to the signal. Single carrier transmissions are particularly prone to
problems caused by delay spread as it normally sets the upper limit on symbol rate.
This is because the bit error rate (BER) increases as the delay spread time becomes a
significant fraction of the symbol time. Simple modulation schemes such as BPSK
can tolerate a delay spread of approximately 10 - 20% of the symbol period; any
more and the BER is too high. However, higher modulation schemes such as 16-
98
QAM, 256-QAM, etc, which have a higher spectral efficiency, are much more
sensitive to ISI and thus the delay spread must be less than several percent of the
symbol period.
Table 3-2 shows the RMS delay spread for a range of propagation environments. For
small indoor environments (< 10 m) the RMS delay spread ranges from 8 – 22 ns,
where as it is higher in larger buildings (up to 100 m) with the RMS delay spread
varying from 20 – 100 ns.
Description Frequency Mean RMS RMS delay

(MHz) Delay spread range
Spread (10% - 90%)
Empty Conference Omni 60 GHz 18.1 ns −
Room Wide (60°) 60 GHz 13.6 ns −
(13.5 x 7.8 x 2.6 m) Standard (10°) 60 GHz 4.7 ns −
LOS, 7 m Tx − Rx Narrow (5°) 60 GHz 1.1 ns −
separation [66]
Residential Room, LOS, (4.6m) [68] 5200 − 8 – 12 ns
Business Room, LOS, (14 m) [68] 5200 − 30 – 50 ns
University Building, 37.2 GHz 10 ns 8.5 − 15.5 ns
LOS, (Average Room Dim: 11 x 5 m)
4 m Avg Tx − Rx separation [67]
Engineering Building, (60 x 44 m) 2400 50 ns 21 – 79 ns
(Mixed LOS and NLOS) [69]
Retail Store, (100 x 108 m) 2400 64 ns 30 – 98 ns
(Mixed LOS and NLOS) [69]
Office Building, 1100 17 ns 12 − 22 ns
(cubicles, offices and laboratory space)
5 m Tx − Rx separation [73]
Office Building, 1100 32 ns 17 − 45 ns
(cubicles, offices and laboratory space)
30 m Tx − Rx separation [73]
City streets in New York, 910 1.2 µs 0.5 – 2.5 µs
NLOS, (2 x 2.5 km) [70]
Table 3-2, Measured delay spread for a variety of environments.
3.5 SHORT TERM FADING
In a multipath environment, the received signal fades with distance due to the
changing phase of the multipath components. Short term fading is caused by the
interference (constructive or destructive) that result from the combination of multiple
received waves. As the receiver or transmitter are moved in space the relative phase
between the different multipath components change, causing the interference to also
change, resulting in fades in the received signal power. At certain locations, the
signal can suffer almost compete cancellation of the signal, resulting in a deep null in
99
the signal. These nulls can be as much as 30 dB. Nulls occur approximately at
intervals of the RF wavelength (30 cm for 1 GHz transmission). The rate of fading
with distance is usually measured using the coherence distance. This is a measure of
the distance over which the radio channel experiences comparable or correlated
fading.
Figure 3-5, Plot of fast fading with distance. This is measured data for a short
indoor link. (script s0030)
See section 3.8.1 for the location of the link. The spacing between fades is
approximately one wavelength, in this case 30cm.
3.6 FREQUENCY SELECTIVE FADING
Multipath also causes fading changes with frequency. This is due to the phase
response of the multipath components varying with frequency. The received phase,
relative to the transmitter, of a multipath component corresponds to the number of
wavelengths the signal has travelled from the transmitter. The wavelength is
inversely proportional to frequency and so for a fixed transmission path the phase
will change with frequency. The path distances of each of the multipath component
is different and so results in a different phase change. Figure 3-6 shows an example
two-path transmission. Path 1 is a direct signal and has a transmission distance of 10
m, while the second path is a reflection with a longer transmission distance of 25 m.
For a wavelength of 1 m each path is an integer number of wavelengths hence the
phase change from transmitter to receiver will be 0° for each path. At this frequency,
100
the two paths will reinforce each other. If we change the frequency to have a
wavelength of 0.9 m then path 1 will be 10/0.9 = 11.111λ, or a phase of 0.111 × 360°
= 40°, while second path will be 25/0.9 = 27.778λ, a phase of 0.778 × 360° = 280°.
This makes the two paths out of phase, which results in a reduction in the signal
amplitude at this frequency.
Reflector
Path 1
8m
Path 2
17m
Receiver
10m
Transmitter
Figure 3-6, Two path transmission to demonstrate frequency selective fading.

(see text)
If we look in terms of the phasors of the multipath components as in Figure 3-3 (b),
each phasor rotates in phase at a different rate, with changes in frequency. The rate of
phase rotation is proportional to the path distance of each multipath component. For
environments with a large number of multipath components, this results in complex
variations in the fading verses frequency. Figure 3-7 shows an example of measured
frequency selective fading within an indoor environment. The signal power varies by
more than 25 dB with frequency, showing that at certain frequencies near complete
signal cancellation is occurring. The frequency selective fading characteristics of a
channel can be summarised by the correlation bandwidth of the channel. This is the
approximate maximum bandwidth or frequency interval over which the fading is
similar and correlated. The exact correlation bandwidth depends on the required level
of correlation. The correlation bandwidth is inversely proportional to the channel
delay spread.
101
3.5m from receiver toward entrance door
-45
Power (dB)
-50
-55
-60
8 8.5 9 9.5 10
Frequency (Hz) 8
x 10
0
Phase (Radians)
-50
-100
-150
8 8.5 9 9.5 10
Frequency (Hz) 8
x 10
Figure 3-7, Frequency selective fading for a short indoor link. (script s0028)
3.7 FREQUENCY-SPACE FADING
As we have seen previously, in multipath environments, the signal fades with

changes in distance (short term fading) and with frequency (frequency selective
fading). In practice, these effects will always occur simultaneously. Figure 3-8 and
Figure 3-12 show the measured channel response of a short indoor link, as a function
of distance and frequency. From this it can be seen that fading patterns in frequency-
space are very similar. In fact the scaling of the fading rates over both dimensions are
dual problems. The coherence bandwidth (inverse of the fading rate in frequency) is
approximately inversely proportional to the transmission distance, while the
coherence distance (inverse of the fading rate in distance) is approximately inversely
proportional to the carrier frequency. Doubling the transmission distance within a
given transmission environment will approximately double the channel delay spread
and consequently halve the correlation bandwidth. Similarly, doubling the carrier
frequency will approximately halve the correlation distance. These comparisons are
only approximate as they are dependent on the locations and reflective nature of the
objects in the radio channel.
102
Figure 3-8, Short term fading, showing fading dependence on distance and
frequency. (script s0030)
3.7.1 EQUALISATION
One method to overcome the limitations of delay spread for single carrier
transmission is to use equalisation. The aim of equalisation is to find an inverse filter
that compensates for the ISI so that all the multipath signals become shifted and
aligned in time, rather than being spread out. For example, the GSM phone system,
which uses 270 k symbols/s (3.7 µs symbol period), can tolerate a delay spread of up
to 15 µs. This is a delay spread of over four symbol periods. The problem with
equalisation is that it becomes increasing difficult, as the ISI is spread over more
symbols. Errors in equalisation, particularly for delay spreads over more than one
symbol, make it difficult to use higher modulation schemes in multipath
environments.
OFDM systems work by resolving the frequency domain so that the width of the
subcarriers is much narrower than frequency selective fading of the radio channel.
This makes the frequency response over the bandwidth of each subcarrier effectively
flat. Only simple equalisation is required for each subcarrier for data transmission as
the flat fading on each subcarrier only results in an amplitude scaling and a phase
rotation.
103
For coherent transmissions, equalisation is implemented by transmitting reference

pilot symbols or tones, which are set to an amplitude and phase known by both the
transmitter and receiver. The channel response is then estimated by dividing the
received subcarrier IQ vector of the pilot symbol or tone by the known transmitted
vector. This measured channel response is then used to equalise the transmitted data.
No equalisation is needed for transmissions using differential phase modulation.

With this method, the amplitude of the subcarrier is not used for carrying information
and so its value is not important. With differential phase modulation the data is
transferred as a phase difference between successive symbols. This compensates for
any phase offset caused by the propagation channel.
3.8 FREQUENCY SPACE FADING MEASUREMENTS
Equalisation of each of the OFDM subcarriers provides a large amount of

information about the radio channel. It indicates which subcarriers have a high signal
power, and those that are in nulls in the spectrum. This allows the allocation of
subcarriers to users to be optimised, to maximise the performance of the system.
These allocation schemes are outlined in detail in section 4.2, however in order to
study these user allocation schemes a detailed knowledge of the fading
characteristics of radio channels is required. Most traditional methods for modelling
radio channels, statistically describe the fading probabilities. However, to investigate
the user allocation schemes, the direct relationships between frequency fading and
changes in space are required.
An experiment was set up to measure the fading as a function of frequency and space
for an indoor environment. The aim of this experiment was to measure changes in
frequency selective fading as a function of distance.
3.8.1 MEASUREMENT LOCATIONS

These measurements were performed in the Electrical Engineering building at James
Cook University. This building has a structure that is fairly typical of a medium sized
office building. The building is three stores high and has outer walls constructed
104
from concrete block. The internal walls were constructed mainly from plasterboard,
with support columns approximately every 6 m. Measurements were taken at six
locations on the second floor, and two locations on the first floor (Rm120 and
Rm122).
Figure 3-9, Plan of measurement locations for frequency space fading

measurements.
Figure 3-10, Set up of equipment for frequency space fading measurements.
Set of the receiver, showing the guide track in the middle of the raised section
of the bench. On this track is the antenna trolley, which pulled along by the
motor on the left. A spectrum analyser was used to measure the power
frequency response at each location along the track, which were logged by a PC
shown on the right.
105
The set up used in the experiment is shown below in Figure 3-11. The signal
generator produced a slow sweep (approximately 3 min per sweep) across the
frequency band of interest (970 - 1040 MHz). This signal appears as a slow moving
continuous wave signal by the spectrum analyser. The maximum hold feature of the
spectrum analyser was then used to trace out the received power at each frequency
across the band. This trace corresponds to the power response of the radio channel
transfer function. This channel response was then logged to a PC and after each
sweep measurement the trolley was moved 1 cm and the process repeated. In all, 240
sweep measurements were done, taking a total time of about 12 hours. The
measurements we very sensitive to any movement within the building and so the
measurements were performed overnight.
Figure 3-11, Experimental set up for measuring the radio channel response as a
function of distance.
These measurements were used for simulating the performance of an adaptive

modulation OFDM system in section 4.2. These measurements give a deterministic
relationship between the fading in the frequency domain, and the fading in the
special domain. This is extremely important for accurately determining the
performance of multiuser OFDM systems using adaptive modulation.
106
3.8.2 MEASUREMENT RESULTS

Figure 3-12 shows an example of one of the measurements taken. The remainder of
the results are shown in Appendix C. The result shown is for a non-LOS transmission
over a distance of 14.9 m. The transmission signal passes through three walls, and as
a result the average path loss (74.9 dB) is significantly higher than free space loss
(55.9 dB) for this same distance.
The results of the measurements show that there is a strong correlation between
fading as a function of distance and fading as a function of frequency. The fading
pattern shows regions of strong constructive interference, with some of these peaks
reaching 6 dB lower path loss than the mean. Nulls caused by destructive
interference occur between these peaks, and are typically relatively small in distance
and bandwidth. Although the nulls are small, they sometimes form clusters resulting
in large suppressed regions.
Figure 3-12, Measured frequency space measurement for a 14.9 m non-LOS

indoor link. (script s0030)
Same data as in Figure 3-8 except as a 2D surface plot. The brightness indicates
the received power for a particular position in space and frequency. Note the
dotted contour shows the mean path loss.
107
Figure 3-13 shows the path loss for all the measurements taken. The path loss shown
was taken as the average over the entire data set for each location. The path loss
exceeds free space loss for all but one location. The best-fit path loss exponent (á)
for all the measurement locations was 2.6, which compares well with previously
published results for office buildings of 2.4 – 3.0, see Table 3-1. The variations in the
path loss closely correlate with environment for each transmission. The LOS
locations have a lower path loss, closely matching free space, with the exception of
location 5. The path loss for this measurement was higher due to its location being
directly above the measurement track, which corresponds to a null in the antenna
pattern. This null is a result of the use of monopole antennas. Most of the other
locations fall around the α = 2.5 line, with the largest deviations occurring for
location 7 and 8. These locations correspond to the next floor down in the building,
resulting in an additional 6 - 9 dB of attenuation. Measurements that were taken at
bench height have a higher path loss, due to the larger number of objects in the
transmission path.
-40 Meas. LOS

Meas. NLOS
-45 5 C α= 2
-50 4 C α= 2 . 5
4 B α= 3
-55
Path Gain (dB)
-60
3 C
-65 6 C
8 B,floor 1
-70
-75 2 B
7 C,floor 1
-80 1 C
-85
1 B
2 4 6 8 10 12 14 16 18
Transmission Distance (m)
Figure 3-13, Path loss of the measurements taken. (script s0030)
The number beside each data location corresponds to the position number, while
‘B’ indicates that the measurement was taken at bench height, and ‘C’
corresponds to ceiling height. Two of the data points were taken on the next
floor down show by the ‘floor 1’ tag. Lines shown represent the expected path
gain for different path loss exponents.
108
100
% Prob. Rx. Signal Strength < Abscissa

Rayleigh Fading
Measured
10
0.1
-30 -25 -20 -15 -10 -5 0 5 10
Received Signal Strength (dB about median)
Figure 3-14, Fading probability distribution for each of the measurement

locations. (script s0030)
Figure 3-14 shows the fading distribution for each of the measurement locations.
They all closely follow a Rayleigh fading distribution [119], indicating that the direct
signal energy is significantly smaller than the multipath energy. This was even true
for LOS measurements.
3.9 RATE OF CHANGE OF RADIO CHANNELS
The aim of this set of experiments is to address the issue of how fast do the
characteristics of indoor radio channels change, due to movement of people and
movement of the transmitter. These measurements were performed to provide a
guide to how fast radio channels change. Numerous studies have been performed to
describe the statistical distribution of narrow band fading [62]. However, not many
studies have been performed to investigate the effects of moving bodies around
transmitter antennas [74]. This experiment simulates the variations in the channel
response that would be experienced in a Wireless LAN application.
This thesis outlines several new adaptive schemes (see Chapter 4) for maximising the
link capacity of an OFDM system. One of the reasons these techniques have not been
used before in mobile applications is that, they must track the response of the radio
channel continuously. This tracking adds complexity to the system and more
109
importantly, data overhead. An adaptive technique for maximising system capacity is

not very useful if all the benefits are outweighed by the overheads. The overhead
required for tracking of a changing channel depends primarily on how fast it changes
and by how much. This experiment aims to address some of these questions.
In any radio channel the amount of disturbance an object makes to the channel
response depends on several factors. These include, the size of the object, the
distance it is from the transmitter and receiver, its reflectiveness, and how much
fading the transmission was suffering from before moving the object. If the channel
response is in a deep fade, only a very small change in the physical channel is
required to cause a large change in channel response. For this reason the channel
disturbance measurements were taken under two different conditions. The first test
involved positioning the transmit antenna so that the overall channel response, before
any object movement, was at a local maximum in the frequency fading spectrum.
The other test was done when the position of the transmitter resulted in a local
minimum (deep fade) in the channel response.
In this experiment, the complex propagation characteristics (magnitude and phase

change due to the radio channel) of a short link were measured using a network
analyser. Three sets of measurements were performed. The first test shows the effect
of a person walking toward the transmitter antenna. The second test shows the effect
of a person walking up to the transmitter antenna, sitting down and typing on a
keyboard near the antenna. The third test shows the result of a person walking around
in the same room as the transmitter and receiver.
All measurements were performed at 940MHz, which is lower in frequency to most

wireless LAN systems (2.4GHz and 5.2GHz). However it was chosen due to
limitations of the network analyser and the availability of a suitable filter for the pre-
amplifier stage.
3.9.1 EXPERIMENTAL SET UP

The propagation characteristics across the room were measured using an analog
network analyser. This set up is shown in Figure 3-15. The output sweep signal from
the network analyser was feed back via a power splitter to the reference input. The
110
network analyser calculates the transfer function of the channel by comparing the
received amplitude and phase to this reference signal. The channel measurement
taken was the combined the transfer function of the coax, antennas, radio channel,
preamplifiers and receiver band pass filter. The transfer function of these elements is
however fixed, and since the aim of the experiment is to analyse the changing nature
of the radio channel these static elements do not affect the results. A digital
oscilloscope was used to capture the real and imaginary output waveforms from the
network analyser. Before each measurement, a zero point measurement was made to
ensure offsets in the IQ measurement taken by the Network Analyser were
compensated for. The captured results were processed further on a PC using Matlab
scripts 2.
The simulated performance of a channel response tracking system was performed on

each of the measured responses. This was done to estimate the tracking error when
an adaptive modulation system is utilised as outlined in section 4.2. The tracking
error was measured by taking the time dependent power response of the channel,
applying different sampling rates, then finding the error between the sample points
and the signal in between the sample points. This information gives an indication of
the amount of channel response error that can be expected for a given sample rate of
the channel response.
If we assume that the received power level primarily determines the SNR of the
channel, then from this data we can estimate how fast the modulation scheme must
be updated in an adaptive modulation system. For example, for square QAM
modulation schemes, the SNR threshold changes by 3 dB for each 1 b/s/Hz change in
spectral efficiency. Thus if the channel power level changes by more than 3 dB then
the modulation scheme should also change, otherwise the BER will become large if
the SNR dropped, or the spectral efficiency should increase by 1 b/s/Hz if the SNR
went up.
2 Code\IQchannelmeas\s0028_plotdata.m
111
Tx antenna
moved
9 dB Loss Monopole
Horz Out Power

Multipath
(I) splitter
Vert
Ref Propagation
(Q) In (5.7 m)
Network Analyser 19 dB 19 dB
Gain Gain Monopole
935 MHz
Digital Oscilloscope Bandpass
Tektronix, TDS380 15 MHz BW
2 dB Loss
Figure 3-15, Experimental set up. The Network Analyser was used to measure
the amplitude and phase of the channel. The Digital Oscilloscope was used to
capture the results for further processing.
3.9.2 SCOPE AND SCALING OF THE RESULTS

The scope of this experiment is limited in that only one room was measured and at
one radio frequency. However, these results do give a strong indication of the
magnitude of effects.
The results can be roughly scaled in frequency by changing the time scale. Fading
effects and multipath are proportional to the RF wavelength, and thus the higher
frequency the shorter the distance between signal fades. The results shown here are
measured in time and not distance, and so for a fixed velocity, changing the time
scale will effectively give an indication of the fading rate for a different RF
frequency. For example for HiperLAN2 that operates at 5.2 GHz, the time scale on
the plots needs to be reduced by 5.2/0.94 = 5.53 times. Thus tracking of fading at 5.2
GHz has to be 5.53 times faster than at 940 MHz to maintain the same performance.
The actual effect of objects moving within the lab will vary slightly with frequency
because of the size of the moving object relative to the wavelength of the radio
carrier will be different.
112
3.9.3 MEASUREMENT LOCATION

The measurements were performed in one of the engineering labs at James Cook
University. A photo of the lab is shown in Figure 3-16. The wall on the left is the
outer wall of the building, which has embedded wire meshing, and thus would be a
good reflector. The other three walls are made of plasterboard. The room contains a
large number of objects, creating a scattering environment. Figure 3-17 shows a plan
layout of the room, indicating the transmitter and receiver location.
Figure 3-16, Photo of the room where the propagation measurements were
taken.
0.6 1.0 2.7 0.8
Bench 0.9
Chairs
1.8
Cupboard
Receiver
0.8
2.0
Cupboard
Bench 2.0
2.8 5.7m
1.3
0.8
Transmitter Bench 0.9
0.4
0.2 1.5 0.4 2.1
Figure 3-17, Plan view of the test room, showing the layout of the benches, and
the position of the transmitter and receiver. Note: All lengths are in metres.
113
3.9.4 EFFECT OF WALKING TOWARD THE TRANSMITTER

CHANNEL RESPONSE
This test involved walking toward the transmitter, pausing then returning to the start
location. This test demonstrates the effect of a person walking around a room for a
short range LAN system. The path of movement is shown in Figure 3-18. Two repeat
measurements were performed. For the first measurement, the transmitter was moved
to a location that resulted in the receiver being located in a local minimum (null in
the spectrum). The second measurement was taken where the receiver was in a local
maximum. The transmitter location was moved 30cm to achieve this. The walking
pattern was the same in both tests.
Receiver
Person
2.1
0.6
Transmitter
0.9
Figure 3-18, Movement of the person in the test room during the measurement
period
Figure 3-19 shows the measured channel response during the movement of the
person toward and away from the transmitter. The results show that the movement of
the person has a significant effect on the channel response, particularly if the
subcarrier signals are in a null in the spectrum. The magnitude of the disturbance
from the person was about the same in both tests. The ripple in the measured IQ
channel response caused by a person walking was about 2x10-4 (Unit-less due to the
relative testing) in both cases. This disturbance had a much greater effect when the
transmission was in a local minimum due to the disturbance being a large percentage
114
of the transmission power. The average power in the local minimum case is about 10
dB lower than the local maximum measurement. The Doppler shift was 10 times
higher for the local minimum case and the disturbance to the power was about 14 dB
compared with 2 dB for the local maximum.
These results show that the energy reflected from the person walking around was not
sufficient to cause complete fading for frequencies that are at a local maximum. This
means for an adaptive modulation system, which tracks the peaks in the spectrum,
only small disturbances are caused by the movement of a person. Thus if the adaptive
modulation system used a link margin of greater than 2 dB it would not have had to
track the fine structure of the movement of the user. The link margin is difference
between the channel SNR and the minimum SNR required for the modulation
scheme used and the BER threshold chosen.
For the adaptive modulation scheme, as proposed in section 4.2, the modulation
needs to be updated for each 3 dB change in the channel response. From Figure 3-19
(e) we can see that when the transmission is at a local maximum that a tracking rate
of 4 Hz is sufficient to maintain a tracking error smaller than 1.2 dB for 99.9% of the
time. Even when the transmission is in a null a tracking rate of 8 Hz will ensure that
the tracking error is less than 3 dB for 98% of the time (Figure 3-19 (f)).
115
Local Maximum Local Minimum

(Local Maxima) Walked to tx, stopped, walked back to start (Local Minima) Walked to tx, stopped, walked back to start
-65
-56
Power (dB)
Power (dB)
paused -70
-57
pause walk to tx walk back to start -75

pause walk to tx paused walk back to start pause
-58
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Time (sec) Time (sec)
-0.4 2
Phase (Radians)
Phase (Radians)
1
-0.6
0
-0.8 -1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(a) (b)
0.4
4
Doppler (Hz)
Doppler (Hz)
0.2
2
0
0
-0.2
-2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
-4 Time (sec)
x 10 -4 Time (sec)
x 10
real 6
Rx Amplitude
10 real
Rx Amplitude
imag
imag
5 4
0 2
-5
0
1 2 3 4 5 6 7 8 9 10
Time (sec) 1 2 3 4 5 6 7 8 9 10
Time (sec)
(c)
(d)
1.2
90% 90%
7
98% 98%
1 99.5 99.5
99.9% 6
99.9%
Tracking Error (dB)
Tracking Error (dB)
0.8 5
4
0.6
3
0.4
2
0.2 1
4 5 6 7 8 10 15 20 30 50
Tracking Sample Rate (Hz) 4 5 6 7 8 10 15 20 30 50
Tracking Sample Rate (Hz)
(e) (f)
Figure 3-19, Radio channel response variation due to a person walking toward
the transmitter. (At 940 MHz) (script s0028)
116
3.9.5 EFFECT OF TYPING ON A KEYBOARD NEAR THE

TRANSMITTER ON THE CHANNEL RESPONSE
Figure 3-20 shows the effect of a person walking up to the transmitter antenna,
sitting in front of it, and pretending to type (as though on a keyboard). This test was
performed to simulate the effect of a WLAN transmitter attached to a notebook
computer. The transmitter antenna was located 15 cm from the hands of the person
typing. Figure 3-20 shows the results of this experiment. The measurement was
performed twice, once with the transmitter located so that the transmission was in a
local maximum in the channel response, and the other with it located in a local
minimum.
The largest disturbance in the channel response occurred after the person sat down in
front of the antenna and lent forward. This is a result of the large displacement of the
body so close to the antenna. The typing in front of the antenna had very little effect
on the channel response, resulting in only a 1 dB ripple for the antenna in a local
maximum, and a 3 dB ripple for the antenna in a local minimum. From Figure 3-20
(e) we can see that only a very low tracking rate (> 4 Hz) is required for an adaptive
modulation scheme, provided the signal is in a local maximum. However if the signal
is in a local minimum the tracking rate needs to be 15 – 30 Hz to ensure that the
tracking error is smaller than 3 dB.
117

(Local Maxima) Walked to tx, sat down and pretended to type (Local Minima) Walked to tx, sat down and pretended to type
-70
-58 lent forward
Power (dB)
Power (dB)
-80
-60
-90
-62 -100
Walk to tx Walk to tx lent paused pretend typing
& sat down paused pretend typing & sat down forward
-64 -110
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
-0.5 5
Phase (Radians)
Phase (Radians)
-1 0
-1.5 -5
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(a) (b)
0.3 8
Doppler (Hz)
Doppler (Hz)
0.2 6
0.1 4
2
0
0
-0.1 -2
-4
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
-3 Time (sec) Time (sec)
x 10 x 10
-4
1
real
Rx Amplitude
real
Rx Amplitude
imag 4
imag
0
2
0
-1
-2
1 2 3 4 5 6 7 8 9 10
Time (sec) 1 2 3 4 5 6 7 8 9 10
Time (sec)
(c)
(d)
90% 35 90%
98% 98%
2 99.5 99.5
30
99.9% 99.9%
Tracking Error (dB)
Tracking Error (dB)
25
1.5
20
1 15
10
0.5
5
4 5 6 7 8 10 15 20 30 50
4 5 6 7 8 10 15 20 30 50
(e) (f)
Figure 3-20, Fading causes by a person pretending to type on a keyboard 15 cm
from the transmitter antenna. (script s0028)
118
3.9.6 EFFECT OF A MOVING PERSON ON THE CHANNEL

RESPONSE
This test shows the effect on the channel response of a person walking through a lab,
with a transmission across the room. The test was performed twice. One with the
transmitter positioned so that the receiver was in a null for the person at the starting
position (local minimum), and the other for the transmitter placed to have the
transmission at a peak (local maximum). The magnitude of the ripple in the received
amplitude was about the same in both tests, as expected. However, since the signal
power for the transmission in the local minimum is low this disturbance ripple
created a large variation in the received power. The power fluctuation for the local
minimum case was 15 dB, as compared with only 4 dB for the local maximum.
From these results we can estimate the tracking rate required for implementation of
adaptive modulation. If a passive user allocation scheme (FDMA, TDMA, Random,
Comb, see section 4.2) than the user is just as likely to be allocated a local minimum
as a local maximum. If we assume a maximum tracking error of 3 dB, and a 98%
tracking probability then a tracking rate of 12 Hz is required in the local minimum to
cope with the disturbance of the person moving in the lab.
If adaptive user allocation is used then users are allocated peaks in the channel
response (local maximum) and so the disturbance is smaller and requires less
tracking. From Figure 3-22 (e) we can see that for a local maximum the disturbance
is small enough that only a very low tracking rate is required (< 4 Hz) in order to
maintain a tracking error of less than 3 dB.
Although a tracking rate of 4 Hz will be sufficient when the receiver and transmitter
are stationary, it will not be enough if the transmitter or receiver is moved. Section
4.4 shows that when the antenna (transmitter or receiver) is moving a tracking rate of
around 100 Hz is needed for walking speed at 1 GHz.
119
(3)
Receiver
5.1
(2) Person (1)

2.1
Transmitter
2.7 0.9
Figure 3-21, Movement of the person relative to the transmitter and receiver
120

(Local Maxima) Walking in the same room as Tx and Rx in an L shape (Local Minima) Walking in the same room as Tx and Rx in an L shape
-57 Walk from (1) to (2)

-65
Power (dB)
Power (dB)
-58
-59 -70
-60 -75
Walk from (1) to (2) Walk from (2) to (3) pause P pause
Walk from (2) to (3)
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
-0.5 2
Phase (Radians)
Phase (Radians)
1
-1
0
-1.5 -1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(a) (b)
0.5
4
Doppler (Hz)
Doppler (Hz)
2
0
0
-2
-0.5
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
-3 Time (sec) -4 Time (sec)
x 10 x 10
1
real 6 real
Rx Amplitude
Rx Amplitude
imag imag
4
0
2
0
-1 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
(c) (d)
2.5
90% 90%
98% 10 98%
99.5 99.5
2 99.9% 99.9%
Tracking Error (dB)
Tracking Error (dB)
1.5
6
1
4
0.5 2
4 5 6 7 8 10 15 20 30 50 4 5 6 7 8 10 15 20 30 50
Tracking Sample Rate (Hz) Tracking Sample Rate (Hz)
(e) (f)
Figure 3-22, Fading caused by a person walking in the same room as the
transmitter and receiver. (script s0028)
121
Chapter 4 Multiuser OFDM
CHAPTER 4 MULTIUSER OFDM
OFDM has been used successfully in several broadcast systems, namely DAB and
DVB, and for point-to-point communications [80]. However, its application in
multiuser two-way communication systems has been limited. Only recently has it
been applied to Wireless LAN applications, with the development of HiperLAN2
[97] and IEEE 802.11a [99].
OFDM is a good contender for the RF interface in 4th generation mobile systems,
however not much work has been done to date in addressing the issues of applying
OFDM in large scale multiuser applications. Multiuser OFDM provides a very
flexible RF interface allowing users to be allocated using Frequency Division
Multiplexing (FDM), by Time Division Multiplexing (TDM) or a hybrid of
FDM/TDM. In addition to this, the multicarrier nature of OFDM allows the radio
channel to be characterised and monitored quickly and easily, presenting numerous
opportunities for optimising the overall system performance, such as:
• Allocating user subcarriers so as to minimise Signal to Interference Ratio

(SIR) in cellular systems.
• Allocate subcarriers to minimise the effects of frequency selective fading.
• Dynamically allocate the modulation scheme on an individual subcarrier

basis to match the current channel conditions.
• Dynamically change the bandwidth of each user based on the link quality.
This allows the bandwidth of weak users to be reduced so that their energy
spectral density remains sufficiently high to maintain communications.
122
4.1 NETWORK STRUCTURE
4.1.1 PEER-TO-PEER NETWORKING

Most current WLAN systems, such as networks using the IEEE 802.11b standard (11
Mbps DSSS), use peer-to-peer networking. This allows computers to communicate
directly to each other. Computers on this type of network, typically connect to the
rest of the world via a single computer connected via a wired Internet connection.
This computer effectively acts as a virtual base station, as all external data traffic is
routed through it. This type of networking is simple to set up and doesn’t require any
special hardware for the virtual base station. One problem with peer-to-peer
networking is that some of the stations can become hidden from each other due to
shadowing or distance, preventing communication. This problem can be overcome
with various methods of multiple hopping of data packets, but this can cause system
complexity to be greatly increased, especially if stations are mobile.
Wired
external Data
External Data
(Internet)
Internal Data
(File Sharing)
Bad Link
Figure 4-1, Peer-to-peer network. Users communicate directly to each other,

with no centralised control.
Peer-to-peer networking requires that all stations can communicate with each other.
As a result most systems use half duplex communications and transmit and receive at
the same frequency.
4.1.2 SINGLE CELL BASE STATION

One subset of peer-to-peer networking is to use a fixed base station as shown in
Figure 4-2. All data traffic is routed through the base station, simplifying routing as
123
mobiles always send data to the same computer. This set up is more suited to larger
WLAN systems and has effectively the same structure as an isolated cell from a
cellular network. One of the main advantages of having a base station is that
additional resources can be added to it, such as improved antennas, diversity
reception and multiple channel processing. These resources on the base station can
significantly improve the performance of the entire network. Additionally due to the
centralised control, resources such as time and frequency allocations can be
optimised.
Wireless
External Data
(Internet)
Wired
Base Station
External Data
Wireless
Internal Data
(File Sharing)
Mobile
Station
Figure 4-2, Base Station connected system. All data passes through the base
station.
As all traffic passes through the Base Station (BS), the communication channel
consists of a forward and reverse channel.
4.1.3 FORWARD CHANNEL

The forward channel corresponds to the signal transmitted from the BS to the remote
users. The signal from the BS is received by all users, which then demodulate the
proportion of the signal relevant to the user. Although the signal from the BS is
common to each user, the propagation path to each user is different. The signal from
the BS is subjected to the transfer function of the propagation channel. In a multipath
environment this results in frequency selective fading, which will be different for
each user.
One method for providing multiuser access is using Frequency Division

Multiplexing (FDM) by allocating different subcarriers to each user. This however
124
requires that the signals for all users are time and frequency synchronised to each
other.
In the forward link all user subcarriers are transmitted from the base station, and thus
all subcarriers can be transmitted with perfect frequency and time synchronisation
with respect to each other. However in the reverse link, subcarriers from each user
are transmitted from different sources, leading to possible inter-user interference.
Rx. User 2
Channel 2
Base Station
Transmit
Channel 1 Spectrum
Rx. User 1 Channel 3 Rx. User 3
Figure 4-3, Forward Link in a single cell system.
4.1.4 REVERSE CHANNEL

The reverse channel corresponds to the transmission from the remote users to the
base station. This link is more difficult than the forward link, as all the transmissions
originate from multiple sources. It is therefore difficult to maintain orthogonality
between all signals, as the propagation effects such as path loss, frequency selective
fading, delay spread and Doppler shift are different for all users.
Due to the spectral overlapping nature of OFDM any loss of orthogonality can result
in high levels of inter-carrier interference. Frequency and time synchronisation errors
result in loss of orthogonality between subcarriers. A frequency offset error of 1 - 2
% of the subcarrier spacing results in a subcarrier power to interference ratio of 20
dB [85].
125
Remote
Station 2
Transmit Base Station

Spectrum
Channel 2
Received
Spectrum
Channel 1 Channel 3
Remote Remote
Station 1 Station 3
Transmit Transmit
Spectrum Spectrum
Figure 4-4, Reverse link of an OFDM system using a single cell network and
FDM for user allocation.
Distortion in the transmission causes spectral spreading due to inter-modulation

distortion. This can be a significant problem if users are of different received powers,
as the distortion products from a strong user can swamp the signal from a weak user.
This causes a problem known as near-far effect, as users that are close to the base
station can swamp the signal from distant users. CDMA systems are very sensitive to
near-far effects, requiring that all signals arrive at the base station with equal power
to within 0.5 - 2 dB of each other. In the reverse link of a CDMA system, the users
are not orthogonal to each other, and so appear as noise to each other. It is the level
of this noise that sets the limit to the capacity of the system. If one user has a signal
that is twice the power (3 dB) of the other users, it is equivalent to 2 users. This
problem is overcome in CDMA systems by using very fast power control (typically
1500 updates per second). Multiuser OFDM systems are much less sensitive to this
problem, as the power level difference between users is limited by spectral
spreading, rather than inter-user interference as with CDMA. Therefore, if the
transmission is free of distortion, then there are no near-far problems. In a typical
system the power levels between users will need to be approximately 20 - 30 dB of
each other.
126
4.2 ADAPTIVE MODULATION
Adaptive modulation is a powerful technique for maximising the data throughput of

subcarriers allocated to a user. Adaptive modulation involves measuring the SNR of
each subcarrier in the transmission, then selecting a modulation scheme that will
maximise the spectral efficiency, while maintaining an acceptable BER. This
technique has been used in Asymmetric Digital Subscriber Line (ADSL) [86], [87],
[88], to maximise the system throughput. ADSL uses OFDM transmission over
copper telephone cables. The channel frequency response of copper cables is
relatively constant and so reallocation of the modulation scheme does not need to be
performed very often, as a result the benefit greatly out ways the overhead required
for measuring of the channel response. Using adaptive modulation in a wireless
environment is much more difficult as the channel response and SNR can change
very rapidly, requiring frequent updates to track these changes.
Adaptive modulation has not been used extensively in wireless applications due to
the difficulty in tracking the radio channel effectively. Work has been done studying
the use of adaptive modulation in single carrier systems [90] - [93], however not
many works have been published on use of adaptive modulation in OFDM systems.
BeamReach is a new communications company that formed in March 2000. They are
utilising adaptive modulation in conjunction with OFDM and beam forming [95].
Unfortunately little information has been released due to the propriety nature of the
research. In [94] the effectiveness of a multiuser OFDM system using an adaptive
subcarrier, bit and power allocation was investigated. Optimisation of the
transmission was achieved by minimising the power requirement for a given
transmission channel and user data rate. It was found that the use of adaptive
modulation, and adaptive user allocation reduced the required transmitter power by 5
- 10 dB. The work in [94] didn’t however investigate the effects of channel tracking
errors on the BER performance.
This chapter outlines adaptive modulation and adaptive user allocation, with a focus
on maximising the user data rate and the QOS (connection access probability).
Section 4.3 shows the simulated performance of an adaptive multiuser OFDM
127
system. This research also investigates the effects of channel tracking errors on the
BER performance.
Work in this chapter, demonstrates the effectiveness of using adaptive modulation in

conjunction with different user allocation schemes. It also provides insight into the
required tracking rate of the radio channel. Previous published work has shown the
effectiveness of adaptive modulation using narrow bandwidth fading models. This
chapter extends this to show simulations over wide bandwidth transmissions.
Most OFDM systems use a fixed modulation scheme over all subcarriers for
simplicity. However each subcarrier in a multiuser OFDM system can potentially
have a different modulation scheme depending on the channel conditions. Any
coherent or differential, phase or amplitude modulation scheme can be used
including BPSK, QPSK, 8-PSK, 16-QAM, 64-QAM, etc, each providing a trade off
between spectral efficiency and the bit error rate. The spectral efficiency can be
maximised by choosing the highest modulation scheme that will give an acceptable
Bit Error Rate (BER).
In a multipath radio channel, frequency selective fading can result in large variations
in the received power of each subcarrier. For a channel with no direct signal path this
variation can be as much as 30 dB in the received power resulting in a similar
variation in the SNR. In addition to this, interference from neighbouring cells can
cause the SNR to vary significantly over the system bandwidth. To cope with this
large variation in SNR over the system subcarriers, it is possible to adaptively
allocate the subcarrier modulation scheme, so that the spectral efficiency is
maximised while maintaining an acceptable BER.
Figure 4-5 shows an example of applying adaptive modulation to an individual

subcarrier as the channel SNR varies with time.
128
Threshold
Levels
32QAM
16QAM
SNR (dB)
16QAM
8QAM
QPSK
8QAM
8QAM
QPSK
QPSK
BPSK BPSK
Time Excess SNR
Figure 4-5, Adaptive Modulation. The modulation scheme is set based on the
SNR of the channel.
The SNR must be greater than the threshold (chosen from Table 4-1) to
maintain a maximum BER. Excess SNR results in the BER being lower than the
BER threshold. This diagram assumes that the modulation scheme is updated
continuously and with no delay.
Using adaptive modulation has a number of key advantages over using static
modulation. In systems that use a fixed modulation scheme the subcarrier modulation
must be designed to provide an acceptable BER under the worst channel conditions.
This results in most systems using BPSK or QPSK. However these modulation
schemes give a poor spectral efficiency (1 - 2 b/s/Hz) and result in an excess link
margin most of the time. Using adaptive modulation, the remote stations can use a
much higher modulation scheme when the radio channel is good. Thus as a remote
station approaches the base station the modulation can be increased from 1 b/s/Hz
(BPSK) up to 4 - 8 b/s/Hz (16-QAM – 256-QAM), significantly increasing the
spectral efficiency of the overall system. Using adaptive modulation can effectively
control the BER of the transmission, as subcarriers that have a poor SNR can be
allocated a low modulation scheme such as BPSK, or none at all, rather than causing
large amounts of errors with a fixed modulation scheme. This significantly reduces
the need for Forward Error Correction.
129
4.2.1 LIMITATIONS OF ADAPTIVE MODULATION

There are several limitations with adaptive modulation. Overhead information needs
to be transferred, as both the transmitter and receiver must know what modulation is
currently being used. Also as the mobility of the remote station is increased, the
adaptive modulation process requires regular updates, further increasing the
overhead.
There is a trade off between power control and adaptive modulation. If a remote
station has a good channel path the transmitted power can be maintained and a high
modulation scheme used (i.e. 64-QAM), or the power can be reduced and the
modulation scheme reduced accordingly (i.e. QPSK).
Distortion, frequency error and the maximum allowable power variation between
users limit the maximum modulation scheme that can be used. The received power
for neighbouring subcarriers must have no more than 20 - 30 dB variation at the base
station, as large variations can result in strong signals swamping weaker subcarriers.
Inter-modulation distortion results from any non-linear components in the
transmission, and causes a higher noise floor in the transmission band, limiting the
maximum SNR to typically 30 - 60 dB. Frequency errors in the transmission due to
synchronisation errors and Doppler shift result in a loss of orthogonality between the
subcarriers. A frequency offset of only 1 - 2 % of the subcarrier spacing results in the
effective SNR being limited to 20 dB [85]. The limited SNR restricts the maximum
spectral efficiency to approximately 5 - 10 b/s/Hz.
130
4.2.2 EFFECTIVE SNR MEASUREMENT

Delay through process of N
(Channel symbols
Compensation) Output
a Demodulate Forward Data
a (Map from IQ Error
Subcarrier Vector b
to Data) Correction
(After FFT) b
Channel Delay
Response N symbols
Modulate Data
2 2 Pilot Symbol
x x Vector
b (Noise Power)
a a SNR
b
(Signal Power)
Figure 4-6, Algorithm for estimating the signal plus noise, to noise ratio
((S+N)/N) for a subcarrier. This approximates the SNR when the noise is
significantly smaller than the signal power.
This is for coherent modulation schemes. The switch selects whether to use
blind feed forward estimation, or to use a known pilot symbol. To use the pilot
symbol, the SNR estimation must be synchronised with the received pilot
symbol. |x| is the absolute of the complex vector, <x> is the average.
Adaptive modulation requires accurate knowledge of the radio channel. Any errors in
this knowledge can result in large increases in the BER, due to the small link margin
used. The effective SNR of each subcarrier can be estimated by calculating the noise
vector on the received signal after channel response compensation. Figure 4-6 shows
a block diagram of an algorithm for estimating the SNR. The first step is to equalise
the received data vector by dividing it by the measured channel response. This
algorithm assumes that the noise on this channel response measurement is zero. In
practice this is not the case as channel measurements will be subjected to the same
noise as we are trying to measure. However the noise on the channel measurements
can be minimised in several ways including boosting the power of the pilot tones or
symbols, and combining results from multiple measurements.
131
After equalisation, the noise in the signal is found by taking the difference between
the known reference IQ vector for each subcarrier and the received IQ vector. The
SNR is then found by finding the ratio between the signal power and the noise
power. These powers must be averaged over multiple symbols to smooth out the
variations in the noise.
This type of measurement will treat any form of signal degradation as noise, such as
AWGN, distortion, residual frequency errors, Doppler Spread, etc. Transmitting a
symbol with known data allows the phase error to be estimated, giving the Signal to
Interference, Noise and Distortion Ratio (SINADR) of each subcarrier. This
SINADR can then be used to select the modulation scheme. The SINADR is referred
to here as the effective SNR.
4.2.2.1 NOISE VECTOR ESTIMATION

To estimate the noise vector we need to know what the correct vector should have
been. This can be achieved in two ways. One method is to use pilot symbols, which
have a precisely known transmission vector, thus allowing the noise vector to be
measured under all conditions. This method is the most reliable and fast method for
characterisation.
Modulation Spectral Required SNR (dB)

Scheme Efficiency BER < BER < BER < BER <
(Coherent) (b/s/Hz) 1x10-2 1x10-4 1x10-5 1x10-6
BPSK 1 4.32 8.41 9.61 10.42
QPSK 2 7.33 11.41 12.58 13.48
8-QAM 3 11.38 15.30 16.45 17.35
16-QAM 4 13.90 18.22 19.46 20.43
32-QAM 5 17.75 21.58 22.74 23.58
64-QAM 6 19.94 24.39 25.60 26.52
128-QAM 7 23.62 27.60 28.76 29.83
256-QAM 8 25.74 30.34 31.60 32.61
512-QAM 9 29.44 33.54 34.75 35.65
1024-QAM 10 31.54 36.29 37.58 38.59
2048-QAM 11 35.23 39.49 40.67 41.75
Table 4-1, Required SNR to maintain a BER below a given threshold.
This is for an OFDM transmission with AWGN, using coherent modulation,

with perfect channel characterisation. This table was derived from the results
shown in Section 2.6. (script snrthresh.m)
Another method is to use the demodulated data as though it was known data, which
is approximately true provided that the BER is low. The demodulated data is
132
remodulated to act as a reference allowing the noise vector to be estimated. Using

this type of measurement allows the SNR to be tracked continuously without the
need for pilot tones or symbols. There are however limitations to this method. The
BER must be low in order for the SNR estimation to be correct, thus the minimum
SNR that can be measured accurately depends on the modulation scheme being used.
Table 4-1 shows the SNR to maintain the BER below three different thresholds, from
this we can estimate the minimum SNR that can be measured for each modulation
scheme. Each time there is a bit error in received signal, it will tend to make the
noise level appear smaller than it really is. However, if averaging is used the chance
of having more than one estimation error in the result is low, even if the BER is
moderately high (around 10-2 ). The lower limit to accurately estimating the SNR for
a particular modulation scheme will depend on the required measurement accuracy,
the amount of averaging used, and the forward error correction method. In a typical
system the lower limit to estimating the SNR will be the SNR required to
maintaining a BER of 1x10-2 , which can be found in Table 4-1.
Another method for estimating the subcarrier SNR is to measure the receiver power
on each subcarrier. The channel equalisation information can be used to measure the
received power on each subcarrier. The SNR can then be estimated by taking the
ratio of the received power to the average noise floor. The level of noise in most
transmissions will be relatively constant, and so a separate measurement of the noise
floor can be made. This method can estimate the SNR very quickly as it does not
require averaging over multiple symbols. The noise floor can be estimated by
averaging the noise vector over many symbols, thus allowing it to roughly track the
noise of the channel.
4.2.3 SINGLE USER ADAPTIVE MODULATION

The performance of a single user adaptive modulation system was simulated over a
wide bandwidth fading channel. The path loss measurement used for the simulation
was one of the measured channels shown in section 3.8. The power used in the
transmission was set so that the average SNR was 17.41 dB. This SNR was
iteratively chosen so that the average spectral efficiency of the adaptive modulation
scheme was 2.0 b/s/Hz. This is so that it could be compared directly with a fixed
133
QPSK transmission. Figure 4-7 shows the resulting SNR of the channel, which varies
from 0 dB up to 25 dB depending on the fading depth. The signal bandwidth of 65
MHz was divided into 200 subcarriers, and the adaptive modulation was updated
every 1 cm, of travel. This corresponds to a velocity of 54 km/hr at an update rate of
1500 Hz. A delay of one update period (1 cm in distance travelled) in the adaptive
modulation was used to simulate the time required for the measurement and
reallocation of the subcarrier modulation schemes.
Figure 4-7, Signal to Noise Ratio of wide band radio channel used for single
user adaptive modulation simulation. (script s0059)
In order to set up an adaptive modulation system an acceptable BER for the system
must be decided. This BER threshold is used for deciding the SNR thresholds for
each of the modulation schemes. To complicate the matter slightly, the overall BER
of the system will be higher than the chosen BER threshold as a consequence of
movement of the user and objects in the environment, and the delay between channel
response measurement and the reallocation of the subcarrier modulation schemes.
For this experiment the system BER was chosen to be 1×10-4 as this was considered
to be low enough for most forward error correction schemes to work effectively. The
BER threshold used for the adaptive modulation allocation was chosen to be 2×10-6
to ensure that final BER of the system maintained a BER of below 1×10-4 . Figure 4-8
shows the resulting modulation scheme applied to each subcarrier as a function of the
position moved by the user. The modulation scheme allocated varied from no
modulation up to 5 b/s/Hz (32-QAM).
134
Figure 4-8, Modulation scheme allocated to each subcarrier. (script s0059)
Note: Dark blue means no allocation applied to that subcarrier.
The BER of the modulation allocation was estimated based on the delay in the
allocation, and the SNR of the channel between the modulation updates. The BER
was solely estimated based on the modulation scheme currently in use and the
channel SNR. Other detrimental effects were not simulated, including Doppler
spread, ISI, time and frequency sync errors and effects of distortion. This
simplification, should however still give accurate results, as the above detrimental
effects can be minimised with suitable system design. The BER as a function of
modulation scheme and channel SNR was estimated using the simulated results in
section 2.6. The channel SNR was calculated with a distance sample rate 10 times
the update rate to ensure that fading between allocation updates was modelled
accurately. Figure 4-9 shows the location of the bit errors as a function of frequency
and space. This is very similar to a contour plot of the channel SNR, with more
errors occurring around the transitions between modulation scheme allocations. The
bit errors are also clustered around the falling edges of the channel SNR, where the
modulation is being reduced from a high to a lower modulation scheme. The errors
tend to occur here due to the delay in the allocation scheme. The BER threshold used
for modulation allocation was 2×10-6 , which is 20 times lower than the resulting
BER of 4.5×10-5 . This is also due to the delay in the allocation.
135
Figure 4-9, Location of the bit errors for the adaptive modulation simulation.
(script s0059)
In addition to the adaptive modulation simulation, the performance of a fixed

modulation scheme was simulated under the same conditions. The average spectral
efficiency of the adaptive modulation scheme was 2.0 b/s/Hz, and so the fixed
modulation was simulated using QPSK, to give the same spectral efficiency. The
location of the bit errors is shown in Figure 4-10. As expected we can see that most
of the errors occur in the deep fades in the channel SNR, and any region where the
SNR was less than 10 dB. Regions of high SNR have an error rate of virtually zero,
showing that they are under-utilised. The overall BER corresponded to 0.011, which
is extremely high, and approximately 250 times worse than the adaptive modulation
scheme.
The fixed modulation simulation was repeated, but with a higher SNR. It was found
that the SNR had to be raised to 34 dB before the BER dropped below the BER of
the adaptive modulation scheme at a SNR of 17.4 dB. This shows that the adaptive
modulation dramatically decreased the effects of fading, allowing the SNR to be
lowered by 16.5 dB as compared with a fixed modulation scheme.
136
Figure 4-10, Location of bit errors for fixed modulation of QPSK, for the same
channel as used for the adaptive modulation simulation (script s0059).
137
4.3 MULTIUSER SUBCARRIER ALLOCATION
There are several methods for allocating subcarriers to users in a multiuser OFDM
system. The main five schemes are to use a group of subcarriers with a fixed
frequency, randomly hopped subcarriers, using TDMA, spread out subcarriers in a
comb pattern and adaptive user allocation.
The performance of these five different user allocation schemes were simulated using
the wide band radio measurements presented in Chapter 3 section 3.8. These data
sets are a measurement of the path loss to 10 locations with in the Electrical and
Computer Engineering building. The details of each measurement are shown in
Appendix C. The path loss was measured as a function of frequency and distance,
with it showing the dependence of frequency selective fading with space.
The multiuser allocation simulation was performed using 10 users. The propagation
characteristics for each user were obtained by matching each user to one of the
propagation measurements shown in Appendix C. The velocity of all the users was
set to 3 m/s and the user allocation was performed over a movement path of 2 m. All
five simulations used adaptive modulation in order to determine the modulation for
each subcarrier in each time slot, with the only difference being the subcarrier
frequencies allocated to each user. The adaptive modulation and the subcarrier
frequency allocation was updated at a rate of 250 Hz, which corresponds to a
distance travelled between updates of 4% of wavelength, for the carrier frequency of
1005 MHz used. The frequency allocations we subdivided into 100 subcarriers over
the system bandwidth of 70 MHz. In practice the number of subcarriers would be
more than this, however the number was kept reasonably low so that the frequency
allocations could be seen in the simulation plots.
A BER threshold of 1×10-6 was used to decide on the SNR thresholds for the
adaptive modulation allocation. The choice of this BER threshold is arbitrary and
dependent on requirements of the final application of the system. Changing this
threshold would result in a slight change in data throughput, however the general
conclusions derived from the comparison of the different techniques would be the
same. In this simulation the adaptive modulation, and user allocation were performed
138
with zero delay. The effect of adding delay to the allocation scheme is investigated in
section 4.4.
Perfect channel estimation was assumed in this simulation. This would enhance the
performance of the simulated system as compared with a practical system. In a
practical system the effects of noise in the channel estimation can be minimised
using several techniques discussed in section 2.7, and so the absolute error
introduced by assuming perfect channel estimation will be less than 0.5 - 3 dB,
depending on the system implementation. Since all the user allocation schemes will
be affected in a similar manner the error in the comparison between these schemes
will be much smaller (about 0.2 – 0.5 dB).
Two-dimensional quadratic interpolation was used to find the channel response at

times matching the adaptive modulation tracking times. Finer interpolation was used
to find the response of the channel between the tracking times. This was then used to
estimate the BER due to channel response tracking errors of the system.
For each user allocation scheme simulation, two tests were performed. The first
simulation measured the performance when all the transmitters in the system used a
fixed transmitter power of –10 dBm. Since the path loss to each of the users varied
by 40 dB the average received SNR also varied by about 40 dB. The second
simulation measured the performance when power control was used to set the
average SNR of all users to 20 dB. This power control was performed by calculating
the average path loss of the channel from the base station to each user, then setting
the transmitter power so that the average SNR at the base station was 20 dB. This
calculated transmitter power was fixed over the 2 m path moved by each user.
In each of these simulations multiple users are transmitting at the same time in the
same frequency band. These multiple transmissions form a single OFDM signal at
the base station’s receiver. In order for these signals to remain orthogonal to each
other, they must all be frequency and time synchronised with each other. The time
synchronisation must be accurate to within the effective guard period length (shorter
then the actual guard period length due to channel delay spread), while frequency
synchronisation must be sufficiently accurate as to maintain a sufficiently high
effective SNR as to use the modulation schemes used in the simulation. The
139
maximum spectrum efficiency used in the adaptive modulation scheme used was
256-QAM, which requires an effective SNR of at least 32 dB (see Figure 2-31). If we
assume that the period between channel characterisation pilot symbols is 16 symbols,
then from Figure 2-46 we can estimate that a frequency offset accuracy of less than
0.02% of the subcarrier spacing is required to ensure that the SNR is not degraded by
frequency errors. For a system with subcarrier spacings the same as HiperLAN2
(312.5 kHz) the maximum allowable frequency offset would be 62.5 Hz. At the
transmission frequency of HiperLAN2 (5.2 GHz) a velocity of 3.6 m/s would result
in a Doppler spread of 62.5 Hz. In order to use 256-QAM at higher velocities a
multiuser Doppler spread correction technique would need to be implemented. The
best method for implementing this is currently unknown. The frequency error
tolerance of the system improves when the modulation scheme is lowered or the
number of pilot symbols or tones is increased.
4.3.1 FIXED FREQUENCY GROUPED SUBCARRIERS

The simplest user subcarrier allocation scheme is to assign each user a group of fixed
frequency subcarriers. Grouping the subcarriers will tend to minimise the level of
inter-user interference due to distortion, power level variation and frequency errors.
However, having a fixed group of subcarriers makes the transmission susceptible to
fading, as the whole group of subcarriers can be lost in a null in the spectrum. Time
interleaving with forward error correction can improve fading performance of a
moving station. However for stationary applications, static nulls can greatly degrade
performance. One method for overcoming the problems of fading is to use diversity
reception. A study of diversity using the same propagation measurements is outlined
in Appendix C.1.
Each user in this simulation was allocated a continuous block of spectrum, 7 MHz in
bandwidth. The frequencies were fixed over the simulated path and showup as
horizontal stripes in Figure 4-11.
140
Figure 4-11, Frequency allocations for the fixed frequency group of subcarriers
simulation. (script s0059)
Each of the 10 users was allocated 7 MHz bandwidth, shown as different

colours and tones in the figure.
Figure 4-12 shows the simulated performance for the fixed user allocation scheme.
For the simulation using a fixed transmitter power of –10 dBm, several of the users
have insufficient power to support the bandwidth allocated, and so their SNR is
insufficient to support even BPSK. These users have a data rate of 0 bps. Several of
the users have a very high SNR and so end up continually operating at the maximum
spectral efficiency, which in this simulation was set to 8 b/s/Hz.
The second simulation shows the performance when the path loss between the users
is normalised using a rough power control. In this simulation the average SNR of
each user was set to 20 dB. Each of the ten users suffers from frequency selective
fading causing their data rate to vary significantly with time. Each of the users
experienced fading of sufficient enough depth to interrupt the data transfer
(modulation efficiency of 0 b/s/Hz). The length of these nulls was typically 5 cm in
length.
It can be seen from Figure 4-12 (g) and (h) that the overall system throughput was
higher for the slow power control experiment. This is simply due to a higher average
SNR and does not imply that slow power control is better than a fixed transmitter
power. The absolute throughput of the slow power control simulation was not
intended to be compared directly with the fixed power simulation, but more to
highlight the different properties of these techniques.
141
Fixed Transmitter Power Slow Power Control
(-10dBm) (Normalised Power to give SNR of 20dB)
60 30
50 25
Mean SNR for Each User (dB)

40
20
30
15
20
10
10
5
0
0
-10
-20 -5
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Distance (m) Distance (m)
(a) (b)
(c) (d)
Percentage Time User Data > 0 Mbps, for all users : 79.39% Percentage Time User Data > 0 Mbps, for all users : 98.48%
60 45
40
50
35
User Data Rate (Mbps)
40 30
25
30
20
20 15
10
10
5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(e) (f)
Average System Throughput: 250.8 Mbps Average System Throughput: 196.8 Mbps
320 280
Total System Data Rate (Mbps)
260
300
240
280
220
260
200
240
180
220
160
200 140
180 120
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(g) (h)
Figure 4-12, Simulated performance from allocating a fixed frequency group of
subcarriers to each user. (script s0059)
142
4.3.2 RANDOM FREQUENCY HOPPING

Fixed wireless systems suffer from problems caused by static fading. In a static radio
environment the frequency selective fading is constant resulting in static fading. In a
mobile channel time interleaving can be used to cope with the brief periods when the
entire signal is faded, however for a static channel the fading period becomes
indefinitely long in time, making time interleaving ineffective.
This problem of static fading can be partly overcome by randomly frequency

hopping the subcarriers over a wide bandwidth. In the user allocation scheme
described by [84], groups of subcarriers are transmitted in short time blocks. These
blocks are randomly frequency hopped to ensure that the time period spent in a null
would is relatively short, approximately 11 symbols. To recover data lost during a
null, time interleaving and forward error correction is used. These come at the cost of
reduced system data capacity and increased delay. Grouping of the subcarriers allows
the correlation between the subcarriers to be utilised.
A multiuser OFDM system using random hopping subcarrier allocation was

simulated using the same conditions as the fixed frequency grouped carriers in
section 4.3.1 . The performance of the random hopping system is shown in Figure
4-14. The quality of service for the random hopping is better than the fixed frequency
group allocation. This is because the random allocation tended to spread the user
allocation over the entire system bandwidth, minimising the probability of all the
subcarriers suffering deep fading at the same time.
143
Figure 4-13, Frequency allocations for each user with random frequency
hopping. (script s0059)
144
50 26
24
40
22

20
30
18
20 16
14
10
12
10
0
8
-10 6
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(a) (b)
(c) (d)
Percentage Time User Data > 0 Mbps, for all users : 85.55% Percentage Time User Data > 0 Mbps, for all users : 100%
60 40
35
50
30
40
25
30 20
15
20
10
10
5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(e) (f)
290 260
280
240
270
220
260
250 200
240
180
230
160
220
210 140
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(g) (h)
Figure 4-14, Simulated performance for randomly hopped groups of subcarriers.
(script s0059)
145
4.3.3 TIME DIVISION MULTIPLE ACCESS ALLOCATION

Using TDMA to separate users is one of the simplest methods for obtaining multiple
user access. This is implemented by allocating each user unique time slots, during
which only a single user accesses the channel. The allocation of time slots to
individual users can be achieved in several ways, with the most suitable depending
on the application and the type of data being transferred. With voice
communications, video and other real time constant bit rate applications, it is better
to pre-allocate users fixed time slots, then cycle between them in a round robin
fashion. This ensures each user obtains the bit rate required and prevents collisions
between transmissions. However for most applications the data traffic tends to arrive
in random bursts. Data traffic tends to have similar statistical properties over a range
of time scales, from milliseconds, seconds, minutes, hours, and even weeks. This
means that data tends to cluster, and that these clusters also tend to cluster [89]. This
can make pre-allocating of time slots inefficient, as most of the time no data is being
transferred. To partly overcome these difficulties, it is likely that the most efficient
allocation scheme would be a dynamic hybrid of pre-allocating users that have a
constant bit rate and dynamically allocating users with a variable bit rate.
The performance of a round robin TDMA user allocation scheme is shown in Figure
4-16. In this simulation each user was allocated the full bandwidth of the channel.
The 10 users were then allocated evenly spaced time slots in a round robin fashion.
The peak transmitter power was boosted by 10 dB to ensure that the average power
was the same as the other user allocation tests.
Figure 4-16 (a) shows the SNR verse distance travelled for all 10 users. The SNR is
discontinuous due to the switching of time slots between users.
The synchronisation requirements for TDMA allocation is reduced compared with

allocation schemes that allow multiple users to transmit at the same time. These
include fixed frequency group allocation, random allocation, comb allocation, and
adaptive user allocation. The overlapping of the subcarriers in the frequency domain
requires very tight frequency synchronisation between multiple users transmitting at
the same time. TDMA allocation removes the need for users to be frequency
synchronised to each other.
146
Figure 4-15, Frequency allocations for each user with TDMA. (script s0059)
Each user is shown as a different colour and tone. The users are allocated time
slots in a round robin fashion, during which the single user allocated the time
slot uses all the system subcarriers.
147
50 26
24
40

22
30
20
20 18
16
10
14
0
12
-10 10
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(a) (b)
(c) (d)
Percentage Time User Data > 0 Mbps, for all users : 100%
Percentage Time User Data > 0 Mbps, for all users : 88.41%
400
600
350
500
300
400
250
300 200
150
200
100
100
50
0
0 0.5 1 1.5 2 0
0 0.5 1 1.5 2
Distance (m)
Distance (m)
(e) (f)
600 400
350
500
300
400
250
300
200
200
150
100
100
0 50
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(g) (h)
Figure 4-16, Simulated performance for TDMA allocation. (script s0059)
148
4.3.4 COMB SPREAD SUBCARRIERS

Static nulls can occur for fixed wireless applications causing problems if the
transmission uses a single group of subcarriers with a fixed frequency. Instead,
subcarriers can be allocated in a fixed comb pattern, spreading them over the entire
system bandwidth. This improves the frequency diversity, preventing all the
subcarriers used by a user being lost in a single null in the spectrum.
Transmitting as a comb pattern requires user subcarriers to be interleaved with one

another, resulting in a large amount of overlapping energy between the users. Any
slight loss of orthogonality due to frequency or timing errors can result in significant
inter-user interference. By comparison, grouping the user subcarriers will slightly
reduce the energy overlap between users, thus reducing inter-user interference.
Despite these potential problems this type of user allocation is useful in applications
that cannot use adaptive hopping or random hopping, due to the added complexity.
Figure 4-18 shows the simulated performance of a 10 user multiuser system using
adaptive modulation and a frequency allocation using a comb pattern. The frequency
allocation is shown in Figure 4-17. Using a comb pattern results in a high frequency
diversity of the signal. The average SNR of the subcarriers allocated to each user (see
Figure 4-18 (a)) is very small with relative small fading (< 10 dB).
Figure 4-17, Frequency allocations for each user with a comb frequency pattern.
(script s0059)
149
Fixed Transmitter Power (-10 dBm) Slow Power Control (Normalised Power to
give SNR of 20 dB)
50 26
24
40

22
30 20
18
20
16
10 14
12
0
10
-10
0 0.5 1 1.5 2 8
Distance (m) 0 0.5 1 1.5 2
Distance (m)
(a) (b)
(c) (d)
Percentage Time User Data > 0 Mbps, for all users : 85.79% Percentage Time User Data > 0 Mbps, for all users : 100%
60 40
50 35
30
40
25
30
20
20
15
10
10
0 5
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(e) (f)
280 230
220
270
210
260
200
250 190
180
240
170
230
160
220 150
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(g) (h)
Figure 4-18, System performance when using adaptive modulation and comb
frequency allocation. (script s0059)
150
4.3.5 ADAPTIVE USER ALLOCATION AND ADAPTIVE

BANDWIDTH
A new adaptive frequency hopping technique is proposed for multiuser OFDM
systems such that blocks of subcarriers are hopped based on the current channel
conditions. After the radio channel has been characterised, each of the users is
allocated subcarriers that have a high SNR for that user. Since each user will be in a
different physical location their fading pattern will be different. The strongest
subcarriers for one user are likely to be different from the other users. Similarly the
subcarriers that are nulls for one user are likely to be strong in at least one other user
in the system. Thus in most cases users can be allocated subcarriers with a high SNR,
allowing the entire system bandwidth to be used without any nulls.
All the previous user allocation schemes outlined from section 4.3.1 - 4.3.4 have
allocated a fixed amount of bandwidth to each user, regardless of the received signal
power. This has led to problems for users that have low received signal strength. The
SNR of these users was insufficient to support communications even using BPSK.
The SNR seen at the receiver is dependent on the signal bandwidth, and so reducing
the bandwidth while using the same transmitter power increases the SNR of the
signal. For example, reducing the signal bandwidth by 10 times, allows the full
transmitter power to be concentrated into one tenth the bandwidth, increasing the
transmitted power spectral density by 10 fold, resulting in an improved received SNR
of 10 dB.
The main aim of adaptive bandwidth allocation is to maintain communications with

users that have low received signal strength. This is achieved by reducing their
bandwidth to the point where the transmitted power spectral density is high enough
to support communications at a low data rate. This can be used as a method for
improving the quality of service (i.e. decreasing the outage probability).
Adaptive bandwidth by itself will not be suitable for all applications, especially those
that required a fixed data rate such as streaming video and audio. In these
applications a joint optimisation of bandwidth and modulation scheme could be
preformed to maintain a fixed data rate, while minimising the amount of bandwidth
used at any one time. This could be achieved by allocating both the user bandwidth
and modulation scheme so that the spectral efficiency multiplied by the user
151
bandwidth results in the required data rate. This way, as the signal strength becomes
weaker, the amount of bandwidth allocated to that user increases to compensate. This
fixed data rate optimisation is not included in the presented adaptive user allocation
algorithm, and could be researched as future work.
In order for the adaptive techniques (adaptive user allocation, adaptive modulation
and adaptive bandwidth) to work effectively all users in the system must be
frequency and time synchronised to each other in order to maintain subcarrier
orthogonality in the reverse link to the base station. In addition to this, the base
station requires complete knowledge of the channel response to all the users in the
system. This makes full implementation of such a technique difficult, as the overhead
in requiring this information can be very high, particularly if the mobile users are
moving or the system bandwidth is low. The principles involved in adaptive user
allocation and adaptive bandwidth can however be reduced to fit various
applications, while still providing some optimisation of system efficiency.
4.3.5.1 ADAPTIVE BANDWIDTH ALGORITHM

An algorithm for implementing adaptive bandwidth allocation was developed to
investigate its possible benefits. While the algorithm developed performs reasonably
well, it is not optimal. The joint optimisation of adaptive bandwidth allocation,
adaptive user allocation and adaptive modulation allocation is reasonably complex.
One method would be to check each combination of user allocations using brute
force, however such a scheme would be extremely inefficient, especially when the
number of users and subcarriers is large. The adaptive bandwidth algorithm
presented uses an iterative process allocating users a given bandwidth; if this
allocation results in the SNR being too low due to a low power spectral density, then
the bandwidth to that user is decreased. The freed bandwidth is then re-allocated to
other users. The algorithm is performed as follows:
Adaptive Bandwidth algorithm used in simulations:
(1) Initially allocate all users an equal number of subcarriers, using the adaptive user
allocation algorithm (see algorithm shown in section 4.3.5.2).
(2) Calculate the SNR of all the subcarriers for a given user, based on the bandwidth
allocated and the particular subcarriers that were allocated in step 1.
152
(3) Find the minimum SNR of the subcarriers allocated to each user.
(4) For each user check that every subcarrier allocated to that user, has a SNR greater
than a given threshold. This is done to ensure that each subcarrier has a
sufficiently high SNR to support at least BPSK modulation. A threshold of 12 dB
was used in the simulations shown in Figure 4-19 and Figure 4-20. If a user has
any subcarriers below this threshold then reduce the BW (number of subcarriers)
allocated to that user. This frees up BW, making it available to other users. Make
sure that each user is allocated at least one subcarrier. This prevents the
degenerate case where a user is allocated zero subcarriers, which in turn makes
the estimated SNR infinite (The SNR is inversely proportional to the signal
bandwidth).
If all subcarriers meet the minimum SNR threshold, then the user allocation is
complete, so apply adaptive modulation to each subcarrier and exit. If the SNR of
one or more subcarriers is below the threshold then continue onto step 5.
(5) Redistribute the free BW by allocating it to users that have a SNR significantly
greater than the minimum SNR threshold. Transmitting over a wider BW results
in a lower transmitted spectral density and received SNR. Thus there is no use
reallocating the extra BW to a user that is just above the SNR threshold, as this
would result in their SNR dropping below the threshold, defeating the purpose of
the reallocation. In the simulations the extra BW was only allocated to users that
were at least 3 dB above the minimum SNR threshold.
To help with data load levelling, the extra BW should be allocated to users that
have a low SNR first, as users with a high SNR will already have a
comparatively high data rate and thus shouldn’t need more BW.
(5) Based on the new user BW allocations, reallocate the subcarriers to users by
repeating the adaptive user allocation algorithm (see algorithm shown in section
4.3.5.2).
(6) Repeat from (2) until all allocated subcarriers have a SNR above the required
threshold.
The exit for this algorithm is in step (4). It usually takes 3-5 iterations, however
under certain channel conditions there will be no solution for meeting the minimum
153
SNR criteria, causing it to iterate infinitely, so limit the number of iterations to

approximately 10 loops. No solution will be found when the received energy for one
or more users is insufficient to support a SNR above the threshold, even when being
allocated a single subcarrier.
4.3.5.2 ADAPTIVE USER ALLOCATION ALGORITHM

This algorithm allocates system subcarriers to each of the users based on the
bandwidth allocated to each user, and the SNR of the user channels. Subcarriers are
allocated to each user based on the frequency selective fading of the channel for that
user. Ideally a system wide optimisation would result in all users being allocated
subcarriers that correspond to peaks in the frequency response for each of the users.
A fully optimised solution would however require an exhaustive search of all
subcarrier-to-user allocation combinations, and thus would be very computationally
intensive. A simple algorithm is proposed that allocates subcarriers to users with the
lowest average SNR, thereby giving them first pick at the best subcarriers. This
algorithm will tend to maximise the quality of service rather than maximise system
spectral efficiency.
The algorithm outlined below performs relatively poorly if all of the users have a
similar average SNR, as the user with the highest SNR (even if it is only 0.1 dB
higher than the other users) will get last pick of the subcarriers. As a consequence it
may end up with subcarriers that are all in nulls. A fairer algorithm in this case might
be to allocate single subcarriers from each user in a round robin fashion.
Optimisation of the algorithm is left to future research.
Adaptive User allocation algorithm used in simulations:
(1) Find the mean SNR over the entire system BW for each user.
(2) Perform allocation of subcarriers to users in-order, from, lowest mean SNR, to
highest mean SNR. This helps to ensure that weak users get access to the best
subcarriers.
(3) Sort the SNR response for the user being allocated, removing any subcarriers that
have already been allocated to other users. The SNR response is the SNR of each
subcarrier as seen by that user. This will be different for each user, due to
propagation variations.
154
(4) Allocate from the sorted SNR response subcarriers in descending order from best
SNR to worst SNR subcarriers to meet the BW required for the user. Repeat from
step 2 until all users have been allocated.
4.3.5.3 COMBINED ADAPTIVE USER ALLOCATION, ADAPTIVE

BANDWIDTH RESULTS
The simulated performance of a combined adaptive user allocation, adaptive
modulation and adaptive bandwidth system was measured under the same channel
conditions as previous simulations outlined in section 4.3.1 through to 4.3.4 . In this
simulation the adaptive modulation, adaptive bandwidth and adaptive user allocation
scheme was updated for each 0.04λ of movement, where λ is the wavelength of the
RF carrier. The performance of the system was measured at a finer spacing of 0.005λ
in order to show the effect of channel change between modulation updates.
Figure 4-19 (a) and (c) show the performance of a system with a fixed user
transmission power of (-10 dBm). From this we can see that the bandwidth allocated
to each user varies due to the adaptive bandwidth scheme. In previous simulations
several of the users could not support any data transmission due to an insufficient
SNR. With adaptive bandwidth the weakest user was only allocated a bandwidth of
less than 1 MHz, instead of the usual 7 MHz, (see Figure 4-20 (c)). From Figure 4-20
(e) we can see that the percentage time where the data rate is above 0 bps, is 100%,
which is much better than other allocations schemes (fixed bandwidth group 79.4%,
Random frequency hopping 85.6%, TDMA 88.4% and Comb 85.8). This is a direct
result of the adaptive bandwidth allocation. A SNR threshold of 12 dB was used to
determine the maximum bandwidth that could be supported by each user. If their
SNR was below 12 dB then the user bandwidth was reduced until the SNR of their
weakest subcarrier achieved a SNR above 12 dB. The effect of this algorithm can be
seen in Figure 4-20 (a) where the SNR of the weaker users hovers around the
threshold of 12 dB. Their SNR dips below the 12 dB threshold several times due to
the slow tracking rate of the adaptive user allocation (0.04λ).
For the simulation using power control, Figure 4-19 (b) and (d), the modulation
applied is very consistent, with little evidence of fading.
The system data rate with adaptive user allocation is much higher than that achieved
155
by any passive allocation scheme. For the fixed transmitter power the average system
data rate was 348 Mbps, which is 40% higher than the passive allocation schemes
that all achieved a data rate of approximately 250 Mbps.
Adaptive user allocation combined with adaptive bandwidth and adaptive modulation
produces a system that has both a very high quality of service and a high spectral
efficiency.
Fixed User Power of –10 dBm Slow Power Control, All users average power
normalised to give SNR of 20 dB
(a) (b)
(c) (d)
Figure 4-19, Simulated adaptive user allocation, utilising adaptive bandwidth,
adaptive frequency, and adaptive modulation. (script s0059)
Results shown in (a) and (c) are for each user having a fixed transmission
power. The transmitter power in (b) and (d) were set so that the average
received power would result in an average SNR of 20 dB. (a) and (b) show the
frequency allocations for each of the users. Each user is shown as a different
colour. (c) and (d) show the modulation scheme used at each frequency. This
simulation assumed perfect channel knowledge for each user.
156
Fixed User Power of –10 dBm Slow Power Control, All users average power
normalised to give SNR of 20 dB
50 35
45
30

40
35
25
30
20
25
20
15
15
10 10
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(a) (b)
20 14
18
12
16
14 10
User BW (MHz)
User BW (MHz)
12
8
10
6
8
6 4
4
2
2
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(c) (d)
Percentage Time User Data > 0 Mbps, for all users : 100% Percentage Time User Data > 0 Mbps, for all users : 100%
120 50
100
40
80
30
60
20
40
10
20
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(e) (f)
Average System Throughput: 348.2 Mbps Average System Throughput: 288 Mbps
420 340
400 320
380 300
360 280
340 260
320 240
300 220
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(g) (h)
Figure 4-20, Continuation of results from Figure 4-19
157
(a) and (b) show the SNR for each user, (c) and (d) shows the bandwidth
allocated to each user, (e) and (f) shows the resulting data rate for each user, and
(g) and (h) showing the overall system data rate. (script s0059)
4.3.6 SUMMARY OF USER ALLOCATION SCHEME

PERFORMANCE
The data throughput and QOS for a number of user allocation schemes were
simulated. These schemes can be logically grouped into two types based on whether
the user allocation was based on current channel conditions are not. Fixed frequency
grouped allocation, TDMA allocation, Random allocation, and Comb allocation all
assign the user subcarriers with no consideration of the current channel conditions,
and so these can be grouped and are referred to as passive allocation. As a contrast to
these schemes, adaptive user allocation allocates the frequency and time slots based
on the current channel conditions, and so can be considered to be an active allocation
scheme.
Fixed User Power (-10 dBm) Slow Power Control (Avg. SNR
User 20 dB)
Allocation System QOS BER System QOS BER
Data Rate (% User Data (x10-5 ) Data Rate (% User Data (x10-5 )
(Mbps) > 0 Mbps) (Mbps) > 0 Mbps)
Fixed Group 250.8 79.4 0.95 196.8 98.5 0.61
TDMA 245.5 88.4 1.1 192.2 100 0.60
Random 249.3 85.6 1.4 192.6 100 0.59
Comb 249.8 85.8 1.3 192.2 100 0.62
Adapt. User 348.2 100.0 0.4 288.0 100 0.27
Table 4-2, Summary of system performance when using different user
allocation schemes. Note: all schemes used adaptive modulation.
It can be seen from Table 4-2 that the system data rate for all of the passive user
allocation schemes is approximately the same at around 250 Mbps. This is a result of
the uniform probability distribution of all radio channels. All of the radio channels in
the simulation had a Rayleigh fading probability distribution (see section 3.8). As a
result any passive allocation scheme will also suffer from Rayleigh fading, regardless
of how the frequency-time slots are allocated. This means that the system throughput
of all passive allocation schemes will be approximately the same, as each of them
will have approximately the same SNR distribution and corresponding spectral
efficiency.
158
The only scheme that performed better is adaptive user allocation. It resulted in a
higher system data rate, a higher QOS, and a lower BER, all of which can be
attributed to the optimisation of the user frequency allocation. The higher data rate is
a result of the users being allocated peaks in the channel response, which reduced the
path loss and consequently resulted in a higher average SNR and data rate. The
higher QOS was a result of the adaptive bandwidth algorithm used. The lower BER
of adaptive user allocation as compared with passive allocation schemes is a result of
the users being allocated peaks in the channel response. The channel response
(magnitude of the channel response in log scale) changes slower at the peaks than the
nulls and as a result the channel tracking for adaptive modulation is more accurate.
This reduces tracking error and results in a reduction in the BER.
4.4 TRACKING RATE REQUIRED FOR ADAPTIVE

MODULATION AND ADAPTIVE USER
ALLOCATION
Adaptive modulation and adaptive user allocation are dynamic techniques that
require continual tracking of the changing channel conditions in order to perform
well. The reallocation of the modulation and the user subcarriers have to be refreshed
many times per second in order to maintain a high performance.
Two different user allocation schemes were simulated (adaptive user allocation and
fixed frequency group allocation) at different tracking rates, both using adaptive
modulation.
Adaptive user allocation has a different performance to passive allocation schemes

such as random frequency hopping, fixed group or comb allocation, as it aims to
steer clear of nulls in the spectrum. This improves the SNR of the subcarriers and
lowers the required tracking rate. This is because the rapid changes in the channel
response occur around the nulls, thus by avoiding them the channel appears to
change more slowly, consequently lowering the required tracking rate.
159
-5
Adaptive Modulation allocation threshold BER of 1x10
-2
10
-4
10
Bit Error Rate

-6
10
-8
10 Adapt, Margin 0 dB, 304.1 Mbps
Adapt, Margin 2 dB, 260.3 Mbps
Fixed, Margin 0 dB, 211.4 Mbps
-10 Fixed, Margin 2 dB, 172.1 Mbps
10
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Distance between Tracking (Fraction of wavelength)
Figure 4-21, BER as a function of the distance between tracking points for a
multiuser system using adaptive modulation and adaptive user allocation. (script
s0061)
Figure 4-21 shows the BER as a function of the update rate. For adaptive user
allocation both the subcarriers allocated to each user and the modulation scheme
used were reallocated at each update. For the fixed group of subcarriers, only the
modulation scheme was updated for each subcarrier. This simulation assumed that
the BER of the link was determined only by the modulation scheme used and the
current SNR of each subcarrier. Perfect channel knowledge was also assumed and
that the delay between the channel response measurement and reallocation of the
modulation scheme was zero.
Although the time between changing the modulation scheme was potentially large
(up to 0.16λ) in this simulation it was assumed that the equalisation of OFDM
receiver would be updated sufficiently fast to not cause a detrimental effect on the
communication link. Also other detrimental effects such as distortion, frequency
errors, Doppler spread, etc were not considered. Also overheads required
implementing an adaptive user allocation scheme, or adaptive modulation was not
included in the system data rate results. The main aim of this simulation was to focus
on the potential effectiveness of both adaptive modulation and adaptive user
allocation.
In Figure 4-21 the distance has been normalised as a fraction of the wavelength of
the RF carrier.
160
VT
D= (4-1)
λ
Where D is the normalised distance as a fraction of the wavelength, V is the velocity

of the user in m/s, T is the time between adaptive modulation tracking (sec), λ is the
wavelength in m.
In the legend for Figure 4-21, the right most set of numbers shows the average
system capacity for each simulation result. The system capacity was nearly constant
as a function of update distance, with it varying by less than 1%. Adaptive user
allocation, with a link margin of 0 dB, achieved the highest system data rate of 304.1
Mbps, (an average of 4.34 b/s/Hz). The fixed subcarrier allocation method only
achieved 211.4 Mbps (3.02 b/s/Hz) for the same transmission power and
communication channels. Adaptive user allocation achieves a higher capacity by
exploiting the frequency selective fading so that users are allocated peaks in the
frequency response, improving the average received power by about 4 dB. If the
radio channels were completely flat with no fading, both fixed frequency allocation
and adaptive user allocation would have performed equally well. In fact the adaptive
user allocation would have been worse since it requires overhead to perform the
adaptive allocation. Note however that the above simulation has not taken this
overhead into account.
A link margin was added to the SNR thresholds determined by the BER. This link
margin would increase the required SNR for each of the modulation schemes,
resulting in the adaptive modulation erring on the side of caution. A positive link
margin consequently results in a reduction in the BER and system spectral efficiency.
The SNR thresholds used to decide on the modulation allocations were chosen to
give a maximum BER of 1×10-5 . These thresholds can be found in Table 4-1.
However, at low speed, with a 0 dB link margin, the BER ends up being lower than
this threshold (approx. 2×10-6 ), which is a result of quantisation of the modulation
allocation.
Another simulation was done showing the BER when a link margin of 2 dB was
added to the SNR thresholds for a BER of 1×10-5 . This is also shown in Figure 4-21.
The increased SNR threshold dramatically reduced the BER, but can at the cost of a
161
reduced overall system capacity. The system data rate for adaptive user allocation
dropped from 304.1 Mbps (4.34 b/s/Hz) when no link margin was used, to 260.3
Mbps (3.71 b/s/Hz) for a 2 dB margin. This drop in the data rate is to be expected.
For coherent QAM the spectral efficiency can be increased by 1 b/s/Hz for every 3
dB increased in SNR. Thus for a 2 dB margin we would expect 2/3 = 0.67 b/s/Hz
drop in the system spectral efficiency. The simulated results showed a drop of 4.34-
3.71 = 0.63 b/s/Hz, closely matching the expected value of 0.67 b/s/Hz. The
advantage of this added link margin is that it allows the tracking rate to be lowered,
while still maintaining the BER below the required threshold.
The lower limit on the BER that can be expected for adaptive modulation for a
chosen BER threshold can be calculated. The easiest case is when the tracking rate is
sufficiently high that it appears to be continuous. This is the case in Figure 4-21
where the distance between subsequent modulation updates is less than 0.01 × λ. If
the adaptive modulation is updated on a continuous basis then SNR of the channel
will always exceed the minimum SNR to achieve the chosen BER threshold. This
excess in SNR, above the minimum threshold, results in the BER being lower than
the chosen BER threshold.
If we assume that the SNR of the channel varies in a random manner, then the excess
SNR will be approximately uniformly distributed between 0 and the SNR difference
between the modulation scheme thresholds. For coherent QAM modulation this
difference corresponds to 3 dB for each 1 bit/s/Hz increase. The average BER for a
particular modulation scheme can be calculated by taking the average of the BER
from the minimum SNR up to the threshold for the next modulation scheme. In most
systems the SNR will vary sufficiently such that all modulation schemes will be
used, thus the overall BER will be the average of average BER results for all
modulation schemes.
162
BER Theoretical Resulting BER Effective Margin

Threshold (at high update rate) (dB)
1×10-5 1.4×10-6 0.82
1×10-4 1.7×10-5 0.94
1×10-3 2.4×10-4 1.08
1×10-2 3.8×10-3 1.24
Figure 4-22, Calculated BER for adaptive modulation. (script s0065)
This is for a high update rate (distance between updates less than λ/100), using
modulation schemes from BPSK through to 1024-QAM. The BER verses SNR
for each modulation was found using simulation result shown in Section 2.6.
4.4.1 EFFECT OF DELAY ON THE PERFORMANCE OF

ADAPTIVE MODULATION
The effect of delay on the performance of a single user adaptive modulation system
was investigated. The set up of the system was the same as that in section 4.2.3 . The
delay between the channel measurement and the modulation allocation update was
varied and the resulting BER recorded. The adaptive modulation was updated 50
times for every wavelength of movement of the user. The simulation was performed
twice, at an average SNR of 17 dB and 30 dB. Figure 4-23 shows the simulated
performance. At 30 dB the BER is more susceptible to allocation delay than at 17
dB. The reason for this is that at a low SNR the nulls in the channel are not used due
their low SNR, and consequently are not allocated data. As a result the subcarriers
that are in nulls do not contribute any bit errors. However when the SNR is high
enough (such as in the 30 dB simulation) all of the subcarriers have a high enough
SNR to be allocated data, including the nulls in the channel. As a consequence of all
the subcarriers being used, the system becomes most susceptible to allocation delay.
The channel response changes very rapidly in the nulls and so even small delays in
the allocation result in large errors in the channel tracking. This results in the
majority of the errors occurring in these nulls. Figure 4-24 shows the location of the
bit errors for the single user adaptive modulation scheme at a SNR of 42 dB. It can
be seen that most of the errors occur as the channel is fading into nulls.
163
B E R t h r e s h o l d 1 x 1 0 - 6, M o d . U p d a t e R a t e : 5 0 p e r w a v e l e n g t h
-1
10
SNR: 17 dB
SNR: 30 dB
-2
10
-3
10
BER
-4
10
-5
10
-6
10
0.01 0.1
Delay (wavelengths)
Figure 4-23, BER as a function of the delay between channel measurement and
adaptive modulation allocation (script s0085)
Figure 4-24, Location of the bit errors in a single user adaptive modulation
scheme at a SNR of 42 dB. Channel measurement Figure A - 12 was used for
this simulation. (script s0059)
164
Chapter 5 Access Point Repeaters
CHAPTER 5 ACCESS POINT

REPEATERS
Wireless networking is an emerging technology allowing users the freedom of

movement. The aim of WLAN systems is to provide users with a data rate
comparable with wired networks within a limited geographic area. Currently most
WLAN products use Direct Sequence Spread Spectrum (DSSS) techniques, based on
the IEEE 802.11b standard, providing a data rate of 11 Mbps in the 2.4 GHz ISM
band [96]. The next generation of WLAN systems will be based on two similar
WLAN standards known as: HiperLAN2 (Europe), and IEEE802.11a (US) [97].
These support a physical layer transmission rate of up to 54 Mbps and use OFDM for
the physical layer implementation.
OFDM is a multi-carrier modulation scheme, which has a high immunity to

multipath effects, and allows a wide range of carrier modulations schemes to be used.
HiperLAN2 adaptively changes the forward error correcting coding rate and the
carrier modulation scheme (BPSK, QPSK, 16-QAM, 64-QAM) allowing the data
rate to be maximised based on the current radio channel characteristics. Using a
higher spectral efficiency modulation scheme, such as 64-QAM, allows the data rate
to be increased, but requires a higher Signal to Noise Ratio (SNR) for a fixed error
rate. Minimising the path loss, allows the SNR and corresponding data rate to be
maximised, as described in section 4.2.
Currently the number of WLAN systems is relatively low and thus interference
between most systems is low. In addition to this, most operate within buildings,
which provide significant interference shielding by the outer walls. This results in the
SNR being primarily limited by transmission power not intercellular interference. If
we can therefore minimise the path loss over the coverage area of the WLAN, we
can therefore maximise the SNR, and the corresponding data rate.
165
In a free space environment, the path loss is proportional to the transmission distance
squared. However, for indoor environments the path loss typically increases much
faster with distance due of shadowing caused by absorption from obstructing objects,
such as walls, cabinets, doors etc.
Shadowing can result in numerous regions of the building having an inadequate

coverage, resulting in a poor Quality Of Service (QOS). This is a significant problem
as the QOS of WLAN systems is very important if they are to replace wired
networking. Shadowing also increases the average path loss compared with free
space, requiring more transmission power, in order to maintain communications.
High quality coverage of a building can be obtained by using a cellular system as

shown in Figure 5-1 (a). Each base station forms a cell, covering a small area of the
building. These cells operate at different frequencies to prevent interference between
them. The path loss is minimised, as the distance from the mobile station to the
closest base station is minimised. Additionally if the transmission path to the closest
base station is blocked, the mobile can connect to another cell. However, this type of
implementation can be expensive, due to the requirement of a number of base
stations. Additionally it requires multiple frequency bands for cellular
implementation and hand-offs between cells, increasing the system complexity. A
new method is proposed to augment the performance of a single cell. This technique
is a simpler and cheaper method for obtaining reliable coverage, in an environment
that suffers from shadowing.
APR
BS BS
Network BS PS
BS BS
(a) (b)
Figure 5-1, Methods for obtaining spatial diversity.
APR - Access Point Repeater, BS - Base Station, PS - Power Splitter /

Combiner. (a) Cellular system with four cells controlled by four base stations,
connected with a wired network for hand off between cells. (b) Single multiple
transmission cell using Access Point Repeaters.
166
5.1 SPATIAL DIVERSITY BY MULTIPLE REPEAT

TRANSMISSIONS
A form of spatial diversity can also be achieved by splitting the signal from a single
central base station and transmitting the same signal from multiple locations around
the building to be covered. This is shown in Figure 5-1 (b). Shadowing is reduced
because there are multiple opportunities to receive the signal. Even if several paths
are blocked by walls or buildings, others may have a reasonable transmission path.
Using multiple transmitters effectively reduces the transmission distance, as the
closest transmitter dominates the signal power.
(a)
(b)
Figure 5-2, Pictorial demonstration showing the difference in shadowing when
using a single transmitter (a), and 4 transmitters (b).
The total power in both cases is the same. With 4 transmitters the energy is
spread more evenly. (Produced using Corel 3D Dream 6.0)
167
An example of how spatial diversity maximises power efficiency is lighting of a

building. If you were trying to illuminate a typical large building with a single light
bulb, the high opacity of the walls would result in many deep shadowed regions. The
amount of power that would be required, for all parts of the building to be reasonably
bright, would be extremely high, as shadowed regions would be lit solely from
reflections and gaps in obstructing walls, such as doorways. However, by using one
light per room, the illumination will be much more even, with minimal shadowing
and a reduced total lighting power. The same is also true for radio signals, thus by
using multiple transmitters spread over the area of a building we can reduce the
average path loss.
Most communication systems need two-way communications using a forward and

reverse link. Due to the reciprocal nature of radio propagation, the properties of the
reverse link will be the same as the forward link. Thus, the path loss for the reverse
link can also be minimised by using multiple receive antennas connected to a single
central base station. The receive and transmit antennas can be combined into one
unit, referred to here as an Access Point Repeater (APR). The use of APRs can
provide a significant improvement with minimal expense.
The problem with using APRs is the mobile receiver will see signals from all APR
transmitters. The propagation delay from each transmitter will be different, and so
will arrive at the receiver with a different delay. This is equivalent to receiving the
signal distorted by strong multipath. This causes frequency selective fading, and an
increase in the delay spread of the transmission, which can result in Inter-Symbol
Interference (ISI). This is particularly a problem for modulation schemes such as
Frequency Shift Keying (FSK), as the maximum symbol rate is limited by ISI caused
by delay spread.
Direct Sequence Spread Spectrum (DSSS) systems have a high multipath tolerance
when a RAKE receiver is used [102]. A RAKE receiver uses correlation to resolve
delayed copies of the signal caused by multipath propagation. These are then aligned
in time and combined. This allows the signals from each of the transmitters to be
resolved and combined. The maximum delay spread that can be tolerated by an
IEEE802.11b DSSS WLAN system, using a 16-tap RAKE receiver, is 125 ns at 11
Mbps and 250 ns at 5.5 Mbps [103]. This level of multipath tolerance should allow
168
APRs to be spaced out with a maximum diameter of about 20 - 35 m, although

problems might arise due to the increased delay spread. This small antenna spacing
means that there would be little advantage in using multiple APR with an
IEEE802.11b system as sufficient coverage can easily be obtained over 20 - 35 m
with a single transmitter.
OFDM has a higher multipath tolerance than DSSS, and allows for a much greater
range of modulation schemes to be used. Minimising the path loss allows the SNR to
be increased, provided inter-cellular interference is low. This improved SNR can be
utilised by adaptively setting the modulation scheme based on the SNR, allowing the
throughput of the system to be increased.
The high multipath tolerance of OFDM is a result of the low symbol rate used, and
the use of a guard period between symbols. The guard period is a cyclic extension of
each OFDM symbol, giving protection against ISI, provided it is longer than the
delay spread of the radio channel.
Multipath causes frequency selective fading, which can result in subcarriers of the
OFDM signal being lost in nulls in the spectrum. These nulls are typically handled
by including forward error correction to compensate for the lost data. Adding
multiple transmitters will result in an increased delay spread, resulting in a decrease
in the correlation bandwidth of the channel. This will not cause any detrimental
effects on the OFDM transmission, provided that enough subcarriers are used.
Indoor propagation normally suffers from a combination of Rician and Rayleigh

fading [74]. Using multiple APRs increases the level of multipath energy, as a result
of the receiver seeing the direct signal from each APR and the reflections from each
APR (see Figure 5-3). This increase will tend to result in more areas in the building
suffering from Rayleigh fading. However this increase in fading should be more than
compensated for by the decreased path loss due to the multiple APRs. More research
is needed to establish the extent of the additional multipath when using APRs.
169
Object
Reflected Signal 3 Txs

from same
BS
Transmitter
Receiver Receiver
Impulse Response Impulse Response

Power
Power
Time Time
Figure 5-3, Increase in the multipath delay spread as a result of using Access
Point Repeaters.
The overall impulse response with APRs corresponds to the sum of the impulse
responses from each APR antenna.
The use of multiple transmitters to obtain improved coverage was first introduced in
the Digital Audio Broadcasting (DAB) system [20]. DAB uses OFDM with a low
symbol rate and a long guard period. With DAB, it is possible for all transmitters to
use the same frequency and transmit copies of the same signal, referred to as a Single
Frequency Network (SFN). For VHF band transmissions, DAB uses 1536 subcarriers
with a symbol time of 1 ms, and a guard period of 246 µs, allowing transmitters to be
up 96 km apart in a SFN, before the delay spread is too large.
5.1.1 APR SPACING FOR HIPERLAN2 SYSTEM

The maximum diameter for a HiperLAN2 system, using APRs, can be estimated
based on the length of the guard period used in HiperLAN2. If we assume that there
are two APRs in a uniform environment (uniform multipath scattering) the worst-
case multipath will occur at a location approximately three quarters between the APR
antennas. At this distance the signal power from both transmitters is reasonably
similar, resulting in significant frequency selective fading. Also at this distance the
time delay between the two paths is close to the maximum possible. If the receiver
170
moves much closer to either of the transmitters then the signal power will be
dominated by just one of the transmitters reducing the multipath fading.
The guard period for HiperLAN2 is 800 ns, and so the first approximation for
maximum spacing between the transmitters is:
TGF × c 800 × 10− 9 × 3 × 108

D= = = 320m
0.75 0.75
where D is the space between the transmitter antennas, TGF is the length of the flat
section of the guard period and c is the speed of light.
This estimate is however overly optimistic for several reasons. Some of the guard
period is needed to allow for error in the accuracy of the time synchronisation,
shortening the allowable delay spread. Also band pass filtering in the transmitter and
receiver can introduce additional delay spread. In addition to this, the result would
only be valid if we were only using BPSK or QPSK. These modulation schemes are
robust and can tolerate a large amount of ISI energy. However, HiperLAN2 uses
higher modulation schemes such as 16-QAM and 64-QAM, which are more sensitive
to ISI caused by delay spread, and as a consequence the maximum antenna
separation must be reduced. This will reduce the channel delay spread and the
resulting ISI. In a typical building the RMS delay spread is in the order of 20 - 60 ns
(see Table 3-2). This is much shorter than the length of the HiperLAN2 guard period,
however the RMS delay spread is only a good indicator for robust modulation
schemes such as BPSK and QPSK. The RMS delay spread can be though of as
approximately the time period over which half the channel impulse energy arrives.
This means that the other 50% of the energy is spread over a longer period. For 64-
QAM, it can only tolerate a very small amount of ISI energy. At a BER of 1×10-5 ,
64-QAM requires an effective SNR of 25.6 dB. Channel noise, IMD, frequency and
time synchronisation errors, Doppler spread, channel equalisation noise and ISI all
degrade the effective SNR of the OFDM transmission, and so the degradation from
just ISI must result in an effective SNR of significantly greater than 25.6 dB. For this
reason it has been estimated that the level of ISI must be result in an effective SNR
of greater than 30 dB.
In order to maintain an effective SNR of greater than 30 dB the level of ISI must be
171
less than 0.1% of the symbol power, thus the guard period must be sufficiently long
so that it covers the period over which 99.9% of the multipath impulse energy
arrives. Most publications on multipath characterisation on have focused primarily
on the RMS delay spread [69], [71], [73], [76], [77], however what is needed is the
time period over which 99.9% of the energy arrives. This time spread can be
estimated approximately by looking at the time spread over which the impulse
reflections have died away to below 30 dB the normalised power. This type of
measurement is known as the profile width and has been used in [77], [78], and [79].
Unfortunately all these papers are for urban radio propagation, making there results
difficult to apply to indoor environments. However they do allow the relationship
between the RMS delay spread and the profile width to be estimated. From these
results it was estimated that the profile width with a cut off of 30 dB below the
normalised power is approximately 6 - 8 times that of the RMS delay spread.
Obviously this will vary based on the actual environment but it does allow an
estimate of the guard period used up by multipath in a typical building environment.
If the RMS delay spread is 20 - 60 ns than period over which 99.9% of the energy
arrives will be about 160 - 480 ns.
As well as protecting against building multipath, the guard period in a HiperLAN2

system also protects against timing offset errors and allows for some side-lobe
suppression by using a raised cosine windowing of 100 ns. Since the raised cosine
section of the guard period overlaps with the previous symbol its effective guard
period length is reduced. So if we assume that the effective Hiperlan/2 guard period
for multipath protection is reduced by 150 ns due to these factors than we get an
effective guard period of 650 ns. Taking into account the multipath spread of 480 ns
we are left with 650-480 ns = 170 ns. From this we can estimate the maximum
spacing between the APRs to be approximately:
170 × 10−9 × 3 × 108

D= = 68 m
0.75
Several other factors haven’t been included in this estimate, such as the possible
detrimental effect of the decreased channel coherence bandwidth on channel
equalisation and degraded time synchronisation accuracy due to the larger delay
spread. As a result of these degradations, it is estimated that the maximum allowable
172
range spacing for a HiperLAN2 network using APRs is 40 - 60 m. Additional

research is needed to verify this.
5.2 IMPLEMENTATION OF AN ACCESS POINT

REPEATER
All APRs within the same cell transmit and receive the same signal, thus only one
base station (BS) is needed per cell. For transmission, the signal generated by a base
station is split N ways, where N is the number of APRs. Coaxial cable or some other
medium, such as optic fibre, is used to deliver the signal to each APR. The signal is
then amplified at the APR to compensate for any losses in the coaxial cable
transmission. For reception, the reverse process is used. Each APR has a Low Noise
Amplifier (LNA) to compensate for losses in the coax, in order to maintain a low
noise figure. The received signals from all APRs are combined and demodulated at
the base station. Phase differences between the APRs have little or no effect as they
are corrected for, in the OFDM demodulation process.
TX To BS
PA
combiner
RX
LNA
Access Point Repeater
(a) Simple APR
RF filter IF filter
TX
PA (5 GHz) (1 GHz)
LNA
RX
Freq.
Clock filter
Synth (10MHz) To BS
Clock combiner
Access Point Repeater Reference
(b) High frequency APR where coax losses are too large. The signals to and
from the BS are sent on the coax at a suitable IF frequency.
Figure 5-4, Possible implementation of an Access Point Repeater.

173
Figure 5-4 (a) shows a simple implementation of an APR, which can be used when
the loss in the coaxial cable is low (< 20 dB). The mast-head amplifier operates at the
RF frequency and so no frequency translation is required. A low frequency control
signal can be used to select whether to transmit or receive. For a full duplex system,
the RF switches can be replaced with diplexers.
In larger systems where the cable losses are too large, the signal can be transmitted
on the cable at an IF frequency. Figure 5-4 (b) shows one possible implementation
for an IF APR. A common clock reference must be used by all APRs to ensure
frequency synchronisation, otherwise frequency errors will reduce the orthogonality
of the OFDM, causing ICI. A suitable out of band clock reference must be
transmitted from the base station. This common clock can then be frequency
multiplied using a Phase Locked Loop (PLL) to generate the Local Oscillator (LO)
for the IF-RF mixing. For such a design, care must be taken to ensure the phase noise
of the clock reference is low and that the clock signal does not interfere with the
received signal.
Another method for distributing high frequency RF signals is to use optic fibre. The
RF signals can be converted to optical signals and transmitted over long distances
with optic fibre. This type of technology is currently being developed for use in
mobile phone applications as a method for producing pico sized cells [101].
5.3 EXPERIMENTAL SET UP
An experiment was set up to test the effectiveness of using multiple transmitters

(APR) to minimise shadowing in an indoor environment. The forward link from the
base station to the mobile stations was tested, by measuring the effective path loss
from fixed transmitters to a mobile receiver. The path loss from a single transmitter
(simulating a single Access Point) was compared with the path loss when using two
transmitters (simulating two APRs).
The path loss was measured at 235 locations on the second floor of the Electrical and
Computer Engineering (ECE) building at James Cook University. The measurement
locations and the building layout are shown in Figure 5-5. The internal walls are
shown as dark lines. The selection of the measurement locations, the position of the
174
transmitters, and the construction of the building can easily influence the overall path
loss probability distribution. However, the results presented will be representative of
the improvement that could be expected in a typical building of similar construction
to the one the measurements were taken in.
Figure 5-5, Path loss measurement locations in the ECE building for APR
testing.
Lines represent internal walls. The thin lines are plaster board walls, and the
thick ones are concrete walls.
The transmitter set up is shown in Figure 5-7. Simple monopole antennas were used
and were assumed to have an effective gain of 0 dBi, including matching losses. The
uncertainty in the gain of the antennas resulted in an absolute error in the path loss
measurements of approximately 3 dB and a differential error between the single and
dual transmitter measurements of 1 dB.
The signal strength at each location was measured using a spectrum analyser on a
trolley, as shown in Figure 5-6, and converted to path loss by compensating for
transmitter power and receiver gain. A slightly different frequency was used for the
single and dual transmitters to allow simultaneous measurements of the path loss.
The receiver antenna was placed on a rotating platform, as shown in Figure 5-8,
which was turned during the received power measurement. Video averaging on the
spectrum analyser was used to average the power over a 0.6 m circular path swept by
the antenna. This was done to remove the effects of frequency selective fading, and
obtain an accurate measure of the path loss. Even though the path loss was measured
using a continuous wave transmission, the results are still valid for a wide bandwidth
signal.
175
Figure 5-6, Photo of the mobile receiver trolley, showing the rotating antenna
on top, spectrum analyser in the middle, and the battery for powering the pre-
amp.
Due to limitations of the equipment available, the measurements were performed at 1

GHz, rather than 5 GHz. This will tend to reduce the effects of shadowing, as the
opacity of most materials is lower at lower RF frequencies, thus we would expect the
effectiveness of using APRs to improve at higher RF frequencies.
TX2a TX1 TX2b

+1.4 dBm +4.4 dBm +1.4 dBm
Power Splitter
Signal Generator 1
990.2 MHz
Signal Generator 2
990.0 MHz
Figure 5-7, Transmitter set up simulating Access Point Repeaters. The total
transmitter power from Tx2a and Tx2b were matched to the Tx1 power.
176
Top view of trolley Side view of trolley

Antenna
Antenna Rotating platform
0.6 m
Spectrum
Analyser
Movement of Antenna
Figure 5-8, The receiver antenna was swept over a circle to average out effect of
fading.
5.4 DISCUSSION OF MEASURED RESULTS
Figure 5-9 shows the path loss for the single transmitter. The path loss increases
rapidly with distance from the transmitter. The path loss in the bottom left corner of
Figure 5-9 at location 5 m, 5 m, has a high path loss of over 100 dB, which is
equivalent to 1 km of free space loss, even though it is less than 20 m from the
transmitter. Figure 5-10 shows the path loss when two transmitters were used. In this
case the path loss is lower and much more evenly distributed.
Figure 5-9, Measured path loss with one transmitter over the area of the ECE
building at 990 MHz. (script s0011)
The dark lines are the internal walls. The × shows the location of the
transmitters. (script s0011)
177
Figure 5-10, Measured path loss for two transmitters over the area of the ECE
building at 990 MHz. (script s0011)
This simulates two APRs. Note the total transmitter power is the same as for the
one transmitter measurements. The × shows the location of the transmitters.
(script s0011)
Figure 5-11 shows the measured and simulated probability distribution of the path
loss. It shows that the measured path loss is more than 7 dB lower when using two
transmitters as compared with one. This result is to be expected, as the longest
distance to a transmitter was reduced by 1.62 times. The typical path loss exponent
of an obstructed path within a building is 4 to 6 [119], as compared to a path loss
exponent of 2 for free space loss. We would therefore expect the path loss to be
reduced by 8.4 to 12.6 dB for a reduction in distance of 1.6 times. Since the
transmission power is split over two transmitters this will reduce the received power
by 3 dB, thus resulting in an overall improvement of 5.4 to 9.6 dB. This compares
well with the measured improvement of 7 dB. Even though there are two
transmitters, the received signal power tends to be dominated by the closest
transmitter due to the rapid fall of power with distance. This is why the improvement
in signal power can be estimated with reasonable accuracy by considering the
reduction in distance to the transmitter.
178
100
Probability of worse pathloss (%)

30
10
Single Tx
Dual Tx
1
65 70 75 80 85 90 95 100
Pathloss (dB)
Figure 5-11, Probability of path loss over the measured area of the building.
(script s0011)
5.5 REDUCTION IN CELLULAR INTERFERENCE
At a large distance, multiple APRs appear as a point source and so the received
power is the same as when using a single Access Point, where the total transmitted
power is the same. This means that the interference to neighbouring systems is
approximately the same regardless of the number of APRs used, for fixed total
transmission power. The near field path loss can be reduced by 7 dB using two
APRs, allowing the total transmitted power to be reduced by up to 7 dB for the same
SNR. This will reduce the overall interference to distant (distance is greater than the
diameter of the APR WLAN system) WLAN systems by 7 dB. Further
improvements could be made using a higher number of APRs. Using one APR per
room could potentially decrease the near field path loss by 20 - 30 dB, allowing a
large reduction in external interference with suitable power control. The reduced
interference will allow a better frequency reuse in a cellular system.
5.6 SHADOW MODELLING
A simple ray trace path loss model was developed to allow continued investigation
into the use of multiple access point repeaters. This is similar to the model used in
[75]. A 2D model was used instead of a 3D one for simplicity and to reduce the
179
amount of simulation time. The 2D model calculates the path loss along radial rays,
which are converted to a fixed rectangular data grid using interpolation. The
rectangular grid allows the signal power from multiple transmitter sources to be
added together.
(a)
Accumulated
0 1 2 3 4
Object Loss
(dB)
Object Loss
(b) Reflection
Loss (dB)
Adjusted
Space
Free
(c)
Attenuation
Total
(dB)
(d)
Figure 5-12, Attenuation calculations for a ray passing through two objects.
The path loss along each ray is found by calculating its intercept to each object in the
environment. The loss through each object is calculated based on the reflection
coefficient of the object surface and the distance through the object. Each surface the
ray intercepts results in energy being reflected causing a jump in the accumulated
attenuation. As shown in Figure 5-12 (a), surface 1, 2, 3, 4 result in jumps in the
accumulated object loss shown in Figure 5-12 (b). In this model, the amount of
reflected energy is constant regardless of the angle of the ray. This reflected energy is
not calculated as another ray, and is ignored to simplify the model.
In addition to reflected losses, each object absorbs energy as the ray passes through
it. This is calculated as a loss proportional, in dB, to the distance travelled through
the object medium. The loss along the ray is accumulated as the ray passes through
multiple objects. This loss is then added to the loss calculated by the standard free
space radio path loss equation to obtain the overall attenuation with distance. The
radio path loss equation is
180
  λ α 
PL = −10 log 10   
  4πR  
(5-1)
 
where PL is the path loss (dB), λ is the wavelength of the radio signal (m), R is the
distance from the transmitter (m), and α is the path loss exponent. For free space
propagation, α = 2. The simulation calculates the path loss due to objects, and so
ideally, the path loss exponent used in the simulation should be two. However due to
the simplicity of the model, a path loss exponent of 2.65 was found to best match the
measured results.
5.6.1 INTERCEPT CALCULATIONS

Figure 5-13 shows the method used for finding the intercepts between the ray being
calculated, and the objects in the environment. Each object is represented as a set of
vertices. The intercept points were found by calculating the intercept between in the
line equations that join the object vertices and the line equation for the ray. If this
intercept lies within the bounds of the two lines then this is taken as an object
intercept. The process is repeated for each line that makes up the surface of each
object and then for all objects in the environment. After all object intercepts have
been found the attenuation along the ray is calculated in a piece wise manner. The
ray is initially assumed to start on the outside of all objects, and so the first object
intercept indicates the point where the ray enters into an object. Each pair of object
intercepts represent the entering and exiting of objects. This model assumes that the
objects do not overlap. The attenuation caused by each object is found by
multiplying the object absorption coefficient (in dB/m) and the distance between the
entry and exit points of the object. In addition to this a fixed amount of energy is also
assumed to be reflected from each surface of the object.
181
Line 3 & Ray intercept,

Within line boundaries Ray End
∴ Object intercept
Line 3
Line 4 Line 2
Line 1
Object vertex
Line 4 & Ray intercept,

Outside Line 4 boundary
∴ No object intercept
Ray Start
Figure 5-13, Diagram showing the method for find the intercepts between a ray
and an object.
Once the path loss has been calculated along all rays, it is converted to a rectangular
grid system using 2D cubic interpolation. This normalisation of the grid locations
allows the signal power from multiple simulations to be combined. This allows for
simulation of multiple transmitters, and the investigation of interference between
transmitters in neighbouring cells. Figure 5-14 shows the conversion from the polar
grid to rectangular.
Rectangular Grid
(used for plotting)
Polar Grid (pathloss
found at locations on
Ray (line of pathloss
calculation)
Figure 5-14, Polar to rectangular conversion used for plotting of simulated path
loss.
The path loss is calculated along at each point on the polar grid. These results
are then converted to a 2D rectangular grid using 2D cubic interpolation.
182
5.6.2 SPACIAL ANTI-ALIASING

Aliasing effects can occur in the path loss simulation due to the finite number of rays
used. This problem becomes more acute when the number of rays is small. This
problem in shown in Figure 5-15. Averaging multiple simulations, each with the
transmitter moved slightly, can significantly reduce this problem. Good results were
obtained by moving the transmitter in a circular area, with a radius equal to the
wavelength. Averaging of the transmitter movement smoothes shadowing, reducing
the fine shadows due to gaps and overlapping joins between objects. This also gives
a crude approximation of diffraction for objects close to the transmitter.
Figure 5-15, Example of spatial aliasing due to finite number of rays used.
A ray that just misses an object, or one that just intersects an object results in
shadowing which is disproportionate to the objects.
5.6.3 MODEL VERIFICATION

The radio model was compared to the measured results to verify the model accuracy
and to adjust the model parameters to give realistic results. The material properties of
the walls were measured directly and are shown in Table 1. These values compare
well with values measured in [75]. The walls were found to be non homogeneous,
caused by wiring, reinforcing, studs, and near by objects, making accurate
measurements difficult.
In the simulation, the material properties of the walls were optimised to obtain a
good match between the simulated and measured building path loss shown in Figure
183
5-9 and Figure 5-10. The resulting simulation parameters were found to closely
match the direct wall loss measurements, as shown in Table 5-1. Figure 5-16 shows
the resulting match between the measured and simulated results. This figure shows
the percentage area of the building (Y-axis), which has a path loss worse than a
specified path loss in the X-axis. For example, 20% of the building has a path loss of
greater than 90dB when using a single transmitter, where as it is only 3% for two
transmitters.
Wall Type Thickness Total Attenuation Direct Measurement of

(m) (dB) wall loss (dB)
Plaster Board 0.13 1.45 1.5 ± 1
Concrete 0.2 3.80 4 ± 1.5
Table 5-1, Material properties of the internal walls.
This shows direct measurements and wall loss found by simulation optimisation
0
10
1 Tx (Meas)
2 Tx (Meas)
Probability of worse pathloss
1 Tx (Sim)
2 Tx (Sim)
-1
10
-2
10
60 70 80 90 100
Pathloss (dB)
Figure 5-16, Cumulative Distribution Function of the path loss within the
measured and simulated building. (script s0011)
5.6.4 SIMULATED RESULTS

Figure 5-17 and Figure 5-18 show the simulated path loss for a single and double
transmitter respectively, for the same building and transmitter location as measured
data shown in Figure 5-9 and Figure 5-10. The probability distribution of the path
loss is shown in Figure 5-16. The main difference between the simulated and
184
measured results is that the simulated results are smoother, which is the result of the
simple environment model.
In addition to simulating one and two transmitters, the path loss was simulated with
four and eight transmitters. The transmitters were positioned in the simulation to
approximately maximise their effectiveness. Figure 5-19 and Figure 5-20 shows the
simulated results for 4 APR and 8 APR respectively. These show that the path loss
continues to decrease significantly as the number of repeaters is increased. This is
despite the fact that the energy being transmitted by each repeater becomes smaller
and smaller, as the total transmission power is fixed. Figure 5-21 shows the path loss
distribution for all the simulated results. The path loss for the worst 2% of the area of
the building was improved by 7 dB when using two transmitters, 16 dB for four
transmitters and 22 dB for eight transmitters.
Figure 5-17, Simulated path loss for one transmitter at 990 MHz. (script s0011)
The × shows the transmitter location.
185
Figure 5-18, Simulated path loss for two transmitters at 990 MHz. (script
s0011)
The × shows the location of the transmitters.
Figure 5-19, Simulated path loss for four transmitters for 990 MHz. (script
s0011)
The × shows the location of the transmitters.
186
Figure 5-20, Simulated path loss for 8 transmitters for 990 MHz. (script s0011)
Note the total transmission power is the same as for one transmitter. The ×
shows the location of the transmitters.
0
10
1 Tx
2 Tx
Probability of worse pathloss
4 Tx
8 Tx
-1
10
-2
10
60 70 80 90 100
Pathloss (dB)
Figure 5-21, Cumulative Probability Distribution for simulated results, showing

the result for 1, 2, 4 and 8 transmitters. (script s0011)
187
5.7 ACCESS POINT REPEATER PERFORMANCE WITH

A MOVING INTERFERER
Previous simulations and measurements show that adding multiple transmitters

significantly reduces the path loss within a building, improving the capacity and
quality of service. This reduced path loss should also improve the tolerance to
outside signals, caused by inter-cellular interference or a mobile station operating at
the same frequency near by. To investigate this possibility a simulation was set up
using the radio model developed in section 5.6.
Two buildings were set up side by side to simulate neighbouring buildings. Both
buildings in the simulation used the same lay out as the Electrical and Computer
Engineering building at JCU. This model was used as it has been previously verified
with measurement. If a different building model were used, it would result in a
different shadowing pattern and thus a different result, however the general results
would be much the same.
The top building was set up as a fixed wireless LAN system using 1, 2, 4, or 8 APRs.
The interference between this WLAN and a moving interferer was simulated. The
interferer was moved from the outside of lower building then moved in a straight-line
passing through the lower building and out the other side. The signal to noise ratio
over the area of the top building was calculated for each movement of the interferer.
5.7.1 SNR CALCULATION

The Signal to Noise Ratio (SNR) at each point in space was calculated based on the
transmission power of the APRs and mobile interferer, the path loss with shadowing
from the buildings, and the thermal and electrical noise in the receivers based on the
transmission bandwidth.
SNR = PLAN − ( PI + N P ) (5-2)
Where PLAN is the power received from the fixed wireless LAN system (from the
APRs) in dBm, PI is the power from the interferer in dBm, and NP is the noise power
from thermal and electrical noise, also in dBm.
188
The received power for multiple APRs was calculated by adding the power received
from each APR.
 A 
PLAN = 10 log 10  ∑ (Tk ⋅ H k ) ( dBm ) (5-3)
 k =1 
Where A is the number of APRs, Tk is the transmitted power from each APR in mW,
and Hk is the channel gain from the transmitter to the receiver. The channel gain is
the ratio of the transmitted power to the received power and was found using the ray
trace model, and is the inverse of the path loss.
For the interferer the calculation was the same except that there was only a single
transmitter.
PI = 10 log 10 (TI ⋅ H I ) ( dBm ) (5-4)
Where TI is the transmitted power from the interferer in mW, and HI is the channel
gain from the interferer to the receiver.
In addition to interference noise, thermal and electrical noise was included in the
SNR calculation:
N P = 10 log 10 ( KTB) + NF ( dBm ) (5-5)
Where K is Boltzmann’s constant (1.38x10-23 J/K), T is the antenna temperature in K,

and B is the bandwidth of the system in Hz. NF is the noise figure of the receiver (in
dB), which is a measure of the added electrical noise by the receiver electronics.
For the simulations the bandwidth was set to 20 MHz, the NF set to 10 dB, the
antenna temperature set to 300 K, the interferer power set to 100 mW, and the total
APR transmitter power to 100 mW.
5.7.2 RESULTS
Figure 5-22 shows the signal power for the interferer and APRs, from which the SNR
was calculated. The performance of the simulated WLAN system was measured by
taking the average SNR over the area of the top building. From this calculated SNR
the spectral efficiency of an OFDM system was estimated for a system using
adaptive modulation. Coherent QAM from 1 b/s/Hz to 10 b/s/Hz were used and
189
allocated based on a BER threshold of 1×10-4 . Figure 5-23 shows the resulting
spectral efficiency as the interferer moves through the lower building. The spectral
efficiency is more than double for 8 transmitters compared with a single transmitter,
corresponding to an improvement of 10 dB. The interferer has the most detrimental
effect when it is outside the lower building, which is to be expected as the building
will tend to shield and contain the interference.
As well as spectral efficiency, the QOS was estimated by measuring the percentage
area of the building where the SNR was insufficient to support a spectral efficiency
of at least 1 b/s/Hz (SNR > 8.4 dB). Figure 5-24 shows the QOS of the WLAN
system in the top building as the interferer is moved. For the WLAN system using a
single APR between 25 - 40 % of the building is not covered indicating a poor
performance. However, increasing the number of APRs improves the performance
considerably, with both 4 and 8 APRs resulting in an average outage of less than 1%.
190
Interferer (passes through the lower buildi ng) Wanted signal, for 1, 2, 4 and 8 APRs.
(each × is an APR)
Figure 5-22, Received signal power for interferer and wanted signal. (script
s0010)
The interferer column shows the results at four points in the simulation. The
APR column shows the result for 4 separate simulations. (colourbar in dBm)
191
Average Spectal Efficiency

6
1 transmitter
3
2 transmitters
4 transmitters
8 transmitters
2
-10 0 10 20 30 40 50 60 70
Distance travelled by interfering mobile user (m)
Figure 5-23, Average spectral efficiency over the area of the top building, as a
function of the movement of the interferer. (script s0009)
The boundary of the building that the interferer passes though is from 0 m
through to 53 m. (y-axis units are in b/s/Hz)
45
1 transmitter
40 2 transmitters
Percentage Area < 1 bits/Hz
4 transmitters
35 8 transmitters
30
25
20
15
10
0
-10 0 10 20 30 40 50 60 70
Distance travelled by interfering mobile user (m)
Figure 5-24, Quality of service provided in the top building in the simulation.
(script s0009)
192
5.8 CONCLUSION
Using multiple Access Point Repeaters is a low cost method that can be used to
reduce shadowing and near field path loss within a building. This allows a reduction
in intercellular interference with suitable power control. It was shown experimentally
to decrease the near field path loss by 7 dB @ 1 GHz within an indoor environment,
using two repeaters as compared with a single transmitter. Simulated results show
that the improvement can be as much as 20 dB for eight transmitters. Using multiple
APRs will result in an increase in the received multipath, however OFDM systems,
including HiperLAN2 and IEEE802.11a, should be able to tolerate this without
detrimental effects. This tolerance should allow a maximum spacing of 40 - 60 m
between any two APRs in a HiperLAN2 cell. Using APRs can be used to minimise
the effects of interference, by improving the coverage within a building. This allows
a reduction in the frequency reuse in cellular and neighbouring WLAN networks
improving the overall spectral efficiency.
193
Chapter 6 Generation of low Crest Factor Multi-Tone Signals
CHAPTER 6 GENERATION OF LOW

CREST FACTOR MULTI-TONE
SIGNALS
One disadvantage of OFDM is that the peak of the signal can be up to N times the
average power (where N is the number of subcarriers). These large peaks increase
the amount of inter-modulation distortion resulting in an increase in the error rate.
The average signal power must be kept low in order to prevent the transmitter
amplifier limiting. The variability of a signal is normally measured by its crest factor
(CF), which corresponds to a measure of the peak to average envelope power of the
modulated RF carrier. Minimising the CF allows a higher average power to be
transmitted for a fixed peak power, improving the overall signal to noise ratio at the
receiver.
This chapter describes how the CF of a sampled signal can be measured accurately,
and outlines several techniques for minimising the crest factor of multi-tone signals.
The term Discrete Multi-Tone (DMT) is used throughout this chapter to describe the
OFDM signals under investigation. Although these terms refer to the same signals,
OFDM implies data communications, while DMT simply describes the signal
structure. The term DMT is used here to emphasise the focus on the multi-tone
nature of the signals rather than the use for communications.
194
26 tones, Carrier Offset : 2 tone spacings 26 tones, Carrier Offset : 2 tone spacings
1 50
Amplitude (Normalised to peak)
0
0.5
-50
-100
Power (dB)
0
-150
-0.5 -200
-250
-1
Real DMT symbol -300
Magnitude of Complex Base Band DMT symbol
-350
0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 250 300
Time (Normalised to symbol period) Frequency (Normalised to the Carrier Spacing)
(a) (b)
1 50
0
0.5
-50
-100
Power (dB)
0
-150
-0.5 -200
-250
-1
-350
0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 250 300
(c) (d)
1 50
0
0.5
-50
-100
Power (dB)
0
-150
-0.5 -200
-250
-1
-350
0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 250 300
(e) (f)
Figure 6-1, The magnitude of the complex base band DMT signal (shown in
red) represents the envelope. (script s0001)
195
6.1 CREST FACTOR MEASUREMENT
The CF is a measure of the peak to average power ratio of the RF envelope. We

however want to be able to calculate the CF from the base band signal, as simulating
an RF waveform requires an extremely high sample rate, resulting in excessive
computational overhead. Figure 6-1 shows the same DMT signal generated at
different centre frequencies using a real waveform. As the centre frequency is
increased, the envelope caused by DMT modulation can be clearly seen. Another
method for finding the envelope accurately is to generate the DMT centred on DC,
using a complex waveform rather than real waveform. Figure 6-1 also shows the
magnitude of the DMT calculated at base band using complex waveform. It can
clearly be seen that the magnitude of the complex waveform corresponds to the
envelope of the carrier. Let the complex base band signal be defined as in.
~s (t ) = s (t ) + j ⋅ s (t ) (6-1)
I Q
When this is quadrature modulated to RF the signal can be written in polar form as:
s rf (t ) = a (t ) ⋅ cos (2π ⋅ f c t + θ (t )) (6-2)
where a(t) is the amplitude and θ(t) is the phase of the signal. Thus:
a (t ) = s I (t ) + sQ (t )
2 2
(6-3)
The CF is the ratio of the peak envelope power to the mean envelope power, taken
over a given time period. Since we are interested in the CF of individual DMT
symbols the time period is taken to be the symbol time of the DMT signal. The
definition of the CF is shown in Equation (6-4), where T is the DMT symbol period
[108].
( ) 
 max a2 (t )
CFdB = 10 log 10  0 < t < T 2 (6-4)
( ) 
 mean a (t )
 0< t < T
196
6
low CF:0.896dB
random phase:7.86dB
Envelope (Normalised Power)

5 zero phase:14.1dB
0
0 0.2 0.4 0.6 0.8 1
Time (Normalised to symbol period)
Figure 6-2, Envelope for 26 tone DMT symbol for different phasing schemes.
(script s0005)
Low CF uses the phasing method described later in this chapter (see section
6.2).
Figure 6-2 shows three DMT symbols. These symbols have equal power on each of
their subcarriers. The difference between them is that the phase angle of the
subcarriers is different. The CF of a DMT is dependent on the phase angle of the
subcarriers. If the phase of the subcarriers is set to zero, or a linear ramp, the
envelope will be the same as a band limited impulse. This has the worst case CF,
which corresponds to N for N subcarriers, thus in dB, for the 26 subcarriers the CF
will be:
CF = 10 log 10 (N ) = 10 log 10 (26) = 14.14dB (6-5)
6.1.1 RANDOM PHASE DMT SYMBOLS

Using a random phasing scheme results in a variable CF depending on the phase
sequence. Some sequences result in a low CF while others give a high CF. The
distribution of the CF follows approximately a lognormal Probability Distribution
Function (PDF). This means that the PDF of the log of the CF (i.e. CF in dB) follows
a normal distribution function (bell curve). This can be seen in Figure 6-3. This type
of distribution results in most of the symbols having a moderate CF, typically around
7 - 10dB, with only a very small number having a very high or low CF. As the
197
number of tones is increased the mean CF also increases, however the standard
deviation decreases, thus the spread of the CF distribution is decreased.
0.16 8 T, mean:7.09dB, STD:0.91dB

Cummulative Prob Distribution
32 T, mean:8.26dB, STD:0.621dB
0.14 128 T, mean:8.97dB, STD:0.444dB
0.12
0.1
0.08
0.06
0.04
0.02
0
4 5 6 7 8 9 10
CF (dB)
Figure 6-3, Crest Factor probability distribution function (split into 40 bins) for
simulated random phase DMT signals, with 8, 32 and 128 tones (script s0032)
Studying random phase DMT symbols is also useful for analysing the CF
distribution for digital data transmission using OFDM. If we assume that the data is
random and that the transmission uses a phase modulation scheme such as QPSK,
16-PSK, etc, the distribution of the CF of the data symbols can be considered to be
close to a random phase distribution. Figure 6-4 shows the distribution for random
phase DMT symbols. This plots shows that for data transmissions the CF will only
very rarely go over 11 dB (less than 0.1% of the time) even for a large number of
subcarriers such as 1000. This CF is much lower than the worst-case CF of
10×log10 (N), where N is the number of subcarriers. For 1000 subcarriers the worst-
case CF is 10×log10 (1000) = 30 dB.
198
10
CF (dB) 7
6 99.9%
99%
5 90%
mean
4 10%
1%
3 0.1%
1 2 3
10 10 10
Number of tones
Figure 6-4, CF distribution of phase modulated OFDM signals verses the

number of tones. (script s0032)
Note: most (99.9%) random phase DMT symbols have a CF much less than the
worst case of 10 log10 (number of tones).
6.1.2 MEASUREMENT OF CREST FACTOR

Care must be taken when estimating the CF of a waveform, particularly for a discrete
time signal. Accurate measurement of the CF requires that the true peak of the
waveform be accurately measured. If the signal under investigation occupies close to
the full nyquist bandwidth, then there will be few samples representing the fast
transitions in the waveform. This can lead to large errors in peak estimation, as
shown in Figure 6-5.
One method for measuring the peaks accurately is to interpolate the signal, and then
measuring the peak of the resulting signal. The greater the signal is interpolated the
better the estimate of the true peak and consequently the higher the accuracy of the
CF measurement. This method assumes that the reconstructed signal will be perfectly
frequency band limited. In practice when the signal is converted back to analogue,
non-ideal reconstruction, such as aliasing non-linear amplification, and non-ideal
frequency response, will result in slight variations in the peaks of the waveform, and
consequently changes in the CF. This amount of deviation from ideal is however
typically small (0.01 - 0.1 dB).
199
Using interpolation to find the peak accurately
Original
1 1.0000
Interpolated
0.8332
0.5
Amplitude
0
-0.5
-1
1 2 3 4 5 6 7 8
Time (samples)
Figure 6-5, Accurate measurement of CF requires accurate measurement of the

peaks in the waveform, which can be found using interpolation. (script s0004)
6.1.3 ESTIMATION OF ERROR IN CF MEASUREMENTS

The accuracy of the CF measurement of a waveform can be estimated theoretically.
If we assume that the modulating waveform is a sine wave, then we can calculate the
amount of error in the CF measurement as a function of the over-sampling of the
waveform. Obviously, a multi-tone waveform is not the same as a sine wave. For an
arbitrary phase modulated DMT signal the peak estimation will tend to be worst for
sections of the waveform that are changing quickly. The higher the frequency of a
signal, the faster it will change, thus a sine wave at the nyquist rate (frequency = half
the sample rate) should approximate the fast changes in the DMT waveforms.
yp
ye
A
t
Carrier Envelope
Amplitude
0
Time
Figure 6-6, Worst case peak estimation sampling.
200
Figure 6-6 shows the waveform for the CF error estimation, which is a sine wave
with amplitude the same as the carrier. The sampled locations are shown as red dots.
The worst peak estimation will occur when the two sample locations around the true
peak are the same in amplitude. Due to the symmetrical nature of the sine wave, this
will occur when the true peak is exactly half way between samples. Let y(t) be the
modulating sine wave, with y p the true peak of the waveform and y e be the
estimated peak, based on the sampled waveform. Let N be the number of samples per
half cycle of the sine wave. This means that N is the interpolation rate. For the
minimum interpolation rate of one (N = 1), there would be two samples per cycle of
the sine wave, hence the frequency would correspond to the nyquist rate. t is the
sample time, thus in terms of phase:
π
t= (6-6)
N
Since N is the number of samples per half cycle and there are π radians per half
cycle.
The estimated peak can be found from the samples that are centred one either side of
the true peak. The amplitude of these samples is equal, and so the estimated
amplitude can be found using the sample to the left of the peak.
π t
y e = A + A sin  −  (6-7)
2 2
Thus the estimated peak, as a function of the over sampling rate N is,
π  1 
y e = A + A sin  1 −   (6-8)
 2  N 
To find the error in the CF measurement for the sample signal, we need to find the
CF for the continuous signal and the sampled signal.
peak power max y (t ) ( 2

)
CF = =
mean power mean y (t ) 2 ( ) (6-9)
The CF for the continuous waveform is:
201
max ( A + A sin (θ ))
θ =2 π
CFp =
mean( A + A sin (θ )) θ =0
(6-10)
=
(2 A ) 2
=
8
2
1.5 A 3
The CF for the sampled signal will be:
ye
CFe =
1.5 A 2
2
 π  1  (6-11)
 A + A sin   1 −   
 2 N 
=
1.5 A 2
Cancelling out A and expanding the squared top term we get:
π  1  π  1 
1 + 2 sin  1 −   + sin 2  1 −  
 2  N   2  N 
CFe = (6-12)
1.5
Thus the error in the CF estimation can be determined by the difference between the
CF of the continuous waveform and the sampled waveform. The CF calculated
previously was in linear scale, however the error in dB is what is of interest, thus CF
error can be found by difference between CFp and CFe after converting to dB. Thus
the error in the CF estimation for the sampled system is:
CFerror = 10 log 10 CFp − 10 log 10 CFe (6-13)
 π 1  π 1 
 1 + 2 sin  1 −   + sin 2  1 −   
 8   2  N   2  N  
CFerror = 10 log 10   − 10 log 10  
 3 
1.5

 
 
 π  1   π  1  
= 10 log 10 (4) − 10 log 10  1 + 2 sin  1 −   + sin 2  1 −   
  2  N   2  N  
(6-14)
202
6.1.4 SIMULATED RESULTS

A simulation was set up to verify the theoretical error estimate, and to find how
much effect having a DMT signal instead of a sine wave envelope has on the CF
measurement. The simulation was performed by generating a large number of
random phase, complex base-band, DMT signals. These were generated at different
amounts of interpolation. The interpolation was done in the frequency domain by
increasing the size of the IFFT used for transform from the frequency domain to the
time waveform. The unused subcarriers were set to zero. The CF for each DMT
symbol was measured at different amounts of interpolation, from 1 to 128. The
interpolated result with the highest CF was assumed to be best estimate of the true
CF for that waveform. The CF error was found by taking the difference between the
highest CF and the CF found for all other rates of interpolation. This process was
repeated with different random phase DMT symbols, and the highest error for each
level of interpolation was recorded. Figure 6-7 shows the simulated results. The
predicted results match the simulation results closely provided the number of carriers
is greater than 2, and the interpolation rate is greater than 1.
(Search of 20000 random phase DMT symbols)

Worst Crest Factor Estimation Error (dB)
2 carriers
0
8 carriers
10 256 carriers
Predicted
-1
10
-2
10
-3
10
0 1 2
10 10 10
Oversample rate
Figure 6-7, Maximum error in CF measurements as a function of over sampling

of the DMT signals. (script s0002)
The over sampling is the simulated bandwidth used divided by the minimum
nyquist bandwidth for the signal. This is a plot of simulated DMT results, and
theoretical results for a sine wave modulated subcarrier.
203
6.1.5 CF ERRORS IN PRACTICE

The CF can be theoretically calculated very accurately for a waveform using
interpolation. However, when it comes time to generate the signal in practice, several
factors will make the CF different from the theoretical. The main two effects that
significantly change the CF are the frequency response and distortion of the
transmitter. Most low CF DMT generation algorithms assume that the amplitude of
all subcarriers is equal, and that the phase angle of these subcarriers is used to
generate a low CF. However in practice no system will have a perfectly flat
frequency response, resulting in some of the subcarriers having more power than was
assumed in the phasing scheme. This modifies the waveform and will typically result
in a worse CF. Distortion will also change the CF, particularly clipping of the signal,
which tends to reduce the CF of the signal.
6.2 CREST FACTOR MINIMISATION USING GENETIC

ALGORITHMS
This section outlines a new phasing method for generating ultra low Crest Factor
(CF) OFDM symbols. This phasing scheme uses genetic algorithms to optimise the
phase of the subcarriers to minimise the CF of the waveform. The result of this
process is that symbols with a CF as low as 0.65 dB are possible, which is 0.7 dB
lower than any previously known algorithm. These ultra low CF symbols can be used
for channel estimation due to the known phase angle of the subcarriers transmitted.
This makes them useful for pilot symbols in coherent modulation or for adaptive
modulation to allow tracking of the radio channel.
Because of their ultra low CF, the average power of these symbols can be boosted
significantly (~6 dB) while still maintaining the same peak power as data symbols.
This allows the channel estimation to have an improved SNR and thus improve
channel estimation and speed.
Several phasing schemes have been previously developed, producing low overall
CFs. We start out by looking at the performance of these techniques.
204
6.2.1 SHAPIRO-RUDIN PHASES

Shapiro and Rudin (1954) [104] showed that for multi-tone signal with N tones (N is
a power of 2) that by setting the phase of the subcarriers to 0 or pi, in a particular
sequence, that this could yield a crest factor of 2 (3 dB). The sequence is defined as
follows. We start with the string p=11, and repeatedly concatenate to p a copy of p
with its second half negated. For example:
Length Shapiro-Rudin Phase Sequence

2 s = {1 1}
4 s = {1 1 1 –1}
8 s = {1 1 1 –1 1 1 –1 1}
16 s = {1 1 1 –1 1 1 –1 1 1 1 1 –1 –1 –1 1 –1 }
32 s = {1 1 1 –1 1 –1 1 1 1 1 –1 –1 –1 1 –1 1 1 1 –1 1 1 –1 1 –1 –1 –1 1 1 1 –1 1}
Table 6-1, Shapiro-Rudin phase sequences for generating low CF DMT signals
The phase of the subcarriers is then set to 0 if sk = 1, and π if sk = -1.
 0, sk = 1
θk =  (6-15)
π, s k = −1
If N is not a power of 2 the crest factor is still low, with it dipping to 2 when the
number of tones is a power of two. To generate a non-power of two number of tones,
the phase sequence is truncated from both directions. Figure 6-8 shows the crest
factor as a function of the number of tones for Shaprio-Rudin's phasing scheme.
Shapiro-Rudin Phase scheme

6
5.5
5
Crest Factor (dB)
4.5
3.5
2.5
2
1 2 3
10 10 10
Number of Tones
Figure 6-8, Crest Factor of a multitone signal when using Shaprio-Rudin's

phasing scheme. (script s0067)
205
Time Envelope of the signal Spectrogram showing time dependence of the
frequency response
Shapiro-Rudin Phase scheme
0.04
0.035
Power Envelope (linear)
0.03
0.025
0.02
0.015
0.01
0.005
0
0 0.2 0.4 0.6 0.8 1
Time (Normalised to Symbol Length)
(b)
(a)
Figure 6-9, Time envelope and spectrogram for DMT based Shaprio-Rudin's
phasing scheme. (script s0055)
This is for a 224-tone signal.
Shapiro and Rudin's scheme is interesting in that the phase angles used are discrete,
and semi-random. This differs from the other schemes such as Newmann [105] and
Narahashi [106] (as described later) as these schemes vary the phase continuously in
a quadratic form.
6.2.2 NEWMANN PHASES

D. J. Newmann proposed varying the phase in a quadratic fashion in 1965 [105].
Equation (6-16) calculates Newmann phases where N is the number of subcarriers
and k is the index of the particular subcarrier. Figure 6-10 shows the crest factor as
the number of tones is varied. The worst case CF is 3.6 dB, which is for 3 tones. For
greater than 10 tones the CF is less than 2.8 dB. The CF approaches an asymptote of
2.6 dB as the number of tones tends toward infinity.
π (k − 1) 2
θ= (6-16)
N
206
Newmanns phasing scheme

6
5.5
Crest Factor (dB)

4.5
3.5
2.5
2
1 2 3
10 10 10
Number of Tones
Figure 6-10, Crest Factor for Newmann's Phasing scheme. (script s0067)
Figure 6-11 shows the time waveform and spectrograph of a 224-tone signal using
Newmann’s phasing scheme. From the spectrograph it can be seen that the signal is
very similar to a linear frequency sweep. If we take a single continuous tone it will
have a constant envelope power, resulting in a CF of 0 dB. If this tone is then swept
slowly, then we end up with a signal similar to a DMT signal produced by
Newmann’s phasing scheme. The main difference is the transients at the start and
end of the sweep. The FFT of the DMT will be perfectly flat in the frequency
domain, and will have a small amount of ripple in the time domain, resulting in a CF
of greater than 0 dB. Conversely, the linear frequency sweep will be perfectly flat in
the time domain, but have a slight ripple in the frequency domain.

frequency response
Newmanns phasing scheme
0.04
0.035
0.03
0.025
0.02
0.015
0.01
0.005
0
0 0.2 0.4 0.6 0.8 1
(a) (b)
Figure 6-11, Time envelope and spectrogram for DMT based on Newmann’s
quadratic phasing scheme. (script s0055)
207
6.2.3 S. NARAHASHI AND T. NOJIMA PHASING SCHEME

In 1994 S. Narahashi and T. Nojima [106] proposed a new phasing scheme based on
a result similar to Newmann. The phase of each of the subcarriers is set as in
equation (6-17). The resulting crest factor is very similar to Newmann's phasing
scheme except that it performs slightly better than Newmann's method for small
number of tones (< 6). However, Newmann's method is consistently better for N >
24.
θ=
(k − 1)(k − 2 ) π (6-17)
N −1
S. Narahashi and T. Nojima phasing scheme

6
5.5
5
Crest Factor (dB)
4.5
3.5
2.5
2
1 2 3
10 10 10
Number of Tones
Figure 6-12, Crest Factor as a function of the number of tones for S. Narahashi
and T. Nojima phasing scheme [106] (script s0067)
The time waveform and spectrograph of a 224-tone DMT signal generated using S.
Narahashi and T. Nojima’s phasing scheme is shown in Figure 6-13. Not surprisingly
it is almost identical to that obtained using Newmann’s method, as both schemes use
a quadratic phase equation of approximately the same gain, and both give about the
same CF.
208
frequency response
S. Narahashi and T. Nojima phasing scheme
0.04
0.035
0.03
0.025
0.02
0.015
0.01
0.005
0
0 0.2 0.4 0.6 0.8 1
(a) (b)
Figure 6-13, Time envelope and spectrogram for DMT based on S. Narahashi
and T. Nojima’s quadratic phasing scheme. (script s0055)
In 1995, S. Narahashi presented a new method for producing low CF symbols [107].
Instead of developing an equation to minimise the CF, an iterative optimisation
procedure was performed. The algorithm starts with a random phase sequence, and
then uses steepest descent minimisation on the variance of the instantaneous
envelope. This process is repeated for many random phase sequences and the best
resulting sequence is kept. This algorithm results in a much lower CF symbols than
previous methods. Figure 6-14 shows the resulting CF using this algorithm compared
with the GA method presented in this chapter.
3
Steepest Descent
2.8 Genetic Algorithm Optimised
2.6
2.4
Crest Factor (dB)
2.2
1.8
1.6
1.4
1.2
2 4 6 8 10 12 14 16 18 20 22 24
Number of tones
Figure 6-14, Crest Factor for the steepest descent optimised phasing scheme
proposed in [107]. (script s0034)
Also shown is a comparison with the result obtained using the proposed
technique presented in section 6.2.4 , using genetic algorithms to optimise the
phase sequence.
209
6.2.4 GENETIC ALGORITHM PHASE OPTIMISATION

The envelope of a DMT signal is extremely complex as the phase angles of each of
the tones in the signal interact to give constructive and destructive interference at
different points along the resulting time domain signal. It is easy to obtain the
resulting time domain waveform for any given subcarrier phase sequence, however it
is not easy to obtain a method for optimising the phase sequence to minimise the CF
of the resulting DMT waveform. The addition of each new subcarrier into a DMT
signal, effectively adds another dimension to the problem, and so to optimise the CF,
the searching must span a large multi-dimensional space to find the optimal solution.
This makes it impossible to perform a brute force numerical optimisation, as the
number of optimisation variable is too large. For example a 100-tone DMT signal
has 100 variables that need optimising. To make this even worse, the space defining
the relationship between the phase sequence and the resulting CF is fractal in nature.
This means that there are many similar good solutions, making it difficult to ever
find any global minimum.
There is no known direct analytical method for calculating a phase sequence that will
result in the minimum CF possible [107], and so previous published research has
focused obtaining a minimisation of the CF by using a mathematical approximation
for the envelope of the DMT symbols [105], [106]. Two of the most important
phasing schemes developed using this technique have been outlined in section 6.2.2
and 6.2.3 . CF minimisation was achieved by using analytical minimisation of
envelope approximation. This method resulted in good solutions (CF of 2.6 dB), but
does not approach the lowest CF possible.
This thesis presents a new method for obtaining phasing schemes that result in DMT
signals with a very low CF. The method presented uses Genetic Algorithms (GA) to
perform a searching algorithm to find low CF phase sequences. This method does not
rely on any approximations for the signal envelope and so results in a much better
end solution. The lowest CF obtained using this method was 0.65 dB, which was for
64-tone signal, making this 2 dB better than analytical techniques. This process uses
numerical searching to obtain good solutions and so the performance of the end
result is dependent on the amount of computer processing applied to the problem,
210
and so it is still an open question as to how low the CF can be made for a multi-tone
signal.
Genetic Algorithms solve problems in a completely different manner to analytical

minimisation. GA allows the optimisation of arbitrarily complex systems without
requiring a mathematical understanding of how the optimisation is to be done. It is
also very well suited to optimising problems that have a very large number of
variables that require optimisation, as in this case. The use of GAs has gained
popularity in the last few years as a method for optimising complex problems [114] -
[116].
GAs are based on the principles of Darwinian evolution. To use genetic algorithms to
solve a problem, the solution is represented as a sequence of variables known as a
genome (or chromosome). In the case of optimising the phase scheme for generation
of low CF DMT symbols, the genome corresponds to the sequence of phase angles
applied to each tone in the DMT signal. Multiple genome sequences are used,
making up a population. The genomes within a population are all different,
representing genetic diversity. The actual problem domain to be optimised is defined
by the transformation from each genome to its performance or fitness. For the case
under investigation, the fitness corresponds to calculating the CF of the DMT for
each of the phase sequences, where the lower the CF, the higher the fitness of the
phase sequence. Optimisation occurs by selecting the highest performing phase
sequences, then using these to create a new generation by mutating these sequences.
Figure 6-15 shows the algorithm used to perform the CF minimisation.
6.2.4.1 GENETIC ALGORITHM USED FOR CREST FACTOR

MINIMISATION OF DMT SIGNALS
The simulation starts with an initial population of phase sequences. The length of
each sequence corresponds to the number of tones being simulated. The fitness of
each phase sequence is measured by calculating the crest factor of the DMT signal
with subcarrier phase angles set to the phase sequence. The sequences are then sorted
by fitness. A percentage of the best sequences are kept for reproduction to make the
next generation, and rest removed, simulating death. The surviving sequences are
paired up randomly, simulating sexual reproduction. Enough pairs are made to
restore the population back to its original size. Each pair forms one resulting phase
211
sequence, by using a random splicing location. To this new sequence a small amount
of random noise is added to simulate mutations in reproduction. The amount of
random noise added is initially relatively large, but is decayed with the generations.
This simulates a form of settled annealing. A large amount of mutation noise allows
the problem domains to be widely searched; however it tends to prevent any settling
into minima, resulting in a poor performance in the long run. Using a small amount
of mutation noise results in a very slow progression, requiring a large number of
generations to be simulated. In addition to this, it also tends to result in the process
lodging in a local minima, leading to a poor performance. Starting with a large
amount of mutation noise, then reducing it with the number of generations, results in
good overall performance. In the simulations performed an exponential decay was
used, with a decay time constant of 400 generations. In each of the simulations a
population size of 1000 was used, and 200 of the best sequences were selected to
create the next generation.
Figure 6-16 shows the CF as a function of the number of tones in the signal when
using the GA outlined above. For symbols with more than 12 tones the CF is below
1.4 dB, and for 60 – 350 tones the CF is below 1.0 dB. For greater than 200 tones,
the crest factor is artifically high due to insufficient number of generations simulated,
as a result of the large number of computations required. For up to 56 tones the
number of generations simulated was 5000. Above this, the number decreased
gradulally to 1200 generations for 512 tones. Most of the simulations were
performed on the university supercomputer, with an approximate total simulation
time of 10 CPU months (233 MHz processor). This large amount of simulation time
was required due to the large number of generations and the large number of
different test set as the number of tones in the signal was varied from 2 to 512. The
number of generations needed to approach the minimum possible CF, varies as a
function of the size of the population used, the amount of mutation noise, the starting
phase sequence and the number of sequences kept to make the next generation. This
simulation time could have been significantly reduced had the GA parameters been
more optimised. This prompted the research on parameter optimisation shown in
section 6.2.4.2.
Appendix D contains tables showing the optimised subcarrier phasing.

212
Calculate the CF of the DMT signal

(waveforms shown are the phase of the
DMT carriers)
8.3 dB 4.1 dB 5.5 dB 8.9 dB 5.8 dB
Sort by fitness (CF)

4.1 dB 5.5 dB 5.8 dB 8.3 dB 8.9 dB
Keep only a faction of the best population,

Typically keep 25-70% ( Death)
4.1 dB 5.5 dB 5.8 dB
Randomly pair sequences, back

to the original population
Randomly splice pairs together.

(Sexual Reproduction)
Add small random noise

(Genetic mutations, Asexual Reproduction)
New Generation
Figure 6-15, Genetic Algorithm used to minimise the CF of the DMT

waveforms
213
Evolved Crest Factor verses number of tones
3
2.8
2.6
2.4
Crest Factor (dB)

2.2
2
1.8
1.6
1.4
1.2
1
0.8
0.6
2 3 4 5 6 7 8 10 20 30 40 60 100 200 300 500
Number of tones
Figure 6-16, Best crest factor found using GA verses the number of tones in the
signal. (script s0020)
Cumulative Power (fraction of total power)
Number of tones: 224

1
0.9
0.8
0.7
0.6
0.5
0.4 S.N. T.N

G.A. optimised
0.3
Random
0.2 Flatness Optimised
Newmann
0.1
Shapiro-Rudin
0
-3 -2 -1 0 1 2 3 4 5 6 7 8
Signal Power (dB above mean symbol power)
Figure 6-17, Power distribution of the time waveform for DMT signals
generated using a range of phasing schemes. (script s0055)
S.N. T.N. quadratic phasing scheme (see section 6.2.3 ), Flatness Optimised
details are outlined in section 6.3.
214
6.2.4.2 PARAMETRIC INVESTIGATION OF USING GENETIC

ALGORITHMS FOR CREST FACTOR MINIMISATION
Using genetic algorithms is an effective optimisation technique for crest factor
minimisation. However, the there are many simulation parameters associated with
the GA process that must be set in order to apply the genetic algorithm, such as the
population size, the percentage to cull after each generation, the amount of mutation
noise to add, etc. The choice of these parameters effects the number of generations
required to achieve a good solution, and the quality of the ultimate solution. In order
to address some of these uncertainties, an investigation into the effect of each
simulation parameter was performed.
Figure 6-18, Crest factor distribution as a function of the simulated generations.

(script s0021)
This is for a 26 tone signal. The simulation was repeated 100 times, showing the
variation in the evolution process.
Each generation consists of a population of phase sequences. These phase sequences

are tested to find the crest factor of the resulting DMT signal. Within each generation
the crest factor varies between individual sequences within the population. As the
number of generations simulated increases the overall crest factor reduces due to the
optimisation of the genetic algorithm. The performance of the genetic algorithm was
tested by monitoring the crest factor of the best sequence in each generation,
producing a plot of crest factor as a function of the number of generations simulated.
Repeating the simulation restarts the evolution process, which can result in a
different end solution. This is due to the random searching nature of genetic
215
algorithms. Figure 6-18 shows the crest factor verses the number of generations
simulated. The solution obtained varies from simulation to simulation due to the
random searching nature of genetic algorithms. Many of the simulations do not
obtain a solution that is close to the global minimum. This is due to fact that the
problem domain has many dimensions (1 dimension for each tone) and has many
local minima. The search tends to get lodged in a local minima, thus rerunning the
simulation multiple times tends to find a better solution.
Figure 6-19 shows the effect of changing the population on the performance of the
GA. The results shown are the average of ten repetitions of the GA optimisation. As
the population is increased the rate of optimisation is slower at the start of the GA
optimisation, however after a sufficient number of generations is simulated a better
end result is achieved. Using a large population improves the genetic diversity,
which in turn improves the end solution.
5
Pop = 100
Pop = 600
4.5
Mean Crest Factor (dB)
3.5
2.5
1.5
0 1 2 3
10 10 10 10
Generation
Figure 6-19, Effect of population size for genetic algorithm optimisation of CF,
for 26 tone signal. (script s0021)
The mean CF was found by repeating the evolution simulation 10 times. (Keep
= 50% for next generation)
Figure 6-20 shows the effect of changing the percentage number of the population to
use for breeding of the next generation. Keeping the top 30% of the population for
breeding the next generation results in the fastest progression to a good solution (CF
< 3 dB), however it produced worst results than keeping 50% and 70% after 2000
generations. If too few of the sequences are used from one generation to the next
216
than genetic diversity is reduced resulting in the solution tending to get lodged in
local minima is the optimisation space, rather then continually optimising to the
global minimum.
5
Keep = 30%
Keep = 50%
4.5
Keep = 70%
4
3.5
2.5
1.5
0 1 2 3
10 10 10 10
Generation
Figure 6-20, Effect of percentage population used to breed the next generation
for genetic algorithm optimisation of CF. (population = 100). (script s0021)
5
Seq Noise = 4
Seq Noise = 2
4.5
Seq Noise = 0.5
3.5
2.5
1.5
0 1 2 3
10 10 10 10
Generation
Figure 6-21, Effect of changing the amplitude of the mutation noise for genetic
algorithm optimisation of CF. (script s0021)
Figure 6-21 shows the effect of changing the amount of mutation noise added to the
phase sequences as part of calculating the next generation. For all the GA
simulations performed the level of noise added to the sequences was decayed with
the generation number, so that the level of noise was high at the start of the evolution
and low toward the end on the simulation. In these simulations the level of the noise
217
added at each generation was calculated based on:
Noise = SeqNoise × Generation −0 .6 .
As the mutation noise is increased if increases the genetic diversity and hence the
searching capability of the GA. It prevents the solution getting lodged in local
minima, although too much noise tends to prevent any progression, as the solution
never settles down.
Figure 6-22, Effect of initial phase sequence used for genetic algorithm
optimisation of the CF. (script s0021)
Figure 6-22 shows the effect of changing the initial population of phase sequences.
The random simulation started with an initial population with purely random phase
sequences. The other two tests started with the phase sequence calculated by
Newmann’s phasing scheme (see section 6.2.2 ) and Narahashi & Nojima’s phasing
scheme (see section 6.2.3 ). Newmann’s and Narahashi phasing schemes generate
low CF signals using a quadratic phase equation. Both these schemes perform much
better than using a random phase sequence. This is probably due to the initial
population starting with a result that is close to the global minimum (best solution),
where as when a random phase sequence is used a much larger solution space needs
to be searched before the global minimum is found.
218
6.3 PHASING SCHEMES FOR NULL SYMBOL

SYNCHRONISATION
One method for time synchronising an OFDM signal is to insert a null symbol (zero
power transmitted for a complete symbol period) at the start of each frame. This null
symbol can be detected at the receiver using a relatively simple envelope detector.
This form of time synchronisation is used in the DAB system [20] for its simplicity.
The best way to detect the null symbol is to measure the instantaneous power of the
received signal, and then apply a moving average filter that averages over the null
symbol length. The output of this detector is approximately constant when the
OFDM signal is present, however when a null symbol is received the output drops
will a minimum occurring at the start of the first OFDM symbol in the frame (Figure
6-28 (d) shows a good example of this). The first symbol in the OFDM frame is
usually a reference symbol for channel equalisation for coherent modulation, or as a
phase reference for differential modulation. This symbol contains a known (by the
transmitter and receiver) phase reference on each of the subcarriers, allowing the
receiver to measure the phase rotation and amplitude scaling of the radio channel.
This phase reference does not transmit any data information due to its predetermined
phase sequence.
When using a null symbol for synchronisation, the waveform of the first symbol in
the frame determines the performance of the time synchroniser under fading
conditions. The question is: how do we pick a suitable reference symbol? To make
the reference symbol effective all of the subcarriers should have an equal
transmission power. Ideally this symbol should have a low CF, so that it can be
boosted in power without causing clipping distortion. It should also give a sharp time
response when using a null symbol synchroniser. The synchroniser basically uses
envelope detection, and so we want a symbol that has a consistently flat envelope,
even over short periods of time (fraction of the symbol time). Thus a phasing scheme
is required so that the CF is minimised and the time response is sharp.
One possible candidate is to use a low CF phasing scheme, such as those developed
by S. Narahashi, T. Nojima [106] or by Newmann [105] or the Genetic Algorithm
219
Optimised low CF developed in Section 6.2. All these phasing schemes have very
low CFs, and have a time response that is very similar to a frequency sweep.
Although this type of phase reference has a low CF, its time domain response can
cause a problem in a multipath environment. Frequency selective fading causes the
filtering of the signal, resulting in some frequencies being boosted in power and
others being reduced. The low CF symbols have a frequency sweep time domain
waveform, and as a result the envelope follows the frequency response of the
channel. This causes the envelope of the received symbol to be non-flat, degrading
the performance of the null symbol time synchroniser. An example of this can be
seen in Figure 6-28 (b).
Ideally the reference symbol should have a low CF, but be spectrally flat regardless
of the time interval over which it is viewed. That way the detected envelope will be
flat regardless of the frequency selective fading of the channel. As a starting place it
would seem logical to start with a low CF phasing scheme. These schemes result in a
single frequency sweep over the period of the symbol. If the rate of the sweep could
be increased so that it repeats many times per symbol then the symbol would appear
to be more spectrally flat, as taking a sub-section of the symbol would contain a
complete sweep, covering all frequencies.
Both Newmann’s (equation (6-16)) and S. Narahashi, T. Nojima’s (equation (6-17))

phasing schemes set the phase angle as a quadratic of the subcarrier frequency. If we
take a single carrier and vary the phase in a quadratic manner then the result is a
frequency sweep. This is because the frequency of a carrier corresponds to the rate of
change of the phase. If we take the differential of a quadratic the result is a linear
sweep. The result is similar for a multicarrier signal.
We can approximate both Newmann’s and Narahashi’s phase schemes as a quadratic

that is a function of subcarrier number, multiplied by a scaling coefficient, as shown
in equation (6-18).
θ = Ak 2 (6-18)
where A is a constant coefficient and k is the subcarrier number. For example for a
48-tone signal k would vary from 1 to 48. Varying the coefficient A changes the rate
of the resulting frequency sweep in the generated OFDM signal.
220
6.3.1 SPECTRAL FLATNESS

To measure the effectiveness of the phasing scheme the time spectral flatness and CF
of the resulting symbol was measured. Time spectral flatness was determined by
taking a time overlapping spectrogram of the time waveform then measuring the
peak to average ratio of the spectrogram. To ensure accurate time and frequency
measurements four times interpolation was performed in both the time and frequency
domain. The spectrogram breaks the input signal into time slices then performs an
FFT of the slice to estimate the time varying frequency response. Interpolation in the
time domain was performed generating the OFDM signal at a sample rate four times
the amount required by the nyquist bandwidth of the signal. This was achieved by
zero padding in the frequency domain by adding zeroed subcarriers to the outer
edges of the OFDM signal. After the OFDM signal was converted to the time domain
the signal was analysed using a spectrogram using a sliding FFT. Zero padding was
added to each time slice before the FFT was performed. This resulted in interpolation
in the frequency domain.
Measuring the spectral flatness by looking at the peak to average power ratio of the
spectrogram is effective as it shows the variation in the frequencies generated in each
subsection of the symbol. If the signal was perfectly flat over all time slices then the
peak to average power ratio of the spectrogram would be 0 dB.
The aim of producing a low CF reference signal is so that the power of this symbol
can be boosted without suffering significant additional distortion. The amount of
interference caused by clipping of peaks in the signal is dependent on the height of
the peak and the percentage time the clipping occurred for. For this reason the peak
to average power ratio of the spectrogram was estimated using the 99% percentile of
the waveform distribution instead of the true peak of the spectrogram. To improve
the quality of the measurement, the length of the FFT used in the spectrogram was
varied from half the symbol length to 1/16 the symbol length in four steps. This
effectively changes the time/frequency resolution over a wide range, removing any
assumptions about the time window used. The worst peak to average power ratio for
these four different results was taken as the measured time spectral flatness.
221
In addition to measuring the time spectral flatness, the CF was measured as well, as
the required symbol must perform well in the frequency domain and time domain.
The results are shown in Figure 6-23. The crest factor function is an extremely
complex function, with small changes in the gain coefficient A resulting in large
variations in the crest factor. Both equations (6-16) and (6-17) divide the phase by
the total number of tones in the signal (N). This scales the CF function so that all
peaks align even for different number of tones and gain coefficients. By removing
this division the CF function becomes more random, as can be seen in Figure
6-23(a). The problem with leaving this term in, is that it causes problems with the
time spectral flatness. Leaving it in results in the time spectral flatness function
varying with the number of tones used. By removing it, the function becomes
normalised, making it less randomised. The spectral flatness is the most important
part of the optimisation and so it is more important that the time spectral flatness is
normalised. Although the CF is relatively randomised there are regions that have a
generally lower crest factor. The time spectral flatness is more regular than the CF,
but is also a complex function. The time spectral flatness is fractal in nature in that
the structure is self-similar.
Measured for 15 different carrier sizes, from 16 to 256 Measured for 15 different carrier sizes, from 16 to 256
18 16
Time Spectral Flatness (dB)
16 14
Crest Factor (dB)
14
12
12
10
10
8
8
6 6
4 4
0.5 1 1.5 2 2.5 3 3.5 0.5 1 1.5 2 2.5 3 3.5
P h a s e E q u a t i o n C o e f f . A , P h = A * X2 Phase Equation Coeff. A, Ph = A*X 2
Figure 6-23, (a) Crest factor and (b) time spectral flatness as a function of the
phase equation coefficient A. (script s0056)
The results are for 15 different number tones from 16 to 256. The final equation
has to work effectively for any number of tones, hence the need to analyse
range of combinations.
In order to choose a phasing scheme that would perform well over a wide range of
tones, a fitness criteria was decided based on the CF and the time spectral flatness.
For a particular number of tones the fitness criteria was based on adding one quarter
222
of the CF to the time spectral flatness, where both are in linear scale and not dB. This
fitness was then calculated for a wide range in the number of tones. The mean fitness
of all the tone tests was taken as the estimate of the fitness. Figure 6-24 shows the
fitness criteria, from has a local minimum for A of 3.6315. The performance of the
fitness approximately repeats every π.
To verify the effectiveness of the phasing scheme, the crest factor and time spectral
flatness was measured as a function of the number of tones.
θ = 3.6315k 2 (6-19)
Fitness (lower is better) 10log (CF/4+Flat)
(CF/4+Flat)
Mean Fitness for 15 different carrier sizes, from 16 to 256 Mean Fitness for 30 different carrier sizes, from 16 to 384
10
7.5
10
10
Fitness (lower is better) 10log

7
8
6.5
7 6
6 5.5
5
3.45 3.5 3.55 3.6 3.65 3.61 3.615 3.62 3.625 3.63 3.635 3.64 3.645 3.65
Phase Equation Coeff. A, Ph = A*X2 Phase Equation Coeff. A, Ph = A*X 2
(a) (b)
Figure 6-24, Fitness of the phase scheme for null symbol time synchronisation
as a function of the phase gain coefficient A. (script s0056)
(a) shows a wide view of the fitness, while (b) shows a zoomed up section,
indicating that the local minimum is at a value of 3.6315.
Figure 6-25 shows the CF and time spectral flatness as the number of tones is varied.
The CF varies between 3.3 dB and 6.8 dB, which corresponds to a CF that is
significantly lower than 99.9% of random phase symbols (see Figure 6-4). The time
spectral flatness is very low for signals containing more than 40 tones, which
remains below 3.6 dB up to the testing limit of 2000 tones.
223
Phase = 3.6315*X2 Phase = 3.6315*X 2
6
6.5
Time Spectral Flatness (dB)

Symbol Crest Factor (dB)
6 5.5
5.5
5
5
4.5
4.5
4
4
3.5 3.5
1 2 3 1 2 3
10 10 10 10 10 10
Number of Carriers Number of Carriers
Figure 6-25, Crest Factor and Time spectral flatness as a function of the number
of tones. (script s0056)
Figure 6-26 shows the time waveform and the spectrogram of a 256 tone OFDM signal
using the time spectral flatness optimised phasing scheme. The spectrogram shows
that the signal is very flat and consistent over the entire symbol period. The time
waveform is semi-repetitive, with around 13 similar sections.
Time Spectral Flatness Optimised, Ph=3.6315*X 2

0.04
0.035
0.03
0.025
0.02
0.015
0.01
0.005
0
0 0.2 0.4 0.6 0.8 1
Figure 6-26, Time waveform and spectrogram of a 224 tone OFDM signal using
the time spectral flatness optimised phasing scheme. (script s0055)
6.3.2 EXPERIMENTAL VERIFICATION

An experiment was performed to verify the effectiveness of the time spectral flatness
optimised phasing scheme for use in null symbol time synchronisation. The
performance was compared with a synchroniser using Narahashi low CF reference
symbols (see section 6.2.3 ). The performance of this time synchronisation method is
affected by multipath, as this filters the received signal, which can result in a non-
uniform signal envelope. The amount of non-uniformity of the envelope will depend
on the time spectral flatness of the symbols immediately before and after the null
224
synchronisation symbol.
Figure 6-27 shows the test set up. An OFDM generator was implemented on a
SHARC DSP development board. The generated signal consisted of a 192 subcarrier
OFDM signal using a bandwidth of 18 kHz, with a null symbol inserted every 30
symbols for time synchronisation. A real output signal was generated using a 512-
point IFFT with a guard period of 64 samples.
The receiver’s DSP performed time synchronisation on the signal. The receiver
generated two output results, one being the null symbol time synchroniser and the
other a straight though, showing the signal as seen by the receiver. The null symbol
synchroniser consisted of taking the magnitude of the input signal and applying a
moving average filter equal in length to the null symbol, corresponding to the IFFT
size plus the guard period (512+64 = 576 samples). The resulting signal is high
during the presence of the OFDM signal, changing to a 'V' shape during the null
symbol. The minimum of this 'V' corresponds to the start of the OFDM frame. In
order for a receiver to use this synchronisation signal, further processing would need
to be applied to detect the minimum in the 'V'.
Test 1
SHARC SHARC
ADSP-21061 ADSP-21061
EZ-KIT EZ-KIT
DSP Test 2 DSP
OFDM Generator OFDM Receiver/
Synchroniser
Talk Through Sync Signal
Tektronix
Digital CRO
Figure 6-27, Experimental set up for investigating the performance of the time
synchronisation method.
Test 2 is a straight through test. Test 1 shows the performance with simulated
channel filtering.
Test 2 showed the performance of the synchroniser with no detrimental effects. Test
1 applied a low pass filter to the transmitted OFDM signal. The signal was filtered
225
with a 3rd order passive filter to simulate a distorted frequency response. This filter
was a highly resonant low pass with a 6 dB peak just before the cut off frequency.
The filter removed approximately 50% of the signal bandwidth. This filtering
represents an extreme example of multipath frequency selective fading.
Figure 6-28 (a) – (c) show measured results for time synchronisation when using the
S. Narahashi, T. Nojima phasing scheme. When no filtering was applied the
performance is adequate, with the time synchroniser forming a reasonably sharp 'V'.
The problem occurs when filtering is applied since this results in the signal envelope
changing dramatically due to the frequency sweep nature of the signal. The envelope
in fact tracks the frequency response of the filter. The resulting time synchronisation
waveform is very poor, with no clear minimum occurring at the start of the frame. If
this signal had a higher amount of noise the performance would be very poor.
Figure 6-28 (d)-(f) shows the same test using the phasing scheme developed. The 'V'
formed by the synchroniser is very sharp with no filter, and remains sharp even when
subjected to the low pass filtering. This verifies that the phasing scheme performs
very well regardless of channel filtering.
226
S.N. T.N Phasing Scheme Phasing scheme developed
(k − 1)( k − 2) θ = 3.6315k 2
θ= π
( N − 1)
(a) (d)
No filtering No filtering
(b) (e)
Filtered response Filtered response
(c) (f)
Zoom up for the filtered response Zoom up for the filtered response
Figure 6-28, Measured time synchronisation accuracy for different reference
symbols.
The bottom trace shows the OFDM signal, and the top signal is the frame
detection signal. The actual time synchronisation is found by finding the
minimum of this waveform. The filtered response was for a passive low pass
filter removing 50% of the OFDM spectrum.
227
Chapter 7 Crest Factor Minimisation of OFDM Data Symbols
CHAPTER 7 CREST FACTOR

MINIMISATION OF OFDM DATA
SYMBOLS
A technique is described for a better than 5 dB reduction in the crest factor for data
carrying OFDM signals. This improvement is achieved by inserting additional
subcarriers into the signal referred to as peak reduction carriers (PRC). These PRC
are set in phase and amplitude, using a codebook, to minimise the overall symbol CF.
The codebook is obtained using a search of all possible signal combinations. This
technique is optimised for signals using a low number of subcarriers, and low
modulation schemes such as BPSK and QPSK, due to the brute force searching
technique employed.
7.1 BACKGROUND
The CF of an OFDM signal can be reduced in several ways. Selective mapping

[108], [109] involves generating a large set of data vectors all representing the same
information. The data vector with the lowest resulting CF is selected. Information
about which particular data vector was used is sent as additional subcarriers.
However, there may be potential problems with decoding the signal in the presence
of noise with selective mapping. Errors in the reverse mapping would result in whole
symbols being lost.
Another technique similar to selective mapping is to use Golay sequences [113].

Information is transmitted by mapping each data word with a Golay sequence. Using
Golay sequences result in a low CF, typically 3 - 6 dB, however the coding rate is
poor, typically half, resulting in a large bandwidth increase. Cyclic coding [110]
228
involves adding extra subcarriers in which the phase of every fourth subcarrier is
calculated based on an algorithm using the phase of the previous three information
subcarriers. This method is similar to the presented technique except that it gives sub
optimal results.
7.2 PEAK REDUCTION CARRIERS
This chapter presents a technique that combines selective mapping and cyclic coding.
A reduction in the CF is achieved by adding extra subcarriers referred to as Peak
Reduction Carriers (PRC). The phase and amplitude of the PRCs is varied to
minimise the overall CF. The original information subcarriers are unaffected and can
be decoded normally. The receiver can disregard the PRCs, or they can be used for
error detection. The frequency of PRCs, or relative positioning of the PRCs can be
varied with respect to the information subcarriers depending on the application. The
results presented were found using a computationally intensive exhaustive search to
find the optimal setting for the PRCs. However it is assumed that further work will
allow a more efficient algorithm to be found.
An optimal setting for the PRCs corresponds to the combination of phase and
amplitude that achieves the lowest CF of the overall OFDM symbol (information
subcarriers and PRCs). In this paper the phase and amplitude of the PRCs was set in
a coarse quantised manner to minimise the number of combinations needed to be
searched. The phase of the PRCs was set to 0° or 180° and the subcarriers were
turned on or off. There are therefore 3M combinations for the PRCs for each
information code word (where M is the number of PRCs). This level of quantisation
was found to be appropriate for BPSK information subcarriers. Finer quantisation
may produce improved results for higher modulation schemes.
An exhaustive search of all combinations of allowable phase and amplitude gives

optimal PRCs, but is computationally intensive. This method can be used for small
numbers of subcarriers where the optimal PRC coding can be stored in a look up
table or code-book. This is impractical for more than 16 information subcarriers or
for more than 10 PRCs as the number of combinations becomes too large to store and
calculate. However for some multi-user OFDM applications 16 or less subcarriers
229
per user is sufficient.
The results shown were calculated based on all combinations of information code
words, thus will give a good indication of the practical CF improvement.
For each experiment the inverse fast Fourier transform (IFFT) of the subcarrier
configuration was used to give a complex base band signal.
For the simulations carried out, the base band subcarriers were centred on DC and
the size of the IFFT was made at least 8 times greater than the number of subcarriers,
oversampling the time domain signal. This ensures that peaks in the signal were
accurately represented to get an accurate CF [111], [112].
Figure 7-1, CF verses number of edge grouped PRCs (8 BPSK data subcarriers)
7.3 RESULTS
The simplest arrangement for the relative positioning of the data and PRCs is to have
a block of data subcarriers immediately followed by a block of PRCs. This
arrangement was used for the results shown in Figure 7-1 through to Figure 7-3.
Figure 7-1 shows the worst-case CF and the 90% point in the cumulative distribution
of CF as the number of PRCs is increased. The maximum CF for the 8 information
subcarriers is reduced by > 5.5 dB for the addition of 10 PRCs. Selecting the optimal
amplitude and phase of the PRC improves the performance significantly as compared
230
with only setting the phase as used in cyclic coding [110]. For this reason phase and
amplitude modulation of the PRCs was used in all later experiments due to the
improved performance.
Figure 7-2, Maximum CF verses number of data subcarriers and edge grouped
PRCs, where M is the number of PRCs.
Figure 7-2 shows the effect of adding PRCs to the CF as the number of information
subcarriers is varied. The improvement in CF remains relatively constant as the
number of information subcarriers is increased. This shows that this technique gives
consistent performance gains as the number of information subcarriers is varied.
Adding PRCs reduces the CF at the expense of additional transmission power and
bandwidth. Figure 7-3 shows the net improvement in CF due to the addition of
PRCs. The CF reduction was calculated as the difference between the CF results for
zero PRCs and the CF results with the addition of PRCs. The loss in signal power
due to the PRCs was then subtracted from the CF reduction in order to give the net
CF improvement. If the data signal power lost due to the transmission of the PRCs
was more than the CF gain then there would be little point in adding the PRCs. It can
be seen that for 10 BPSK subcarriers there is little improvement in adding more than
2 PRCs. In fact adding more than 5 PRCs results in a worsening of the average
(50%) CF. This is due to the power cost of the PRCs.
231
Figure 7-3, Net CF improvement verses number of edge grouped PRCs. (10
BPSK data subcarriers)
7.4 EFFECT OF PRC POSITION
Previous results are shown for grouped PRCs that were positioned immediately after
the data subcarriers as shown in Figure 7-4 (a). Two different positioning tests were
performed. One test kept the PRCs grouped together as in Figure 7-4 (a), however
they were moved with respect to the data subcarriers as in Figure 7-4 (b). The second
test positioned the PRCs in a spread out manner. The best spread pattern was
established using a randomised search.
Edge Grouped PRCs

(a)
Grouped PRCs Position = 0

(b)
External Grouped PRCs
Position = 7
(c)
Spread PRCs
(d)
1 2 3 ... . . . 13 14
Frequency
Data Carrier PRC
Figure 7-4, PRC position combinations
232
Adding PRCs use a significant amount of additional bandwidth. It is therefore

important to minimise the number used, or to position the PRCs so that the
bandwidth can be reused. For example, in a multi-user OFDM system where each
user transmits a block of subcarriers, the PRCs can be overlapped, i.e. they are
transmitted at the same frequency, effectively halving the bandwidth used by the
PRCs.
Figure 7-5, CF verses position of 2 grouped PRCs (10 BPSK data subcarriers)
Figure 7-6, CF verses position of 4 grouped PRCs (10 BPSK data subcarriers)
233
7.5 GROUPED PRCS
In this scheme the PRCs were maintained as a group of subcarriers. They were
repositioned by sliding them with respect to the data subcarriers. Figure 7-5 and
Figure 7-6 show the effect of the position on effectiveness of the PRCs. Figure 7-5
shows that for a small number of PRCs the performance is not particularly dependent
on the position of the PRCs within the data subcarriers. However with 4 or more
PRCs the position has a significant effect on the performance of the PRCs. Placing
the PRCs within the data subcarriers with an off centre of 3 subcarriers gives the best
results. This gives a further reduction of 1 dB as compared with edge grouped PRCs.
Not having edge grouped PRCs prevents the overlapping of the PRCs from different
users in a multi-user OFDM system. As a consequence the bandwidth used by the
PRCs will be effectively doubled.
7.6 SPREAD PRCS
The position of the 4 grouped PRCs had a significant effect on the PRC performance,
thus it seemed likely that spreading the PRCs out might lead to further
improvements. The exact relationship between the position of the PRCs and the CF
distribution is currently unknown and so a random search was used for optimisation.
The PRCs and data subcarriers were positioned randomly to form a block of
subcarriers with no gaps as shown in Figure 7-4 (d). For each position combination
the CF distribution was found and the combination that resulted in the lowest
maximum CF was selected as the optimised PRCs position.
234
Figure 7-7, CF verses the number of spread PRCs (10 BPSK data subcarriers)
The CF distribution was found by testing all combinations of the data code words.
For each data code word combination the optimum PRCs were found as described in
section 3. The CF distribution verses the number of PRCs is shown in Figure 7-7.
This result is for 10 data subcarriers and shows that spreading the PRCs can result in
large reductions in the CF of the OFDM symbol. A reduction of greater than 6 dB is
possible.
Figure 7-8 shows the overall net improvement in the CF using position optimised
PRC. This can be directly compared to Figure 7-3 which shows the results for edge
grouped PRCs. The maximum net gain for position optimised PRCs is approximately
2 dB better than that of the edge grouped PRCs. Figure 7-8 shows that the net CF
gain increases rapidly up to 4 PRCs, after which the gain is minimal. Thus the
optimal number of PRCs would be 4 for 10 data subcarriers. Other tests also show
that the number of PRCs needs to be approximately 40% of the number of data
subcarriers in order to get significant improvements in the CF.
235
Figure 7-8, Net improvement in CF, position optimised PRCs (10 BPSK data
subcarriers)
Overlapping of the spread PRCs in a multi-user OFDM system is more difficult as

most of the PRCs will be bounded with data subcarriers. Thus simple overlapping
may not be possible. This would be the case if the data and PRC are grouped.
However if the subcarriers for each user are spread out, it might be possible to have
spread out PRCs that overlap between the users, but still provide a large reduction in
the CF.
The addition of 4 PRCs with 10 data subcarriers results is a large net gain of 4.5 dB,
allowing more power to be transmitted. For a transmission with no PRCs at an error
rate of 1×10-3 , adding the PRCs and maintaining the same peak power the error rate
would be decreased to 1×10-7 .[117], [118] This is more efficient than adding simple
error correcting bits at the same coding rate. For example Hamming coding at a rate
of 4 parity bits for 11 data bits gives a gain of only 1.2 dB at a bit error rate of 1×10-4
which is significantly less than 4.5 dB.
Table 1 shows the number PRC positions tested, and the best combination found.
236
No. Position Possible Best Position

PRCs tests position Found, see fig. 4d
simulated combinations for label method
1 11 11 1
2 66 66 2,3
3 286 286 3,4,6
4 262 1001 4,10,12,13
5 156 3003 1,3,7,10,15
6 68 8008 1,2,7,10,14,16
7 36 19448 1,2,7,9,11,14, 17
8 23 43758 1,2,3,4,11,12,16,18
Table 7-1, Optimised positions found for 10 BPSK data subcarriers
7.7 CONCLUSION
Adding peak reduction subcarriers can significantly improve the CF of an OFDM

signal. The PRCs can result in a reduction of greater than 6 dB in the maximum CF
and a net reduction of greater than 4.5 dB when the additional power for PRCs is
taken into account. It was found that varying the amplitude as well as phase for the
PRCs gave improved performance over just phase variation. It was also found that
spreading the position of the PRCs resulted in a better performance than grouped
PRCs. Adding more PRCs results in a lower CF, however the use of large numbers
of PRCs is limited by the cost of additional transmission power, bandwidth and
complexity limits.
237
Chapter 8 Summary
CHAPTER 8 SUMMARY
8.1 FUTURE WORK
OFDM has a promising future for use in wireless LAN networks and for fixed and
mobile communications. Its use is likely to be significantly increased over the next
10 years, due to its high spectral efficiency and flexibility in spectrum allocation and
modulation. OFDM is mostly suited to low mobility high data rate networks, making
it difficult to apply to mobile phone networks. At low user data rates, much of the
adaptive techniques would not be suitable due to the high overhead of implementing
such techniques. As a result, in multiuser applications at low data rates the overall
efficiency of OFDM is much poorer with little advantage over CDMA techniques.
Mobile networks of the future will shift away from low data rate voice applications
to become primarily data terminals, striving for high data rates.
If OFDM is to be used in 4th generation mobile networks, a significant amount of

research needs to be done in applying OFDM to cellular networks. Issues such as
inter-cellular interference need to be studied, in the context of identifying the best
frequency / time user allocation scheme. Ideally the networks should be self-
organising in frequency reuse, and so an effective means of implementing this needs
to be developed. Most current multiuser OFDM systems use TDMA as a method for
providing multiple accesses. However the greatest performance gains can be made
by jointly allocating users based on frequency and time. Due to the overlapping
nature of subcarriers in OFDM signals, allocating users different subcarriers in the
reverse link creates problems in maintaining subcarrier orthogonality. Each mobile
user will have a different propagation delay and Doppler spread, making
compensation for these factors in the reverse link a non-trivial problem. It is possible
that the most robust schemes will use blocks of OFDM signals, which are band pass
filtered to help separate signals from different users.
238
Chapter 8 Summary
The effective use of adaptive modulation and adaptive user allocation will depend on
the overhead required for implementation. Protocols need to be developed that will
allow these techniques to be implemented efficiently. Research is needed to
investigate the trade off between overhead and technique efficiency, so the optimal
amount of overhead for each application can be established. Research is also needed
for optimisation of the adaptive user allocation algorithm so as to satisfy the
requirements of the different user in the system. This includes optimising so that the
system throughput and quality of service is maximised. The optimisation will need to
cope with variable and fixed data rate services. Additionally the effect of multi-
cellular user allocation needs to be investigated.
Much of the development of effective allocation schemes will be dependent on

improved modelling of wideband RF propagation. Much of the focus on propagation
modelling has been on path loss modelling, statistical descriptions of fading, and
discrete approximations of channel impulse using ray tracing. Ray tracing methods
are improving rapidly, with much of the focus being on computation efficiency, and
accurate prediction of path loss. Research needs to be done on wideband models that
show the response of the radio channel with small changes in space, similar to the
measurements presented in this thesis. These models need to be smooth in space to
allow testing of multiuser user allocation schemes in terms of subcarrier SNR, the
effects of Doppler spread, and propagation time. With accurate models it may be
possible to develop a technique for predicting the channel response for a small time
ahead, allowing adaptive techniques to cope with higher user mobility.
With the advent of OFDM, higher spectral efficiency modulation schemes (16-QAM,
64-QAM, etc) are now being used in terrestrial wireless networks. These modulation
schemes are more susceptible to the effects of ISI caused by multipath propagation
than previously used low spectral efficiency modulation schemes such as BPSK and
QPSK. The use of OFDM helps to reduce the level of ISI by using a low symbol rate,
and the use of a guard period. The guard period represents a significant time
overhead and so should be minimised in length. The most suitable length depends on
the multipath in the environment and the modulation scheme used. Traditionally the
length of the required guard period has been estimated using the RMS delay spread
of the propagation channel. This measurement is accurate for predicting the guard
239
Chapter 8 Summary
period length required for BPSK and QPSK modulation schemes, but si inadequate
for estimating the performance of high spectral efficiency modulation schemes. The
estimated guard period length is too short for these modulation schemes, resulting in
significant remaining ISI, resulting in degraded BER. The delay spread of indoor
environments needs to measured using a more appropriate multipath estimate than
the RMS delay spread. Such a measurement might be, the delay spread over which a
given percentage of the impulse energy arrives. For example if the length of the
guard period matches the time over which 99.9% of the impulse energy arrives, then
the ISI should only be 0.1%, resulting an effective SNR of 30 dB, which is suitable
for 64-QAM. A set of propagation experiments is needed using such a multipath
measurement so the performance of OFDM systems using higher modulation
schemes can more accurately be assessed.
This thesis presented the use of Access Point Repeaters as a method for low cost
maximisation of signal strength in indoor environments. This technique could be
extended to cellular networks as a method for controlling shape of cells, and for
minimising shadowing within a cell. The effectiveness of using this technique in a
cellular environment depends on the cost of installing extra base stations as
compared with APRs, and with potential problems with combining APR with smart
antenna technology. Research needs to be done to establish the amount of additional
multipath created by using APRs, and the practical effectiveness of the technique.
Much work has been done on minimising the crest factor of data carrying OFDM
signals. The high CF of OFDM signals requires a moderate power back off (6 - 9 dB)
in the transmitting power amplifiers. However OFDM is very efficient in terms of
the amount of power required for transmission, making this less significant. Most of
the focus on minimisation techniques has been on minimising the worse case CF.
However with sufficient data whitening the probability of very large CF symbols is
so low that this is not a problem. Focus needs to be placed on CF minimisation
techniques that will reduce the output power back off required. Currently published
CF minimisation techniques only give minimal overall improvement, with most
requiring large overheads in either complexity or throughput. The potential
improvement for CF minimisation techniques is relatively small (2 – 4 dB) and so
overhead in implementation must be small to justify its use.
240
Chapter 8 Summary
8.2 CONCLUSION
This thesis has presented an investigation into the application of OFDM in multiuser
systems, and has focused on techniques for improving the system spectral efficiency.
Most current communication systems operate at a very low spectral efficiency,
resulting in under utilisation of the radio spectrum. Future systems will have to
improve the spectral efficiency to achieve the capacities required. This thesis has
investigated techniques that exploit the flexibility of OFDM to maintain a maximum
spectral efficiency, by matching the system parameters, such as subcarrier
modulation and frequency, based on current conditions of the radio channel. The aim
of this work is for the communication system to approach the maximum theoretical
channel capacity given the constraints of the radio channel, transmitter power and
quality of service. While many cellular systems are primarily limited by inter-cellular
interference, this thesis has primarily focused on the performance of isolated
communication networks. This work is thus appropriate for applications such as
Wireless LAN systems, which tend to be partially shielded from interference by the
outer walls of indoor environments.
With bi-directional communications, information from the receiver can be used to

adaptively exploit the radio channel characteristics. The application of adaptive
modulation was investigated and found to provide an effective means to mitigate the
effects of fading. Adaptive modulation greatly decreases the BER, with it requiring
12 - 16 dB less SNR to achieve the same BER as compared with a fixed modulation
system. In addition to this, it allows the data rate of the system to improve when the
channel SNR is high. When using coherent QAM, every 3 dB improvement in SNR
allows the spectral efficiency of the system to improve by 1 b/s/Hz. Implementation
of adaptive modulation requires some information overhead as both the transmitter
and receiver must know the modulation scheme currently being used on each
subcarrier, and the subcarrier modulation must be updated to track the radio channel.
The required tracking rate for mobile systems employing adaptive modulation was
found to be in the order of 0.05 - 0.1λ, with the actual required rate depending on the
required BER, the amount of link margin used in the modulation allocation, the delay
in the allocation, and the subcarrier allocation scheme. This technique is suited to
241
Chapter 8 Summary
pedestrian velocities as the overhead in modulation update becomes excessive at high

velocities. In a mobile application at 5.2 GHz and at a velocity of 50 km/hr, a
tracking rate of 2400 - 4800 Hz is required. This would result in an information
overhead in the order of 5 - 10% for a system similar to HiperLAN2.
Adaptive User Allocation was presented as a method for optimising the subcarrier
frequency allocation in multi-user applications. This technique exploits the variation
in channel response between users, caused by frequency selective fading. This
technique also reduces the effects of fading as users tend to be allocated strong
subcarrier frequencies, with nulls in the frequency response of the channel being
avoided. This technique can be used in conjunction with adaptive modulation, to
provide an additional signal power gain of around 3 - 5 dB. This technique requires
complete knowledge of the channel response to each user, and so implementation
requires significant overheads, making this technique only suitable for relatively
fixed wireless applications such as wireless LAN systems.
This thesis has also presented a method for maximising the signal strength within
buildings, by using transmission repeaters. The implementation complexity and cost
is low, as multiple repeaters are distributed from a single base station. This technique
is similar to Single Frequency Networks used in DAB and DVB, except at a much
smaller scale. This technique increases the multipath in most environments, but this
is compensated for by the high multipath tolerance of OFDM. The performance of
using Access Point Repeaters (APR) was measured by finding the reduction in path
loss in a simulated WLAN system using a simple ray trace model. This was verified
against direct measurement of the building that was simulated. It was found to
decrease the path loss by 7 dB for a system with two APRs, and up to 20 dB for eight
APRs.
In addition, two techniques were presented for reducing the Crest Factor (peak to
average power ratio of the RF signal envelope) of OFDM signals.
The first technique is a phasing scheme for OFDM pilot symbols, which uses genetic
algorithms to optimise the phase angle of each subcarrier to achieve very low CF
waveforms. This technique achieves a lower CF than any previously published
techniques, obtaining a CF as low as 0.65 dB, which is 2 dB lower than commonly
242
Chapter 8 Summary
used techniques. These low CF symbols can be used for reference pilot symbols. The
low CF of these symbols allows the average power to be boosted (up to 6 dB) while
maintaining a peak power significantly lower than the data symbols. This boosted
power reduces the noise in the channel equalisation, improving the system
performance.
The second technique reduces the CF of data carrying symbols, by including

additional subcarriers that are optimised in amplitude and phase to cancel out the
peaks in the overall OFDM symbol. This was found to produce a net improvement of
4 dB to the worst-case symbol CF.
243
References
REFERENCES
UTMS AND THIRD GENERATION SYSTEMS

[1] K. W. Richard, “UMTS overview”, IEE Electronics and Communication
Engineering Journal, Vol. 12, No. 3, June 2000, pp. 93 - 100
[2] Ermanno Berruto, Giovanni Colombo, Pantelis Monogioudis, Antonella Napolitano,

Kyriacos Sabatakakis, “Architectural Aspects for the Evolution of Mobile
Communications Toward UMTS”, IEEE Journal on Selected Areas in
Communications, Vol. 15, No. 8, October 1997,pp. 1477 - 1487
[3] Erik Dahlman, Björn Gudmundson, Mats Nilsson, Johan Sköld, “UMTS/IMT-2000
Based on Wideband CDMA”, IEEE Communications Magazine, September 1998,
pp. 70 - 80
[4] Fumiyuki Adachi, Mamoru Sawahasi, Hirohito Suda, “Wideband DS-CDMA for
Next-Generation Mobile Communications Systems”, IEEE Communications
Magazine, September 1998, pp. 56 - 69
[5] Josef Huber, “3G Forum Leveraging Spectum Capacity In Accordance with 3G
Deployment”, 3G Forum 17th -18th April 2001, Washington D. C., Online:
http://www.umts-forum.org/ipapers/Leveraging_Spectrum_Capacity.pdf
[6] Douglas N. Knisely, Sarath Kumar, Subhasis Laha, Sanjiv Nanda, “Evolution of
Wireless Data Services: IS-95 to cdma2000”, IEEE Communications Magazine,
October 1998, pp. 140 - 149
[7] “Third Generation Mobile Phone Licensing in Europe”, TIA on line, Online:
http://www.tiaonline.org/international/regional/nis/licensing.cfm, October 2000.
[8] “Auction of Third Generation Mobile Telecommunications Licences in the UK,

Frequently Asked Questions”, Radiocommunications Agency of UK, Online:
http://www.spectrumauctions.gov.uk/documents/faq2.htm,
244
References
[9] “Spectrum Auctions”, Radiocommunications Agency of UK, Online:
http://www.spectrumauctions.gov.uk/auction/auction_index.htm
[10] S. Dehghan, D. Lister, R. Owen, P. Jones, “W-CDMA capacity and planning issues”,
IEE Electronics and Communications Engineering Journal, Vol. 12, No. 3, June
2000, pp. 101 - 118
[11] Sanjiv Nanda, Krishna Balachandran, and Sarath Kumar, “Adaptation Techniques in
Wireless Packet Data Services”, IEEE Communications Magazine, January 2000, pp.
54 - 64
4TH GENERATION SYSTEMS

[12] Terje Tjelta, Agne Nordbotten, Marco Annoni, Enrico Scarrone, Simone Bizzarri,
Laurissa Tokarchuk, John Bigham, Chris Adams, Ken Craig, Manuel Dinis, “Future
Broadband Radio Access Systems for Integrated Services with Flexible Resource
Management”, IEEE Communications Magazine, August 2001, pp. 56 - 63
[13] NTT DoCoMo Advertisement, “The Path to 4G Mobile”, IEEE Communications

Magazine, March 2000, pp. 38 - 41
DIGITAL VIDEO BROADCASTING

[14] “Digital Video Broadcasting”, ETSI, Online: http://www.etsi.org/broadcast/dvb.htm,
October 2000
[15] “A Guideline for the use of DVB specifications and standards”, DVB Blue Book
A20, pp. 9, Online: www.dvb.org/news/pdf/dvb_cook.pdf
[16] ETSI EN 300 421, “Framing Structure, channel coding and modulation for 11/12
GHz satellite services”, August 1997, pp. 6, Online: http://www.etsi.org
[17] Ulrich Reimers, “Digital Video Broadcasting”, IEEE Communications Magazine,

June 1998, pp. 104 - 110
[18] ETSI ETS 300 744, “Framing Structure, Channel Coding and Modulation for Digital
Terrestrial Television”, March 1997, Online: http://www.etsi.org
DIGITAL AUDIO BROADCASTING

[19] Louis Thibault, Minh Thien Le, “Performance Evaluation of COFDM for Digital
Audio Broadcasting Part I: Parametric Study”, IEEE Transactions on Broadcasting,
Vol. 43, No.1, March 1997, pp. 64 – 75
245
References
[20] “Digital Audio Broadcasting - Overview and Summary of the DAB System”, World
DAB Forum,
Online: http://www.worlddab.org/public_documents/eureka_brochure.pdf
[21] ETSI EN 300 401, “Radio Broadcasting Systems; Digital Audio Broadcasting (DAB)
to mobile, portable and fixed receivers”, May 2001, Online: http://www.etsi.org
GSM NETWORKS
[22] Michel Mouly, Marie-Bernadette Pautet, “The GSM System for Mobile
Communications”, M. Mouly et Marie-B. Pautet, France, 1992, ISBN: 2-9507190-0-
7
[23] Siegmund Redl, Matthias Weber, Malcom Oliphant, “GSM and Personal
Communications Handbook”, Artech House Publishers, Norwood, 1998, ISBN: 0-
89006-957-3
[24] Telstra, “Flexi-Plans”,
Online: http://www.telstra.com.au/mobilenet/flexipln/dflex.cfm, November 2001
[25] Optus, “Rate Plans”, Online: http://optusdirect.optus.com.au/shop/rateplans,

November 2001
CDMA OVERVIEW
[26] William C. Y. Lee, “Overview of Cellular CDMA”, IEEE Transactions on Vehicular
Technology, Vol. 40, No. 2, May 1991, pp. 291 - 302
[27] Ryuji Kohno, Reuven Meidan and Laurence B. Milstein, “Spread Spectrum Access
Methods for Wireless Communications”, IEEE Communications Magazine, January
1995, pp. 58 - 67
OFDM
[28] Yiyan Wu, William Y. Zou, “Orthogonal Frequency Division Multiplexing: A Multi-
Carrier Modulation Scheme”, IEEE Transaction on Consumer Electronics, Vol. 41,
No. 3, August 1995, pp. 392 - 399
[29] William Y. Zou, Yiyan Wu, “COFDM: An Overview”, IEEE Transactions on

Broadcasting, Vol. 41, No. 1, March 1995, pp. 1 - 8
[30] R. R. Mosier and R. G. Clabaugh, “Kineplex, a bandwidth-efficient binary

transmission system”, AIEE Transactions, Vol. 76, January 1958, pp. 723 - 728
246
References
[31] Robert Chang, “Synthesis of Band-Limited Orthogonal Signals for Multichannel
Data Transmission”, The Bell System Technical Journal, December 1966, pp. 1775 -
1796
[32] Robert Chang, “Orthogonal frequency division multiplexing”, US. Patent 3,488445,
filed November 14, 1966, issued January 6, 1970
[33] S. B. Weinstein, Paul M. Ebert, “Data Transmission by Frequency-Division

Multiplexing Using the Discrete Fourier Transform”, IEEE Transactions on
Communication Technology, Vol. COM-19, No. 5, October 1971, pp. 628 - 634
[34] Carl Magnus Frodigh, Perols Leif Mikael Gudmundson, “Adaptive channel
allocation in a frequency division multiplexed system”, US. Patent 5,726,978, Filed:
June 22, 1995, Issued: March 10, 1998
SYNCHRONISATION FOR OFDM

[35] J. Beek, M. Sandell, P. Borjesson, “ML Estimation of Time and Frequency Offset in
OFDM Systems”, IEEE Transactions on Signal Processing, Vol. 45, No. 7, July
1997, pp. 1800 - 1805
[36] P. Moose, “A Technique for Orthogonal Frequency Division Multiplexing Frequency

Offset Correction”, IEEE Transactions on Communications, Vol. 42, No. 10,
October 1994, pp. 2908 - 2914
[37] L. Wei, C. Schlegel, “Synchronization Requirements for Multi-user OFDM on

Satellite Mobile and Two-path Rayleigh Fading Channels”, IEEE Transactions on
Communications, Vol. 43, No. 2/3/4, February/March/April 1995, pp. 887 - 895
[38] Meng-Han Hsieh, Che-Ho Wei, “Channel Estimation for OFDM systems based on
Comb-Type Pilot Arrangement in Frequency Selective Fading Channels”, IEEE
Transactions on Consumer Electronics, Vol. 44, No. 1, February 1998, pp. 217 - 225
[39] Ye Li, Nambirajan Seshadri, Sirikiat Ariyavisitakul, “Channel Estimation for OFDM
Systems with Transmitter Diversity in Mobile Wireless Channels”, IEEE Journal on
Selected Area in Communications, Vol. 17, No. 3, March 1999, pp. 461 - 471
[40] Marco Luise, Ruggero Reggiannini, “Carrier Frequency Acquisition and Tracking of
OFDM Systems”, IEEE Transactions on Communications, Vol. 44, No. 11,
November 1996
247
References
[41] Y. H. Ng, P. M. Grant, R. A. Stirling-Gallacher, “Carrier tracking techniques for
OFDM signal transmissions”, Electronic Letters, Vol. 32, No. 22, 24th October 1996,
pp. 2047 - 2048
[42] T. M. Schmidl, D. C. Cox, “Blind synchronization for OFDM”, Electronic Letters,

Vol. 33, No. 2, 16th January 1997, pp. 113-114
[43] Uwe Lambrette, Michael Speth, Heinrich Meyr, “OFDM Burst Frequency
Synchronisation by Single Carrier Training Data”, IEEE Communication Letters,
Vol. 1, No. 2, March 1997
[44] Jean Armstrong, “Analysis of New and Existing Methods of Reducing Intercarrier
Interference Due to Carrier Frequency Offset in OFDM”, IEEE Transactions on
Communications, Vol. 47, No. 3, March 1999, pp. 365 - 369
[45] Hui Liu Ufuk Tureli, “A High-Efficiency Carrier Estimator for OFDM
Communications”, IEEE Communications Letters, Vol. 2, No. 4, pp. 104 - 106
[46] M. J. Fernández-Getino, S. Zazo, J. M. Páez-Borrallo, “Pilot patterns for channel

estimation in OFDM”, Electronic Letters, Vol. 36, No. 12, 8th June 2000, pp. 1049
[47] William Warner, Cyril Leung, “OFDM/FM Frame Synchronization for Mobile Radio
Data Communication”, IEEE Transactions on Vehicular Technology, Vol. 42, No. 3,
August 1993, pp. 302 - 313
[48] J. Ahn, H. S. Lee, “Frequency Domain Equalisation of OFDM signals over

Frequency Nonselective Rayleigh Fading Channel”, Electronics Letters, Vol. 29, No.
16, 5th August 1993, pp. 1476 - 1477
[49] Marco Luise, Ruggero Reggiannini, Giorgio Vitetta, “Blind Equalization/Detection

for OFDM Signals over Frequency-Selective Channels”, IEEE Journal on Selected
Areas in Communications, Vol. 16, No. 8, October 1998, pp. 1568 - 1577
[50] Donghoon Lee, Kyungwhoon Cheun, “A New Symbol Timing Recovery Algorithm
for OFDM systems”, IEEE Transactions on Consumer Electronics, Vol. 43, No. 3,
pp. 767 - 775
[51] Ufuk Tureli, Didem Kivanc, Hui Liu, “Experimential and Analytical Studies on a
High-Resolution OFDM Carrier Frequency Offset Estimator”, IEEE Transactions on
Vehicular Technology, Vol. 50, No. 2, March 2001, pp. 629 – 643
248
References
[52] Yang-Seok Choi, Peter J. Voltz, Frank A. Cassara, “ML Esimation of Carrier
Frequency Offset for Multicarrier Signals in Rayleigh Fading Channels”, IEEE
Transactions on Vehicular Technology, Vol. 50, No. 2, March 2001, pp. 644-655
INFORMATION THEORY
[53] C. E. Shannon, “A Mathematical Theory of Communication”, 1949, Online:
http://www.math.psu.edu/gunesch/Entropy/shannon.ps
EFFECT OF DISTORTION ON OFDM

[54] Minh T. Le, Louis Thibault, “Performance Evaluation of COFDM for Digital Audio
Broadcasting Part II: Effects of HPA Nonlinearities”, IEEE Transactions on
Broadcasting, Vol. 44, No. 2, June 1998, pp. 165 - 171
[55] S. Merchán, A. García Armada, J. L. García, “OFDM Performance in Amplifier

Nonlinearity”, IEEE Transactions on Broadcasting, Vol. 44, No. 1, March 1998, pp.
106 - 113
[56] Qun Shi, “OFDM in Bandpass Nonlinearity”, IEEE Transactions on Consumer

Electronics, Vol. 42, No. 3, August 1996, pp. 253 - 258
[57] Maria-Gabriella Di Benedetto, Paolo Mandarini, “An Application of MMSE

Predistortion to OFDM Systems”, IEEE Transactions on Communications, Vol. 44,
No. 11, November 1996, pp. 1417 - 1420
[58] Hyun Woo Kang, Yong Soo Cho, Dae Hee Youn, “On Compensating Nonlinear
Distortions of an OFDM System Using an Efficienct Adaptive Predistorter”, IEEE
Transactions on Communications, Vol. 47, No. 4, April 1999, pp. 522 - 526
[59] Won Gi Jeon, Kyung Hi Chang, Yong Soo Cho, “An Adaptive Data Predistorter for
Compensation of Nonlinear Distortion in OFDM Systems”, IEEE Transactions on
Communications, Vol. 45, No. 10, October 1997, pp. 1167 - 1171
[60] James K. Cavers, “Amplifier Linearization Using a Digital Predistorter with Fast
Adaption and Low Memory Requirements”, IEEE Transactions on Vehicular
Technology, Vol. 39, No. 4, November 1990, pp. 374 - 382
[61] Sante Andreoli, Howard G. McClure, Paolo Banelli, Saverio Cacopardi, “Digital
Linearizer for RF Amplifiers”, IEEE Transactions on Broadcasting, Vol. 43, No. 1,
March 1997, pp. 12 - 19
249
References
RADIO PROPAGATION
[62] Fechtel, Stefan A., “A Novel Approach to Modelling and Efficient Simulation of
Frequency-Selective Fading Radio Channels”, IEEE Journal on Selected in
Communications, Vol. 11, April 1993, pp. 422 - 431
[63] Andersen Jørgen, Rappaport Theodore S., Yoshida Susumu, “Propagation

Measurements and Models for Wireless Communications Channels”, IEEE
Communications Magazine, January 1995, pp. 42 - 49
[64] Durgin Greg, Rappaport Theodore S., Xu Hao, “5.85 GHz Padio Path Loss and
Penetration Loss Measurements In and Around Homes and Trees”, IEEE
Communications Letters, Vol. 2, No. 3, March 1998, pp. 70 - 72
[65] Talbi L., “Effect of frequency carrier on indoor propagation channel”, Electronic
Letters, Vol. 36, No. 15, 20th July 2000, pp. 1309 - 1310
[66] Takeshi Manabe, Yuko Miura, Toshio Ihara, “Effects of Antenna Directivity and
Polarization on Indoor Multipath Propagation Characteristics at 60 GHz”, IEEE
Journal on Selected Areas in Communications, Vol. 14, No. 3, April 1996, pp. 441 -
447
[67] Larbi Talbi, Gilles Y. Delisle, “Experimental Characterisation of EHF Multipath

Indoor Radio Channels”, IEEE Journal on Selected Areas in Communications, Vol.
14, No. 3, April 1996
[68] McDonnel J.T.E, Spiller T.P., Wilkinson T.A., “RMS delay spread in indoor LOS
environments at 5.2 GHz”, Electronic Letters, Vol. 34, No. 11, 28th May 1998, pp.
1149 - 1150
[69] Seong-Cheol Kim, Henry L. Bertoni, Miklos Stern, “Pulse Propagation

Characteristics at 2.4 GHz Inside Buildings”, IEEE Transactions on Vehicular
Technology, Vol. 45, No. 3., August 1996, pp. 579 - 592
[70] Donald C. Cox, Robert P. Leck, “Correlation Bandwidth and Delay Spread Multipath
Propagation Statistics for 910 MHz Urban Mobile Radio Channels”, IEEE
Transactions on Communications, Vol. COM-23, No. 11, November 1975, pp. 1271
- 1280
250
References
[71] Donald C. Cox, Robert P. Leck, “Distributions of Multipath Delay Spread and
Average Excess Delay for 910-MHz Urban Mobile Radio Paths”, IEEE Transactions
on Antennas and Propagations, Vol. AP-23, No. 2, March 1975, pp. 206 - 213
[72] Donald C. Cox, “Delay Doppler characteristics of multipath propagation at 910 MHz
in a suburban mobile radio environment”, IEEE Transactions on Antennas and
Propagation, Vol. AP-20, September 1972
[73] Homayoun Hashemi, David Tholl, “Statistical Modelling and Simulation of the RMS
Delay Spread of Indoor Radio Propagation Channels”, IEEE Transactions of
Vehicular Technology, Vol. 43, No. 1 February, 1994, pp. 110 - 119
[74] Robert J. C. Bultitude, Samy A. Mahmoud, William A. Sullivan, "A Comparison of

Indoor Radio Propagation Characteristics at 910 MHz and 1.75 GHz", IEEE Journal
on Selected Areas in Communications, Vol. 7, No. 1, January 1989, pp. 20 - 39
[75] S. Y. Seidel and T. S. Rappaport, “914 MHz path loss prediction models for indoor
communications in multifloored buildings”, IEEE Transactions on Antennas and
Propagation, Vol. 40, No. 2, February 1992, pp. 207 - 217
[76] Laura Dossi, Guido Tartara, Fiorenzo Tallone, “Statistical Analysis of Measured
Impulse Response Functions of 2.0 GHz Indoor Radio Channels”, IEEE Journal on
Selected Areas in Communications, Vol. 14, No. 3, April 1996, pp. 405 - 410
[77] Theodore S. Rappaport, Scott Y. Seidel, Rajendra Singh, “900-MHz Multipath

Propagation Measurements for U.S. Digital Cellular Radiotelephone”, IEEE
Transactions on Vehicular Technology, Vol. 39, No. 2, May 1990, pp. 132 - 139
[78] Elvino S. Sousa, Vladan M. Jovanoviæ, Christian Daigneault, “Delay Spread

Measurements for the Digital Cellular Channel in Toronto”, IEEE Transactions on
Vehicular Technology, Vol. 43, No. 4, November 1994, pp. 837 - 847
[79] Robert J. C. Bultitude, G. Keith Bedal, “Propagation Characteristics on Microcellular

Urban mobile Radio Channels at 910 MHz”, IEEE Journal on Selected Areas in
Communications, Vol. 7, No. 1, January 1989, pp. 31 - 39
MULTIUSER OFDM
[80] “Overcoming Multipath in Non-Line-of-Sight High-Speed Microwave
Communications Links”, Cisco Systems, Online:
http://www.cisco.com/warp/public/cc/pd/witc/wt2700/mulpt_wp.htm
251
References
[81] M. Fattouche, “Method and Apparatus for Multiple Access between Transceivers in
Wireless Communications using OFDM spread spectrum”, U.S. Patent 5,282,222,
Filed Mar. 31, 1992, Issued Jan. 25, 1994
[82] T. Williams, R. Prodan, “Multiple User Access Method Using OFDM”, U. S. Patent
5,815,488, Filed Sep. 28, 1995, Issued Sep. 29, 1998
[83] L. Cimini, J. Chuang, N. Sollenberger, “Advanced Cellular Internet Service (ACIS)”,

IEEE Communications Magazine, Oct. 1998, pp. 150 - 159
[84] M. Wahlqvist, C. Östberg, J. Beek, O. Edfors, P. Börjesson, “A Conceptual Study of

OFDM-based Multiple Acess Schemes”, Technical Report Tdoc 117/96, ETSI STC
SMG2 meeting no 18, Helsinki, Finland, May 1996, Online:
http://www.sm.luth.se/csee/sp/publications
[85] L. Wei, C. Schlegel, “ Synchronization Requirements for multi-user OFDM on

Satellite Mobile and Two-path Rayleigh Fading Channels”, IEEE Transaction on
Communications, Vol. 43, No. 2/3/4, February/ March/ April 1995, pp. 887-895
ASDL TECHNOLOGY
[86] Z. Papir, A. Simmonds, “Competing for Throughput in Local Loop”, IEEE
Communications Magazine, May 1999, pp. 61 - 66
[87] M. Barton, L. Chang, T. R. Hsing, “Performance Study of High-Speed Asymmetric

Digital Subscriber Lines Technology”, IEEE Transactions on Communications, Vol.
44, No. 2, Feb 1996, pp. 156 - 157
[88] J. Chow, J. C. Tu, J. M. Cioffi, “A Discrete Multitone Transceiver System for HDSL
Applications”, IEEE Journal on Selected Areas in Communications, Vol. 9, No. 6,
Aug 1991
NETWORK DATA TRAFFIC

[89] W. E. Leland, M. S. Taqqu, W. Willinger, and D. V. Wilson, "On the Self-Similar
Nature of Ethernal Traffic", IEEE Transactions on Networking, Vol. 2, No. 1,
February 1994, pp. 1 - 15
ADAPTIVE MODULATION
[90] K. J. Hole, G. E. Øien, “Spectral Efficiency of Adaptive Coded Modula tion in Urban
Microcellular Networks”, IEEE Transactions on Vehicular Technology, Vol. 50, No.
1, pp. 205 - 222.
252
References
[91] M. C. Aguayo-Torres, J. T. Entrambasaguas, F. Ruiz, “Variable rate multitone

system for multipath fading channels”, Electronic Letters, 11th Jun 1998, Vol. 34,
No. 12, pp. 1203 - 1204
[92] Nilo C. Ericsson, Sorour Falahati, Anders Ahlén, and Arne Svensson, "Hybrid type-
II ARQ/AMS supported by Channel Predictive Scheduling in a Multi-User
Scenario", IEEE Vehicular Technology Conference – VTC'00-Fall, Boston, MA,
USA, September 24-28, 2000, pp. 1804 - 1811
Online: http://www.signal.uu.se/Publications/abstracts/c0015.html
[93] Mohamed-Slim Alouini, "Adaptive and Diversity Techniques for Wireless Digital
Communications over Fading Channels", Ph.D Thesis, Department of Electrical
Engineering, California Institute of Technology, Pasadena, California, June 1998.
Online: http://systems.stanford.edu/Publications/Alouini/alouini_phd.ps
[94] Cheong Yui Wong, Roger S. Cheng, Khaled Ben Letaief, Ross D. Murch, “Multiuser
OFDM with Adaptive Subcarrier, Bit, and Power Allocation”, IEEE Journal on
Selected Areas in Communications, Vol. 17, No. 10, October 1999, pp. 1747-1757
[95] BeamReach Networks, "BeamReach's Adaptive MultiBeam OFDM Technology",

Online: http://www.beamreachnetworks.com/brn/technology/tech_descr.html
HIPERLAN2 AND WLAN SYSTEMS

[96] B. Crow, I. Widjaja, J. G. Kim, P. Sakai, “IEEE 802.11 Wireless Local Area
Networks”, IEEE Communications Magazine, pp 116 - 126, September 1997
[97] M. Johnsson, “HiperLAN/2 – The Broadband Radio Transmission Technology

Operating in the 5 GHz Frequency Band”,
Online: http://www.hiperlan2.com/site/specific/specmain/specwh.htm, 1999
[98] ETSI, “Hiperlan/2-Technical Overview”,
Online: http://www.etsi.org/technicalactiv/Hiperlan/hiperlan2tech.htm
[99] ETSI TS 101 475 V1.1.1, “Broadband Radio Access Networks (BRAN);
HIPERLAN Type 2; Physical (PHY) layer”, April 2000, Online: http://www.etsi.org
[100] “Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)
specifications: High-speed Physical Layer in the 5 GHZ Band”, IEEE Std 802.11a-
1999, September 1999
253
References
[101] C. Lim, A. Nirmalathas, D. Novak, “Dynamic Range Of A Multi-Section Laser In A
Millimetre-Wave Fibre-Wireless Uplink”, Proc. Asia Pacific Microwave Conference
2000, Sydney, Australia, December 2000
[102] U. Grob, A. L. Welti, E. Zollinger, R. Küng, H. Kaufmann, “Microcellular Direct-

Sequence Spread-Spectrum Radio System Using N-Path RAKE Receiver”, IEEE
Journal on Selected Areas in Communications, Vol. 8, No. 5, pp. 772 - 779, June
1990
[103] Intersil, (May, 2000), Direct Sequence Spread Spectrum Baseband Processor with
Rake receiver and Equalizer, HFA3863 Data sheet,
Online: http://www.intersil.com/data/FN/FN4/FN4856/FN4856.pdf
PHASING SCHEMES FOR LOW CREST FACTOR MULTITONE

SIGNALS
[104] S. Boyd, “Multitone signals with low crest factor”, IEEE Transactions on Circuits
and Systems, vol. CAS-33, pp. 1018 - 1022, Oct. 1986
[105] D. J. Newmann, "An L1 extremal problem for polynomials," Proc. Amer. Math. Soc.,
vol. 16, pp. 1287-1290, Dec. 1965
[106] S. Narahashi, T. Nojima, “New phaseing scheme of N-multiple carriers for reducing
peak-to-average power ratio”, Electronic Letters, Vol. 30, No. 17, pp. 1382 - 1383,
Aug. 1994
[107] S. Narahashi, K. Kumagai and T. Nojima, “Minimising peak to average power ratio
of multitone signals using steepest descent method”, Electronic Letters, Vol. 31, No.
18, 31st August 1995, pp. 1552 - 1554
CREST FACTOR MINIMISATION OF OFDM DATA SYMBOLS

[108] R.W. Bauml, R.F.H. Fischer, J.B. Huber, “Reducing the peak-to-average power ratio
of multicarrier modulation by selected mapping”, Electronic Letters, 1996, Vol. 32,
pp. 2056 - 2057
[109] Eetvelt J.Van, G. Wade, M. Tomlinson, “Peak to average power reduction for OFDM
schemes by selective scrambling”, Electronic Letters, 1996, Vol. 32, pp. 1963 - 1964
[110] D. Wulich, “Reduction of peak to mean ratio of multicarrier modulation using cyclic
coding”, Electronic Letters, 1996, Vol. 32, pp. 432 - 433
[111] C. Tellambura, “Use of m-sequences for OFDM peak-to-average power ratio
254
References
reduction”, Electronic Letters, 1997, Vol. 33, pp. 1300 - 1301
[112] C. Tellambura, “Phase optimisation criterion for reducing peak-to-average power

ratio in OFDM”, Electronic Letters, 1998, Vol. 34, pp. 169 - 170
[113] J.A. Davis, J. Jedwab, “Peak-to-mean power control and error correction for OFDM
transmission using Golay sequences and Reed-Muller codes”, Electronic Letters,
1997, Vol. 33, pp. 267 - 268
APPLICATION OF GENETIC ALGORITHMS

[114] Moshe Sipper, Edmund M. A. Ronald, “A new species of hardware”, IEEE
Spectrum, Vol. 37, No. 3, March 2000, pp. 59 - 64
[115] “The Genetic Algorithms Archive”, Navy Center for Applied Research in Artificial
Intelligence, March 2001, Online: http://www.aic.nrl.navy.mil/galist
[116] “The Hitch-Hiker’s Guide to Evolutionary Computation”, Online:

ftp://ftp.cerias.purdue.edu/pub/doc/EC/Welcome.html or
ftp://ftp.cs.wayne.edu/pub/EC/Welcome.html
GENERAL COMMUNICATION TEXTBOOKS

[117] Gordon L. Stüber, “Principles of Mobile Communication”, Kluwer Academic
Publishers, 1996, pp. 225 - 238
[118] B. Sklar, “Digital Communications Fundamentals and Applications”, Prentice Hall,

1988, pp. 300
[119] J. Gibson, “The Mobile Communications Handbook”, CRC Press, 1996, pp. 355 -
369
[120] Emmanuel C. Ifeachor, Barrie W. Jervis, “Digital Signal Processing, A Practical

Approach”, Addison-Wesley, 1993, ISBN: 0-201-54413-X
[121] Martin S. Roden, “Analog and Digital Communication Systems”, Prentice-Hall

International, Inc, 4th Edition, 1996, ISBN: 0-13-399965-3
[122] B. P. Lathi, “Modern Digital and Analog Communication Systems”, CBS College
Publishing, 1983, ISBN: 0-03-058969-X
[123] Yoshihiko Akaiwa, “Introduction to Digital Mobile Communication”, John Wiley &
Sons, Inc, 1997, ISBN: 0-471-17545-5
255
References
PUBLICATIONS RESULTING FROM

THIS PHD
Eric Lawrey, Cornelis J. Kikkert, "Adaptive Frequency Hopping for Multiuser OFDM",
Second International Conference on Information, Communications and Signal
Processing, ICICS’99, Singapore, Dec 7-10, 1999, Proceedings: paper number 361,
Publisher NTU, http://www.ntu.edu.sg/eee/conferences/icics/icics.html.
Eric Lawrey, Cornelis J. Kikkert, "Peak to average power ratio reduction of OFDM signals
using peak reduction carriers", Fifth International Symposium on Signal Processing
and its Applications, ISSPA99, Brisbane, Aug. 1999, pp. 737-740.
Eric Lawrey, "Multiuser OFDM", Fifth International Symposium on Signal Processing and
its Applications, ISSPA99, Brisbane, Aug. 1999, pp. 761-764.
Eric Lawrey, Cornelis J. Kikkert, "Maximising Signal Strength inside Buildings for
wireless LAN Systems using OFDM", Asia Pacific Microwave Conference, APMC
2000, Sydney, Dec 3-6, 2000, pp. 257-260
Eric Lawrey, Cornelis J. Kikkert, "Maximising Signal Strength for OFDM inside
buildings", IEEE Transactions on Microwave Theory and Techniques, November
2001, Special Mini-issue from APMC 2000
A copy of these papers is attached to the end of this thesis.
256
Appendix A Window functions
APPENDIX A WINDOW FUNCTIONS
Window functions are used for frequency analysis and for designing FIR filters. For
spectral analysis the time waveform is usually multiplied by a window function to
reduce the effects of spectral spreading. Spectral analysis is normally performed
using a Fast Fourier Transform (FFT). This calculates the frequency response of the
cyclic extension of the time waveform. As a consequence if the time waveform is
non-cyclic it results in a transient between the end of the time waveform and the
start, resulting in spectral spreading. Applying a window function tappers the start
and ends of waveform reducing these transients, and consequently spectral
spreading.
0.8
Amplitude
0.6
0.4
boxcar
triangle
0.2
hanning
blackh4
0
0 0.2 0.4 0.6 0.8 1
Time (Normalised)
Figure A - 1, Time waveform for several common window functions.
(Blackh4 = 4 term Blackman Harris) (script s0072)
The power spectrum of a discrete time waveform s(t), can be estimated using the
following:
N
P( s ) = 20 log 10 ( FFT (s(t ) ⋅ w(t )) ) − 20 log 10   + WL
2
257
Where w(t) is the window function being used, WL is the window loss of the window
function, N is the number of samples in s(t), and P(s) is the power spectrum in dB. N
is also usually the length of the FFT. | x | is the magnitude of the complex value x.
Figure A - 2 shows the application of window functions for spectral estimation. The
spectrum shown is for a four-tone signal with a large variation in tone power. Figure
A - 2 (a) shows the result for a boxcar window function. This is the same as having
no window function at all. This spectral spreading in this result masks all but the
strongest tone. Figure A - 2 (b) shows the result when using a Hanning window
function. This allows for three of the tones to be seen clearly. The Blackman Harris
window function allows the fourth tone to be just seen at a frequency of 0.25,
however it is too close to the strong tone at a frequency of 0.2, to allow accurate
estimation. Figure A - 2 (d) shows the estimation with the new presented Lawrey6
window function. Details of the implementation of this window function are shown
in section A.4. The side-lobes of the window function are sufficiently low to allow
all four tones to be easily distinguished. One other thing that can be seen is that the
width of the main lobe of the tone becomes wider as the side-lobe power is reduced.
Boxcar Hanning
0 0
-20 -20
-40 -40
Power (dB)
Power (dB)
-60 -60
-80 -80
-100 -100
-120 -120
-140 -140
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Frequency (1 = Nyquist) Frequency (1 = Nyquist)
(a) (b)
4 Term Blackman Harris 6 Term Lawrey
0 0
-20 -20
-40 -40
Power (dB)
Power (dB)
-60 -60
-80 -80
-100 -100
-120 -120
-140 -140
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Frequency (1 = Nyquist) Frequency (1 = Nyquist)
(c) (d)
Figure A - 2, Power spectra of a 4 tone signal with power levels varying up upto
100 dB. (script s0074)
258
A.1 WINDOW FUNCTION TRANSITION WIDTH AND

SIDE-LOBE POWER
The performance of a window function is determined largely by two main

parameters. These are the side-lobe power and the transition width of the window
function. The transition width of a window function is measured by taking the
magnitude of the fourier transform of the time waveform of the window function. It
is a measure of how fast the window function rolls off from DC to the first null.
Figure A - 3 shows the spectrum of the Boxcar and Hanning window functions. The
transition width for the Boxcar window function is 1.0 as the first null in the
frequency response corresponds to a frequency that is the inverse of the symbol time.
The Hanning window function has a transition width very close to 2.0.
0
boxcar
hanning
-10
-20
Power (dB)
-30
-40
-50
-60
-70
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Frequency (1 = 1/Symbol Time)
Figure A - 3, Spectrum of Boxcar and Hanning window functions. (script

s0076)
There is a strict trade off between side-lobe power and the transition width of a
window function. Decreasing the side-lobe power increases the transition width, thus
using a window functions with low side-lobe power for spectral analysis results in a
lower frequency resolution. Figure A - 4 shows the side-lobe power for a number of
window functions, as a function of the transition width. The Kaiser window function
incorporates a side-lobe power control parameter, allowing a trade off between side-
lobe power and transition width. See [120] for more details on the Kaiser window
function.
259
Transition Width from peak to first null
-20 Kaiser
-30 Boxcar
First Sidelobe Power (dB)

-40 Hanning
-50 Hamming
Triangle
-60
Blackman
-70
Blackh4
-80 Lawrey5
-90 Lawrey6
-100
y = -25.66.x+12.06 (dB)
-110
-120
-130
-140
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6

Transition width of the window (normalised to boxcar)
Figure A - 4, Transition width for a number of window functions, as a function

of the side-lobe power. (script s0039)
Table A - 1 shows the transition width and side-lobe power for a variety of window
functions. This is the performance that would be seen when using the window
functions for spectral analysis.
Window Transition width of window function Maximum Side-Lobe Power

Function DC to first Null (normalised to 1/ for window function (dBc)
symbol time) (WT)
Boxcar 1.00 -13.3
Hanning 2.00 -31.5
Hamming 2.00 -42.6
Triangle 2.00 -26.5
Blackman 3.03 -58.1
4 term Blackman 4.03 -92.0
Harris
5 term Lawrey 4.67 -109.9
6 term Lawrey 5.81 -139.1
Table A - 1, Transition width and side-lobe power of different window
functions. (script s0039)
A.2 WINDOW FUNCTION PERFORMANCE WHEN

USED FOR FIR FILTERS
Window functions can be used to design FIR filters. This is achieved by taking the
inverse fourier transform of the required frequency response to find the time domain
impulse response of the filter. This impulse response is the coefficients of the FIR
design. If we take for example and ideal brick-wall low pass filter, the impulse
260
response corresponds to a sinc function. The sinc function is however infinite in

length, and so truncation is required for implementation of the filter. However, this
truncation results in spectral spreading. Windowing the truncated impulse response
reduces the spectral spreading at the cost of broadening the transition width of the
filter. For a more detail description of the window method for FIR filter design refer
to [120].
Window Function (b) Maximum Side-lobe Power,

for FIR filter using window
method (dBc)
Boxcar -20.9
Hanning -43.9
Hamming -54.2
Triangle -26.2
Blackman -78.3
4 term Blackman Harris -109.2
5 term Lawrey -127.7
6 term Lawrey -160.9
Table A - 2, Side-lobe power of different window functions when used with the
window method of FIR filter design. (script s0071)
Side-lobe power for FIR using Kaiser window function
Max. side-lobe power of FIR filter (dBc)
-20
-40
-60
-80
-100
-120
-140
-160
1 1.5 2 2.5 3 3.5 4 4.5 5
Window function transition width (Normalised to boxcar)
Figure A - 5, Side-lobe power for Kaiser window function when used with the
window method of FIR filter design. (script s0071)
The x-axis is normalised to the transition width of the window function by

itself, the same as used in Figure A - 4.
When a window function is used for FIR filter design the out of band attenuation is
greater than the side-lobe power of just the window function itself. Table A - 2
shows the maximum side-lobe power for FIR filters designed using the window
method for a range of window functions. Table A - 3 shows the transition width of
261
each of the window functions when used with the window method for FIR filter
design. The transition width has been normalised to the number of taps in the filter.
The transition width of an FIR filter will be approximately:
wt
ft = f
N s
Where f t is the transition frequency width in Hz, wt is the window function transition
width as in Table A - 3, N is the number of taps in the filter, and f s is the sample rate
in Hz. For example if I used a 100 tap filter with a Hanning window function, a
sample rate of 1 kHz, then the transition width from the pass band (6dB down) to the
first null would be:
1.67
ft = 1000 = 16.7 Hz
100
Also shown in Table A - 3 is the relative transition width from –6dBc to –20dBc for
each of the window functions. This represents the ratio of the number of filter taps
required to maintain the same rate of filter roll off and an example is shown in Figure
A - 7.
Window Function Transition Width of FIR filter Transition Width for FIR
(From X dB point to first null) filter from –6 dB to -20
1 dB 3 dB 6 dB dB, normalised to Boxcar
Boxcar 1.05 0.83 0.62 1
Hanning 2.59 2.09 1.67 2.15
Hamming 2.65 2.17 1.76 2.04
Triangle 2.95 2.38 1.95 2.36
Blackman 4.18 3.54 3.03 2.67
4 term Blackman Harris 5.39 4.64 4.03 3.14
5 term Lawrey 6.00 5.19 4.53 3.41
6 term Lawrey 7.45 6.52 5.78 3.85
Table A - 3, Transition of FIR filters using different window functions. (script
s0071)
262
FIR transition width for Kaiser window function
7
Norm. width for FIR filter to 1st Null

1 dB to Null
3 dB to Null
6
6 dB to Null
20 dB to Null
5
0
1 1.5 2 2.5 3 3.5 4 4.5 5
Window function transition width (Normalised to boxcar)
Figure A - 6, Transition of FIR filters using the Kaiser window function.
The x-axis is normalised to the transition width of the window function by

itself, the same as used in Figure A - 4. The y-axis shows the transition width of
FIR filter from X dB point to first null in the filter response.
0 boxcar: 40 taps
blackh4: 126 taps
-50
Power (dB)
-100
-150
0 0.2 0.4 0.6 0.8 1
Frequency (1 = Nyquist)
Figure A - 7, Frequency response of two FIR filters with the same rate of roll
off. The FIR filter using the Blackman Harris window function requires 3.14
times more filter taps than the boxcar filter. (script s0073)
A.3 WINDOW LOSS
Applying a window function as part of spectral analysis tapers the start and end of
the time waveform, resulting in a loss of power seen in the spectrum of the signal.
This power loss is referred to as window loss. Table A - 4 shows the loss for each of
the common window functions.
263
Window Function Window Loss (dB)
Boxcar 0.000
Hanning 6.018
Hamming 5.354
Triangle 6.021
Blackman 7.537
4 term Blackman Harris 8.906
5 term Lawrey 9.583
6 term Lawrey 10.608
Table A - 4, Window loss for different window functions (script s0075)
A.4 LAWREY WINDOW FUNCTION
Window functions with low side-lobe power are especially important for spectral
analysis and FIR filter design. The Kaiser window function is a flexible window
function that can be used to generate side-lobes with arbitrary attenuation, however
Bessel functions are required to calculate the window function. This makes it
difficult to implement on a DSP.
Two window functions were developed to provide lower side-lobes than the 4-term
Blackman Harris window function, and be simple to implement. Both window
functions were calculated by approximating a Kaiser window function using a sum
of sinusoids. The coefficients were optimised to maximise the side-lobe attenuation
for the given number of coefficients.
2π
θ= t = 0, 1, L n − 1
(n − 1) ⋅ t
w(t ) = c1 + c 2 cos(θ ) + c3 cos(2θ ) + L + c m cos((m − 1)θ )
where c is the coefficients of the window function, and m is the number of

coefficients or terms.
264
Coefficient Hanning Hamming Blackman Blackh4 Lawrey5 Lawrey6

c1 0.5 0.54 0.42 0.35875 0.3318676 0.2949200
c2 -0.5 -0.46 -0.50 -0.48829 -0.4766027 -0.4531624
c3   0.08 0.14128 0.1675404 0.2004549
c4    -0.01168 -0.0234285 -0.0467364
c5     0.0006249 0.0046253
c6      -0.0001010
Table A - 5, Coefficients for Lawrey5 and Lawrey6 window functions, and
other common window functions.
All digits are required to maintain window function performance. For Lawrey5
and Lawrey6 single precision representation of the coefficients is sufficient.
0
Lawrey5
-20 Lawrey6
-40
-60
Power (dB)
-80
-100
-120
-140
-160
-180
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Frequency (1 = 1/Symbol Time)
Figure A - 8, Spectrum of Lawrey5 and Lawrey6 window functions.
265
Appendix B BER verses SNR for OFDM in AWGN
APPENDIX B BER VERSES SNR

FOR OFDM IN AWGN
The performance of OFDM in the presence of AWGN was simulated. These tables
show the same results as that shown in Section 2.6.9 but in tabular form.
B.1 COHERENT QUADRATURE AMPLITUDE

MODULATION
EBNR (dB) BPSK QPSK 8QAM 16QAM 32QAM 64QAM

-6.0 2.393E-01 2.389E-01 2.944E-01 2.861E-01 3.801E-01 3.259E-01
-5.5 2.264E-01 2.268E-01 2.836E-01 2.749E-01 3.743E-01 3.165E-01
-5.0 2.137E-01 2.129E-01 2.714E-01 2.624E-01 3.685E-01 3.069E-01
-4.5 1.994E-01 1.993E-01 2.600E-01 2.495E-01 3.625E-01 2.974E-01
-4.0 1.868E-01 1.862E-01 2.475E-01 2.368E-01 3.555E-01 2.878E-01
-3.5 1.726E-01 1.720E-01 2.357E-01 2.245E-01 3.485E-01 2.781E-01
-3.0 1.584E-01 1.584E-01 2.225E-01 2.116E-01 3.404E-01 2.685E-01
-2.5 1.444E-01 1.443E-01 2.086E-01 1.992E-01 3.326E-01 2.588E-01
-2.0 1.310E-01 1.306E-01 1.968E-01 1.872E-01 3.236E-01 2.495E-01
-1.5 1.169E-01 1.171E-01 1.832E-01 1.750E-01 3.140E-01 2.403E-01
-1.0 1.039E-01 1.038E-01 1.694E-01 1.637E-01 3.045E-01 2.304E-01
-0.5 9.114E-02 9.114E-02 1.555E-01 1.523E-01 2.930E-01 2.208E-01
0.0 7.883E-02 7.847E-02 1.428E-01 1.414E-01 2.820E-01 2.113E-01
0.5 6.712E-02 6.700E-02 1.287E-01 1.299E-01 2.693E-01 2.017E-01
1.0 5.627E-02 5.627E-02 1.146E-01 1.188E-01 2.562E-01 1.919E-01
1.5 4.646E-02 4.639E-02 1.017E-01 1.084E-01 2.432E-01 1.824E-01
2.0 3.737E-02 3.756E-02 8.880E-02 9.792E-02 2.281E-01 1.723E-01
2.5 2.972E-02 2.960E-02 7.673E-02 8.739E-02 2.136E-01 1.626E-01
3.0 2.289E-02 2.291E-02 6.476E-02 7.741E-02 1.981E-01 1.534E-01
3.5 1.712E-02 1.724E-02 5.390E-02 6.802E-02 1.825E-01 1.436E-01
4.0 1.250E-02 1.254E-02 4.409E-02 5.847E-02 1.663E-01 1.334E-01
4.5 8.810E-03 8.803E-03 3.530E-02 5.000E-02 1.507E-01 1.243E-01
5.0 5.946E-03 5.968E-03 2.740E-02 4.182E-02 1.344E-01 1.144E-01
5.5 3.857E-03 3.865E-03 2.085E-02 3.440E-02 1.196E-01 1.050E-01
6.0 2.378E-03 2.382E-03 1.528E-02 2.786E-02 1.038E-01 9.582E-02
6.5 1.399E-03 1.412E-03 1.083E-02 2.197E-02 8.913E-02 8.637E-02
7.0 7.817E-04 7.699E-04 7.410E-03 1.708E-02 7.573E-02 7.720E-02
7.5 3.992E-04 3.941E-04 4.833E-03 1.272E-02 6.269E-02 6.836E-02
8.0 1.983E-04 1.872E-04 2.995E-03 9.253E-03 5.071E-02 5.979E-02
266
EBNR (dB) BPSK QPSK 8QAM 16QAM 32QAM 64QAM

8.5 8.630E-05 8.613E-05 1.773E-03 6.510E-03 4.034E-02 5.170E-02
9.0 3.448E-05 3.337E-05 9.704E-04 4.397E-03 3.152E-02 4.398E-02
9.5 1.270E-05 1.186E-05 5.051E-04 2.855E-03 2.357E-02 3.683E-02
10.0 4.163E-06 3.503E-06 2.423E-04 1.744E-03 1.741E-02 3.037E-02
10.5 7.740E-07 9.108E-07 1.057E-04 1.022E-03 1.214E-02 2.444E-02
11.0 2.111E-07 2.452E-07 4.179E-05 5.667E-04 8.342E-03 1.929E-02
11.5 3.528E-08 1.414E-05 2.904E-04 5.409E-03 1.486E-02
12.0 5.439E-06 1.365E-04 3.338E-03 1.109E-02
12.5 1.240E-06 6.156E-05 1.967E-03 8.065E-03
13.0 3.263E-07 2.361E-05 1.083E-03 5.643E-03
13.5 6.124E-08 8.827E-06 5.655E-04 3.803E-03
14.0 2.953E-06 2.736E-04 2.466E-03
14.5 7.995E-07 1.179E-04 1.512E-03
15.0 1.795E-07 4.740E-05 8.812E-04
15.5 1.626E-05 4.860E-04
16.0 6.175E-06 2.448E-04
16.5 1.347E-06 1.190E-04
17.0 2.681E-07 5.137E-05
17.5 8.838E-08 2.054E-05
18.0 6.733E-06
18.5 2.371E-06
19.0 4.025E-07
19.5 1.197E-07
20.0 2.359E-08
20.5 2.540E-08
267
EBNR (dB) 128QAM 256QAM 512QAM 1024QAM 2048QAM 4096QAM

-6.0 4.064E-01 3.529E-01 4.217E-01 3.719E-01 4.312E-01 3.858E-01
-5.5 4.030E-01 3.461E-01 4.188E-01 3.659E-01 4.289E-01 3.807E-01
-5.0 3.994E-01 3.388E-01 4.158E-01 3.604E-01 4.265E-01 3.762E-01
-4.5 3.962E-01 3.316E-01 4.127E-01 3.546E-01 4.235E-01 3.710E-01
-4.0 3.907E-01 3.241E-01 4.099E-01 3.485E-01 4.208E-01 3.661E-01
-3.5 3.870E-01 3.168E-01 4.062E-01 3.428E-01 4.181E-01 3.613E-01
-3.0 3.831E-01 3.095E-01 4.027E-01 3.371E-01 4.151E-01 3.561E-01
-2.5 3.774E-01 3.024E-01 3.987E-01 3.307E-01 4.119E-01 3.505E-01
-2.0 3.728E-01 2.946E-01 3.941E-01 3.250E-01 4.081E-01 3.457E-01
-1.5 3.664E-01 2.868E-01 3.906E-01 3.184E-01 4.048E-01 3.404E-01
-1.0 3.608E-01 2.791E-01 3.855E-01 3.118E-01 4.009E-01 3.352E-01
-0.5 3.529E-01 2.710E-01 3.807E-01 3.057E-01 3.970E-01 3.297E-01
0.0 3.461E-01 2.631E-01 3.764E-01 2.992E-01 3.928E-01 3.248E-01
0.5 3.393E-01 2.554E-01 3.710E-01 2.924E-01 3.882E-01 3.185E-01
1.0 3.311E-01 2.473E-01 3.656E-01 2.855E-01 3.838E-01 3.128E-01
1.5 3.227E-01 2.389E-01 3.601E-01 2.793E-01 3.792E-01 3.075E-01
2.0 3.132E-01 2.305E-01 3.529E-01 2.721E-01 3.741E-01 3.016E-01
2.5 3.039E-01 2.221E-01 3.470E-01 2.653E-01 3.694E-01 2.953E-01
3.0 2.941E-01 2.134E-01 3.406E-01 2.579E-01 3.643E-01 2.900E-01
3.5 2.835E-01 2.049E-01 3.337E-01 2.510E-01 3.590E-01 2.834E-01
4.0 2.717E-01 1.962E-01 3.254E-01 2.435E-01 3.537E-01 2.775E-01
4.5 2.599E-01 1.874E-01 3.187E-01 2.363E-01 3.480E-01 2.717E-01
5.0 2.472E-01 1.787E-01 3.100E-01 2.290E-01 3.426E-01 2.660E-01
5.5 2.345E-01 1.700E-01 3.020E-01 2.215E-01 3.363E-01 2.596E-01
6.0 2.205E-01 1.612E-01 2.931E-01 2.144E-01 3.300E-01 2.536E-01
6.5 2.062E-01 1.527E-01 2.846E-01 2.069E-01 3.238E-01 2.471E-01
7.0 1.924E-01 1.442E-01 2.754E-01 1.996E-01 3.170E-01 2.410E-01
7.5 1.784E-01 1.355E-01 2.663E-01 1.925E-01 3.109E-01 2.348E-01
8.0 1.636E-01 1.272E-01 2.563E-01 1.851E-01 3.043E-01 2.288E-01
8.5 1.493E-01 1.186E-01 2.459E-01 1.777E-01 2.973E-01 2.221E-01
9.0 1.343E-01 1.103E-01 2.355E-01 1.702E-01 2.897E-01 2.158E-01
9.5 1.198E-01 1.022E-01 2.244E-01 1.627E-01 2.822E-01 2.098E-01
10.0 1.057E-01 9.394E-02 2.130E-01 1.553E-01 2.752E-01 2.031E-01
10.5 9.207E-02 8.578E-02 2.007E-01 1.476E-01 2.673E-01 1.965E-01
11.0 7.991E-02 7.796E-02 1.890E-01 1.403E-01 2.595E-01 1.903E-01
11.5 6.718E-02 7.017E-02 1.770E-01 1.326E-01 2.511E-01 1.837E-01
12.0 5.573E-02 6.257E-02 1.639E-01 1.252E-01 2.428E-01 1.773E-01
12.5 4.551E-02 5.494E-02 1.517E-01 1.182E-01 2.347E-01 1.709E-01
13.0 3.650E-02 4.805E-02 1.391E-01 1.105E-01 2.258E-01 1.642E-01
13.5 2.841E-02 4.117E-02 1.256E-01 1.035E-01 2.165E-01 1.578E-01
14.0 2.160E-02 3.485E-02 1.131E-01 9.626E-02 2.074E-01 1.511E-01
14.5 1.594E-02 2.908E-02 1.008E-01 8.921E-02 1.975E-01 1.446E-01
15.0 1.137E-02 2.379E-02 8.868E-02 8.211E-02 1.876E-01 1.378E-01
15.5 7.780E-03 1.904E-02 7.743E-02 7.536E-02 1.775E-01 1.314E-01
268
EBNR (dB) 128QAM 256QAM 512QAM 1024QAM 2048QAM 4096QAM

16.0 5.153E-03 1.482E-02 6.613E-02 6.847E-02 1.672E-01 1.250E-01
16.5 3.254E-03 1.135E-02 5.610E-02 6.153E-02 1.561E-01 1.181E-01
17.0 1.925E-03 8.373E-03 4.612E-02 5.496E-02 1.453E-01 1.119E-01
17.5 1.081E-03 6.015E-03 3.791E-02 4.845E-02 1.342E-01 1.056E-01
18.0 5.841E-04 4.154E-03 3.008E-02 4.225E-02 1.230E-01 9.891E-02
18.5 2.964E-04 2.774E-03 2.327E-02 3.639E-02 1.119E-01 9.265E-02
19.0 1.299E-04 1.754E-03 1.767E-02 3.090E-02 1.010E-01 8.661E-02
19.5 5.496E-05 1.061E-03 1.303E-02 2.580E-02 8.987E-02 7.995E-02
20.0 2.129E-05 6.079E-04 9.309E-03 2.113E-02 7.968E-02 7.424E-02
20.5 6.151E-06 3.288E-04 6.342E-03 1.700E-02 6.929E-02 6.786E-02
21.0 2.154E-06 1.631E-04 4.148E-03 1.331E-02 5.999E-02 6.199E-02
21.5 7.778E-07 7.463E-05 2.628E-03 1.024E-02 5.042E-02 5.584E-02
22.0 2.881E-08 3.149E-05 1.539E-03 7.615E-03 4.229E-02 5.033E-02
22.5 1.136E-05 8.855E-04 5.448E-03 3.448E-02 4.484E-02
23.0 4.251E-06 4.542E-04 3.803E-03 2.778E-02 3.901E-02
23.5 1.363E-06 2.210E-04 2.538E-03 2.161E-02 3.368E-02
24.0 2.121E-07 9.954E-05 1.627E-03 1.661E-02 2.896E-02
24.5 4.079E-08 4.203E-05 9.820E-04 1.226E-02 2.432E-02
25.0 1.554E-05 5.705E-04 8.855E-03 2.005E-02
25.5 5.512E-06 3.051E-04 6.091E-03 1.629E-02
26.0 1.121E-06 1.541E-04 4.100E-03 1.301E-02
26.5 6.662E-07 7.323E-05 2.588E-03 1.002E-02
27.0 2.273E-07 2.959E-05 1.551E-03 7.533E-03
27.5 1.173E-05 8.843E-04 5.548E-03
28.0 4.118E-06 4.735E-04 3.894E-03
28.5 1.299E-06 2.387E-04 2.656E-03
29.0 2.937E-07 1.148E-04 1.739E-03
29.5 1.115E-07 4.632E-05 1.083E-03
30.0 1.838E-05 6.360E-04
30.5 5.618E-06 3.580E-04
31.0 2.522E-06 1.847E-04
31.5 6.329E-07 9.174E-05
32.0 1.686E-07 4.014E-05
32.5 1.662E-05
33.0 6.369E-06
33.5 1.851E-06
34.0 5.385E-07
34.5 8.733E-08
35.0 3.538E-08
269
B.2 COHERENT PHASE SHIFT KEYING
EBNR (dB) 8PSK 16PSK 32PSK 64PSK 128PSK 256PSK

-6.0 2.712E-01 3.058E-01 3.320E-01 3.479E-01 3.612E-01 3.734E-01
-5.5 2.596E-01 2.964E-01 3.229E-01 3.412E-01 3.552E-01 3.670E-01
-5.0 2.473E-01 2.862E-01 3.142E-01 3.342E-01 3.495E-01 3.620E-01
-4.5 2.330E-01 2.744E-01 3.051E-01 3.272E-01 3.431E-01 3.565E-01
-4.0 2.214E-01 2.634E-01 2.962E-01 3.190E-01 3.366E-01 3.511E-01
-3.5 2.093E-01 2.529E-01 2.866E-01 3.119E-01 3.302E-01 3.451E-01
-3.0 1.960E-01 2.416E-01 2.788E-01 3.050E-01 3.243E-01 3.403E-01
-2.5 1.830E-01 2.296E-01 2.692E-01 2.975E-01 3.180E-01 3.340E-01
-2.0 1.715E-01 2.188E-01 2.604E-01 2.895E-01 3.115E-01 3.289E-01
-1.5 1.582E-01 2.074E-01 2.518E-01 2.823E-01 3.058E-01 3.235E-01
-1.0 1.467E-01 1.963E-01 2.429E-01 2.750E-01 2.993E-01 3.181E-01
-0.5 1.339E-01 1.855E-01 2.338E-01 2.682E-01 2.935E-01 3.129E-01
0.0 1.226E-01 1.742E-01 2.251E-01 2.603E-01 2.873E-01 3.076E-01
0.5 1.116E-01 1.636E-01 2.159E-01 2.534E-01 2.808E-01 3.021E-01
1.0 1.007E-01 1.535E-01 2.066E-01 2.463E-01 2.750E-01 2.967E-01
1.5 9.031E-02 1.435E-01 1.979E-01 2.394E-01 2.687E-01 2.916E-01
2.0 8.062E-02 1.336E-01 1.890E-01 2.312E-01 2.625E-01 2.864E-01
2.5 7.141E-02 1.249E-01 1.797E-01 2.250E-01 2.567E-01 2.808E-01
3.0 6.221E-02 1.156E-01 1.706E-01 2.175E-01 2.504E-01 2.758E-01
3.5 5.423E-02 1.069E-01 1.628E-01 2.103E-01 2.444E-01 2.706E-01
4.0 4.587E-02 9.898E-02 1.535E-01 2.031E-01 2.384E-01 2.652E-01
4.5 3.839E-02 9.070E-02 1.453E-01 1.961E-01 2.325E-01 2.603E-01
5.0 3.219E-02 8.290E-02 1.372E-01 1.891E-01 2.265E-01 2.552E-01
5.5 2.590E-02 7.549E-02 1.286E-01 1.818E-01 2.200E-01 2.497E-01
6.0 2.040E-02 6.808E-02 1.207E-01 1.743E-01 2.145E-01 2.445E-01
6.5 1.585E-02 6.110E-02 1.131E-01 1.673E-01 2.086E-01 2.393E-01
7.0 1.197E-02 5.421E-02 1.055E-01 1.599E-01 2.022E-01 2.338E-01
7.5 8.597E-03 4.790E-02 9.864E-02 1.523E-01 1.960E-01 2.283E-01
8.0 6.231E-03 4.135E-02 9.141E-02 1.449E-01 1.902E-01 2.232E-01
8.5 4.164E-03 3.542E-02 8.476E-02 1.379E-01 1.841E-01 2.181E-01
9.0 2.755E-03 3.007E-02 7.862E-02 1.303E-01 1.782E-01 2.131E-01
9.5 1.696E-03 2.495E-02 7.194E-02 1.234E-01 1.719E-01 2.080E-01
10.0 1.001E-03 2.022E-02 6.611E-02 1.163E-01 1.657E-01 2.027E-01
10.5 5.531E-04 1.608E-02 6.008E-02 1.098E-01 1.598E-01 1.972E-01
11.0 2.925E-04 1.264E-02 5.436E-02 1.031E-01 1.533E-01 1.919E-01
11.5 1.430E-04 9.483E-03 4.884E-02 9.638E-02 1.475E-01 1.866E-01
12.0 6.302E-05 7.074E-03 4.356E-02 9.034E-02 1.414E-01 1.817E-01
12.5 2.686E-05 4.908E-03 3.836E-02 8.426E-02 1.348E-01 1.765E-01
13.0 8.881E-06 3.388E-03 3.319E-02 7.825E-02 1.286E-01 1.710E-01
13.5 3.354E-06 2.296E-03 2.859E-02 7.256E-02 1.225E-01 1.656E-01
14.0 8.503E-07 1.434E-03 2.406E-02 6.769E-02 1.160E-01 1.605E-01
14.5 3.779E-07 8.333E-04 2.009E-02 6.229E-02 1.097E-01 1.552E-01
15.0 4.840E-04 1.621E-02 5.714E-02 1.039E-01 1.499E-01
15.5 2.506E-04 1.297E-02 5.204E-02 9.782E-02 1.443E-01
16.0 1.252E-04 1.021E-02 4.746E-02 9.213E-02 1.389E-01
16.5 5.547E-05 7.691E-03 4.282E-02 8.647E-02 1.340E-01
270
EBNR (dB) 8PSK 16PSK 32PSK 64PSK 128PSK 256PSK

17.0 2.300E-05 5.634E-03 3.846E-02 8.074E-02 1.280E-01
17.5 9.212E-06 4.022E-03 3.381E-02 7.588E-02 1.228E-01
18.0 3.224E-06 2.754E-03 2.958E-02 7.072E-02 1.174E-01
18.5 8.858E-07 1.832E-03 2.549E-02 6.591E-02 1.117E-01
19.0 2.126E-07 1.158E-03 2.170E-02 6.082E-02 1.065E-01
19.5 3.543E-08 6.865E-04 1.825E-02 5.664E-02 1.011E-01
20.0 3.884E-04 1.483E-02 5.208E-02 9.547E-02
20.5 2.029E-04 1.205E-02 4.784E-02 9.013E-02
21.0 1.024E-04 9.369E-03 4.377E-02 8.515E-02
21.5 4.569E-05 7.275E-03 3.950E-02 7.994E-02
22.0 1.919E-05 5.406E-03 3.539E-02 7.505E-02
22.5 7.228E-06 3.929E-03 3.151E-02 7.030E-02
23.0 2.494E-06 2.703E-03 2.777E-02 6.584E-02
23.5 8.362E-07 1.842E-03 2.425E-02 6.108E-02
24.0 1.984E-07 1.189E-03 2.087E-02 5.724E-02
24.5 7.086E-08 7.163E-04 1.760E-02 5.279E-02
25.0 4.189E-04 1.459E-02 4.885E-02
25.5 2.253E-04 1.205E-02 4.526E-02
26.0 1.169E-04 9.515E-03 4.139E-02
26.5 5.604E-05 7.431E-03 3.784E-02
27.0 2.361E-05 5.598E-03 3.404E-02
27.5 8.846E-06 4.180E-03 3.058E-02
28.0 3.319E-06 2.998E-03 2.733E-02
28.5 9.803E-07 2.061E-03 2.395E-02
29.0 3.897E-07 1.366E-03 2.093E-02
29.5 8.267E-08 8.787E-04 1.788E-02
30.0 2.062E-08 5.236E-04 1.516E-02
30.5 2.961E-04 1.249E-02
31.0 1.595E-04 1.028E-02
31.5 8.013E-05 8.145E-03
32.0 3.717E-05 6.336E-03
32.5 1.597E-05 4.816E-03
33.0 6.195E-06 3.538E-03
33.5 1.832E-06 2.537E-03
34.0 5.467E-07 1.758E-03
34.5 7.086E-08 1.141E-03
35.0 4.049E-08 7.230E-04
35.5 4.301E-04
36.0 2.424E-04
36.5 1.297E-04
37.0 6.474E-05
37.5 2.847E-05
38.0 1.262E-05
38.5 4.765E-06
39.0 1.444E-06
39.5 4.606E-07
40.0 1.143E-07
40.5 3.376E-08
271
B.3 DIFFERENTIAL QAM
EBNR (dB) DBPSK DQPSK D16QAM D64QAM D256QAM D1024QAM D4096QAM

-6.0 3.885E-01 4.150E-01 4.890E-01 4.835E-01 4.799E-01 4.773E-01 4.765E-01
-5.5 3.775E-01 4.051E-01 4.856E-01 4.800E-01 4.758E-01 4.739E-01 4.730E-01
-5.0 3.651E-01 3.949E-01 4.804E-01 4.748E-01 4.718E-01 4.700E-01 4.694E-01
-4.5 3.510E-01 3.837E-01 4.741E-01 4.696E-01 4.667E-01 4.660E-01 4.657E-01
-4.0 3.357E-01 3.702E-01 4.674E-01 4.636E-01 4.616E-01 4.612E-01 4.616E-01
-3.5 3.208E-01 3.554E-01 4.596E-01 4.578E-01 4.565E-01 4.565E-01 4.574E-01
-3.0 3.021E-01 3.418E-01 4.510E-01 4.508E-01 4.510E-01 4.519E-01 4.535E-01
-2.5 2.850E-01 3.253E-01 4.414E-01 4.438E-01 4.450E-01 4.468E-01 4.485E-01
-2.0 2.664E-01 3.079E-01 4.312E-01 4.364E-01 4.386E-01 4.413E-01 4.438E-01
-1.5 2.455E-01 2.896E-01 4.200E-01 4.288E-01 4.323E-01 4.361E-01 4.396E-01
-1.0 2.260E-01 2.699E-01 4.082E-01 4.199E-01 4.257E-01 4.302E-01 4.344E-01
-0.5 2.045E-01 2.502E-01 3.951E-01 4.111E-01 4.187E-01 4.247E-01 4.296E-01
0.0 1.846E-01 2.293E-01 3.815E-01 4.014E-01 4.117E-01 4.187E-01 4.249E-01
0.5 1.630E-01 2.080E-01 3.659E-01 3.914E-01 4.046E-01 4.126E-01 4.193E-01
1.0 1.415E-01 1.874E-01 3.508E-01 3.820E-01 3.967E-01 4.068E-01 4.140E-01
1.5 1.215E-01 1.666E-01 3.334E-01 3.715E-01 3.893E-01 4.003E-01 4.086E-01
2.0 1.036E-01 1.457E-01 3.152E-01 3.611E-01 3.810E-01 3.934E-01 4.035E-01
2.5 8.454E-02 1.264E-01 2.966E-01 3.504E-01 3.733E-01 3.873E-01 3.979E-01
3.0 6.804E-02 1.068E-01 2.770E-01 3.389E-01 3.654E-01 3.809E-01 3.923E-01
3.5 5.326E-02 8.935E-02 2.568E-01 3.273E-01 3.569E-01 3.738E-01 3.863E-01
4.0 4.047E-02 7.264E-02 2.365E-01 3.154E-01 3.479E-01 3.671E-01 3.806E-01
4.5 2.966E-02 5.857E-02 2.162E-01 3.020E-01 3.393E-01 3.601E-01 3.745E-01
5.0 2.105E-02 4.578E-02 1.951E-01 2.892E-01 3.299E-01 3.531E-01 3.690E-01
5.5 1.434E-02 3.497E-02 1.745E-01 2.754E-01 3.214E-01 3.461E-01 3.627E-01
6.0 9.401E-03 2.598E-02 1.551E-01 2.609E-01 3.118E-01 3.385E-01 3.569E-01
6.5 5.793E-03 1.853E-02 1.359E-01 2.463E-01 3.017E-01 3.313E-01 3.506E-01
7.0 3.349E-03 1.289E-02 1.174E-01 2.307E-01 2.924E-01 3.236E-01 3.446E-01
7.5 1.789E-03 8.499E-03 9.992E-02 2.153E-01 2.827E-01 3.165E-01 3.384E-01
8.0 9.148E-04 5.500E-03 8.415E-02 1.994E-01 2.730E-01 3.092E-01 3.322E-01
8.5 4.147E-04 3.258E-03 6.927E-02 1.832E-01 2.625E-01 3.016E-01 3.263E-01
9.0 1.800E-04 1.898E-03 5.609E-02 1.674E-01 2.519E-01 2.939E-01 3.199E-01
9.5 6.823E-05 1.026E-03 4.415E-02 1.512E-01 2.408E-01 2.859E-01 3.132E-01
10.0 2.277E-05 5.138E-04 3.414E-02 1.353E-01 2.297E-01 2.780E-01 3.069E-01
10.5 6.388E-06 2.456E-04 2.563E-02 1.197E-01 2.185E-01 2.695E-01 3.001E-01
11.0 1.699E-06 1.021E-04 1.863E-02 1.047E-01 2.062E-01 2.617E-01 2.937E-01
11.5 3.664E-07 3.802E-05 1.309E-02 9.063E-02 1.939E-01 2.536E-01 2.871E-01
12.0 1.360E-05 8.761E-03 7.729E-02 1.815E-01 2.450E-01 2.806E-01
12.5 4.424E-06 5.688E-03 6.456E-02 1.687E-01 2.367E-01 2.740E-01
13.0 1.019E-06 3.537E-03 5.355E-02 1.558E-01 2.279E-01 2.675E-01
13.5 2.446E-07 2.097E-03 4.308E-02 1.429E-01 2.194E-01 2.609E-01
14.0 1.142E-03 3.374E-02 1.298E-01 2.101E-01 2.536E-01
14.5 5.867E-04 2.627E-02 1.169E-01 2.010E-01 2.471E-01
15.0 2.804E-04 1.965E-02 1.043E-01 1.916E-01 2.402E-01
272
EBNR (dB) DBPSK DQPSK D16QAM D64QAM D256QAM D1024QAM D4096QAM

15.5 1.209E-04 1.426E-02 9.213E-02 1.818E-01 2.330E-01
16.0 4.838E-05 9.889E-03 8.025E-02 1.722E-01 2.261E-01
16.5 1.787E-05 6.775E-03 6.891E-02 1.618E-01 2.190E-01
17.0 5.227E-06 4.335E-03 5.826E-02 1.513E-01 2.118E-01
17.5 1.687E-06 2.686E-03 4.840E-02 1.406E-01 2.046E-01
18.0 2.852E-07 1.560E-03 3.967E-02 1.303E-01 1.970E-01
18.5 2.480E-08 8.550E-04 3.178E-02 1.196E-01 1.899E-01
19.0 4.353E-04 2.482E-02 1.087E-01 1.821E-01
19.5 2.106E-04 1.916E-02 9.772E-02 1.743E-01
20.0 9.327E-05 1.415E-02 8.759E-02 1.664E-01
20.5 3.600E-05 1.010E-02 7.714E-02 1.582E-01
21.0 1.240E-05 6.994E-03 6.782E-02 1.498E-01
21.5 4.295E-06 4.635E-03 5.813E-02 1.410E-01
22.0 1.133E-06 2.959E-03 4.928E-02 1.324E-01
22.5 2.852E-07 1.794E-03 4.135E-02 1.234E-01
23.0 7.949E-08 1.021E-03 3.384E-02 1.144E-01
23.5 8.585E-09 5.423E-04 2.714E-02 1.055E-01
24.0 2.714E-04 2.108E-02 9.607E-02
24.5 1.279E-04 1.625E-02 8.699E-02
25.0 5.255E-05 1.226E-02 7.844E-02
25.5 2.075E-05 8.682E-03 6.928E-02
26.0 6.969E-06 6.092E-03 6.095E-02
26.5 2.052E-06 4.029E-03 5.278E-02
27.0 5.022E-07 2.627E-03 4.523E-02
27.5 9.074E-08 1.582E-03 3.809E-02
28.0 9.098E-04 3.134E-02
28.5 4.976E-04 2.541E-02
29.0 2.545E-04 2.018E-02
29.5 1.149E-04 1.579E-02
30.0 4.812E-05 1.188E-02
30.5 1.983E-05 8.584E-03
31.0 6.665E-06 6.072E-03
31.5 2.054E-06 4.198E-03
32.0 4.859E-07 2.731E-03
32.5 1.196E-07 1.692E-03
33.0 1.014E-03
33.5 5.607E-04
34.0 2.929E-04
34.5 1.416E-04
35.0 6.349E-05
35.5 2.548E-05
36.0 9.663E-06
36.5 3.142E-06
37.0 1.026E-06
37.5 2.501E-07
273
B.4 DIFFERENTIAL PHASE SHIFT KEYING
EBNR (dB) D8PSK D16PSK D32PSK D64PSK D128PSK D256PSK

-6.0 4.285E-01 4.341E-01 4.372E-01 4.402E-01 4.427E-01 4.451E-01
-5.5 4.204E-01 4.271E-01 4.313E-01 4.353E-01 4.384E-01 4.414E-01
-5.0 4.127E-01 4.203E-01 4.257E-01 4.299E-01 4.334E-01 4.373E-01
-4.5 4.050E-01 4.134E-01 4.198E-01 4.242E-01 4.289E-01 4.327E-01
-4.0 3.946E-01 4.060E-01 4.131E-01 4.183E-01 4.239E-01 4.279E-01
-3.5 3.840E-01 3.977E-01 4.062E-01 4.127E-01 4.185E-01 4.232E-01
-3.0 3.736E-01 3.892E-01 3.997E-01 4.069E-01 4.132E-01 4.195E-01
-2.5 3.625E-01 3.809E-01 3.919E-01 4.009E-01 4.080E-01 4.141E-01
-2.0 3.514E-01 3.718E-01 3.851E-01 3.951E-01 4.029E-01 4.100E-01
-1.5 3.387E-01 3.629E-01 3.778E-01 3.886E-01 3.969E-01 4.053E-01
-1.0 3.255E-01 3.535E-01 3.703E-01 3.827E-01 3.921E-01 4.004E-01
-0.5 3.128E-01 3.436E-01 3.628E-01 3.762E-01 3.868E-01 3.955E-01
0.0 2.981E-01 3.351E-01 3.550E-01 3.697E-01 3.811E-01 3.908E-01
0.5 2.834E-01 3.244E-01 3.476E-01 3.636E-01 3.759E-01 3.858E-01
1.0 2.681E-01 3.143E-01 3.400E-01 3.575E-01 3.710E-01 3.816E-01
1.5 2.524E-01 3.042E-01 3.320E-01 3.506E-01 3.652E-01 3.769E-01
2.0 2.370E-01 2.938E-01 3.241E-01 3.443E-01 3.598E-01 3.720E-01
2.5 2.209E-01 2.837E-01 3.164E-01 3.384E-01 3.545E-01 3.668E-01
3.0 2.045E-01 2.729E-01 3.085E-01 3.316E-01 3.484E-01 3.617E-01
3.5 1.876E-01 2.625E-01 3.013E-01 3.252E-01 3.431E-01 3.570E-01
4.0 1.701E-01 2.513E-01 2.931E-01 3.185E-01 3.375E-01 3.526E-01
4.5 1.545E-01 2.401E-01 2.850E-01 3.120E-01 3.320E-01 3.473E-01
5.0 1.377E-01 2.290E-01 2.767E-01 3.055E-01 3.267E-01 3.425E-01
5.5 1.216E-01 2.168E-01 2.684E-01 2.986E-01 3.208E-01 3.378E-01
6.0 1.063E-01 2.050E-01 2.603E-01 2.925E-01 3.148E-01 3.328E-01
6.5 9.163E-02 1.927E-01 2.516E-01 2.860E-01 3.097E-01 3.279E-01
7.0 7.798E-02 1.795E-01 2.428E-01 2.788E-01 3.038E-01 3.230E-01
7.5 6.511E-02 1.663E-01 2.344E-01 2.725E-01 2.982E-01 3.179E-01
8.0 5.365E-02 1.535E-01 2.260E-01 2.656E-01 2.925E-01 3.133E-01
8.5 4.324E-02 1.400E-01 2.169E-01 2.590E-01 2.866E-01 3.077E-01
9.0 3.368E-02 1.269E-01 2.078E-01 2.517E-01 2.813E-01 3.032E-01
9.5 2.594E-02 1.138E-01 1.984E-01 2.451E-01 2.749E-01 2.978E-01
10.0 1.950E-02 1.011E-01 1.894E-01 2.380E-01 2.690E-01 2.929E-01
10.5 1.416E-02 8.868E-02 1.792E-01 2.311E-01 2.638E-01 2.882E-01
11.0 9.755E-03 7.768E-02 1.689E-01 2.235E-01 2.574E-01 2.828E-01
11.5 6.644E-03 6.663E-02 1.585E-01 2.167E-01 2.520E-01 2.782E-01
12.0 4.256E-03 5.575E-02 1.480E-01 2.099E-01 2.462E-01 2.727E-01
12.5 2.599E-03 4.632E-02 1.380E-01 2.019E-01 2.403E-01 2.678E-01
13.0 1.524E-03 3.752E-02 1.274E-01 1.949E-01 2.344E-01 2.624E-01
13.5 8.250E-04 3.002E-02 1.157E-01 1.869E-01 2.284E-01 2.577E-01
14.0 4.195E-04 2.321E-02 1.053E-01 1.797E-01 2.224E-01 2.522E-01
14.5 1.977E-04 1.764E-02 9.424E-02 1.715E-01 2.165E-01 2.470E-01
15.0 8.560E-05 1.295E-02 8.412E-02 1.635E-01 2.105E-01 2.424E-01
274

15.5 3.468E-05 9.227E-03 7.447E-02 1.554E-01 2.045E-01 2.372E-01
16.0 1.207E-05 6.288E-03 6.441E-02 1.466E-01 1.976E-01 2.316E-01
16.5 3.695E-06 4.154E-03 5.550E-02 1.385E-01 1.918E-01 2.265E-01
17.0 8.434E-07 2.597E-03 4.707E-02 1.295E-01 1.857E-01 2.215E-01
17.5 4.712E-07 1.545E-03 3.874E-02 1.199E-01 1.793E-01 2.163E-01
18.0 8.655E-04 3.183E-02 1.117E-01 1.730E-01 2.109E-01
18.5 4.559E-04 2.523E-02 1.023E-01 1.664E-01 2.057E-01
19.0 2.224E-04 1.958E-02 9.335E-02 1.599E-01 2.004E-01
19.5 1.022E-04 1.483E-02 8.417E-02 1.535E-01 1.951E-01
20.0 4.181E-05 1.086E-02 7.507E-02 1.465E-01 1.899E-01
20.5 1.507E-05 7.756E-03 6.665E-02 1.397E-01 1.841E-01
21.0 4.881E-06 5.328E-03 5.806E-02 1.324E-01 1.789E-01
21.5 1.596E-06 3.544E-03 5.047E-02 1.246E-01 1.732E-01
22.0 4.313E-07 2.226E-03 4.241E-02 1.171E-01 1.678E-01
22.5 4.372E-08 1.355E-03 3.591E-02 1.091E-01 1.626E-01
23.0 7.485E-04 2.931E-02 1.018E-01 1.569E-01
23.5 4.003E-04 2.363E-02 9.400E-02 1.514E-01
24.0 1.939E-04 1.867E-02 8.575E-02 1.451E-01
24.5 8.853E-05 1.435E-02 7.795E-02 1.397E-01
25.0 3.670E-05 1.072E-02 7.050E-02 1.341E-01
25.5 1.346E-05 7.758E-03 6.267E-02 1.284E-01
26.0 4.681E-06 5.449E-03 5.519E-02 1.215E-01
26.5 1.166E-06 3.633E-03 4.822E-02 1.154E-01
27.0 4.211E-07 2.326E-03 4.153E-02 1.090E-01
27.5 9.718E-08 1.412E-03 3.505E-02 1.022E-01
28.0 8.387E-04 2.897E-02 9.537E-02
28.5 4.499E-04 2.392E-02 8.847E-02
29.0 2.309E-04 1.920E-02 8.185E-02
29.5 1.092E-04 1.491E-02 7.498E-02
30.0 4.736E-05 1.128E-02 6.802E-02
30.5 1.779E-05 8.368E-03 6.079E-02
31.0 5.840E-06 6.014E-03 5.439E-02
31.5 2.055E-06 4.169E-03 4.791E-02
32.0 6.684E-07 2.760E-03 4.160E-02
32.5 2.814E-08 1.750E-03 3.563E-02
33.0 1.437E-08 1.052E-03 3.020E-02
33.5 5.993E-04 2.497E-02
34.0 3.186E-04 2.035E-02
34.5 1.581E-04 1.620E-02
35.0 7.317E-05 1.278E-02
35.5 3.121E-05 9.642E-03
36.0 1.143E-05 7.112E-03
275

36.5 4.004E-06 5.069E-03
37.0 1.335E-06 3.518E-03
37.5 3.720E-07 2.292E-03
38.0 4.310E-08 1.452E-03
38.5 8.686E-04
39.0 4.952E-04
39.5 2.643E-04
40.0 1.283E-04
40.5 5.971E-05
41.0 2.333E-05
41.5 9.001E-06
42.0 3.023E-06
42.5 7.854E-07
43.0 2.377E-07
43.5 3.100E-08
276
Appendix C RF Measurements
APPENDIX C RF MEASUREMENTS
These results are from the measurements presented in section 3.8 starting page 104.
The location of each of the measurements is shown in Figure 3-9.
The frequency response of the channel measurement set up is shown in Figure A - 9.

The combined frequency response of the two antennas and the preamplifier was
difficult to measure due to the effects of multipath, causing frequency selective
fading. To overcome this, the frequency response was estimated by averaging the
radio channel response from all the measurement sets taken. This frequency response
was used to calibrate the measurements taken. The error in calibration is about ± 1
dB.
Figure A - 9, Frequency response of the radio channel measurement set up,

including antennas, and pre-amplifier.
277
Figure A - 10, Wide bandwidth frequency fading over distance for room 205 to
210.
210.
Streaks in the response are due to narrow band interfering transmissions.
278
210.
210.
279
210. This is a LOS link, as a result the correlation bandwidth is wider.
210. This is a LOS link.
280
Figure A - 16, Wide bandwidth frequency fading over distance for room 210
above the centre of the receiver track to the guide track also in room 210.
210.
281
210.
Rm120 is on the next floor down. From 155 –170 cm the data is blurred due to
movement within the building during the measurement.
210.
Rm122 is on the next floor down, resulting in additional path loss and a lower
correlation bandwidth for the short transmission distance.
282
C.1 DIVERSITY
The performance of an optimal power combining diversity receiver was simulated

for the channel response measurements taken in Appendix C. The performance of the
diversity receiver was estimated for different antenna spacings. The measurements
shown in Appendix C were taken along a linear track 2.4 m in length. As part of the
simulation virtual antennas were placed along this measurement track, and the power
from these antennas were combined to obtain the estimated performance of a
diversity receiver. These virtual antennas were slid along the measurement track in
order to obtain the fading nature of the simulated diversity receiver.
The probability distribution of the fading depth was found by taking the statistics
across each of the frequency bins and each of the different measurement locations.
The results from all 10 measurement-sets in Appendix C were used to estimate the
performance.
Separation : 20.3cm
% Prob. Rx. Signal Strength < Abscissa
100
Measured
Rayleigh
10
No Diversity
1
Diversity 2 Diversity 4
0.1
-28 -24 -20 -16 -12 -8 -4 0 4
Received Signal Strength (dB about median)
Figure A - 20, Simulated fading distribution for optimal power combining

diversity reception.
Note: The data has been normalised about the median. The average power for
an antenna diversity of 2 was 3.0 dB higher, and for an antenna diversity of 4 it
was 6.0 dB higher. (script s0030)
Figure A - 20 shows the fading probability distribution as the amount of diversity is

increased. The “No diversity” measurement shows the fading probability for a single
antenna. The “Diversity 2” results show the performance of a simulated diversity
283
receiver using the measurement sets taken. The distance between the two antennas in
this result were 20.3 cm. The “Diversity 4” result shows the result for the fading
depth when the power is combined from 4 virtual antennas spaced 20.3 cm apart.
Also shown on the plot is the fading depth that would be expected in a Rayleigh
fading channel with zero correlation between the diversity antennas. The simulated
diversity performance based on the measured channel fading show a greater fading
depth than Rayleigh fading. This is likely to be partly due to correlation between the
antennas spaced at 20.3 cm apart. This simulation was repeated for an antenna
separation of 40 cm with similar results. The increased fading in the measured results
compared with Rayleigh fading may also be due to small amounts of shadowing
along the path of the measurement.
Figure A - 21 shows the fading depth as a function of the antenna separation for a
receiver using an antenna diversity of 4. This shows that for antenna separations
must be greater than approximately 10 cm (1/3 wavelength) in order to achieve the
best performance from the diversity receiver.
Diversity: 4, Frequency: 1005 MHz
-2
-4
Fading Depth (dB from median)
-6
-8
-10
-12
-14 10%
3%
-16
1%
-18 0.3%
0.1%
-20
1 2 3 4 5 6 7 8 910 15 20 25 30 40
Separation between antennas (cm)
Figure A - 21, Effect of antenna separation on the fading depth for diversity
reception. (script s0030)
The five results show the probability of the fading exceeding the value shown.
For example, for a separation of 10 cm there is a 0.1% chance of the fading
exceeding 10.1 dB. The non-smooth characteristics of the 0.1% and 0.3%
probability curves are due to measurement variation.
284
Appendix D Phase tables for evolved low Crest Factor Discrete Multi-Tone signals
APPENDIX D PHASE TABLES FOR

EVOLVED LOW CREST FACTOR
DISCRETE MULTI-TONE SIGNALS
The following tables show the phase angles of each subcarrier in radians for the
phase sequence found to have the lowest crest factor. These phase sequences were
developed using the genetic algorithm optimisation algorithm outlined in section 6.2.
Each column in the table represents the phase sequence for a given number of tones
in the signal. Each row specifies the phase angle for each of the tones (c1, c2, etc) in
the signal. For example: for a DMT signal with 9 tones the best phase sequences is:
{-0.1506, 0.8477, 1.1603, 2.7360, -0.3339, 2.0445, -0.2213, -1.2250, -2.9157}
where the lowest frequency tone (c1) is set to a phase of –0.1506 radians, the next
tone (c2) is set to 0.8477 radians, through to the highest frequency tone set to a phase
of –2.9157 radians (c9). All the tones must have the same amplitude.
285
N 2 3 4 5 6 7 8 9 10 11 12 13
CF (dB) 3.0096 2.2180 1.8401 1.7562 1.9353 1.9959 1.1357 0.9195 1.3760 2.0681 1.6585 1.2855
c1 0.3227 0.1739 0.0698 -0.0337 -0.2617 0.3423 0.1494 -0.1506 -0.4931 -0.0515 -0.2348 -0.1128
c2 1.4762 0.7015 0.6043 0.3935 -0.1001 0.6874 1.3464 0.8477 0.5232 0.2956 -0.2604 0.9819
c3 -1.9125 2.9478 3.0420 2.9791 1.5422 1.1741 1.1603 1.2104 1.1168 1.0707 1.5980
c4 0.8174 -0.8563 -0.6664 -2.4205 3.0161 2.7360 2.6591 2.6873 -2.9699 2.2874
c5 -2.5333 1.5292 0.6219 -0.5100 -0.3339 -0.7343 -1.9001 -1.4402 -2.4297
c6 0.3593 -1.1533 -3.1208 2.0445 2.0989 1.0755 -0.3012 0.6287
c7 -2.4180 2.5664 -0.2213 -1.4154 -2.4557 -2.1887 1.9462
c8 0.6028 -1.2250 2.7908 1.5700 0.2226 -1.2270
c9 -2.9157 1.4838 -0.5681 -1.8615 1.8849
c10 -0.1529 -1.9320 1.7537 0.2902
c11 -2.8432 1.0931 -1.1668
c12 0.0312 -1.2351
c13 -2.7297
N 14 15 16 17 18 19 20 21 22 23 24 25
CF (dB) 1.5296 1.4061 1.0676 1.3035 1.1744 1.2271 1.3548 1.3244 1.0552 1.2040 1.1719 0.9441
c1 -0.3143 -0.1263 -0.5219 -0.4220 -0.3229 -0.5961 -0.4055 -0.0107 -0.4534 0.2200 -0.1952 -0.2114
c2 0.2432 0.4711 0.2380 0.3865 0.2420 0.2429 0.2536 1.1275 0.6360 0.8695 0.2656 0.8631
c3 0.5139 1.2162 1.2461 0.8953 0.0416 0.4724 0.2400 0.8363 0.4350 0.9187 1.0006 0.5844
c4 1.7704 1.5999 1.2962 1.6693 1.3563 0.8890 1.0576 1.4033 1.0431 1.4434 1.0848 1.2006
c5 -2.3605 2.9087 2.6942 2.4072 -2.8477 3.0993 2.7334 2.1464 2.0145 1.7812 1.7638 1.7043
c6 -0.6212 -0.2779 -0.4868 -1.6116 -1.8718 -2.2509 -2.3239 -3.1319 -2.9506 2.7395 2.9505 -3.0657
c7 1.9108 1.5417 1.0382 1.1731 0.1778 -0.5172 -0.8969 -0.7769 -0.9944 -1.6769 -1.4679 -1.6283
c8 -1.4077 -2.7183 -2.9410 2.6807 2.3801 2.6164 1.5054 0.8011 0.9142 0.1213 0.1237 -0.5484
c9 1.8683 1.1185 0.0036 -0.8592 -1.3292 -1.8447 -1.6609 2.8528 2.3265 1.9701 2.4190 2.0344
c10 -0.3490 -1.1290 -2.6962 2.3154 1.6415 1.1984 -0.2321 -0.5192 -0.8118 -1.9034 -2.1511 -2.6010
c11 -2.9433 1.6382 1.6725 0.4448 -1.2756 -2.1785 -2.3851 2.8782 2.3203 1.8072 0.4510 -0.8429
c12 1.7131 -0.0948 -1.8216 -2.7057 2.4617 1.9484 0.0099 -1.0587 -0.9372 -1.2398 -2.1815 2.6975
c13 0.9508 -0.8998 2.6614 0.9486 0.0636 -1.5241 -2.0682 2.3418 1.9614 0.7807 0.8952 -0.7629
c14 -0.0831 -2.0682 2.2164 -0.1557 -1.2543 2.6941 1.6879 0.3070 -1.5622 -2.4343 -3.0688 2.1955
c15 -3.0921 0.8151 -1.2969 2.6074 1.4298 -0.0698 -1.8807 2.9414 1.8431 0.3025 -1.0723
c16 -0.3348 -2.1686 0.9460 -1.1177 -2.1905 2.0249 0.9088 -0.3910 -1.7525 -2.7590
c17 2.9411 0.7987 -1.8713 1.8195 0.3572 -1.3860 -2.5124 2.0835 0.8732
c18 -0.1094 -2.4346 1.3435 -0.4166 -2.9621 1.4247 0.1917 -1.5739
c19 2.6738 0.4657 -1.7183 2.0559 0.3541 -1.9411 -2.8132
c20 -0.4869 -1.8388 1.2277 -1.1698 2.8961 1.5308
c21 -3.0810 1.1041 -0.8913 1.9613 -0.1880
c22 -0.1812 -2.3530 1.6802 -0.8995
c23 -3.1265 0.7232 -2.1706
c24 -0.0529 -1.9480
c25 2.8559
286
N 26 27 28 29 30 31 32 34 36 38 40 42
CF (dB) 0.8956 1.0133 1.1251 1.1699 1.2577 1.0325 1.1387 0.9430 0.9542 1.1888 1.1130 0.9188
c1 -0.5829 -0.9571 -0.7045 -0.4533 -0.3604 -0.5336 -0.4014 -0.3149 -0.8210 -0.5677 -0.4621 -0.6650
c2 0.3906 -0.0206 0.6325 0.6797 0.4653 0.3320 0.3673 0.4108 0.0075 0.3619 0.3729 0.2501
c3 0.3867 0.1552 0.3764 0.4762 0.1261 0.6501 0.7934 0.5717 0.2890 0.3591 0.6553 0.4720
c4 1.1177 0.8282 1.3278 1.5319 0.7180 0.5086 1.0739 1.1594 0.8382 0.7963 0.4722 0.7403
c5 1.4465 1.3389 1.8584 1.4224 1.2045 1.3927 1.2120 1.2048 1.2137 1.2182 1.4019 0.8690
c6 2.9086 2.8253 -2.9376 2.7094 2.5228 2.2783 2.4187 2.0015 1.5555 1.8393 1.4324 1.5233
c7 -1.7391 -2.1450 -2.2856 -2.9256 -2.3025 -2.2570 -2.6220 -3.1093 2.7923 2.8033 2.8631 2.8752
c8 -0.2711 -0.3521 -0.4919 -1.3678 -1.3458 -1.4864 -0.7328 -1.7771 -2.1231 -2.5376 -2.5514 -2.2075
c9 1.6222 1.4643 2.1462 1.1630 0.4167 -0.1381 -0.0796 -0.0199 -0.7593 -1.1638 -1.1878 -1.0911
c10 -2.3949 -2.7933 2.8563 2.2907 2.2077 1.8730 1.7388 1.2619 1.1484 0.2462 0.0834 -0.6207
c11 -0.7489 -0.5723 -1.5618 -2.1579 -2.0624 -2.5096 3.1500 2.8907 2.7525 2.0819 1.9094 1.1940
c12 2.1981 1.4586 1.3100 0.8547 0.3606 -0.4489 -0.7996 -1.5420 -2.2753 -2.5713 -3.1608 3.1359
c13 -0.8726 -0.9790 -3.1740 2.3575 2.7569 2.3360 1.3554 0.6230 -0.3638 -1.2097 -1.3888 -1.8455
c14 2.1356 1.5357 -0.0985 -0.3045 -0.6869 -0.6500 -2.3914 -3.0363 2.1613 1.3177 1.0749 -0.1139
c15 -1.3128 -1.2534 2.5837 2.0418 1.7489 1.4047 0.9445 -0.4022 -1.1893 -2.8029 2.8850 2.1152
c16 1.7840 0.9763 -0.5141 -0.4666 -1.5344 -2.3154 2.9156 2.3104 1.3805 -0.0982 -0.8644 -1.8366
c17 -0.0779 -1.3895 2.4841 1.8283 2.0342 1.2316 -0.5656 -1.7308 -2.0996 2.5932 1.7838 0.7421
c18 -2.5874 2.5171 0.8816 -0.2637 -0.9305 -1.0011 -3.1489 1.3148 -0.3464 -0.8781 -2.5896 2.8478
c19 1.5684 0.1765 -1.3547 -2.2402 2.7093 1.7453 1.3227 -1.1314 2.8437 1.7511 0.1747 -1.0847
c20 -0.1587 -1.9856 2.3095 1.3832 0.2070 -1.2617 -1.9902 2.2394 0.5848 -1.6820 -3.0595 1.8793
c21 -2.0296 2.5702 1.0751 -0.4963 -2.0972 2.8037 1.6270 -0.5740 -2.2136 1.7913 0.1660 -0.9072
c22 2.5737 0.9878 -0.3762 -2.2303 2.1859 0.6821 -1.0669 2.8915 1.2249 -1.1803 3.0464 2.2257
c23 2.0077 -0.7581 -1.8973 2.4868 0.2709 -1.4959 -2.8259 0.5376 -1.4301 2.5155 0.4225 -1.4130
c24 1.0272 -1.4714 2.7017 0.6037 -0.8273 -3.0624 1.6379 -1.5150 1.8821 0.5050 -2.3310 1.6738
c25 0.7693 -2.4172 1.6071 -0.5767 -2.5348 2.2266 -0.3347 2.9585 -0.0296 -2.7667 1.3548 -0.8591
c26 -0.4273 -2.6937 1.4484 -1.0221 2.2722 0.2583 -1.6185 1.4737 -1.9153 1.4545 -1.5782 -3.0616
c27 2.4435 0.8836 -2.0902 1.4935 -0.7889 -2.8143 -0.4702 2.5881 -0.7649 2.1346 0.5837
c28 -0.0145 -2.2062 0.6615 -1.9006 2.2819 -1.9919 1.2642 -2.6260 0.7646 -2.0106
c29 2.7755 0.7787 -1.8910 1.6006 2.9169 -0.2843 2.2196 -1.3526 1.8169
c30 -0.1720 -2.4314 1.3488 1.9349 -2.5088 1.0373 -3.0078 0.0015
c31 2.7566 0.9095 1.6870 2.3292 -1.3894 1.6859 -1.4456
c32 -0.0790 0.9137 1.9995 -2.6045 0.1360 2.7574
c33 0.5499 1.2610 2.6657 -1.0742 0.7891
c34 -0.3637 0.8912 2.0626 -2.4307 0.0535
c35 0.3888 1.7345 2.4731 -1.1556
c36 -0.5790 1.0586 1.8962 -2.4971
c37 0.5361 1.2238 2.3982
c38 -0.7169 0.8796 1.5205
c39 0.5340 1.2946
c40 -0.3249 0.9101
c41 0.5586
c42 -0.5792
287
N 44 46 48 50 52 54 56 58 60 62 64 68
CF (dB) 1.0488 0.8532 1.0083 0.6722 1.0039 1.0210 1.0769 0.8749 0.9523 0.8729 0.6497 0.9294
c1 -0.3672 -0.7000 -0.4721 -0.7352 -0.6789 -0.7884 -0.4569 -0.6801 -0.7019 -0.6667 -0.7835 -0.5704
c2 0.3432 0.3875 0.1339 0.3779 0.1906 -0.0295 0.3630 0.0328 0.1436 0.2478 0.1563 0.1868
c3 0.5836 0.2467 0.4575 0.3017 0.4815 0.1621 0.5291 0.1928 0.4523 0.2959 0.5840 0.3545
c4 0.7219 0.7115 0.7983 0.9442 0.6655 0.5736 0.4339 0.7313 0.4302 0.7634 0.6903 0.7806
c5 1.2275 0.9359 0.9406 1.0287 0.8392 0.7504 0.8594 0.5182 0.9408 0.7533 1.1734 0.8108
c6 1.3225 1.4730 1.3388 1.4566 1.4530 1.1877 1.1233 1.1379 0.9331 1.1853 1.1812 1.1655
c7 2.5813 2.0500 1.9793 1.9028 1.9162 1.4858 1.5157 1.2884 1.5048 1.4275 1.6980 1.5260
c8 -2.9130 3.0362 2.8013 2.7648 -3.0122 2.8974 2.6563 2.4689 2.2287 2.1635 1.9463 1.9038
c9 -1.6215 -1.9820 -2.1956 -2.3763 -2.3523 -2.3066 -2.5743 -2.9679 3.1185 -2.9599 2.6055 2.8857
c10 -0.1148 -0.6402 -0.6336 -0.8215 -1.1346 -1.5871 -1.5045 -1.9419 -2.1950 -2.2701 -2.7304 -2.6889
c11 0.9026 0.7848 0.2664 0.3426 -0.0766 -0.2847 -0.5095 -0.6344 -1.1982 -0.9832 -1.4055 -1.7162
c12 2.2961 1.9352 1.5782 1.4733 1.4008 0.8171 0.4431 0.1380 0.1677 -0.1305 0.1101 -0.1916
c13 -2.4891 -2.9942 2.9442 2.9686 2.7835 2.3480 1.5955 1.6668 1.2287 1.0528 0.8676 0.5077
c14 -0.6867 -1.0353 -1.8006 -1.8313 -2.2711 -2.6512 -3.0269 2.8013 2.6587 2.4118 2.0974 1.6667
c15 1.3393 0.9616 -0.1107 -0.6294 -0.5667 -1.2032 -1.4292 -2.0510 -2.4811 -2.6299 2.9992 2.8141
c16 -2.5187 2.8628 2.2626 1.5694 0.9741 0.6087 -0.3139 -0.3167 -0.9637 -1.3186 -1.9585 -2.3353
c17 -0.6639 -0.5815 -1.8677 -2.2438 2.7530 2.3404 1.8224 1.4508 0.9293 0.0809 -0.4962 -0.7795
c18 1.8657 1.1936 0.5809 -0.3744 -0.8721 -1.8494 -1.9364 -2.8582 2.8924 2.1213 1.3926 0.7742
c19 -2.3839 3.0835 2.5210 2.1426 1.0696 0.4444 -0.6170 -0.5493 -1.1724 -1.9283 -2.9933 2.4492
c20 0.0889 -0.5971 -1.7114 -2.0635 -3.0606 2.6097 1.7261 1.3563 0.3667 -0.2694 -0.6423 -1.7441
c21 -2.8912 2.0551 0.6757 -0.4308 -0.9262 -1.7029 -2.5369 2.9154 2.3373 1.8720 1.2220 -0.0628
c22 0.3033 -0.9431 -2.6329 2.1787 1.0332 0.3565 -0.7227 -1.1108 -2.1961 -2.5529 2.8977 2.1223
c23 -2.9116 1.8758 0.3225 -1.0803 -2.3703 3.0093 2.1748 0.9232 -0.0318 -0.8347 -1.4798 -2.6220
c24 0.0258 -1.3403 -2.8739 2.2229 0.5155 -0.7822 -0.9733 -2.7273 2.4565 1.3890 0.2800 -0.6854
c25 2.8702 1.9808 -0.0042 -1.2851 -2.9012 2.3798 1.3279 0.3433 -1.0163 -2.0489 2.5534 1.3666
c26 0.2683 -1.4494 -2.8411 1.7615 -0.4085 -1.2429 -2.6372 2.9597 1.6513 0.6896 -0.5654 -2.4149
c27 -1.9066 2.0542 0.1173 -1.2243 2.7218 1.7050 0.4609 -1.1158 -2.0357 2.9681 2.3140 -0.1885
c28 1.7153 -0.7006 -2.9921 1.6631 -0.0010 -1.3516 -3.0237 1.6976 0.6948 -0.6911 -1.5420 2.4267
c29 -0.2730 -2.7771 1.0007 -1.4652 -3.0028 1.8828 -0.1642 -1.6372 -2.9798 2.0081 1.3881 -1.3311
c30 -2.8585 1.6437 -1.0190 2.0946 0.2127 -1.0003 -3.0967 1.3636 0.2506 -1.4256 -2.3820 1.1651
c31 1.2455 -1.3606 2.7344 0.2802 -2.6815 2.3665 0.5005 -1.6701 -3.0891 1.4110 0.1274 -2.3946
c32 -0.7053 2.8467 0.5492 -1.8986 1.4559 -0.3768 -2.6081 1.6232 -0.0492 -1.7298 2.8932 0.3747
c33 -2.3283 0.7961 -1.8102 1.7008 -0.8815 2.9570 1.1446 -0.9255 -2.8405 1.5877 -0.3090 -3.0881
c34 2.4368 -1.3169 2.0218 -0.2990 3.1091 0.6540 -1.8752 2.7487 1.1903 -1.4759 -3.1493 -0.6730
c35 1.2489 -2.8254 0.0873 -2.8069 1.1070 -1.4388 1.8172 -0.1022 -1.5990 1.9953 0.5707 2.5292
c36 -0.3926 2.2126 -1.6138 1.0969 -1.5870 2.4983 -0.0499 -3.0374 1.6111 -0.7370 -2.1429 -0.0798
c37 -1.8174 0.6240 -3.0148 -0.1580 2.6815 0.3348 -2.2540 1.1117 -1.2978 3.0216 1.1638 -2.9887
c38 -2.7548 -0.8536 1.8347 -1.6587 1.1917 -2.1628 1.5175 -1.2671 2.3931 0.2769 -1.3367 0.3024
c39 2.1135 -2.2546 0.6339 2.9571 -0.4992 2.1011 -0.4935 -3.0846 -0.0142 -2.5463 1.9122 -2.3995
c40 1.9022 2.9083 -1.1641 1.6481 -2.1881 0.3947 -2.0663 1.4485 -1.8393 1.3008 -1.2923 1.3895
c41 1.2528 2.2125 -2.3612 0.2191 2.5539 -1.1368 1.7524 -1.0391 2.3669 -0.6000 2.6550 -1.3474
c42 0.9751 1.4862 2.8254 -1.3656 1.0655 -2.7156 0.1328 3.1166 0.3053 -2.3550 0.8180 2.2887
c43 0.5931 1.1656 2.0170 -2.5138 0.1223 1.7299 -1.4355 1.2083 -1.7704 1.5442 -1.2039 -0.0083
c44 -0.2685 0.5731 1.4102 2.7592 -1.1403 0.6352 -2.6827 -0.5498 2.4762 -0.3322 -2.9384 -2.6748
c45 0.5402 0.9778 2.1742 -2.1715 -0.6424 2.3331 -2.1088 0.6573 -2.3587 1.2920 1.5029
c46 -0.6584 0.6993 1.5828 2.6519 -1.5142 0.5475 2.8748 -0.9696 1.6590 -1.1409 -0.4603
c47 0.0871 1.3137 2.0792 -2.6212 -0.5064 1.3968 -2.3428 0.1694 -3.1044 -2.3329
c48 -0.5210 0.5684 1.3803 2.2131 -1.7908 0.3041 2.3522 -1.1659 1.2369 1.7310
c49 0.5436 1.0299 1.7482 -2.5594 -0.9327 1.1951 -2.6036 -0.3547 0.2683
c50 -0.6785 0.7499 1.0925 2.4951 -2.1376 0.2201 2.2904 -1.7418 -1.9861
c51 0.3466 0.9813 1.7183 -3.0444 -1.1460 0.9572 -2.7516 2.3589
c52 -0.5345 0.4964 1.1610 1.9861 -2.3085 0.0232 2.1755 0.8216
c53 0.3370 0.4787 1.7187 2.9152 -1.3000 1.3396 -0.8130
c54 -0.6012 0.5756 1.0270 2.1406 -2.2198 -0.3440 -2.1344
c55 0.1020 1.0400 1.4442 2.9698 -1.7062 2.9409
c56 -0.7951 0.4752 1.4224 2.0668 -2.7104 1.5242
288
N 44 46 48 50 52 54 56 58 60 62 64 68
c57 0.2111 0.8376 1.7619 2.7551 0.8373
c58 -0.6500 0.5449 1.1821 2.4264 -0.8283
c59 0.2177 1.0896 1.7888 -1.7630
c60 -0.6123 0.5940 1.6728 -2.6455
c61 0.4198 1.0345 2.5010
c62 -0.5951 0.7952 2.1024
c63 0.3031 1.5230
c64 -0.6640 1.1232
c65 1.0375
c66 0.5432
c67 0.1909
c68 -0.6021
N 72 76 80 84 88 92 96 100 104 108 112 116

CF (dB) 0.9224 0.9004 0.8051 0.8633 0.9003 0.8837 0.8998 0.6708 0.8852 0.8263 0.8374 0.8069
c1 -0.6861 -0.6972 -0.7469 -0.7355 -0.8858 -0.6389 -0.7659 -1.0036 -0.7377 -0.7658 -0.7753 -0.9469
c2 0.0671 -0.0120 -0.0116 -0.1188 -0.0457 0.0231 0.0154 -0.0002 -0.0764 0.0072 -0.3397 -0.2138
c3 0.4416 0.3344 0.2892 0.5384 0.4424 0.4863 0.2398 0.2398 0.4068 0.4141 0.2850 0.1987
c4 0.6019 0.3765 0.3344 0.4895 0.4715 0.3326 0.4169 0.4514 0.4454 0.5016 0.3990 0.4265
c5 0.6375 0.8151 0.9408 0.7691 0.5208 0.7694 0.6500 0.6684 0.6839 0.7828 0.5517 0.4817
c6 0.9717 0.8289 1.0345 0.8654 0.7716 0.7614 0.9996 0.6667 0.8385 0.7309 0.6932 0.4959
c7 1.3853 1.3681 1.3855 1.2637 0.9632 1.2176 1.0433 1.1146 1.1183 1.0695 1.0401 1.0356
c8 1.6246 1.6048 1.5176 1.5870 1.4363 1.2933 1.3393 1.1450 1.3025 1.2948 0.9633 1.0141
c9 2.7834 2.6674 2.2400 2.0021 1.8708 1.7746 1.6643 1.6031 1.6671 1.6310 1.4882 1.5177
c10 -2.8044 -2.8573 2.9111 2.9602 2.9601 2.4941 2.3653 2.1290 2.0159 1.7216 1.6235 1.5938
c11 -1.8176 -2.0895 -2.5009 -2.8036 -2.6516 -2.8566 3.0443 -3.1611 3.0787 2.8995 2.6346 2.6478
c12 -0.8856 -1.1437 -1.7297 -1.6187 -1.8695 -1.9055 -2.3639 -2.5192 -2.6973 -2.9583 -3.0622 -2.8323
c13 -0.1275 -0.0518 -0.2932 -0.6464 -1.3170 -1.3676 -1.5927 -1.6074 -1.8753 -2.1581 -2.4541 -2.4789
c14 1.2794 0.9746 0.5979 0.2201 -0.3309 -0.2810 -0.3647 -0.7230 -1.0212 -1.3435 -1.5075 -1.6779
c15 2.4086 2.2262 1.5301 1.0814 0.6919 0.2593 0.2070 -0.1738 -0.1728 -0.4016 -0.7752 -0.9228
c16 -2.7184 -3.0929 2.7031 2.2221 2.1507 1.4053 1.0833 0.9185 0.5795 0.3389 0.0317 -0.2864
c17 -1.6562 -1.9219 -2.4549 -3.0267 3.0035 2.4688 2.1560 1.7356 1.5719 1.4006 0.7645 0.7529
c18 -0.2459 -0.6345 -1.4926 -1.8980 -2.4203 -2.4918 -2.9158 2.9797 2.6868 2.0650 1.7852 1.8157
c19 1.6097 0.9672 0.0237 -0.5255 -1.1236 -1.4686 -2.0656 -2.3355 -2.4615 3.0692 2.8636 2.7581
c20 3.1309 2.7993 1.3832 0.5278 0.1325 -0.3694 -0.8161 -1.4017 -1.7914 -2.3070 -2.5992 -2.6080
c21 -1.0041 -1.7787 3.0201 2.4505 1.8730 1.1968 0.3760 -0.2339 -0.5234 -1.1926 -1.6144 -1.8518
c22 0.9375 -0.1004 -1.1431 -2.0812 3.0417 2.6468 1.8348 1.1226 0.4545 0.1007 -0.4089 -0.8077
c23 2.5232 1.9582 0.4427 -0.4302 -1.2424 -2.0269 -2.7773 2.7050 2.0629 1.3485 0.8068 0.4794
c24 -1.9461 -3.0654 2.3819 1.3692 0.2823 -0.2191 -1.0387 -1.8679 -2.7561 2.5806 2.2174 1.7830
c25 -0.1862 -1.2147 -2.4055 2.9360 2.0503 1.0815 0.2426 -0.4825 -1.0706 -1.9250 -2.5544 2.9035
c26 2.0442 0.5070 -0.7708 -1.6896 -2.8649 2.6278 2.1404 1.3058 0.2855 -0.2796 -0.8179 -1.6241
c27 -1.7175 2.6819 1.1059 -0.0968 -0.9999 -2.0802 -2.8470 2.8351 1.9844 1.2717 0.4857 -0.3113
c28 0.8779 -1.1901 3.0395 2.0294 0.3687 -0.4539 -1.4186 -2.1100 -2.8905 2.6578 2.0677 1.3172
c29 -3.1097 1.2490 -0.7513 -1.9350 2.4674 1.4949 0.2562 -0.7079 -1.3576 -2.3257 3.0588 2.7049
c30 -1.1233 -2.8021 1.8023 0.3951 -1.4218 -2.6548 2.2465 0.9628 0.0417 -0.9015 -1.6343 -2.3766
c31 1.5811 -0.4043 -2.1077 2.5590 0.7627 -0.4297 -2.1279 2.8420 1.7231 0.7235 -0.0612 -0.9348
c32 -1.7815 1.9089 0.0959 -1.6143 2.8720 1.7372 -0.0009 -1.0626 -2.6406 2.6697 1.5301 0.5498
c33 0.6051 -1.8087 2.5106 0.5654 -1.1565 -2.4155 2.3330 1.1234 -0.6329 -1.3239 -2.8177 2.3433
c34 -2.9583 0.8689 -1.0872 3.1468 0.7188 -0.5696 -1.9034 3.1033 1.7888 0.4549 -0.4780 -2.0182
c35 0.0248 -2.4081 1.3842 -0.6702 -3.0616 1.8001 0.2340 -1.2758 -2.5175 2.4601 1.4845 -0.0251
c36 2.8981 0.2654 -2.3920 1.9211 -0.3502 -2.3031 2.3810 0.7853 -0.7884 -1.9776 3.0729 2.1948
c37 -0.3168 -3.0333 0.5058 -1.9033 2.1946 -0.0612 -1.6561 2.7275 1.4331 0.0351 -1.1308 -2.2287
c38 3.0766 0.2555 -2.8174 0.3182 -1.4208 2.6383 0.7548 -0.9836 -2.7191 2.1437 0.8961 -0.7003
c39 0.0601 -2.5411 0.0383 -3.1111 1.0603 -1.0572 -3.1186 1.3982 -0.3860 -1.8098 2.6770 1.4583
c40 -2.6936 0.0030 -2.9493 0.2392 -2.6014 1.4675 -0.8389 -2.5771 1.9429 0.2397 -1.1609 -2.8583
c41 0.8634 -3.0807 0.4816 -2.8592 0.4419 -2.0855 2.0238 -0.1267 -2.0419 2.4140 0.8404 -0.7853
289
N 72 76 80 84 88 92 96 100 104 108 112 116

c42 -2.0177 0.3177 -2.4826 -0.1961 -2.8667 1.1797 -1.3930 2.2203 0.4488 -1.5700 -3.1907 1.5227
c43 1.4951 -2.5935 0.4849 2.8889 0.1705 -2.5735 1.4013 -1.3171 2.9062 0.6431 -0.7292 -2.3322
c44 -0.5638 0.8095 -2.9585 0.3204 -2.9636 -0.2011 -2.0799 1.7967 -0.7242 -3.0534 1.5113 -0.3342
c45 -2.7434 -1.6741 0.2981 -2.8779 0.2069 2.9329 0.6749 -1.6232 2.0079 -0.2172 -2.4031 2.0317
c46 0.6716 2.1891 -2.1759 -0.0675 2.8942 0.1070 -2.6298 0.9656 -1.5583 2.4954 0.4096 -1.7712
c47 -1.8162 -0.5177 1.5117 -2.8619 -0.1515 -3.0283 0.1355 -2.4869 1.2521 -1.2666 3.0225 0.7812
c48 2.1157 -2.7164 -1.2387 0.9096 -2.9905 -0.0428 2.9670 0.5701 -2.1251 1.1752 -0.8643 -2.9156
c49 0.1736 1.3620 2.6642 -1.7735 0.7792 -3.1128 -0.2146 -2.8821 0.8185 -2.1634 1.8476 -0.3969
c50 -1.6722 -1.3635 0.5259 1.7702 -2.0721 0.8315 3.0280 -0.4373 -2.6679 0.8705 -1.5430 2.2173
c51 2.8352 2.4864 -2.0268 -0.7533 1.3306 -2.2092 0.1068 2.6013 0.1766 -2.6780 1.1712 -1.2458
c52 0.9984 0.5620 1.6712 3.0899 -1.3773 0.8415 -2.7691 0.0887 -2.9540 -0.0642 -2.3982 1.5444
c53 -1.3271 -1.2187 -0.7918 0.7079 2.3569 -2.0355 0.7887 -2.6666 0.0248 2.9313 0.4401 -2.1686
c54 -3.0749 -2.8662 -2.9573 -1.5378 -0.1726 1.5102 -2.0651 0.5131 3.1644 0.0787 -2.8679 0.2299
c55 1.3259 1.6179 1.2412 2.6066 -2.7266 -1.0009 1.4074 -2.3760 0.3385 -2.8547 0.2759 -3.0749
c56 -0.0329 -0.0716 -0.5799 0.1037 1.3551 2.5854 -1.5262 1.1753 -2.4858 -0.1880 -2.7742 0.3728
c57 -1.3320 -2.0636 -2.2175 -2.3412 -0.8539 0.3665 1.8211 -1.7545 0.6473 2.9078 0.3229 -2.7035
c58 -2.6034 2.2896 2.1795 1.8205 -2.8624 -2.0818 -0.5466 1.4018 -2.2158 0.2763 -2.7532 0.2076
c59 2.4443 0.7491 0.6701 0.0622 1.0095 1.7978 -3.0620 -1.5150 1.2184 -2.5950 0.1274 -3.0063
c60 1.0160 -0.6902 -1.3607 -1.5604 -1.5219 -0.2034 0.8035 2.3252 -1.5287 0.4950 -3.0808 0.2353
c61 0.1063 -2.0320 -3.0140 -3.0502 2.6636 -2.2782 -1.6594 -0.0421 1.9524 -2.2933 0.5267 -2.8597
c62 -0.9924 2.9758 1.4980 1.1936 0.9563 1.4910 2.3577 -2.4315 -0.9064 1.4329 -2.3453 0.0314
c63 -1.9425 1.7720 0.0260 -0.3707 -0.8957 -0.8523 0.3532 1.4451 2.8212 -1.1005 1.0182 -3.1401
c64 -2.8248 0.7931 -1.4147 -2.2732 -2.4266 -2.9009 -1.9252 -1.2333 0.5110 2.4801 -1.8734 0.5307
c65 2.2536 -0.2054 -2.4270 2.0397 2.0508 1.4884 2.0050 2.9692 -2.0299 -0.4942 1.7033 -1.9827
c66 1.8256 -1.2716 2.3824 0.7175 0.4364 0.0002 -0.1110 0.8825 1.7890 -3.0720 -0.8239 1.4337
c67 1.3196 -2.3383 1.7019 -0.5662 -1.7780 -1.8250 -2.1724 -1.2021 -0.4167 0.8239 2.7558 -1.3706
c68 0.9003 3.1519 0.4377 -1.7440 2.9128 3.0465 2.0333 3.0915 -2.7478 -1.4074 0.0251 2.0911
c69 0.7405 2.1388 -0.5969 -2.9219 1.6010 1.3961 0.4020 0.8461 1.3127 2.6906 -2.5482 -0.5203
c70 0.4973 1.7964 -1.8181 2.1911 0.4633 -0.4259 -1.2518 -1.5093 -0.4498 0.0665 1.5492 -2.9986
c71 -0.0178 1.1865 -2.6996 1.1423 -0.9774 -2.1782 -2.5410 2.8179 -2.7783 -1.9632 -1.0056 0.6881
c72 -0.8284 1.2249 2.6187 0.3590 -2.1490 2.5983 1.7239 1.1136 1.2356 2.0999 -3.1290 -1.7378
c73 0.8148 2.2143 -0.7781 2.8014 0.9609 0.4867 -0.3724 -0.7943 0.0586 0.8471 1.9667
c74 0.6875 1.9329 -2.0153 1.7344 -0.2504 -1.3121 -1.7864 -2.8426 -1.8047 -1.5785 -0.2102
c75 0.2836 1.4958 -2.4958 1.0341 -1.3296 -3.0224 2.8255 1.5869 2.4570 2.8117 -2.4847
c76 -0.6144 1.4039 2.6945 0.0060 -2.4597 1.6995 1.1563 0.2840 0.1865 0.9469 1.3983
c77 0.8729 2.3011 -0.7725 2.7181 0.3158 -0.3789 -1.3542 -1.7887 -1.2090 -0.7006
c78 0.7725 1.7931 -1.7108 1.3831 -0.9101 -2.1069 -2.8506 2.6368 -3.0035 -2.8922
c79 0.3563 1.5292 -2.5232 0.5130 -1.9390 2.4379 1.6728 0.9221 1.3405 1.3369
c80 -0.5579 1.0539 2.5431 -0.5725 -3.0024 1.0925 0.5126 -0.6257 -1.0974 -0.3373
c81 1.1664 2.0256 -1.2022 2.0205 -0.1696 -1.5072 -2.0306 -2.9022 -2.2106
c82 0.6653 1.6257 -2.2044 1.1068 -1.1539 -2.8432 2.7829 1.5777 1.8132
c83 0.3064 1.1989 -3.0229 0.2823 -2.1756 1.6894 1.1543 0.1836 -0.0160
c84 -0.6421 1.2513 2.3089 -0.8583 2.8138 0.5859 -0.5095 -1.5793 -2.1995
c85 0.5739 1.8568 -1.7861 1.9468 -0.6071 -2.2028 -2.7414 2.1630
c86 0.6339 1.4073 -2.5308 0.7093 -1.4727 2.7692 2.0058 0.7235
c87 0.1103 1.2557 2.8707 0.0074 -2.7245 1.4961 0.6612 -0.8172
c88 -0.6099 0.8507 2.2122 -0.8430 2.5004 0.1810 -1.2049 -2.2212
c89 0.6870 1.7197 -1.8312 1.4523 -1.0191 -2.5939 2.7018
c90 0.4272 1.4970 -2.4640 0.8080 -2.2010 2.0889 1.0312
c91 0.1431 1.2120 2.7894 -0.3709 -3.0702 0.8514 -0.3606
c92 -0.8474 0.9322 2.1596 -1.0840 2.4456 -0.3341 -2.1550
c93 0.5962 1.6452 -2.0584 1.2618 -1.3403 2.8524
c94 0.4992 1.5605 -2.6752 0.4539 -2.4015 1.4829
c95 0.0482 1.0452 2.5349 -0.4378 2.9214 0.4138
c96 -0.7330 0.9521 2.1728 -1.3209 1.7945 -0.6603
c97 0.7100 1.7766 -2.3394 0.9670 -1.8051
c98 0.4766 1.5803 -2.7186 0.1169 -2.6548
290
N 72 76 80 84 88 92 96 100 104 108 112 116

c99 0.1485 1.1497 2.4057 -0.6887 2.3440
c100 -0.9304 0.9462 2.1433 -1.5434 1.2983
c101 0.6292 1.7351 -2.4048 1.0119
c102 0.5707 1.5692 -2.9808 -0.1187
c103 0.0477 0.9804 2.2931 -0.6227
c104 -0.7345 0.8391 2.0836 -1.4272
c105 0.6945 1.5164 -2.4593
c106 0.5567 1.4124 -3.1239
c107 -0.0242 1.1343 2.2061
c108 -0.7372 0.8605 1.9351
c109 0.8073 1.5512
c110 0.4219 1.5053
c111 0.0069 1.1244
c112 -0.6813 0.9867
c113 0.6648
c114 0.6057
c115 0.1390
c116 -0.6981
291
D.1 PHASE DIFFERENCE BETWEEN EVOLVED LOW

CF DMT SYMBOLS AND NEWMANN’S PHASING
SCHEME
These plots show the difference between the evolved phase sequences outlined in
Appendix D and the original starting point using Newmann’s phasing scheme as
show in equation A3-1
π (k − 1) 2
θ= A3-1
N
The evolved phase sequences correspond to the phase shown in A3-1 plus the phase
difference shown here in D.1. The results show that the difference is a complex
sequence, which is almost fractal in nature. It is important to remember that due to
the process of evolution the results shown may not the global minima but local
minima. In addition, if the evolution was run again the results might vary slightly.
The main variations appear to be mirroring in the X or Y axis, or resulting is a
completely different sequence. Sequences for tones from 2 – 30 have been simulated
twice. The only major variations between the different evolution runs was for 12
tones the sequence was mirrored in the X axis and for 13 tones the sequence was
quite different.
The evolved results give a clue as to the optimal phasing scheme to achieve the
minimum CF for a DMT signal. There may exist a direct mathematical method for
deriving this phasing scheme, which is left for future research.
292
Num Tones: 2, CF: 3.0096 dB, Gen: 5120 Num Tones: 3, CF: 2.218 dB, Gen: 5130
Phase Adjustment (radians)

0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
1 1.2 1.4 1.6 1.8 2 1 1.5 2 2.5 3
Carrier Number Carrier Number

0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
1 1.5 2 2.5 3 3.5 4 1 2 3 4 5
1
0.6
0.5 0.4
0.2
0 0
-0.2
-0.5 -0.4
-0.6
-1 -0.8
1 2 3 4 5 6 1 2 3 4 5 6 7
1
0.6
0.5 0.4
0.2
0 0
-0.2
-0.5 -0.4
-0.6
-1 -0.8
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 9
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
2 4 6 8 10 2 4 6 8 10
293
1 1.5

1
0.5
0.5
0
0
-0.5
-0.5
-1 -1
2 4 6 8 10 12 2 4 6 8 10 12

0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
2 4 6 8 10 12 14 2 4 6 8 10 12 14
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
2 4 6 8 10 12 14 16 2 4 6 8 10 12 14 16
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
2 4 6 8 10 12 14 16 18 5 10 15
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
5 10 15 20 5 10 15 20
294
1

0.6
0.5
0.4
0.2
0
0
-0.2 -0.5
-0.4
-1
-0.6
-0.8 -1.5
5 10 15 20 5 10 15 20
1

0.5
0.5
0
0
-0.5
-0.5
-1 -1
5 10 15 20 5 10 15 20 25
1
0.6
0.5
0.4
0.2
0
0
-0.2 -0.5
-0.4
-1
-0.6
-0.8 -1.5
5 10 15 20 25 5 10 15 20 25
1.5
0.6
1
0.4
0.5
0.2
0 0
-0.2
-0.5
-0.4
-1
-0.6
-1.5 -0.8
5 10 15 20 25 5 10 15 20 25
1
0.6
0.5
0.4
0.2
0
0
-0.2 -0.5
-0.4
-1
-0.6
-0.8 -1.5
5 10 15 20 25 30 5 10 15 20 25 30
295
1

0.6
0.5 0.4
0.2
0 0
-0.2
-0.5 -0.4
-0.6
-1 -0.8
5 10 15 20 25 30 5 10 15 20 25 30
1 1

0.5 0.5
0 0
-0.5 -0.5
-1 -1
5 10 15 20 25 30 35 5 10 15 20 25 30 35
1
0.6
0.4 0.5
0.2
0 0
-0.2
-0.4 -0.5
-0.6
-0.8 -1
5 10 15 20 25 30 35 40 5 10 15 20 25 30 35 40
1
0.6
0.4 0.5
0.2
0 0
-0.2
-0.4 -0.5
-0.6
-0.8 -1
5 10 15 20 25 30 35 40 10 20 30 40
1
0.6
0.4 0.5
0.2
0 0
-0.2
-0.4 -0.5
-0.6
-0.8 -1
10 20 30 40 10 20 30 40 50
296

0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
10 20 30 40 50 10 20 30 40 50
1 1

0.5 0.5
0 0
-0.5 -0.5
-1 -1
10 20 30 40 50 10 20 30 40 50
1
0.6
0.5 0.4
0.2
0 0
-0.2
-0.5 -0.4
-0.6
-1 -0.8
10 20 30 40 50 60 10 20 30 40 50 60
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
10 20 30 40 50 60 10 20 30 40 50 60
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
10 20 30 40 50 60 70 10 20 30 40 50 60 70
297
1

0.6
0.5 0.4
0.2
0 0
-0.2
-0.5 -0.4
-0.6
-1 -0.8
10 20 30 40 50 60 70 80 10 20 30 40 50 60 70 80
1 1

0.5 0.5
0 0
-0.5 -0.5
-1 -1
10 20 30 40 50 60 70 80 20 40 60 80
1.5
0.6
1
0.4
0.2
0.5
0
-0.2 0
-0.4
-0.5
-0.6
-0.8 -1
20 40 60 80 20 40 60 80 100
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
20 40 60 80 100 20 40 60 80 100
1
0.6
0.4
0.5
0.2
0
0
-0.2
-0.4
-0.5
-0.6
-0.8
20 40 60 80 100 20 40 60 80 100
298
1

0.6
0.4
0.5
0.2
0
0
-0.2
-0.4
-0.5
-0.6
-0.8
20 40 60 80 100 120 20 40 60 80 100 120
1.5

0.6
1
0.4
0.2
0.5
0
-0.2 0
-0.4
-0.6 -0.5
-0.8
20 40 60 80 100 120 20 40 60 80 100 120
299

OFDM Thesis Lawrey

Uploaded by

Copyright:

Available Formats

OFDM Thesis Lawrey

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

OFDM Thesis Lawrey

Uploaded by

Copyright:

Available Formats

Adaptive Techniques for Multiuser OFDM

for the degree of Doctor of Philosophy

The demand for high-speed mobile wireless communications is rapidly growing.

An investigation of two adaptive techniques is also presented. These techniques

The simulated performance of the adaptive techniques is presented using a set of

Table of Contents ..................................................................................................... vii

List of Tables............................................................................................................ xiii

Glossary .................................................................................................................. xxiv

Simulation Files ................................................................................................... xxviii

Chapter 3 Radio Propagation ................................................................................. 91

Chapter 4 Multiuser OFDM.................................................................................. 122

Chapter 5 Access Point Repeaters ........................................................................ 165

Chapter 6 Generation of low Crest Factor Multi-Tone Signals......................... 194

Chapter 7 Crest Factor Minimisation of OFDM Data Symbols ........................ 228

Chapter 8 Summary............................................................................................... 238

Publications resulting from this PhD ................................................................... 256

Appendix A Window functions ....................................................................... 257

Appendix B BER verses SNR for OFDM in AWGN .................................... 266

Appendix C RF Measurements ...................................................................... 277

Table 1-2, DVB transmission parameters [18]................................................................... 8

Table 1-4, Summary of characteristics of IEEE802.11b, IEEE802.11a and

Table 2-1, 4 bit Gray coding in binary.............................................................................58

Table 2-3, Phase mapping for differential QPSK..............................................................63

Table 2-4, SNR degradation in dB due to imperfect channel estimation. ............................71

Table 3-2, Measured delay spread for a variety of environments. ......................................99

Table 5-1, Material properties of the internal walls.........................................................184

Table 7-1, Optimised positions found for 10 BPSK data subcarriers................................237

Figure 2-1, Time domain construction of an OFDM signal. ..............................................19

Figure 2-3, Block diagram showing a basic OFDM transceiver. ........................................22

Figure 2-6, OFDM generation, IFFT stage.......................................................................26

Figure 2-11, Construction of the subcarriers for complex signal representation of

Figure 2-12, Addition of a guard period to an OFDM signal .............................................31

Figure 2-17, Spectrum of a 1536 subcarrier OFDM signal with no band-limiting.

Figure 2-23, Construction of a RC guard period. ..............................................................48

Figure 2-29, IQ data mapping for differential 16-QAM. ...................................................64

Figure 3-20, Fading causes by a person pretending to type on a keyboard 15 cm from

Figure 4-3, Forward Link in a single cell system. ...........................................................125

Figure 4-12, Simulated performance from allocating a fixed frequency group of

Figure 4-14, Simulated performance for randomly hopped groups of subcarriers.

Figure 4-19, Simulated adaptive user allocation, utilising adaptive bandwidth,

Figure 5-1, Methods for obtaining spatial diversity. ........................................................166

Figure 5-4, Possible implementation of an Access Point Repeater. ..................................173

Figure 6-5, Accurate measurement of CF requires accurate measurement of the peaks

Figure 6-6, Worst case peak estimation sampling. ..........................................................200

Figure 6-7, Maximum error in CF measurements as a function of over sampling of the

Figure 6-18, Crest factor distribution as a function of the simulated generations.

Figure 7-4, PRC position combinations .........................................................................232

2G Second Generation mobile phone system (GSM, IS-95)

Wireless communications is an emerging field, which has seen enormous growth in

allocated to the service and by improvements in the spectral efficiency. OFDM is a

1.1 THIRD GENERATION WIRELESS SYSTEMS

Third generation mobile systems such as the Universal Mobile Telecommunications

1.2 4TH GENERATION SYSTEMS AND BEYOND

A significant improvement in spectral efficiency will be required in order for 4G

Stationary 2G systems Hiperlan/2

1.3 ORTHOGONAL FREQUENCY DIVISION

Orthogonal Frequency Division Multiplexing (OFDM) is an alternative wireless

1.3.1 DIGITAL AUDIO BROADCASTING

Parameter Transmission Mode

modulation of the OFDM subcarriers, leading to signal degradation. The amount of