Whole PDF

MODELING AND CHARACTERIZATION OF
MULTIPATH FADING CHANNELS IN CELLULAR
MOBILE COMMUNICATION SYSTEMS
A DISSERTATION SUBMITTED IN FULFILMENT OF THE REQUIREMENTS

FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
BY
Noor Muhammad Khan
SCHOOL OF ELECTRICAL ENGINEERING AND TELECOMMUNICATIONS

THE UNIVERSITY OF NEW SOUTH WALES
March 2006
c Noor Muhammad Khan, 2006.
°
Produced in LATEX 2ε .
I dedicate this thesis to my wife Warisa and kids Adeer, Palvasha and
Ariana for their love and support
Acknowledgements
I would like to express my gratitude to my supervisor Associate Professor Dr. Rodica Ramer
for her guidance, support, and encouragement, especially in the last six months of my re-
search tenure. Working with her has been a true privilege and a great experience for me.
Without her guidance, this thesis would never have been completed. I also owe my deepest
appreciation to my ex-supervisor Professor Dr. Predrag Rapajic for his constant encourage-
ment and valuable guidance throughout my PhD study. I would also like to thank him for
his financial support for my travel to several international conferences and workshops. I am
also thankful to the Honorable Minister and staff of the Ministry of Science and Technology,
Pakistan for their TROSS sponsorship to support this PhD program.
I especially thank Professor Victor Solo and Dr. Jinhong Yuan for their useful suggestions in
my research study and thesis proposal. Their valuable advices have always inspired me from
time to time during my research projects. I would also like to thank Mr. Mohammed Simsim
for his contribution to some parts of my research work. I am also thankful to Ms. Parastoo
Sadeghi for having valuable discussions with me during my studies.
I would like to thank all of my colleagues in the Cellular Mobile Communication Group for
creating such a pleasant working environment and for having useful discussions with me. I
also owe many thanks to Mr. Joseph Yiu in the Cellular Mobile Communication Lab for his
prompt help in technical matters.
I would like to thank my affectionate parents and brothers for their continuous support and
encouragement. I also owe my deepest appreciation to my grandfather who always prayed
for me in my hard times. Finally, I would like to express my earnest gratitude to my loving
wife and beloved children whose incessant support and devotion to me and this research
work never once faded even during harsh times.
iv
Related Publications
This thesis is based mainly on the following publications:
Journal Papers
[J01] Noor M. Khan, Mohammed T. Simsim and Predrag B. Rapajic, “A Generalized
Model for the Spatial Characteristics of the Cellular Mobile Channel”, Accepted for
publication in IEEE Trans. Veh. Technol., submitted in April 2005, revised in January
2006, Paper ID: VT-2005-00203. Paper Status: In press
[J02] Noor M. Khan, Mohammed T. Simsim, Predrag B. Rapajic, Rodica Ramer, “Effect
of Scattering around BS on the Azimuthal Distribution of Multipath Signals”, Sub-
mitted to IEEE Trans. Wireless Commun. in December 2005, Paper ID: TW05-988,
Paper Status: With Editor, Prof. Ross Murch
[J03] Noor M. Khan, Rodica Ramer, and Predrag B. Rapajic, “On Characterizing the
Angle Spread in Multipath Fading Channels”, Accepted for publication in IEEE Trans.
Antennas Propagat., submitted in February 2006, Paper ID: AP0602-0095, Paper Sta-
tus: In press
[J04] Noor M. Khan, Mohammed T. Simsim, Rodica Ramer, and Predrag B. Rapajic,
“Effect of Motion on the Spatio-Temporal Characteristics of Cellular Mobile Channel”,
v
Submitted to IEEE Trans. Veh. Technol. in February 2006, Paper ID: VT-2006-00106,
Paper Status: With Associate Editor, Prof. Ali Abdi
[J05] Mohammed T. Simsim, Noor M. Khan, and Predrag B. Rapajic, “Modeling Spatial
Aspects of Mobile Channel for Microcell and picocell Environments using Gaussian
Scattering Distribution”, International J. on Info. Technol. & Telecommun. (IJIT),
vol. 2/2006, pp. 97-105, June 2006.
[J06] Mohammed T. Simsim, Noor M. Khan, and Rodica Ramer, “Multipath Delay due
to Local and Distant Scatterers in Cellular Environments”, Submitted to IEE Proc.
on Microwaves, Antennas and Propagat. in March 2006, Paper ID: MAP-2006-0066,
Paper Status: Under Review
Conference Papers
[C01] Noor M. Khan, and Predrag B. Rapajic, “Use of State-Space Approach and
Kalman Filter Estimation in Channel Modeling for Multiuser Detection in Time-
Varying Environment”, in Proc. International Workshop on Ultra Wideband Systems
(IWUWS’03), Oulu, Finland, June 2003, Paper No. 1052
[C02] Noor M. Khan, Iftikhar A. Sheikh, and Predrag B. Rapajic, “Investigating Dif-
ferent Factors on Linear Channel Compensated MMSE Receiver Using Kalman Filter
Based Channel Estimation”, in Proc. Australian Telecommun. Networks and Applica-

tions Conf. (ATNAC’03), Melbourne, Australia, December 2003
[C03] Iftikhar A. Sheikh, Noor M. Khan, and Predrag B. Rapajic, “Kalman Filter
based Channel Estimation in OFDM Multi-user Single-Input-Multi-Output System”,
in Proc. IEEE International Symp. Spread Spectrum Techniques and Applications
(ISSSTA’04), Sydney, Australia, August 2004, pp. 802-806
vi
[C04] Noor M. Khan, Mohammed T. Simsim, and Predrag B. Rapajic, “A General-
ized Spatial Model for all Cellular Environments”, in Proc. IEEE Symp. Trends in
Commun. (SympoTIC’04), Bratislava, Slovak Republic, October 2004, pp. 33-38
[C05] Mohammed T. Simsim, Noor M. Khan, and Predrag B. Rapajic, “Modeling Spatial
Aspects of Mobile Channel in Uniformly Distributed Scatters”, in Proc. 8th IEEE
International Conf. on Telecommun. (ConTel’05), Zagreb, Croatia, vol. 1, June 2005,
pp. 195-201
[C06] Noor M. Khan, Mohammed T. Simsim, and Predrag B. Rapajic, “Effect of Mobile
Motion on the Spatial Characteristics of Channel”, in Proc. 62nd IEEE Veh. Technol.
Conf. (VTC’05-Fall), Dallas, USA, vol. 4, September 2005, pp. 2201-2205
[C07] Noor M. Khan, and Predrag B. Rapajic, “A State-Space Approach to Semi Blind
Adaptive Multiuser Detection in Time-Varying Environment”, in Proc. 11th IEEE
Asia-Pacific Conf. on Commun. (APCC’05), Perth, Australia, October 2005, pp.
1024-1027
[C08] Noor M. Khan, Mohammed T. Simsim, Predrag B. Rapajic, and Rodica Ramer,
“On the Spatial Characteristics of cellular Mobile Channel in Low Antenna Height
Environments”, Accepted for publication in Proc. the Progress in Electromagnetic
Research Symp. (PIERS’06), Cambridge, USA, March 2006
[C09] Noor M. Khan, and Rodica Ramer, “Effect of Distant Scatterers on MIMO Fading
Channel Tracking”, Accepted for publication in Proc. the Progress in Electromagnetic
[C10] Mohammed T. Simsim, Noor M. Khan, Predrag B. Rapajic, and Rodica Ramer,
“Investigating the Standard Deviation of the Distribution of Scatterers in Cellular
vii
Environments”, Accepted for publication in Proc. the Progress in Electromagnetic
[C11] Mohammed T. Simsim, Noor M. Khan, Rodica Ramer, and Predrag B. Rapajic,
“Effect of Mobile Motion on the Temporal Characteristics of Cellular Mobile Channel”,
Accepted for publication in Proc. 63rd IEEE Veh. Technol. Conf. (VTC’05-Spring),
Melbourne, Australia, May 2006
[C12] Mohammed T. Simsim, Noor M. Khan, Rodica Ramer, and Predrag B. Rapajic,
“Time of Arrival Statistics in Cellular Environments”, Accepted for publication in
Proc. 63rd IEEE Veh. Technol. Conf. (VTC’05-Spring), Melbourne, Australia, May
2006
[C13] Noor M. Khan, Mohammed T. Simsim, Rodica Ramer, and Predrag B. Rapajic,
“Modeling Spatial Aspects of Mobile Channel for Macrocells using Gaussian Scattering
Distribution”, Submitted to the 3rd IEEE Int. Symp. on Wireless Commun. Systems,
Velencia, Spain, September 2006
viii
Abstract
Due to the enormous capacity and performance gains associated with the use of antenna
arrays in wireless multi-input multi-output (MIMO) communication links, it is inevitable
that these technologies will become an integral part of future systems. In order to assess
the potential of such beam-oriented technologies, direct representation of the dispersion of
multipath fading channel in angular and temporal domains is required. This representation
can only be achieved with the use of spatial channel models. This thesis thus focuses on
the issue of spatial channel modeling for cellular systems and on its use in the characteri-
zation of multipath fading channels. The results of this thesis are presented mainly in five
parts: a) modeling of scattering mechanisms, b) derivation of the closed-form expressions for
the spatio-temporal characteristics, c) generalization of the quantitative measure of angular
spread, d) investigation of the effect of mobile motion on the spatio-temporal characteris-
tics, and e) characterization of fast fading channel and its use in the signature sequence
adaptation for direct sequence code division multiple access (DS-CDMA) system.
The thesis begins with an overview of the fundamentals of spatial channel modeling with
regards to the specifics of cellular environments. Previous modeling approaches are dis-
cussed intensively and a generalized spatial channel model, the ‘Eccentro-Scattering Model’
is proposed. Using this model, closed-form mathematical expressions for the distributions
of angle and time of multipath arrival are derived. These theoretical results for the picocell,
microcell and macrocell environments, when compared with previous models and available
measurements, are found to be realistic and generic. In macrocell environment, the model
incorporates the effect of distant scattering structures in addition to the local ones. Since
the angular spread is a key factor of the second order statistics of fading processes in wireless
communications, the thesis proposes a novel generalized method of quantifying the angular
ix
spread of the multipath power distribution. The proposed method provides almost all pa-
rameters about the angular spread, which can be further used for calculating more accurate
spatial correlations and other statistics of multipath fading channels. The degree of accuracy
in such correlation calculations can lead to the computation of exact separation distances
among array elements required for maximizing capacity in MIMO systems or diversity an-
tennas. The proposed method is also helpful in finding the exact standard deviation of
the truncated angular distributions and angular data acquired in measurement campaigns.
This thesis also indicates the significance of the effects of angular distribution truncation
on the angular spread. Due to the importance of angular spread in the fading statistics,
it is proposed as the goodness-of-fit measure in measurement campaigns. In this regard,
comparisons of some notable azimuthal models with the measurement results are shown.
The effect of mobile motion on the spatial and temporal characteristics of the channel is
also discussed. Three mobile motion scenarios are presented, which can be considered to be
responsible for the variations of the spatio-temporal statistical parameters of the multipath
signals. Two different cases are also identified, when the terrain and clutter of the mobile
surroundings have an additional effect on the temporal spread of the channel during mobile
motion. The effect of increasing mobile-base separation on the angular and temporal spreads
is elaborated in detail. The proposed theoretical results in spatial characteristics can be
extended to characterizing and tracking transient behavior of Doppler spread in time-varying
fast fading channels; likewise the proposed theoretical results in temporal characteristics can
be utilized in designing efficient equalizers for combating inter-symbol interference (ISI) in
time-varying frequency-selective fading channels.
In the last part of the thesis, a linear state-space model is developed for signature sequence
adaptation over time-varying fast fading channels in DS-CDMA systems. A decision directed
adaptive algorithm, based on the proposed state-space model and Kalman filter, is presented.
The algorithm outperforms the gradient-based algorithms in tracking the received distorted
signature sequence over time-varying fast fading channels. Simulation results are presented
which show that the performance of a linear adaptive receiver can be improved significantly
with signature tracking on high Doppler spreads in DS-CDMA systems.
x
Contents
Acknowledgements iv
Related Publications v
Abstract ix
Table of Contents xi
List of Tables xviii
List of Figures xix
Acronyms xxiii
Notations xxvii
1 Introduction 1
1.1 History of the Capacity Demands in Wireless Communication Systems . . . 1
1.2 Multiantenna Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Smart Antenna Systems . . . . . . . . . . . . . . . . . . . . . . . . . 4
xi
1.2.2 MIMO Communication Systems . . . . . . . . . . . . . . . . . . . . . 6
1.3 Channel Fading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Mobile Radio Propagation Environment . . . . . . . . . . . . . . . . 6
1.3.2 Small-Scale Multipath Fading . . . . . . . . . . . . . . . . . . . . . . 8
1.3.3 Types of Small-Scale Fading . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.3.1 Based on Multipath Time-Delay Spread . . . . . . . . . . . 11
1.3.3.2 Based on Doppler Spread . . . . . . . . . . . . . . . . . . . 13
1.4 Cellular Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4.1 Picocell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4.2 Microcell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.3 Macrocell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Capacity of MIMO Systems in Fading Channels . . . . . . . . . . . . . . . . 18
1.6 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.7 Scope of the Thesis and Proposed Methodology . . . . . . . . . . . . . . . . 23
1.8 Contributions of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2 A Generalized Spatial Channel Model 30
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1.1 Overview of Physical Spatial Channel Modeling . . . . . . . . . . . . 31
2.1.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.1.3 Contributions in the Chapter . . . . . . . . . . . . . . . . . . . . . . 35
xii
2.2 General Channel Modeling Assumptions . . . . . . . . . . . . . . . . . . . . 36
2.3 General Channel Modeling Parameters . . . . . . . . . . . . . . . . . . . . . 39
2.3.1 Shape of the Scattering Region . . . . . . . . . . . . . . . . . . . . . 39
2.3.1.1 Circular Scattering Model . . . . . . . . . . . . . . . . . . . 40
2.3.1.2 Elliptical Scattering Model . . . . . . . . . . . . . . . . . . 41
2.3.2 Distribution of Scatterers . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.2.1 Uniformly Distributed Scatterers . . . . . . . . . . . . . . . 41
2.3.2.2 Gaussian Distributed Scatterers . . . . . . . . . . . . . . . . 42
2.4 The Proposed Eccentro-Scattering Channel Model . . . . . . . . . . . . . . . 43
2.5 The Proposed Jointly Gaussian Scattering (JGSM) Model . . . . . . . . . . 46
2.6 Application of the Eccentro-Scattering Model . . . . . . . . . . . . . . . . . 48
2.6.1 Picocell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.6.2 Microcell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.6.3 Macrocell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.7 Summary of the Chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3 Modeling Spatial Characteristics of Mobile Channel 54
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 Angle Of Arrival For Picocell And Microcell Environments . . . . . . . . . . 56
3.2.1 Uniformly Distributed Scatterers . . . . . . . . . . . . . . . . . . . . 58
3.2.2 Gaussian Distributed Scatterers . . . . . . . . . . . . . . . . . . . . . 59
xiii
3.3 Angle Of Arrival For Macrocell Environment . . . . . . . . . . . . . . . . . . 67
3.3.1 Uniformly Distributed Scatterers . . . . . . . . . . . . . . . . . . . . 70
3.3.2 Gaussian Distributed Scatterers . . . . . . . . . . . . . . . . . . . . . 72
3.4 Modeling the Impact of Scattering around BS on the AoA Statistics . . . . . 79
3.4.1 Low Antenna-Height Urban Macrocell Model . . . . . . . . . . . . . . 80
3.4.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4 Characterization of Angle Spread 88
4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.2 Angle Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3 The Proposed Method to Quantify Angle Spread . . . . . . . . . . . . . . . 92
4.4 Effect of Distribution Truncation on the Angle Spread . . . . . . . . . . . . . 95
4.5 Factors which Cause Gaussian Distribution Truncation . . . . . . . . . . . . 102
4.6 Angle Spread as the Goodness-of-Fit Measure in Measurement Campaigns . 107
4.6.1 Indoor Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.6.2 Outdoor Environments . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.8 Appendix 4-A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.9 Appendix 4-B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5 Modeling Temporal Characteristics of Mobile Channel 116
xiv
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.1.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.2 Model Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.3 pdf of ToA for Picocells and Microcells . . . . . . . . . . . . . . . . . . . . . 124
5.4 pdf Of ToA for Macrocells . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.4.1 Local Scatterers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.4.2 Distant Scatterers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6 Effect of Mobile Motion on the Spatio-Temporal Characteristics 139
6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.2 Mobile Motion Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.3 Behavior of Spatial Characteristics under Mobile Motion . . . . . . . . . . . 144
6.3.1 Important Spatial Parameters . . . . . . . . . . . . . . . . . . . . . . 144
6.3.2 Effect of Mobile Motion on the Spatial Characteristics of the Channel 147
6.3.2.1 Picocell and Microcell Environments . . . . . . . . . . . . . 147
6.3.2.2 Macrocell Environments . . . . . . . . . . . . . . . . . . . . 151
6.4 Behavior of Temporal Characteristics under Mobile Motion . . . . . . . . . . 159
6.4.1 Temporal Channel Model . . . . . . . . . . . . . . . . . . . . . . . . 159
xv
6.4.2 Some Important Temporal Parameters . . . . . . . . . . . . . . . . . 161
6.4.3 Lifetime of Scatterers . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.4.3.1 Case 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.4.3.2 Case 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.4.4 Effect of Mobile Motion on ToA Statistics . . . . . . . . . . . . . . . 167
6.4.4.1 Picocell and Microcell Environments . . . . . . . . . . . . . 167
6.4.4.2 Macrocell Environments . . . . . . . . . . . . . . . . . . . . 169
6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
7 Fast Fading Channel Modeling for Single-Carrier DS-CDMA Systems 173
7.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.1.1 History of DS-CDMA Detectors . . . . . . . . . . . . . . . . . . . . . 174
7.1.2 Time-Varying Nature of the DS-CDMA Channel . . . . . . . . . . . . 179
7.1.3 Contributions of the Chapter . . . . . . . . . . . . . . . . . . . . . . 180
7.2 State-Space Approach in Multipath Fading Channel Modeling . . . . . . . . 181
7.2.1 Characterization of Multipath Fading Channels . . . . . . . . . . . . 182
7.2.2 State-Space Model of the Communication System over Fast Fading

Multipath Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.3 Kalman Filter Based Adaptive Detection for DS-CDMA System . . . . . . . 189
7.3.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
7.3.1.1 Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . 191
xvi
7.3.1.2 Receiver Model . . . . . . . . . . . . . . . . . . . . . . . . . 194
7.3.2 Kalman Filter-Based Signature Sequence Tracking . . . . . . . . . . . 196
7.3.2.1 State-Space Model for the Adaptive Multiuser Detection . . 199
7.3.2.2 Signature Sequence Estimation Algorithm . . . . . . . . . . 199
7.3.3 Modes of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.3.3.1 Decision Directed (DD) mode . . . . . . . . . . . . . . . . . 201
7.3.3.2 Training Directed (TD) and DD mode . . . . . . . . . . . . 202
7.3.3.3 TD and Non-Estimation (NE) mode . . . . . . . . . . . . . 202
7.3.3.4 Repeated TD and NE mode . . . . . . . . . . . . . . . . . . 202
7.3.4 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
7.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
8 Conclusions and Future Work 209
8.1 Summary of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
8.2 Conclusions and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . 214
xvii
List of Tables
2.1 Typical Values of the Parameters for Picocell and Microcell Environments
Based on the Eccentro-Scattering Model . . . . . . . . . . . . . . . . . . . . 50
2.2 Typical Values of the Parameters for Macrocell Environments Based on the
Eccentro-Scattering Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.1 Summary of the Spatial Channel Models . . . . . . . . . . . . . . . . . . . . 78
3.2 Values of Rnull for different macrocell environments . . . . . . . . . . . . . . 86
xviii
List of Figures
1.1 Typical outdoor multipath propagation environment . . . . . . . . . . . . . . . . 8
1.2 Illustration of Doppler effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Types of small-scale fading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Multipath scattering in (a) Picocell (Indoor), (b) Microcell (Outdoor), and (c)
Macrocell (Outdoor) Environments . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5 Illustration of the Proposed Dissertation Methodology . . . . . . . . . . . . . . . 24
2.1 Typical multipath fading environment . . . . . . . . . . . . . . . . . . . . . . . 36
2.2 Modeling with respect to the shape of the scattering region for uniform scatterer
distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3 Modeling with respect to the Gaussian scatterer distribution . . . . . . . . . . . . 42
2.4 Elliptical diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 Jointly Gaussian Scattering Model (JGSM) . . . . . . . . . . . . . . . . . . . . 47
2.6 Typical (a) Picocell and (b) Microcell Environments in Gaussian distributed scattering 49
3.1 Eccentro-Scattering model in uniform scattering . . . . . . . . . . . . . . . . . . 57
3.2 Eccentro-Scattering model in Gaussian scattering . . . . . . . . . . . . . . . . . 60
xix
3.3 Effect of increasing a with respect to σM S on the pdf of AoA in picocells and mi-
crocells, σBS =0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.4 Effect of increasing σM S with respect to a on the pdf of AoA in picocells and mi-
crocells, σBS =0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.5 Comparison of the pdf in AoA for the Eccentro-Scattering model, and Laplacian
function with measurements [1]. . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.6 Typical macrocell environment . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.7 Simulated and theoretical pdf of AoA for suburban macrocell with d = 2000m, a =
300m, D = 5000m, aD = 150m, and θD = 15◦ . . . . . . . . . . . . . . . . . . . 74
3.8 The proposed scattering model . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.9 pdf of AoA at BS in urban macrocell mobile environments . . . . . . . . . . . . . 84
4.1 Comparison between truncated (θspan = 90◦ ) and untruncated (θspan = 360◦ ) Gaus-
sian density functions, σg = 30◦ , θ̄ = 45◦ . . . . . . . . . . . . . . . . . . . . . 96
4.2 Effect of Gaussian distribution truncation on the angle spread . . . . . . . . . . . 98
4.3 Comparison between truncated (θspan = 90◦ ) and untruncated (θspan = 360◦ ) Lapla-
cian density functions, σl = 30◦ , θ̄ = 45◦ . . . . . . . . . . . . . . . . . . . . . . 99
4.4 Effect of Laplacian distribution truncation on the angle spread . . . . . . . . . . . 101
4.5 The factors which cause truncation of the Gaussian distribution in AoA . . . . . . 104
4.6 Comparison of the distribution in AoA for the candidate models (Eccentro-Scattering
Model [section 3.2], Gaussian [2] and Uniform Elliptical Scattering Model [3]) with
the measurements [1] and simulations in indoor environments . . . . . . . . . . . 108
4.7 Comparison of the distribution in AoA for the candidate models (Eccentro-Scattering
Model [section 3.2], Laplacian [1] and Raised-Laplacian) with the measurements [1]
in indoor environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
xx
4.8 Comparison of the distribution in AoA for the candidate models (GSDM [2] and
JGSM [section 3.4]) with the measurements [4] in outdoor environments . . . . . 110
5.1 Geometry of the proposed temporal channel model . . . . . . . . . . . . . . . . . 121
5.2 Temporal model for a typical multipath fading environment . . . . . . . . . . . . 121
5.3 Proposed temporal channel model for picocells and microcells . . . . . . . . . . . 125
5.4 pdf of ToA for picocell and microcell environments . . . . . . . . . . . . . . . . . 127
5.5 Temporal channel model for macrocells . . . . . . . . . . . . . . . . . . . . . . . 130
5.6 pdf of ToA for macrocells with the effect of distant scatterers . . . . . . . . . . . 136
6.1 Illustration of motion scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.2 Microcell and picocell environments . . . . . . . . . . . . . . . . . . . . . . . . 146
6.3 Effect of MS motion on the pdf of AoA, various line styles show the plots at three
different MS positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.4 Behavior of θspan , θ̄ and σθ under the effect of MS motion in microcell and picocell
environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.5 Behavior of S0 and Λ under the effect of MS motion in microcell and picocell envi-
ronments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.6 Macrocell environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.7 Behavior of θspan , θ̄ and σθ under the effect of MS motion in macrocell environment 154
6.8 Collective effect of distant scattering cluster and MS motion on the spatial spread
of cellular mobile channel in macrocell environment . . . . . . . . . . . . . . . . 156
6.9 Behavior of S0 and Λ under the effect of MS motion in macrocell environment . . 157
xxi
6.10 A general power delay profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
6.11 Illustration of the lifetime of scatterers . . . . . . . . . . . . . . . . . . . . . . . 162
6.12 Behavior of power delay profile under the effect of MS motion . . . . . . . . . . . 164
6.13 Movement of scattering discs due to MS motion in (a) picocells and microcells (b)
macrocells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.14 Behavior of στ , τe , τ0 , τ̄ and τmax in picocells and microcells under the effect of MS
motion when lifetime of scatterers is based on case 1 and case 2 . . . . . . . . . . 167
6.15 Effect of MS motion on the pdf of ToA; various line styles show the plots at different
MS positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.16 Behavior of στ , τe , τ0 , τ̄ and τmax in macrocells under the effect of MS motion

when lifetime of scatterers is based on case 1 and case 2 . . . . . . . . . . . . . . 170
7.1 Typical multipath fading environment . . . . . . . . . . . . . . . . . . . . . . . 184
7.2 Eccentro-Scattering model for picocell and microcell environments . . . . . . . . . 185
7.3 Eccentro-Scattering model for macrocell environments . . . . . . . . . . . . . . . 186
7.4 Discrete-time baseband model for synchronous DS-CDMA system . . . . . . . . . 190
7.5 Proposed model of the receiver structure for the Kalman filter based adaptive mul-
tiuser detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.6 BER versus SNR with N = 16 and K = 4 . . . . . . . . . . . . . . . . . . . . . 203
7.7 The criterion, Jk (i)=tr{P̃k (i)}, for arbitrary user, k, at iteration, i . . . . . . . . 204
7.8 BER performance comparison of different operation modes of the Kalman filter-
based adaptive CDMA detector, N = 31, K = 4 . . . . . . . . . . . . . . . . . . 207
xxii
Acronyms
1G First Generation (of Land Mobile Systems)
2G Second Generation (of Land Mobile Systems)
3G Third Generation (of Land Mobile Systems)
3GPP Third Generation Partnership Project
2-D Two Dimentional
3-D Thee Dimentional
1xRTT 1x Radio Transmission Technology

AMPS Advanced Mobile Phone Services
AoA Angle of Arrival

AR Autoregressive
ARMA Autoregressive Moving-Average
AT&T American Telephone and Telegraph

AWGN Additive White Gaussian Noise
BER Bit Error Rate

BLAST Bell Labs Layered Space-Time
BPSK Binary Phase Shift Keying
BS Base Stations
BW Bandwidth
CCI Co-Channel Interference
xxiii
CDF Cumulative Distribution Function
CDMA Code Division Multiple Access
CSI Channel State Information

DCM Directional Channel Model
DD Decision Directed
DS Delay Spread
DS-CDMA Direct-Sequence Code Division Multiple Access

FDD Frequency Division Duplex
FM Frequency Modulation
FSK Frequency-Shift Keying
GBSBM Geometrically-Based Single Bounce Macrocell

GSDM Gaussian Scatter Density Model
GSM Global System for Mobile
HIPERLAN High Performance Radio Local Area Network
iid independent identically distributed

IMT International Mobile Telecommunications
ISI Inter-Symbol Interference
JGSM Jointly Gaussian Scattering Model
LoS Line of Sight

LMS Least Mean Square
MA Moving-Average
MAI Multiple Access Interference

MAP Maximum A Posteriori
MC Multi-Carrier
MF Matched Filter
MIMO Multiple-Input Multiple-Output
xxiv
MLMR Maximum Likelihood Multiuser Receiver
MMSE Minimum Mean Squared Error
MRC Maximum Ratio Combining

MS Mobile Station
NE Non-Estimation
OFDM Orthogonal Frequency Division Multiplexing
OTD Orthogonal Transmit Diversity

PAS Power Azimuthal Spectrum
pdf Probability Density Function
PSK Phase Shift Keying
QAM Quadrature Amplitude Modulation

QoS Quality of Service
RF Radio Frequency
RLS Recursive Least Squares
rms Root Mean Square

SD Standard Deviation
SDMA Space Division Multiple Access
SINR Signal-to-Interference plus Noise Ratio
SM Spatial Multiplexing
SNR Signal-to-Noise Ratio
ST Space-Time
TD Training Directed
TDD Time Division Duplex
TDMA Time Division Multiple Access
ToA Time of Arrival
UHF Ultra-High Frequency
xxv
UMTS Universal Mobile Telecommunication Systems
V-BLAST Vertical Bell Laboratories Layered Space-Time
WCDMA Wideband Code Division Multiple Access

WLAN Wireless Local Area Network
WLMS Wiener Least Mean Square

WSSUS Wide-Sense Stationary Uncorrelated Scattering
xxvi
Notations
Boldface capital letters matrices
Boldface small letters vectors
Capital letters points on a geometrical diagram
or scalar quantities
Small letters scalar quantities
(·)T transpose
(·)∗ conjugate
(·)H Hermitian transpose
s∗h convolution between s and h

tr(·) trace
E(·) expectation
diag(x) diagonal matrix with x on its diagonal

|·| absolute value of a scalar
or determinant of a matrix
k·k Frobenius-norm
<{·} real part of a complex number
={·} imaginary part of a complex number
IN identity matrix of size N
ai ith element of vector a
xxvii
aij (i, j)th element of matrix A
ak vector a belonging to kth user
Ak matrix A belonging to kth user

R the set of real numbers
C the set of complex numbers
xxviii
Chapter 1
Introduction
This chapter first gives a brief overview of wireless communications. The background and
motivations for the research work in this thesis are described. A concise outline for the thesis
is provided at the end of this chapter.
1.1 History of the Capacity Demands in Wireless Com-
munication Systems
Capacity enhancement has always been an important issue in wireless communication sys-
tems. The demand for more capacity was realized even, when the first 2 MHz land mobile
radiotelephone system was installed by the Detroit Police Department in 1921 for Police car
dispatch. This was the beginning of the civilian use of wireless technology. The problem
1.1 History of the Capacity Demands in Wireless Communication Systems
of the lack of channels in the low frequency band was initially tackled in 1933, with the
use of higher frequency bands and the invention of Frequency Modulation (FM). In 1946, a
Personal Correspondence System introduced by Bell Systems began service and operated at
150 MHz with speech channels 120 KHz apart [5]. As demand for public wireless services
began to grow, the Improved Mobile Telephone Service (IMTS) using FM technology was
developed by AT&T.
The Cellular concept, conceived by D. H. Ring at Bell Laboratories in 1947, paved the
way for the modern cellular mobile communication systems [6] and later inspired AT&T to
propose its first high capacity analog cellular telephone system called the Advanced Mobile
Phone Service (AMPS) in 1968-70. AMPS was the first U.S. cellular telephone system,
and was deployed in late 1983 by Ameritech in Chicago, IL [7, 8]. In AMPS, radio systems
rely on judicious frequency re-use plans and frequency division multiple access (FDMA) to
maximize capacity. The analog cellular mobile systems of that age, altogether, are known
as the First Generation (1G) wireless technologies. Mobile systems have evolved rapidly
since then, incorporating digital communication technology and have thus transformed into
the new era of the Second Generation (2G) wireless technologies. This era includes the
Global System for Mobiles (GSM), IS-136, and IS-95. In order to maximize capacity or to
accommodate a large number of users, the former two standards use time division multiple
access (TDMA), while the latter one uses code division multiple access (CDMA).
With the increasing use of the internet in the late 1990s, the demand for higher spec-
2
1.1 History of the Capacity Demands in Wireless Communication Systems
tral efficiency and data rates has led to the development of the Third Generation (3G)
wireless technologies [5]. 3G offers Universal Mobile Telecommunication Systems (UMTS),
i.e., wideband CDMA and 1xRTT, i.e., CDMA-2000 as the primary standards. The 1X in
1xRTT refers to 1x the number of 1.25MHz channels and RTT in 1xRTT stands for Ra-
dio Transmission Technology. In order to satisfy the higher- rate requirements of modern
data services, the narrow-band 2G systems are being upgraded to 3G cellular systems and
wideband wireless local area networks (WLANs). Limitations in the radio frequency (RF)
spectrum necessitate the use of some innovative techniques to meet the increased demand in
data rate, quality of service (QoS) [5] and capacity. The innovative technique, which opened
a new dimension - space and pledged to improve the performance substantially, is the use of
multiple antennas at the transmitter and/or receiver in a wireless communication link. In
the mid 1990s, multiple transmit and receive antennas were incorporated effectively for the
first time, to establish a highly spectral efficient communication system - BLAST [9]. The
system was highly robust in highly scattering environments, with the assumption that the
signals arriving from the individual transmit antennas at each of the receive antennas are
uncorrelated. Soon V-BLAST was introduced with more improvements [10], but spectral
efficiency was, still dependent on the signal correlations.
3
1.2 Multiantenna Systems
As spectrum became a more and more precious resource, researchers investigated ways of
increasing the capacity of wireless systems without actually increasing the required spectrum.
Multiantenna systems offer such a possibility. Multiantenna systems can be grouped into
two categories on the basis of multiantenna element deployment [11]:
1. Smart Antenna systems, where multiantenna elements are deployed at one link end
only.
2. Multiple-Input Multiple-Output (MIMO) systems, where multiantenna elements are
deployed at both link ends.
1.2.1 Smart Antenna Systems
These are the antennas with multiple elements, where signals from different elements are
combined/created by an adaptive (intelligent) algorithm. When a smart antenna is deployed
at the receiver, the signals are combined ; for the transmitter case, the signals are created [11].
In most practical situations, smart antennas are deployed at the BS. Smart antennas exploit
the directional properties of the channel; hence, they provide spatial diversity with smartness.
Increasing the capacity is the most important application of smart antennas. They
achieve this goal through the following approaches:
4
Spatial Filtering for Interference Reduction: This is used in TDMA/FDMA systems
to reduce the reuse distance. A conventional TDMA/FDMA system cannot reuse
the same frequency in each neighboring cell, since the interference from the adjacent
cells would be too strong [11]. Smart antennas reduce interference and hence help in
reducing the reuse distance. This leads to an improvement in total spectral efficiency.
Space Division Multiple Access (SDMA): In this method, reuse distance remains un-
changed, while the number of users within a given cell is increased. SDMA controls the
radiated energy for each user in space and serves different users by using spot beam
smart antennas. Multiple users can be accommodated on the same time/frequency slot
(TDMA/FDMA), because the BS distinguishes them at different locations by means
of their different spatial signatures. SIR is increased for the desired users and hence
capacity of the whole system is increased [12].
Capacity increase in CDMA systems: In these systems, smart antennas enhance the
received signal power and hence the number of admissible users is increased in the
cell. The number of admissible users in the cell increases linearly with the number of
antenna elements, in ideal channel conditions.
Capacity increase in 3G CDMA systems: In 3G networks, high-rate data transmis-
sion is considered an important application. So users and interferers are usually allo-
cated low spreading factors. Hence, smart antennas have to use the spatial filtering
approach in these systems. Thus, they reduce the interference of the desired user by
5
1.3 Channel Fading
placing a null in the direction of the interferer [11].
1.2.2 MIMO Communication Systems
A large suite of techniques, known collectively as MIMO communications, have been devel-
oped in the past several years to exploit effectively the resulting multi-dimensionality, when
multiple antennas are used at both ends of the wireless communication link.
Since multiple antennas in MIMO systems can also serve as one-sided smart antennas,
MIMO systems are able to accomplish the tasks of smart antennas. Therefore, besides all
those applications meant for smart antennas, MIMO systems can also be used for spatial
multiplexing. Spatial multiplexing allows direct improvement of capacity by simultaneous
transmission of multiple datastreams in parallel.
1.3 Channel Fading
1.3.1 Mobile Radio Propagation Environment
Radio signals generally propagate according to three mechanisms:
1. Reflection
2. Diffraction
6
1.3 Channel Fading
3. Scattering
Reflections arise when plane waves are incident upon a surface with dimensions that are
very large compared to the wavelength. Diffraction occurs according to Huygen’s principle
when there is an obstruction between the transmitter and receiver antennas, and secondary
waves are generated behind the obstructing body. Scattering occurs when the plane waves
are incident upon an object whose dimensions are on the order of a wavelength or less, and
causes the energy to be redirected in many directions [8, 13].
The above three mechanisms give rise to three nearly independent phenomena in radio
channel:
1. Path-loss
2. Shadowing
3. Multipath Fading
Path-loss is the variation of the signal strength with the distance, while shadowing occurs,
when the signal is obstructed by a huge terrain feature such as skyscrapers and hills. Most
cellular systems operate in urban areas where there is no direct line-of-sight (LoS) path be-
tween BS and MS, therefore the presence of high-rise buildings causes severe diffraction loss.
Due to multiple reflections from various objects, the electromagnetic waves travel along dif-
ferent paths of varying lengths. The interaction between these waves causes multipath fading
7
1.3 Channel Fading
MS
BS Local Scattering Points
Figure 1.1: Typical outdoor multipath propagation environment
at a specific location. Each of these phenomena is caused by a different underlying physical
principle and each must be accounted for when designing and evaluating the performance of
a cellular system.
Each of the above mentioned wave propagation phenomena has its own importance in
designing and evaluating the performance of a cellular system. However, the carrier wave-
length used in UHF mobile radio applications typically ranges from 15 to 60 cm. Therefore,
small changes in differential propagation delays due to MS mobility will cause large changes
in the phases of the individually arriving plane waves. Hence, multipath fading can be con-
sidered the most important wave propagation phenomenon, which causes small-scale fading
effects. In Fig. 1.1, a typical outdoor multipath propagation environment is shown, where
MS receives many replicas of the transmitted signal, with different delays.
1.3.2 Small-Scale Multipath Fading
The three most important small-scale fading effects of multipath in the radio channel are [8]:
8
1.3 Channel Fading
1. Rapid changes in signal strength over a small travel distance or time interval
2. Time dispersion (echoes) caused by multipath propagation delays
3. Random frequency modulation due to varying Doppler shifts on different multipath
signals.
Multipath fading results in rapid variations in the envelope of the received signal and is
caused when plane waves arrive from many different directions with random phases and
combine vectorially at the receiver antenna. Typically the received envelope can vary by as
much as 30 to 40 dB over a fraction of a wavelength due to constructive and destructive
addition [13].
Multipath also causes time dispersion, because the multiple replicas of the transmit-
ted signal propagate over different transmission paths and reach the receiver antenna with
different time delays. The most important parameters of time dispersion of the channel
are mean excess delay and rms delay spread. Analogous to the delay spread parameters in
time domain, coherence bandwidth, BC , is used to characterize the channel in the frequency
domain. Coherence bandwidth is a statistical measure of the range of frequencies over which
the channel can be considered flat (i.e., a channel which passes all spectral components with
approximately equal gain and linear phase) [8].
Due to the relative motion between the mobile and the base station, each multipath
wave experiences an apparent shift in frequency. The shift in received signal frequency due to
9
1.3 Channel Fading
y Scattering Point
Ψ
Incoming
multipath
MS v
Line of
Sight
θ
x
BS
Figure 1.2: Illustration of Doppler effect
motion is called the Doppler shift, and is directly proportional to the velocity and direction
of motion of the mobile with respect to the direction of arrival of the received multipath
wave. Thus, the Doppler shift can be written as,
v
fd = cos ψ
λ
vfc
= cos ψ (1.1)
c
where v is the speed of the mobile, ψ is the direction of motion of the mobile with respect
to the direction of arrival of the multipath and λ and fc are the wavelength and carrier
frequency of the radio signal, respectively. Fig. 1.2 illustrates the Doppler effect on the
mobile channel in small-scale fading environment.
Since the angle of arrival of the multipath signal at MS varies constantly, Doppler shift
causes random frequency modulation of the radio signal at each instant of time. When a
pure sinusoidal frequency fc is transmitted, the received signal spectrum, called the Doppler
10
1.3 Channel Fading
spectrum, will have components in the range fc − fd to fc + fd . This spectral broadening can
be quantified by Doppler spread. Doppler spread is the measure of the spectral broadening
caused by the time rate of change of the mobile radio channel and is defined as the range
of frequencies over which the received Doppler spectrum is essentially non-zero [8]. If the
baseband signal bandwidth is much greater than the Doppler spread, the effects of Doppler
spread are negligible at the receiver. This is a slow fading channel. Coherence time, TC ,
is the time domain dual of the Doppler spread and is used to characterize the time-varying
nature of the frequency dispersiveness of the channel in the time domain.
1.3.3 Types of Small-Scale Fading
Since small-scale fading channels are characterized by their time-dispersion and time-varying
nature, small-scale fading can thus be categorized on the basis of delay and Doppler spreads
[8, 11, 13]. However, another type of dispersion called spatial dispersion also has a significant
effect on the mobile channel, especially when a multiantenna system, discussed in the pre-
vious section, is used at one or both link ends. Large multipath angular spreads can induce
space-selective fading in multiantenna or directional communication channels.
1.3.3.1 Based on Multipath Time-Delay Spread
Flat Fading: If the mobile radio channel has a constant gain and linear phase response
over a bandwidth which is greater than the bandwidth of the transmitted signal, then
11
1.3 Channel Fading
the received signal will undergo frequency flat fading or simply, flat fading. This type
of fading is historically the most common type of fading described in the technical
literature. Flat fading channels are also known as amplitude varying channels and are
sometimes referred to as narrowband channels, since the bandwidth of the applied sig-
nal, BS , is narrow as compared to the channel flat fading bandwidth or the Coherence
bandwidth, BC . To summarize, a signal undergoes flat fading if [8]
BS << BC (1.2)
and
TS >> στ (1.3)
where BS and TS are the bandwidth and symbol duration of the transmitted signal
and στ is the rms value of the delay spread.
Frequency Selective Fading: If the channel possesses a constant-gain and linear phase
response over a bandwidth that is smaller than the bandwidth of transmitted signal,
then the channel creates frequency selective fading on the received signal. Under such
conditions, the channel impulse response has a multipath delay spread which is greater
than the reciprocal bandwidth of the transmitted message waveform. When this occurs,
the received signal includes multiple versions of the transmitted waveform which are
attenuated (faded) and delayed in time, and hence the received signal is distorted.
Thus the channel induces intersymbol interference (ISI) [8]. For frequency selective
fading, the spectrum S(f ) of the transmitted signal has a bandwidth which is greater
12
1.3 Channel Fading
than the coherence bandwidth BC of the channel. To summarize, a signal undergoes
frequency selective fading if [8]
BS > B C (1.4)
and
TS < σ τ (1.5)
1.3.3.2 Based on Doppler Spread
Depending on how rapidly the transmitted baseband signal changes as compared to the rate
of change of the channel, a channel may be classified as a fast fading or slow fading channel.
Fast Fading or Time Selective Fading: If the channel impulse response changes rapidly
within the symbol duration, then the channel is said to be a fast fading channel.
Under such conditions, the coherence time of the channel is smaller than the symbol
period of the transmitted signal. This causes frequency dispersion (also called time-
selective fading) due to Doppler spreading, which leads to signal distortion. Viewed
in the frequency domain, signal distortion due to fast fading increases with increasing
Doppler spread relative to the bandwidth of the transmitted signal. Therefore, a signal
undergoes fast fading or time selective fading if [8]
TS > T C (1.6)
13
1.3 Channel Fading
Bandwidth
Flat in Time and Selective in both

Selective in Time and Frequency
Frequency
Bc
Flat in Time Flat in Frequency and
and Frequency Selective in Time
Time
Tc
Figure 1.3: Types of small-scale fading
and
BS < B D (1.7)
where BD is the Doppler spread of the channel.
Slow Fading: In a slow fading or time-flat fading channel, the channel impulse response
changes at a rate much slower than the transmitted baseband signal. In this case, the
channel may be assumed to be static over one or several reciprocal bandwidth intervals.
In the frequency domain, this implies that the Doppler spread of the channel is much
less than the bandwidth of the baseband signal. Therefore, a signal undergoes slow
fading or time-flat fading if [8]
TS << TC (1.8)
and
BS >> BD (1.9)
14
1.4 Cellular Environments
A typical cellular radio system consists of a collection of fixed base stations (BSs) that define
the radio coverage areas or cells1 . The height and placement of the BS antennas affect the
proximity of local scatterers at the BS [13]. Radio environments have extremely different
geographical and electrical features, which in turn lead to different propagation mechanisms.
These propagation mechanisms have been grouped into three categories based on the cell
type, i.e. picocell, microcell, and macrocell. However, there would still be variability within
these categories due to other factors such as the differences in antenna heights, average
height of the surrounding buildings, total number of users, distance between users, size of
the coverage area, and density of scatterers, etc.
1.4.1 Picocell Environment
In the picocell environment, BS and MS, both surrounded by scatterers, are a few meters
away from each other (see Fig. 1.4(a)). The antenna heights are relatively low and the
scatterers are assumed to exist near BS as likely as near MS. This situation mainly occurs
in indoor wireless communication, e.g. offices and factory halls, and it may include street
crossings under some circumstances. Wireless Local Area Networks (WLANs) are also good
examples of picocell environment.

1
In military applications the BSs may be moving
15
Low-Height BS Antenna
BS
MS
MS
Scatterers common to both MS and BS
(Equally Influential) Scatterers local to both MS and BS (Influential)
(a) (b)
Remote Dominant Medium-Distance Dominant

Reflectors/Scatterers Reflectors/Scatterers
(Influential) (Influential)
Elevated BS
Antenna
MS
Scatterers local to BS (Non-influential) Scatterers local to MS (Influential)
(c)
Figure 1.4: Multipath scattering in (a) Picocell (Indoor), (b) Microcell (Outdoor), and (c) Macrocell
(Outdoor) Environments
16
1.4.2 Microcell Environment
In the Microcell environment, the distance between BS and MS is greater than that in
picocells (but less than a hundred meters [14]). Here also antenna heights are relatively low
and multipath scattering is assumed near BS. However BS possesses usually fewer scattering
points in its vicinity compared to MS (see Fig. 1.4(b)). This situation corresponds mostly
to streets, shopping malls, busy roads, and downtown areas.
1.4.3 Macrocell Environment
In the macrocell environment, BS is usually located far from MS in the order of kilometers.
This environment often refers to BS locations over rooftops [14]. The BS antenna height
is greater than the surrounding buildings/scatterers, and therefore, it is viable to assume
no scatterers in the vicinity of BS. Hence, the received signal at BS results predominantly
from the local scattering process in the vicinity of MS [15] and the distant scattering process
in the transmission path between BS and MS. Rural, suburban, and hilly areas are the
best examples of the macrocell environment. Fig. 1.4(b) depicts a typical macrocellular
environment. Here, the spatial parameters are the functions of two major aspects, namely:
Local scattering: This is the scattering process which occurs in the vicinity of MS (it is
normally influenced by MS speed).
Distant or remote scattering: This is the scattering process which results from the dom-
17
1.5 Capacity of MIMO Systems in Fading Channels
inant distant scattering structures far from both BS and MS. This type of scattering
can occur in hilly and suburban areas due to large scattering structures such as moun-
tains and high-rise building clusters, which have a significant influence on the mobile
channel. Even though these large scattering structures are far away from both BS and
MS, they can act as discrete reflectors or clustered reflectors [16]. In a ‘bad urban’
environment, a two-cluster model is more appropriate [14], where distant or remote
scatterers are usually the high-rise buildings.
In the early years of MIMO research, the main emphasis was on information-theoretic limits.
After 2000, emphasis shifted more to the question of how the theoretical gains of MIMO
systems can be realized in practice. Capacity of the MIMO systems is strictly dependent on
the channel fading. Following the landmark work done by Claude E. Shannon [17], Foschini
et al. has derived the capacity equation for MIMO systems in non-fading channels [18]. The
information-theoretic capacity for single antenna system, over AWGN channels, is [11]:
¡ ¢
Cshannon = log2 1 + γ|h|2 (1.10)
where γ is the SNR at the receiver and h is the normalized transfer function from the
transmitter to the receiver. In the frequency-flat case, h is just a scalar number. It is
obvious from equation (1.10) that capacity increases only logarithmically with the SNR, so
18
that boosting the transmitter power is a highly ineffective way of increasing capacity.
Now consider the case of MIMO, where the channel is represented by a complex channel
transfer matrix H, whose entries hij correspond to the response of the ith receiver to the
signal sent by the jth transmitter. Thus the matrix H can be written as,
 
h11 h12 ··· 0
 h21 h22 ··· 0 
 
H= .. .. .. ..  (1.11)
 . . . . 
hNr 1 hNr 2 · · · hNr Nt
where Nr and Nt are the number of receive and transmit antennas. The received signal
vector
r = Hs + n (1.12)
is received by Nr antenna elements, where s is the transmit signal vector and n is the AWGN
noise vector. The capacity equation for the MIMO system shown in (1.12) can be written
as [11],
· µ ¶¸
γ̄
CMIMO = log2 det INr + HRss HH (1.13)
Nt
where INr is the Nr × Nr identity matrix, γ̄ is the mean SNR per receiver antenna element,
and Rss is the correlation matrix of the transmit data. It is evident from equation (1.13)
that capacity increases linearly with min(Nt , Nr , Ns ), where Ns is the number of significant
parallel paths in space on which data-streams from transmit antennas are transmitted to the
receive antennas.
19
It is also evident from equation (1.13) that channel matrix plays an important role in
increasing the capacity of a MIMO communication system. If channel is Rayleigh-fading,
and fading is independent at different antenna elements, the hij entries are iid zero-mean,
circularly symmetric complex Gaussian random variables with unit variance, i.e., the real
and imaginary part each has variance 1/2. Suppose that a large number of multipath compo-
nents of approximately equal strength exist and there is sufficient distance between antenna
elements. Then, the power carried by each hij is chi-square-distributed with 2 degrees of
freedom. Since fading is independent, there is high probability that the channel matrix H is
full rank and the eigenvalues are fairly similar to each other; consequently, capacity increases
linearly with the number of antenna elements. Thus, the existence of heavy multipath, which
is usually considered a drawback, becomes a major advantage in MIMO systems [11].
On the other hand, the high spectral efficiency of the MIMO systems is significantly
reduced if the signals arriving at the receivers are correlated [19]. As the BS is usually placed
above local clutter, the angular spectrum incident on the base is narrow, inducing correlation
among BS signals, which reduces the capacity of the MIMO system. Maximum capacity can
only be achieved when hhij h∗kl i = 0 for i 6= k and for j 6= l.
Unfortunately, all channels occuring in practice are not ideal, as discussed in section
1.3. They are more complicated and their deviations from the idealized assumptions can
have significant impact on the capacity [11, 20].
20
1.6 Problem Formulation
As discussed earlier, correlations of the signals at different antenna elements can significantly
reduce the capacity of a MIMO system. Actually, correlations of the signals force the spread
in the singular values of the channel matrix, H, lowering the SNRs of some of the parallel
channels.
Correlation is influenced by the angular spectrum of the channel as well as the arrange-
ment and spacing of the antenna elements [11, 19–21]. For antennas that are spaced half a
wavelength apart, a uniform angular power spectrum (i.e., Jakes Model [22, 23]) leads ap-
proximately to a decorrelation of incident signals. A smaller angular spread of the channel
leads to an increase in correlation.
In order to overcome correlation, accurate spatial models are required. It means we need
to model and characterize the radio mobile channel, accurately, to find the exact angular
spreads of multipath signals at the BS. These angular spreads can thus be used to find the
exact separation between the antenna elements of MIMO arrays. Another fading effect which
significantly degrades the performance of a wireless communication and hence the capacity,
is the Doppler effect. In fast fading channels, frequency dispersion occurs, which leads to
severe signal distortion. Spatial models can also help in modeling the channel correlation
coefficients which can later be used in designing fast fading channel estimators.
As far as spatial channel models are concerned, many approaches [2, 3, 24–28] are used to
21
model the scatterers in some arbitrary scattering region. These approaches are specific either
to the cellular environment or to the distribution of scatterers. However, no generalized
approach for spatial channel modeling has been presented so far. A generalized spatial
channel modeling approach will certainly lead to the accurate characterization of the channel
in angular and temporal domains.
Another important issue, which has not been addressed well in the literature, is the
generalization of the angular spread measure. Several definitions of angular spread [29–33],
have currently being used in the literature, but none is able to attain unanimous acceptance
in the research community. This situation makes the task of performance comparison of
different detection techniques difficult and even leads to wrongful results in some cases.
Since angular spread is the main cause of correlation between antenna elements, its wrong
definition will result in mismatched separations between antenna elements, decreasing the
capacity of the communication system.
All the above issues of performance degradation and capacity reduction are properly
addressed in their respective chapters in the thesis, where detailed overviews and broad
historical backgrounds present a comprehensive picture of our motivations and work.
22
1.7 Scope of the Thesis and Proposed Methodology
Our methodology to achieve the goals set in the previous section is illustrated in Fig. 1.5.
In the first round of our research, we extensively study the previous spatial channel models
and find that almost all models are very specific to the particular cellular environments
and cannot be generalized. We thus propose a novel spatial channel model, which can be
exploited to model any cellular environment with the use of appropriate parameters. In the
second round, we derive the angular and temporal statistics of the cellular mobile channel
and discuss the impact of local-to-BS scattering on the angular dispersion. In the third
round, we address the issue of unanimous definition for the angular spread and propose a
novel method of quantifying the angular spread. In the fourth round, we investigate the effect
of mobile motion on the spatial and temporal characteristics of the cellular mobile channel,
which can lead to the optimization of spatial and temporal statistical parameters as functions
of BS-MS separation. In the fifth and last round, we propose a single user adaptive receiver
structure for single-carrier DS-CDMA system over time-varying fast fading channel.
Since the modeling approach used in the thesis can easily be extended to address various
issues of capacity maximization and performance improvement of a cellular mobile commu-
nication system, there are several topics of future research work that can be investigated.
For example, the spatial statistics at BS derived in the thesis, can be utilized for the deriva-
tion of signal correlations between antenna elements to maximize the capacity of the system.
Similarly, robust channel equalizers and RAKE receivers can be designed with the help of
23
Geometrically based Spatial Channel

Modeling for all Cellular Environments
Temporal Statistics Mobile Spatial Statistics

(pdf of ToA) Movement (pdf of AoA)
Doppler At MS At BS
Spread
Parameters of Autocorrelation of Angular Spread at

Delay Spread Time-Varying Channel BS Antenna
Coherence Time
Channel Estimator Signal Correlations b/w

or Channel Tracker Antenna Elements
Antenna Coherence
Coherence Rake Diversity Distance
Bandwidth Receiver
CDMA
Channel case
Equalizer Exact Separations b/w
Envelope Antenna Elements
Full Exploitation of Multipath Fading
Power in case of CDMA Compensation
Signal Detection over Time-

Varying Fading Channels in the
presence of ISI Correlation-free Signal
+ Reception for System
ISI-free Signal Capacity Maximization
Maximum Achievable Capacity by
Detection
using MIMO multi-antenna system
Figure 1.5: Illustration of the Proposed Dissertation Methodology
24
1.8 Contributions of the Thesis
temporal statistics, derived in the thesis, for signal detection schemes in frequency selec-
tive fading channels. Thus, in order to complete the picture of a full-bodied, efficient and
high-capacity cellular mobile communication system, Fig. 1.5 provides all necessary links to
various areas of future work, where our proposed results can be exploited.
Following the methodology depicted in Fig. 1.5, we have presented most of our theoretical
and practical findings in various international journals and conferences, at different stages
of our research tenure. The papers listed under the title of ‘Related Publications’ show
such presentations. This dissertation thus presents the systematized and combined version
of all those papers. However, some important additions have made it more vigorous and
comprehensive.
The organization of the rest of the thesis is as follows:
In Chapter 2, we address the issue of physical channel modeling for the cellular mobile
communication system. In the first part of the chapter, we extensively study the previous
approaches used for mobile channel modeling in picocell, microcell, and macrocell environ-
ments, and present necessary channel modeling parameters. In the second part, we propose
a generalized physical channel model, referred to as the ‘Eccentro-Scattering Model’ which
can be applied to any type of cellular environment with appropriate choice of eccentricity,
25
semi-major axis, and distribution of scatterers around MS and/or BS. We also introduce
a more applicable scattering model, the Jointly Gaussian Scattering Model (JGSM), which
consists of two Gaussian functions each for the distribution of scatterers around BS and MS.
We exploit these models (either individually or jointly) later in the thesis, in order to derive
the spatial and temporal characteristics of the cellular mobile channel.
In Chapter 3, we exploit the proposed Eccentro-Scattering Model to derive the general
expressions for the pdf of AoA of the multipath signals at BS in closed form for picocell,
microcell, and macrocell environments, assuming uniform and Gaussian distributed scatter-
ers. Gaussian distributed scatterers are confined, for the first time, within a scattering disc
and the advantages of this technique are discussed. Also, distant scatterers are considered
in macrocell environments and the pdf of AoA of the multipaths at BS due to dominant
distant scattering clusters is derived thereby. Since the AoA statistics are usually affected
by the high-rise scattering structures around BS antenna in urban macrocell environments,
we propose a modified JGSM model for such environments and present the comparison of
its results with the field measurements. The terrain and clutter conditions of the BS sur-
roundings are simply modeled by a scattering-free region around BS. In this way, the radius
of the scattering-free region indicates the extent of the density of the scattering structures
around BS. All presented formulas are compared with the results obtained by the previ-
ous researchers [2, 3, 24, 26–28]. We show that the Eccentro-Scattering model can be used
to model any type of cellular environment, by simply changing some of the modeling pa-
rameters. A table of comparisons is presented which shows the generality of our proposed
26
Eccentro-Scattering model.
In Chapter 4, we discuss all previous definitions of the angle spread and propose
a novel generalized method of quantifying the angle spread of the multipath power. Our
method provides almost all parameters of the dispersion of multipath power in space, which
can be further used for calculating more accurate spatial correlations and other statistics
of multipath fading channels. These proposed parameters are also helpful in finding the
exact standard deviation of truncated or distorted angular distributions as well as of the
angular data acquired in measurement campaigns. The resulting standard deviations can
lead to the computation of the exact separation distances among array elements needed
for the maximization of capacity and the usage of diversity antennas. Keeping the recent
use of truncated Gaussian and Laplacian functions as the distributions of AoA in view,
we indicate the severity of the effects of distribution truncation on the angle spread. We
itemize all those factors which cause such truncations in the AoA distributions and provide
analytical solution to compensate for their effects. Furthermore, we propose the angle spread
as the goodness-of-fit measure in measurement campaigns and show the comparisons of some
notable azimuthal models with the measurement results.
In Chapter 5, we derive simplified closed-form expressions for the ToA distributions
for picocell/microcell and macrocell environments by considering the scatterers confined in
Eccentro-Scattering discs. This is a more general approach from which the results of the
previous models, such as GBSBM model [3], can be deduced as its special cases. For macro-
27
cell environments, we also include the effect of dominant distant scatterers on the temporal
dispersion of the multipath signals. Objects such as hills, mountains, and skyscrapers act as
clustered scatterers/reflectors when they have line of sight (LoS) to both BS and MS [16, 25].
The derived formulas in chapter 5, can easily be used to simulate temporal dispersion of wire-
less signal in several propagation conditions. Besides the handy use of the pdf of ToA in
determining the coherence bandwidth of a particular system, the pdf of ToA also emerges
as a basic characteristic of the system capacity along with the pdf of AoA [22]. Almost
complete description of the wireless system can be achieved if the pdfs of time and angle of
arrival are known.
In Chapter 6, we consider a realistic situation of a moving MS for the characterization
of multipath fading channel and investigate how this motion affects the spatio-temporal
characteristics at BS. We present three mobile motion scenarios that are responsible for the
alterations in the spatio-temporal characteristics and plot the behavior of various spatial and
temporal spread quantifiers during these motion scenarios. We also explain the behavior of
angle spread under the effect of mobile motion, observed in the field measurements [14]. We
show that the model successfully simulates the time-variability of the angle and delay spreads
induced during the course of MS motion. We identify two different cases when the terrain and
clutter of MS surroundings have an additional effect on the temporal spread of the channel
during MS motion. These cases can provide good basis for the performance evaluation in
those wireless systems which employ additional time-delay processing techniques.
28
In Chapter 7, we develop an autoregressive (AR) model of the time-varying flat fading
channel on the basis of its second order fading statistics and utilize it further to establish
a linear state-space equation pair for signature sequence adaptation in direct sequence code
division multiple access (DS-CDMA) system. We then exploit the Kalman filtering approach
to incorporate our proposed state-space equation pair in an algorithm, meant for estimating
channel-distorted received signature sequences. It is an established fact [34] that the Kalman
filter is a good optimal linear minimum mean squared error (MMSE) detector if a first
order linear state-space model is applied to DS-CDMA system. As mentioned earlier, we
also use the Kalman filter as the MMSE solution for signature distortions caused by the
time-varying fading channel. However, different from the model used in [34], where the
unknown transmitted symbol vector has been used as the state-vector, we use channel-
distorted received signature vector as the state-vector in our model. This approach is based
on the fact that the time-varying channel behaves as the AR model depending on its past
values [23, 35]. In our proposed receiver structure, the need for training sequence is bound
to the startup period only. Later on, the receiver adapts itself to the changes of the channel
during data transmission depending on previous decisions. Simulation results show that
being based on the Kalman filter and of non-gradient nature, our proposed algorithm combats
effectively the impairments and fading effects caused by time-varying multipath channel.
Finally, Chapter 8 gives a brief summary of the thesis, discusses the future research
work based on the results of the dissertation and presents the concluding remarks.
29
Chapter 2
A Generalized Spatial Channel Model
This chapter gives a detailed overview of physical spatial channel modeling in its section 2.1.
Section 2.2 and 2.3 describe the general channel modeling assumptions and parameters. In
section 2.4, the proposed generalized spatial channel model, the Eccentro-Scattering Model is
presented, which will be used as a global model throughout the thesis. Section 2.5, presents
another proposed scattering model, the Jointly Gaussian Scattering Model (JGSM), which
models the heavy scattering around BS in low antenna-height environments. Section 2.6
discusses the implementation of these models in various cellular environments, discussed
in the previous chapter. Finally, section 2.6 gives the summary and final remarks of this
chapter.
2.1 Introduction
2.1 Introduction
2.1.1 Overview of Physical Spatial Channel Modeling
With the growing use of antenna arrays in MIMO systems for enhancing system capacity,
signal detection, interference cancellation, and position location comes the need to better
understand the properties of the spatial channel [36]. MIMO systems have been shown to
offer very high information-theoretic capacities [18, 37]. In order to make realistic evaluations
of the capabilities of different system architectures and MIMO schemes, realistic channel
models are required [38].
Although it is very difficult to categorize the previous approaches used in channel modeling,
especially the earlier work [22, 23, 39]; however, Molisch et al. [38] has grouped the recent
generic modeling approaches for MIMO channels into the following two categories:
Non-physical modeling: This approach is used to model the correlations of the fading of
the signals at the antenna elements [40].
Physical Modeling: This approach models the location of scatterers/reflectors, or the di-
rection of multipath components at the transmitter and receiver.
The latter approach has become more and more popular recently [38, 41], and is being used in
many standard models such as COST 259 DCM [14] and 3GPP [42]. The resultant physical
channel models that use physical modeling approach for the location of scatterers/reflectors
31
2.1 Introduction
are also known as spatial channel models or the scattering models. Since scatterer locations
with respect to MS and BS provide the angular and temporal information of the multi-
path signals, the spatial models are very helpful in evaluating the performance of a wireless
communication system using antenna arrays.
For spatial models, it is a well-established fact that the scatterer locations (or equiva-
lently, the angles and delays of multipath components) are not distributed uniformly over
space, but tend to be concentrated in certain regions [38]. It is possible to obtain the pdf of
AoA and power delay profile from the measured data or from site-specific propagation pre-
diction techniques, but this type of data might not always be available. Therefore, physical
channel models providing statistics of the channel in angle and time-delay are very helpful
in characterizing the angular and time domains of the multipath signals. Accurate and, if
possible, simple geometrically-based physical propagation models would lead to an effective
design and evaluation of modern communication systems. Such models are low-cost and
handy means to accurately predict radio wave propagation behavior.
2.1.2 Previous Work
Appreciable contributions in spatial channel modeling have been made in [2, 3, 24–28]. The
two most common modeling parameters, used by the researchers in spatial channel modeling,
involve shape of the scattering region [3, 24–26, 39], and the scatter density in the scatter-
ing region [2, 27, 43]. Different shapes of the scattering region were proposed to model the
32
2.1 Introduction
scattering phenomenon for specific cellular environments. For example, Circular Scattering
Model (CSM) was proposed to model the scattering environment in macrocells [3, 25, 26],
while Elliptical Scattering Model (ESM) was proposed to model the scattering environments
in microcells and picocells [3, 24]. On the other hand, uniform and Gaussian distributions
were the most common approaches proposed for the distribution of scatterers within the
scattering region. In a uniformly distributed scattering region, scatterers are assumed to
have constant density throughout the scattering area, while in a Gaussian distributed scat-
tering region the majority of scattering points are situated close to mobile station (MS)
and the density of scattering points decreases as the distance from MS increases. The uni-
form scattering assumption simplifies the analysis and manipulation whereas the Gaussian
distribution is a more realistic approach especially in macrocell environments where no scat-
terers are assumed around BS, since the scatterers around MS have the greater influence
on the received signal as compared to other scatterers [44]. Furthermore, the Gaussian dis-
tributed scattering assumption provides higher degrees of freedom to the model in its ability
to change the width and density of the scattering area by changing the standard deviation
of the distribution of the scatterers around MS.
As far as validity of a model to a particular environment is concerned, the existing
models are very specific to the corresponding environments either with respect to the shape
of the scattering disc [3, 24–26] or with respect to the standard deviation of the Gaussian
distributed scatterers around MS [2, 27]. In [28], an elliptical model was proposed that
could be used as a circular one as well, with change in eccentricity of the ellipse assuming
33
2.1 Introduction
uniformly distributed scattering environment. The model might be suitable for macrocell
and quasi-macrocell environments, even though the suitable distribution of the scatterers
in macrocells is indeed Gaussian. Nevertheless the model in [28] was not able to explain
the radio wave propagation phenomenon in microcell and picocell environments where MS
is usually not situated at the center of the scattering area and BS possesses scatterers in
addition to those around MS. In such environments the antenna heights, of transmitter and
receiver, are relatively low and multipath scattering is assumed near BS to be as likely as
that around MS.
The study in [2] proposed a Gaussian Scatter Density Model (GSDM) to be applied to
every type of environment by changing only the standard deviation of the scatterers around
MS. Generally in picocell environments, scatterers exist in the vicinity of BS as much as
around MS. Therefore, it is more practical to model the distribution of scatterers around
BS separately from the distribution of scatterers around MS. In other words, a separate
Gaussian function for the distribution of scatterers around BS is also needed in addition to
the Gaussian function for the distribution of scatterers around MS.
In several measurement campaigns aiming to find the actual distribution of the Angle of
Arrival (AoA) of the multipath signals, both in indoor [1, 45–47] and outdoor (microcell [48]
and macrocell [4]) environments, different distinct distributions for the AoA of the multipath
signals at Base Station (BS) were observed on a short range of angular domain on both sides
of the mean AoA, while the distribution over the rest of the angular domain was found to
34
2.1 Introduction
be very uniform in all cases. Both sides of this uniform region are usually referred to as
the tails of the probability density function (pdf) of AoA. The formation of these uniform
tails in the distribution of AoA remained an unsolved issue in the statistical scattering
models [1, 2, 4] proposed so far. Thus a generalized scattering model is needed which can
explain the formation of these tails in the AoA distribution.
2.1.3 Contributions in the Chapter
In this chapter, we address the issue of physical channel modeling for the cellular mobile com-
munication system. We study intensively the previous approaches used for modeling cellular
mobile channel in picocell, microcell, and macrocell environments. We present necessary
channel modeling parameters and propose a generalized physical channel model, referred to
as the ‘Eccentro-Scattering Model’ which can be applied to any type of cellular environment
with appropriate choice of eccentricity, semi-major axis, and distribution of scatterers around
MS and/or BS. We also introduce a more applicable scattering model, the Jointly Gaussian
Scattering Model (JGSM), which consists of two Gaussian functions each for the distribu-
tion of scatterers around BS and MS. We will exploit these models (either individually or
jointly) to derive the spatial and temporal characteristics of the cellular mobile channel in
the following chapters.
35
2.2 General Channel Modeling Assumptions
y Scattering Point
Incoming
Multipath MS
Line of
Sight

x
BS
Figure 2.1: Typical multipath fading environment
Multiple replicas of the transmitted signal are received at the receiver due to the multipath
propagation in the radio environment. These multipath components of the received signal
arrive at the receiver antenna from different azimuth directions about the horizon with
identical or different delays [22, 23, 49]. The distributions of these multipath components in
the azimuth and time are conveniently described by the functions p(θ) and p(τ ), where θ is
the azimuthal angle of arrival (AoA) [32] and τ is the delay of the multipath component in
time domain. Before proceeding to formulate a generalized spatial channel model, we make
the following assumptions:
• Scatterers are confined in an elliptical shaped scattering disc whose eccentricity can be
changed according to the maximum delay and the distance between BS and MS. It is
36
desirable, practically, to consider only those scatterers that have significant influence
on the received signal. For example, multipaths with longer delays experience greater
path loss and hence have relatively low power as compared to those with shorter
delays [24, 44]. In our work, such a scattering disc, which can imitate both circular
and elliptical scattering discs with corresponding choice of eccentricity, is referred to
as the ‘Eccentro-Scattering Disc’. In the next two sections, we will explain the use of
Eccentro-Scattering disc, in detail.
• The received signal at the antenna undergoes no more than one reflection by scatterers
when traveling from transmitter to receiver. Placing a scattering object at the last re-
radiation and approximating the preceding scattering as a stochastic process can retain
some of the properties of multiple bounces while providing for a much simpler model
[50]. Practically, we are considering only the distribution of scatterers contributing to
the last reradiation while the preceding multiple-bounce can be modeled as a stochastic
process which has lognormal shadowing with Nakagami fading [20]. This assumption,
in conjunction with the first assumption, means that we are considering all scatterers
giving rise to a single bounce multipath signal arriving at the receiving antenna before
and up to time, τmax , where τmax is the maximum allowed delay, i.e. the time difference
between the first and the last signal arrivals at the receiving antenna with signal power
greater than some threshold value defined by the system designer.
• Each scatterer is assumed to be an omnidirectional re-radiating element with equal
37
scattering coefficients and uniform random phases. The word ‘scattering’ is not only
used for diffuse scattering but even for those processes that are strictly speaking ‘spec-
ular reflections’ [14].
• The effective antenna patterns are omnidirectional for both transmitter and receiver.
Practically, the derived formulas for the pdf of AoA should be used in conjunction with
the actual antenna radiation pattern.
• BS is set at the origin of the global coordinate system. β is the angle between Line of
Sight (LoS) and x-axis of this coordinate system, see Fig. 2.1. From now onward, the
word ‘x-axis’ refers to the x-axis of the global coordinate system. Hence, the location of
any scattering point, S, in the system is denoted by the Cartesian coordinates (xS ,yS )
or in polar coordinates (rBS ,θ). Furthermore, the angle θ = 0◦ corresponds to the angle
towards MS, i.e. the direction of Line of Sight (LoS). Consequently, MS is located at
(xM ,yM ) in Cartesian coordinates and at (d,0) in polar coordinates.
• All signals received at the antenna are plane waves coming from the horizon, i.e. only
azimuthal coordinates are considered. Some environments such as indoor cellular en-
vironments may require knowledge of the elevation angle while in other environments
such as macrocell environments the waves propagated over rooftops experience stronger
attenuation, therefore, it is reasonable to study only the azimuthal angle in such envi-
ronments [14]. Yet, the established results are helpful in the analysis and performance
assessment of modern wireless techniques, such as smart antenna system, for all cel-
38
2.3 General Channel Modeling Parameters
lular environments. The angle θ in Fig. 2.1 is the azimuthal angle of arrival of the
multipath signal at BS with respect to x-axis.
• A specific signal delay τ defines a set of scatterers bounded by an elliptical region, with
distinct parameters, which will be referred to as the ‘bounding ellipse’ in our work. The
scatterers on the boundary of the bounding ellipse give rise to single bounce multipath
components [24, 51]. In chapter 5, we will explain further the significance and usage of
the bounding ellipse.
As discussed earlier in section 2.1 of this chapter, all previous spatial channel models are
based on specifying either the shape of the scattering region or the scatterer distribution in
the scattering region, in order to model a specific cellular environment.
2.3.1 Shape of the Scattering Region
There are two further models, which are usually used in modeling the scattering environment
around MS, depending on the shape of the scattering region with uniform scatter density.
39
Y S
rBS
rMS
θ X
B d M
(a) Circular Scattering Model

S
rBS
rMS
θ
B C M
d
(b) Elliptical Scattering Model
Figure 2.2: Modeling with respect to the shape of the scattering region for uniform scatterer distri-
bution
2.3.1.1 Circular Scattering Model
This model is specifically used for macrocell environments [3, 25, 26], where BS antenna is
mounted on an elevated place. A typical circular scattering model is shown in Fig. 2.2(a).
It can excellently model villages and sparsely populated small towns. Since the distribution
of scatterers in macrocell environments tend to be Gaussian, giving rise to Gaussian AoA
distribution [4], this model can not be taken as the ultimate solution. However, it is very
simple and less calculations are involved in deriving its spatial statistics and correlations
among antenna elements at BS. However, the derivation of its temporal characteristics at
BS needs complicated calculations.
40
2.3.1.2 Elliptical Scattering Model
This model is specifically used for picocell and microcell environments [3, 24], with low BS an-
tenna heights. A typical elliptical scattering model is shown in Fig. 2.2(b). It can excellently
model the indoor propagation phenomenon but fails in modeling dense urban streets, where
density of scatterers is not uniform. This model is comparatively more complicated than
circular scattering model in deriving its spatial statistics and correlations among antenna
elements at BS. However, the derivation of its temporal characteristics is very easy.
2.3.2 Distribution of Scatterers
The wave propagation path changes in different environments according to scatterers’ density
and distance between BS and MS in regard to the maximum delay. The two most commonly
used approaches in literature for modeling the distribution of scatterers within the scattering
disc are the uniform distribution and the Gaussian distribution.
2.3.2.1 Uniformly Distributed Scatterers
A uniform model of the spatial pdf of scatterers distributed uniformly within an arbitrarily
shaped region, RA with an area A around MS (and BS in case of elliptical scattering model)
can be written as [3], see Fig. 2.2,

(
1
pXS ,YS (xS , yS ) = A , x S , y S ∈ RA (2.1)
0, elsewhere
41
Y
S
θ X
d
BS MS
Figure 2.3: Modeling with respect to the Gaussian scatterer distribution
Or in polar coordinates, [52],

(
krBS k
pRBS ,Θ (rBS , θ) = A , rBS > 0, {rBS , θ} ∈ RA (2.2)
0, elsewhere
where rBS is the position vector of the scattering point S with respect to BS and θ is the
angle which rBS makes with the horizontal (in our case the AoA of the multipath signal
at BS from scattering point S), as shown in Fig. 2.2. Also, kxk denotes the norm of any
position vector x.
2.3.2.2 Gaussian Distributed Scatterers
The Gaussian distribution of scatterers is a better approximation to physical reality [25].
It assumes that the majority of scattering points are clustered together with their density
decreasing as the distance from MS increases [16].
A Gaussian model of the spatial pdf of the scatterers around MS can be written as
42
2.4 The Proposed Eccentro-Scattering Channel Model
" ¯ ¯#
¯ ¯
(M S) 1 ¯(xS − xM )2 + (yS − yM )2 ¯
pXS ,YS (xS , yS ) = 2
exp − 2
(2.3)
2πσM S 2σM S
Or in polar coordinates, (2.3) can be written as [52],

 µ 2
¶
 krBS k exp − krBS − rBM k , r > 0
(M S) 2 2 BS
pRBS ,Θ (rBS , θ) = 2πσM S 2σM S (2.4)

0, elsewhere
where σM S is the standard deviation of the distribution of scatterers around MS and rBM is
the position vector of MS with respect to BS, as shown in Fig. 2.3. The spatial pdf of the
scatterers around MS presented in [25, 27] is deficient in rBS factor, which is the Jacobian of
the transformation from the Cartesian coordinate system to the polar coordinate system.
The so-called Gaussian Scatter Density Model (GSDM) [2] and the other spatial models pre-
sented in [27, 43] are based on the Gaussian distribution of scatterers around MS. Although
Gaussian distribution is a better approximation to the physical scattering phenomenon, the
derivation of spatial statistical formulas in Gaussian distributed scattering is more compli-
cated compared to that in uniform distributed scattering. That is why Gaussian models are
not very popular among the research community.
Consider an ellipse centered at C with foci B and M separated by a distance 2f and with
major and minor axes 2a and 2b respectively, as shown in Fig. 2.4. rBS and rM S satisfy
rBS + rM S = 2a (2.5)
43
rBS rMS
θ
2b
B C M
2f
2a
Figure 2.4: Elliptical diagram
Eccentricity of the ellipse given in Fig. 2.4 is defined as
√
e= 1 − κ2 (2.6)
where κ (the aspect ratio of the ellipse) = b/a. Eccentricity can also be defined in terms of
the distance between the two foci and major axis as
f
e= (2.7)
a
The circle is a special case of the ellipse with e = 0. The elliptical diagram presented in
Fig. 2.4 can easily be implemented to model any cellular mobile environment with scatterers
around MS and/or BS.
Restricting the shape of the scattering disc to be either circular or elliptical alone, does
not allow full flexibility to the physical channel modeling since the shape of the scattering
disc changes in accordance with the terrain, clutter, and location of scattering structures.
44
For example, local scatterers in streets and shopping malls in urban areas [44] and dominant
distant scatterers like hills near villages and towns in rural areas can be best represented by
elliptical scattering discs where the length and width of streets and shopping malls in the
case of urban and those of hills in the case of rural areas determine the eccentricities and
major axes of the elliptical scattering discs. On the other hand, local scatterers around MS
in rural areas are best modeled by circular scattering discs. Therefore, there is a need for a
flexible model capable of modeling every type of cellular environment.
In this research work, we exploit the elliptical diagram depicted in Fig. 2.4 to make it
feasible to be used as a generalized physical channel model, the ‘Eccentro-Scattering Model’,
named due to the fact that it can be applied to any type of cellular environment with
appropriate choice of eccentricity, semi-major axis, and distribution of scatterers around MS
and/or BS. The elliptical disc of scatterers, whose eccentricity can be adjusted with the type
of cellular environment and the conditions of the terrain, is referred to as the ’Eccentro-
Scattering Disc’.
Contrary to the previous physical channel models [2, 3, 24, 26–28], which are very specific
to particular environments, the Eccentro-Scattering model provides a unified approach in
mobile channel modeling. We will use this model throughout our work for deriving spatial
and temporal statistics of the cellular mobile channel.
45
2.5 The Proposed Jointly Gaussian Scattering (JGSM) Model
2.5 The Proposed Jointly Gaussian Scattering (JGSM)
Model
As mentioned earlier, Gaussian assumption for the distribution of scatterers in the scattering
region is considered the most appropriate in the case of urban, suburban and rural macrocell
environments [25]. In these environments, the BS is usually considered free of scatterers.
That is why all the previous Gaussian scattering models assume the Gaussian scattering
cluster centered at MS. However, in some measurement campaigns aiming to find the actual
distribution of the AoA of the multipath signals, both in indoor [1] and outdoor [4] environ-
ments, two distinct distributions for the AoA of the multipath signals at BS were observed
on a short range of angular domain on both sides of the mean AoA, while the distribution
over the rest of the angular domain was found to be very uniform in both cases. Both sides
of this uniform region are usually referred to as the tails of the probability density function
(pdf) of AoA. The formation of these uniform tails in the distribution of AoA still remains
an unsolved issue in the statistical scattering models [1, 2, 4] proposed so far. These tails are
in fact the aftermath of the reflections/scattering of the radio signal from the scatterers that
surround BS, see Fig. 1.4.
Keeping the effect of scattering in the vicinity of BS in view, we propose a novel Gaussian
scattering model for the distribution of scatterers around MS and BS. In our proposed
model, Gaussian distributed scattering points around BS are also taken into consideration
46
2.5 The Proposed Jointly Gaussian Scattering (JGSM) Model
Y
S2 S1
θ2
θ1 X
BS d
MS
Figure 2.5: Jointly Gaussian Scattering Model (JGSM)
in addition to Gaussian distributed scattering points around MS in order to simulate real
scattering phenomenon more realistically. We propose two distinct Gaussian functions for
the distribution of scatterers around BS and MS with different standard deviations. This
model will be referred to as the ‘Jointly Gaussian Scattering Model’ (JGSM), in our work.
This is in contrast to GSDM, where only one Gaussian function is used for the distribution
of scatterers centered at MS. Furthermore, in order to simulate the heaviness of scattering
around BS, we introduce a scattering-free zone around BS whose dimensions depend generally
on the cell size and BS surroundings. We will explain the significance of this scattering-free
region in channel modeling, in the next chapter. Fig. 2.5, depicts JGSM model.
In addition to (2.3), we can write the Gaussian model of the spatial pdf of the scatterers
around BS as, ¯ ¯
" ¯ 2 2¯#
(BS) 1 ¯xS + yS ¯
pXS ,YS (xS , yS ) = 2
exp − 2
(2.8)
2πσBS 2σBS
Or in polar coordinates, (2.8) can be written as

 µ 2
¶
 krBS k exp − krBS k , r > 0
(BS) 2 2 BS
pRBS ,Θ (rBS , θ) = 2πσBS 2σBS (2.9)

0, elsewhere
47
2.6 Application of the Eccentro-Scattering Model
where σBS is the standard deviation of the distribution of scatterers around BS. Other
notations have been explained in section 2.3.2. JGSM along with the Eccentro-Scattering
model provides a generalized approach to characterization of the spatial phenomenon of
multipath signals.
We exploit the Eccentro-Scattering model to represent all cellular environments with corre-
sponding choice of eccentricity (e), semi-major axis (a), and location of the foci and center
points of the ellipse.
2.6.1 Picocell Environment
The antenna heights in picocell environments are very low, therefore, the scatterers are as-
sumed to exist near BS as likely as near MS (see Fig. 2.6a). Thus, according to the Eccentro-
Scattering model, BS and MS are located at the focal points of the Eccentro-Scattering disc.
The choice of the distribution of scatterers depends on the terrain and clutter conditions of
the environment. If a shopping mall or railway station is to be modeled, then Gaussian as-
sumption for the distribution of scatterers is the most appropriate choice, otherwise uniform
assumption is most suitable for offices and factory halls. In case of Gaussian distributed
scatterers, JGSM along with the Eccentro-Scattering model is the best modeling approach.
48
BS MS
(a)
BS MS
(b)
Figure 2.6: Typical (a) Picocell and (b) Microcell Environments in Gaussian distributed scattering
2.6.2 Microcell Environment
In the Microcell environment, the antenna heights are also not very high and multipath
scattering is usually assumed near BS. However BS possesses relatively few scattering points
in its vicinity compared to MS (see Fig. 2.6b). Thus, according to the Eccentro-Scattering
model, BS and MS are located at the focal points of the Eccentro-Scattering disc. As for
the picocell environment, here also the choice of the distribution of scatterers depends on
the terrain and clutter conditions of the environment.
49
Table 2.1: Typical Values of the Parameters for Pico-

cell and Microcell Environments Based on the Eccentro-
Scattering Model
Environment Picocell Microcell

Delay Spread (τmax ), µs ≤ 0.1 0.3
Angle Spread (2θmax ), degrees 360 120
Semi-major axis (a), m 15 < a < 65 85 < a < 545
Rooftop height, m Not Applicable 15
BS antenna height, m 2−3 3 − 10
MS antenna height, m 1.5 1.5
Distance between BS and MS (d), m 1 < d < 30 1.5 < d < 500
Eccentricity (e) 0<e<1 0.33 < e < 1
Numbers of scatterers around BS and NB ≈ NM , NM >> 1 NB < NM , NM >> 1
MS (NB , NM )
Standard Deviations (σBS , σM S ) of σBS = σM S , σBS ≥ σM S ,
scatterers around BS and MS ae ≤ σM S ≤ 2ae 0.4ae ≤ σM S ≤ 0.68ae
50
Table 2.2: Typical Values of the Parameters for Macrocell

Environments Based on the Eccentro-Scattering Model
Environment Rural Typical Urban Suburban or Bad Urban Hilly

Nature of Scatterers Local to Local to Local to Distant Local to Distant
MS only MS only MS MS
Delay Spread, µs 0.5 5 10 2 20 2
Angle Spread, degrees 1 20 20 − 25 2 30 2
Rooftop height, m - 15 30 -
BS antenna height, m 50 30 50 50
MS antenna height, m 1.5 1.5 1.5 1.5
Distance between BS and 5000 500 500 - 5000 -
MS (d), m
Semi-major axis of local 75 750 75 − 750 - 75 − 750 -
scattering disc (a), m
Eccentricity of local 0 ≈ 0.4 0.2 − 0.4 - 0 − 0.4 -
scattering disc (e)
Distance between BS and - - - D ≤ 3000 - D ≤ 5000
MS via center of the dis-
tant scattering disc(D), m
Semi-major axis of distant - - - 75 − 750 - 750 − 1500
scattering disc (aD ), m
Eccentricity of distant - - - ≈ 0.2 - 0 − 0.5
scattering disc (eD )
51
2.6.3 Macrocell Environment
In the macrocell environment, the BS antenna is higher than the surrounding buildings/scatterers,
therefore, it is viable to assume no scatterers in the vicinity of BS. Here, the received signal at
BS results predominantly from the local scattering process in the vicinity of MS [15] and the
distant scattering process in the transmission path between BS and MS. Thus, according to
the Eccentro-Scattering model, BS is located outside the scattering region and MS is located
at the center of the Eccentro-Scattering disc. We also consider another Eccentro-Scattering
disc for the remote scatterers with some arbitrary center, which depends on the geographical
terrain conditions of the scattering structures.
We consider suburban and hilly macrocell environments as having both dominant dis-
tant scatterers and local scatterers [41] as shown in Fig. 1.4(b), while in flat rural and urban
macrocell environments we only consider local scatterers around MS.
Exemplary values of the model parameters are listed in Tables 2.1 and 2.2. The values
in the tables are merely typical and they may differ slightly from more specific ones. For
example, the delay spread in corridors (picocells) may become larger than what is indicated in
Table 2.1. This difference in values usually results from the variety of antennas, measurement
techniques, measurement locations, and various assumptions made by the working groups.
However, we have chosen the most appropriate values for the model parameters listed in
Tables 2.1 and 2.2 from the studies in [2, 11, 14, 16, 53].
52
2.7 Summary of the Chapter
2.7 Summary of the Chapter
In this chapter, we have addressed the issue of physical channel modeling for the cellular
mobile communication system. We have studied intensively the previous approaches used
for modeling cellular mobile channel in picocell, microcell, and macrocell environments. We
have developed necessary channel modeling parameters and proposed a generalized physical
channel model, referred to as the ‘Eccentro-Scattering Model’, which can be applied to any
type of cellular environment with appropriate choice of eccentricity, semi-major axis, and
distribution of scatterers around MS and/or BS. We have also introduced a more appli-
cable scattering model, the Jointly Gaussian Scattering Model (JGSM), which consists of
two Gaussian functions each for the distribution of scatterers around BS and MS. We will
exploit both the JGSM and the Eccentro-Scattering model (either individually or jointly) to
derive the spatial and temporal characteristics of the cellular mobile channel in the following
chapters. The same methodology can also be used to develop a spatial channel model for a
3-D environment.
53
Chapter 3
Modeling Spatial Characteristics of
Mobile Channel
In this chapter, section 3.1 gives overview of the chapter contributions. Section 3.2 and 3.3
present the derivation of the closed-form formulas for the pdf in AoA of multipaths as seen
from BS for picocell/microcell and macrocell environments. Section 3.4 models the impact
of scattering around BS on the AoA statistics of the cellular mobile channel in dense urban
environments. Finally, section 3.5 concludes the paper and presents final remarks.
3.1 Overview
3.1 Overview
In this chapter, we exploit the proposed Eccentro-Scattering Model to derive the general
expressions for the pdf of AoA of the multipaths at BS in closed form for picocell, mi-
crocell, and macrocell environments assuming uniform and Gaussian distributed scatterers.
Gaussian distributed scatterers are confined, for the first time, within a scattering disc and
the advantages of this technique are discussed. Also, distant scatterers are considered in
macrocell environments and the pdf of AoA of the multipaths at BS due to dominant dis-
tant scattering clusters is derived thereby. Since the AoA statistics are usually affected by
the high-rise scattering structures around BS antenna in urban macrocell environments, we
propose a modified JGSM model for such environments and present its comparison with the
field measurements. The terrain and clutter conditions of the BS surroundings are simply
modeled by a scattering-free region around BS. In this way, the radius of the scattering-free
region indicates the extent of the density of the scattering structures around BS. All presented
formulas are compared with the results obtained by the previous researchers [2, 3, 24, 26–28].
We show that the Eccentro-Scattering model can be used to model any type of cellular en-
vironment, by simply changing some of the modeling parameters. A table of comparisons
is presented which shows the generality of the Eccentro-Scattering model, proposed in the
previous chapter.
55
3.2 Angle Of Arrival For Picocell And Microcell Environments
3.2 Angle Of Arrival For Picocell And Microcell Envi-
ronments
In picocell and microcell environments, both MS and BS are considered to have local scatter-
ers in their vicinity [41, 44] as shown in Fig. 2.6. The scatterers are confined in an Eccentro-
Scattering disc whose eccentricity can be changed according to the distance between BS and
MS and the maximum delay. Table 2.1 provides the corresponding choice of parameters.
Using radio signal propagation theory, the relationship among maximum delay τmax , total
path travelled rBS + rM S , and speed of light c can be written as,
rBS + rM S = cτmax (3.1)
The propagated radio signal experiences shorter delays in picocell and microcell environments
as compared to macrocell environment. Criteria for selecting τmax can be found in [24]. From
(2.5) and (3.1), the semi-major axis of the ellipse is,
cτmax
a= (3.2)
2
Here, the distance between BS and MS, d, is equal to 2f , so using (2.6) and (2.7), the semi-
minor axis b and eccentricity e of the Eccentro-Scattering discs in Fig. 3.1(a) and Fig. 3.2(a)
can be written, respectively, as,

1√ 2
b= 4a − d2 (3.3)
2
d
e= (3.4)
2a
56
S
rBS
rMS
θ
Q 2b
B C M
d
2a
(a) microcells and picocells
P V
Y
Distant Scattering Disc
Q CD
rPM
rBP U rMCD
rBCD Local Scattering Disc
S
θ J
θD rBS
rSM
θ X
B H
θL,max G M 2b
2a
d
(b) macrocells
Figure 3.1: Eccentro-Scattering model in uniform scattering
57
The area bounded by the sector BSQ in Fig. 3.1(a) and Fig. 3.2(a) is a function of the
angle θ, with angles between 0◦ and θmax . For picocells, the maximum AoA, θmax is 180◦ [53],
which means multipath signals are arriving at BS antenna from all directions, whereas in
microcells, it was found that most of the multipath signals have maximum AoA, θmax of
60◦ [53]. The distance between BS and MS, d, is larger in the case of microcell environment
as compared to that in the case of picocell environment.
3.2.1 Uniformly Distributed Scatterers
Fig. 3.1(a) represents the Eccentro-Scattering model for uniformly distributed scatterers in
picocells and microcells. Considering (2.2) and the geometry in Fig. 3.1(a), the Cumulative
Distribution Function (CDF) of the scattering points around BS and MS would be,
Z θ Z z2 Z θ
rBS z22 (ζ)
PΘ (θ) = drBS dζ = dζ (3.5)
−θmax 0 Ae −θmax 2Ae
where Ae is the area of the ellipse, ζ is a dummy variable used for the AoA and z2 is the
positive root of the equation defining the ellipse of Fig. 3.1(a) in polar coordinates,
Ã ! Ã !
2 cos2 θ sin2 θ d cos θ d2
rBS + 2 − rBS + −1=0 (3.6)
a2 b a2 4a2
Solving for rBS results in the following two solutions,
4a2 − d2
z1 (θ), z2 (θ) = (3.7)
2(∓2a − d cos θ)
58
So the pdf of AoA of the multipaths from all scattering points within the Eccentro-Scattering
disc, pΘ (θ), would be the derivative of (3.5) with respect to θ,
d z 2 (θ)
pΘ (θ) = PΘ (θ) = 2 (3.8)
dθ 2Ae
πa
√
Since Ae = 2
4a2 − d2 , therefore,
(4a2 − d2 )3/2
pΘ (θ) = (3.9)
4πa(2a − d cos θ)2
Introducing eccentricity into (3.9), we get,
(1 − e2 )3/2
pΘ (θ) = (3.10)
2π(1 − e cos θ)2
It is evident from (3.10) that the pdf of AoA of the multipaths from all scattering points
within the Eccentro-Scattering disc depends mainly on its eccentricity. In other words, the
pdf of AoA of the multipaths at BS in picocell and microcell environments is primarily
determined by the ratio of BS-MS distance to the major axis of the Eccentro-Scattering
disc.
3.2.2 Gaussian Distributed Scatterers
Practically speaking, in picocells, the density of scatterers around MS is almost equal to that
around BS due to having the same scattering environment around their antennas, whereas
in microcells, the density of scatterers around MS is greater than that around BS.
Previously, researchers [2, 27] derived the pdf formulas assuming unbounded Gaussian
scattering points centered at MS. Even though this assumption simplifies the derivation, it
59
S
rBS
rMS
L θ Q
2b
N B C M
d
2a
(a) microcells and picocells
P V
Y Distant Scattering Disc
Q CD
rPM
rBP U rMCD Local Scattering Disc
rBCD
S
θ rBS J
θD rSM
θ X
B
θL,max G M H
2b
d 2a
(b) macrocells
Figure 3.2: Eccentro-Scattering model in Gaussian scattering
60
is impractical because the maximum delay τmax specifies nearly all the power and AoA of
the multipath signals within some region. Hence, multipaths with longer delays experience
greater path loss and, therefore, have relatively low power compared to those with shorter
delays [16, 24]. It is more realistic to confine the scatterers inside a scattering disc as proposed
in Eccentro-Scattering model. Since the scattering points local to BS have a significant effect
on the AoA statistics, it is more viable to use the JGSM discussed earlier. That is why we
are considering two separate Gaussian functions for the distribution of scatterers around BS
and MS.
Considering (2.4) and the geometry in Fig. 3.2(a), the density of the scattering points
around MS can be described by the bivariate Gaussian distribution,

 µ 2 2
¶
 rBS exp − rBS + d − 2rBS d cos θ , r > 0
(M S) 2 2 BS

0, elsewhere
where θ is the AoA at BS measured from the horizontal.
In a similar way, considering (2.9) and the geometry in Fig. 3.2(a), the density of the
scattering points around BS can be found as,

 ³ 2 ´
 rBS exp − rBS , r > 0
(BS) 2 2 BS
pRBS ,Θ (rBS , θ) = 2πσBS 2σBS (3.12)

0, elsewhere
Considering (3.11) and (3.12) and the geometry in Fig. 3.2(a), the CDF of the scattering
points around MS and BS would be, respectively,
61
Z θ Z z2 µ 2
¶
(M S) rBS −rBS − d2 + 2rBS d cos ζ
PΘ (θ) = 2
exp 2
drBS dζ (3.13)
−θmax 0 2πσM S 2σM S
and
Z θ Z z2 µ 2
¶
(BS) rBS −rBS
PΘ (θ) = 2
exp 2
drBS dζ (3.14)
−θmax 0 2πσBS 2σBS
where z2 is the positive root of the equation defining the ellipse of Fig. 3.2(a) and is defined
in (3.7). Since the received signal at the antenna has interacted with only one single scatterer
in the channel, as assumed earlier, then the AoA of the multipaths from scatterers around BS
and MS are two disjoint events. Hence, the pdf of AoA of the multipaths from all scattering
points within the Eccentro-Scattering disc, pΘ (θ), would be basically the addition of the
derivatives of (3.13) and (3.14) with respect to θ,
1 ³ (M S) (BS)
´ 1³ d
(M S) d (BS) ´
pΘ (θ) = pΘ (θ) + pΘ (θ) = PΘ (θ) + PΘ (θ) (3.15)
2 2 dθ dθ
therefore,
Z z2 µ 2
¶ Z z2 µ 2 ¶
rBS −rBS − d2 + 2rBS d cos θ rBS −rBS
pΘ (θ) = 2
exp 2
drBS + 2
exp 2
drBS
0 2πσM S 2σM S 0 2πσBS 2σBS
(3.16)
Substituting the values of z2 from (3.7) into (3.16), we get the pdf of AoA of the
multipaths at BS from Gaussian distributed scatterers around BS and MS confined within
62
0.025
eq(26) with a=0.5σMS
eq(26) with a=σMS
eq(26) with a=2σ
MS
eq(26) with a=4σ
MS
0.02 Janaswamy
0.015
pdf
0.01
0.005
0
−150 −100 −50 0 50 100 150
Angle of Arrival, (Degrees)
Figure 3.3: Effect of increasing a with respect to σM S on the pdf of AoA in picocells and microcells,
σBS =0.
the Eccentro-Scattering disc as,

" µ ¶ µ ¶
Ω d2 (4a2 − 4ad cos θ + d2 )2
pΘ (θ) = 1 + exp − 2 − exp − 2 2
4π 2σM S 8σM S (2a − d cos θ)
√ µ ¶( µ ¶ µ 2 ¶)
πd cos θ d2 sin2 θ d cos θ 4a − 4ad cos θ + d2 cos 2θ
+ √ exp − 2
erf √ + erf √
2σM S 2σM S 2σM S 2 2σM S (2a − d cos θ)
µ ¶#
(4a2 − d2 )2
− exp − 2 (3.17)
8σBS (2a − d cos θ)2
R 2π
where Ω is the normalizing constant such that 0
pΘ (θ)dθ = 1, and erf(x) is the well known
Rx
error function defined as: erf(x) = 0
exp(−t2 )dt.
The following remarks can be made about (3.17):
63
1. If we consider σM S = ∞ and σBS = 0, (3.17) approaches a uniform distribution for the
AoA of the multipaths at BS.
2. If we consider a = 4σM S and σBS = 0, (3.17) approaches the result found by Janaswamy
[2] for unbounded Gaussian distributed scatterers around MS only. This result implies
because, at a = 4σM S , the Eccentro-Scattering disc would confine almost all (99.99%)
of the scattering points. Whereas, practically, a must be smaller than 4σM S . In
Fig. 3.3, we observe that when a is increased, the pdf of AoA curve tends to overlap
the corresponding curve obtained in case of unbounded scatterers [2]. At a = 4σM S ,
the exact overlap occurs, and the effect of increasing a on the pdf of AoA curve stops
for all values of a greater than 4σM S . In other words, we can say that the pdf of AoA
of the multipaths at BS depends on the value of a such that a ≤ 4σM S . Therefore it is
more realistic to bound the scatterers inside some scattering disc according to terrain
conditions, i.e. in an Eccentro-Scattering disc.
3. If we consider σM S ≥ 3a and σBS = 0, (3.17) approaches the results found originally
by Liberti [24] and derived again in a compact form by Ertel [3] for bounded uniformly
distributed scatterers confined in an elliptical scattering disc. Fig. 3.4 shows the effect
of using large values for σM S on the pdf of AoA of the multipaths at BS. The pdf
of AoA curves are plotted for different values of σM S . For the sake of convenience,
the values of σM S are given in terms of a. We observe that as we increase σM S , the
pdf of AoA curve tends to overlap the curve obtained in case of uniformly distributed
64
0.025
eq(26) with σMS=a
eq(26) with σMS=2a
eq(26) with σMS=10a
Ertel
0.02
0.015
pdf
0.01
0.005
0
−150 −100 −50 0 50 100 150
Figure 3.4: Effect of increasing σM S with respect to a on the pdf of AoA in picocells and microcells,
σBS =0.
scatterers bounded in an elliptical disc [3]. We thus conclude that Gaussian distributed
scatterers confined in an Eccentro-Scattering disc with large values of σM S show the
same results as uniformly distributed scatterers confined within the disc.
These results illustrate the generality of the Eccentro-Scattering model.
In picocell or indoor environments, the effect of standard deviation of the scatterers
is not so visible because BS and MS are usually located very close to each other, while in
microcell environment this effect is more obvious because BS and MS are relatively far from
each other and the scattering disc gets larger values of eccentricity, e (closer to 1).
65
0
10
°
Laplacian Function (σ = 25.5 )
l
Eccentro−Scattering Model
−1 Indoor Measurements
10
−2
10
No. of Occurrences
−3
10
−4
10
−5
10
−6
10
−180 −120 −60 0 60 120 180
Figure 3.5: Comparison of the pdf in AoA for the Eccentro-Scattering model, and Laplacian func-
tion with measurements [1].
In the indoor measurement campaigns [1, 45–47], the formation of tails in the pdf of AoA
curves is actually due to the scattering effect of the radio waves by the scatterers closer to BS.
This phenomenon strengthens the assumption of additional Gaussian distributed scatterers
around BS. Fig. 3.5 shows the comparison of the Eccentro-Scattering model with the field
measurements obtained by Spencer [1]. The pdf of AoA curves using the Eccentro-Scattering
model and Laplacian function (suggested by [1]) are plotted against measurement data on
logarithmic scale. Eccentro-Scattering model shows the best fit for the data, especially for
the tails of the distribution. The tails of the other measurements can also be dealt with
using the same approach.
66
3.3 Angle Of Arrival For Macrocell Environment
Usually in picocell environment, the curve of the azimuthal distribution of AoA at BS is
same as that at MS in shape. However, this phenomenon slightly differs in microcell urban
environments, where streets behave as wave-guides giving different shape to the azimuthal
distribution at MS. Anyhow, the azimuthal distribution of AoA at BS retains its common
shape, i.e. sharp rise on both sides of the mean AoA along LOS and uniform tails on the
rest of the domain. In [48], 3-D high-resolution AoA distribution plots have been presented
for different urban scenarios which reveal the formation of tails in their azimuthal domain
in most of the cases depending on the orientation of the streets. The Eccentro-Scattering
model provides the best modeling approach to represent widths and lengths of the streets in
microcells with the help of corresponding change in the eccentricity e.
In this section, we derive the pdf of AoA of the multipaths at BS for macrocell environments.
As mentioned earlier, BS is at a large distance from MS in macrocell environments and no
scattering is assumed in its vicinity. Further, we consider distant scattering discs in hilly
and suburban areas to represent large scattering structures such as mountains and high rise
building clusters.
Fig. 3.1(b) and Fig. 3.2(b) are modified representations of the Eccentro-Scattering model
for macrocell environment, in uniformly and Gaussian distributed scattering regions respec-
67
Local Scattering Disc θL,max BS
θD
MS
θD,max
Dominant Distant Scatterers
Figure 3.6: Typical macrocell environment
tively. Consider the ellipses centered at M (MS) with semi-major and semi-minor axes a
and b, respectively, as given in Fig. 3.1(b) and Fig. 3.2(b). d is the distance between B (BS)
and M (MS) and S is a scattering point in the vicinity of MS. θ is the angle between the
position vector of any scattering point and x-axis, i.e. the AoA of the multipaths at BS,
taking values between −θL,max and θL,max , where θL,max is the maximum AoA at BS in local
Eccentro-Scattering case defined as,
µ µ ³ a ´¶¶
θL,max = arctan κ tan arcsin (3.18)
d
where κ (the aspect ratio of the local Eccentro-Scattering disc) = b/a.
In macrocells, the eccentricity of the local Eccentro-Scattering disc does not depend on
d, because BS is located outside the scattering region, however it depends on the terrain and
clutter of the scattering environment.
Likewise for the distant Eccentro-Scattering disc, aD , and bD are the semi-major and
68
semi-minor axes of the distant Eccentro-Scattering disc, P is a scattering point in that disc,
D is the total distance between BS and MS via CD , the center of the distant scattering
disc, i.e., D = krBCD k + krM CD k where krBCD k and krM CD k are shown in Fig. 3.1(b) and
Fig. 3.2(b). dD is the distance between BS and CD , i.e., dD = krBCD k, θD is the angle
of krBCD k with x-axis. We assume the major axis of the distant Eccentro-Scattering disc
is parallel to rBCD . This assumption is just for the sake of making the modeling easy,
otherwise any orientation of the distant Eccentro-Scattering disc can be taken according to
the physical location of distant scattering structures. For the distant scattering points, θ
takes values between θD − θD,max and θD + θD,max , where θD,max is the maximum AoA for
distant Eccentro-Scattering case defined as,
µ µ ³ a ´¶¶
D
θD,max = arctan κD tan arcsin (3.19)
dD
where κD (the aspect ratio of the distant Eccentro-Scattering disc)= bD /aD . From the
geometry of Fig. 3.1(b) and Fig. 3.2(b), the angular spread, 2θmax , is defined as,
2θmax = θL,max + θD,max + θD (3.20)
In the case of using two directional antennas to illuminate both local and distant scat-
terers, 2θL,max and 2θD,max would be treated as separate angular spreads each for individual
antenna as shown in Fig. 3.6.
In Fig. 3.1(b) and Fig. 3.2(b), where a dominant distant Eccentro-Scattering disc exists
in addition to the local Eccentro-Scattering disc, the area bounded by the sector GJSH in
69
the local Eccentro-Scattering disc is a function of the angle θ, with angles between 0 and
θL,max , and the area bounded by the sector U QP V in the distant Eccentro-Scattering disc
is a function of the angle θ, with angles between θD and θD + θD,max . Table 2.2 specifies the
choice of the Eccentro-Scattering model parameters for macrocell environment in detail.
3.3.1 Uniformly Distributed Scatterers
Considering (2.2) and the geometry in Fig. 3.1(b), the CDFs of the scattering points for the
local and distant Eccentro-Scattering discs will be, respectively,

Z θ Z z2 Z θ
(L) rBS z22 (ζ) − z12 (ζ)
PΘ (θ) = drBS dζ = dζ (3.21)
−θL,max z1 AL −θL,max 2AL
and
Z θD +θ Z w2 Z θD +θ
(D) rBP w22 (ζ) − w12 (ζ)
PΘ (θ) = drBP dζ = dζ (3.22)
θD −θD,max w1 AD θD −θD,max 2AD
where AL = πab, AD = πaD bD , and z1 , z2 and w1 , w2 are two pairs of roots for the equations
defining the local and distant Eccentro-Scattering discs, respectively, in polar coordinates,
Ã ! Ã !
2 2
2 cos θ sin θ 2d cos θ d2
rBS + − r BS + −1=0 (3.23)
a2 b2 a2 a2
and
Ã ! Ã ! Ã !
2 cos2 θ sin2 θ cos θ cos θD sin θ sin θD cos 2
θD sin 2
θD
rBP + 2 −2dD rBP + +d2D + 2 −1 = 0
a2D bD a2D b2D a2D bD
(3.24)
with roots,
p
b2 d cos θ ∓ a2 sin2 θ + b2 cos2 θ − d2 sin2 θ
z1 , z 2 = (3.25)
a2 sin2 θ + b2 cos2 θ
70
and
¡ ¢ p
dD a2D sin θ sin θD + b2D cos θ cos θD ∓ aD bD a2D sin2 θ + b2D cos2 θ − d2D sin2 (θ − θD )
w 1 , w2 =
a2D sin2 θ + b2D cos2 θ
(3.26)
The above-mentioned quantities are z1 = |BJ|, z2 = |BS|, w1 = |BQ| and w2 = |BP | as
shown in Fig. 3.1(b).
Substituting the values of z1 , z2 and w1 , w2 from (3.25) and (3.26) into (3.21) and (3.22)
and differentiating with respect to θ, we get the pdf of AoA of the multipaths at BS from
the scatterers confined in the local and distant Eccentro-Scattering discs as follows,
 p
2

 2κ E cos θ ¡ sin2 θ + κ2 cos2 θ¢− E 2 sin2 θ , −θL,max ≤ θ ≤ θL,max

 2

 π sin2 θ + κ2 cos2 θ









 0, θL,max < θ < θD − θD,max



pΘ (θ) = ¡ 2
¢q 2


 2ED sin θ sin θD + κD cos θ cos θD sin θ + κ2D cos2 θ − ED
2
sin2 (θ − θD )

 ¡ ¢2 ,

 π sin 2
θ + κ2
cos2
θ

 D

 θD − θD,max ≤ θ ≤ θD + θD,max








0, elsewhere
(3.27)
where E = d/a and ED = dD /aD are important ratios that help in designing directional an-
tennas for local and dominant distant scattering environments especially in urban, suburban,
rural, and hilly areas.
71
3.3.2 Gaussian Distributed Scatterers
Once again we say that Gaussian distributed scatterers represent a better assumption to the
real situations in macrocell environments. Here also, we confine the Gaussian distributed
scatterers, both local and distant, inside the Eccentro-Scattering discs since it is more prac-
tical as discussed earlier.
Considering (2.4) and the geometry in Fig. 3.2(b), the CDFs of the scattering points
for the local and distant Eccentro-Scattering discs will be, respectively,
Z θ Z z2 µ 2
¶
(L) rBS −rBS − d2 + 2rBS d cos ζ
PΘ (θ) = 2
exp 2
drBS dζ (3.28)
−θL,max z1 2πσM S 2σM S
and
Z θD +θ Z w2 µ 2
¶
(D) rBP −rBP − d2D + 2rBP dD cos(ζ − θD )
PΘ (θ) = 2
exp 2
drBP dζ (3.29)
θD −θD,max w1 2πσD 2σD
where z1 , z2 and w1 , w2 are as defined in (3.25) and (3.26).
To simplify the derivation, we assume a circular disc for the distant scatterers, i.e.
κD = 1, so, substituting aD and bD by RD in (3.24) and solving for rBP , results in the
following two solutions,

q
w1 , w2 = dD cos(θ − θD ) ∓ 2
RD − d2D sin2 (θ − θD ) (3.30)
Substituting the values of z1 , z2 from (3.25) and w1 , w2 from (3.30) into (3.28) and
(3.29) and differentiating with respect to θ, we get the pdf of AoA of the multipaths at BS
72
from the scatterers confined in the local and distant Eccentro-Scattering discs as follows,
pΘ (θ) =
 ³ 2 2 ´· ³ ´ ¸

 −κ f1 + 2κf1 f2 − E 2 sin2 θ(sin2 θ + κ4 cos2 θ) −4κf1 f2

 K1 exp exp −1

 f32· ¸f32

 ³ ´ ³ ´ ³ ´

 a2 E 2 sin2 θ erf κf1 + f2 + erf κf1 − f2 ,

 +K 2 E cos θ exp −

 2σM 2
S
f3 f3



 −θL,max ≤ θ ≤ θL,max







0, θL,max < θ < θD − θD,max



 q



 ³ 2 2 ´ ³ RD 1 − ED 2
sin2 (θ − θD ) ´

 RD ED√cos(θ − θD ) RD ED sin2 (θ − θD ) √

 Ω exp 2 erf ,

 2πσD 2σD 2σD









 0, elsewhere
(3.31)
p
where K1 = Ω
2π
and K2 = Ωa√
2σM S 2π
are constants, f1 = sin2 θ + κ2 cos2 θ − E 2 sin2 θ, f2 =
√
2σM S
¡ ¢
E(1 − κ2 ) cos θ sin2 θ and f3 = a
sin2 θ + κ2 cos2 θ are some functions of θ, erf(x) is
Rx dD
the well known error function defined as erf(x) = 0
exp(−t2 )dt, ED = RD
and Ω is the
R 2π
normalizing constant such that 0
pΘ (θ)dθ = 1.
In (3.31), the semi-major axis for the local Eccentro-Scattering disc a and the radius of
the distant scattering disc RD do not change with distances d and dD and remain constant
as far as the terrain and clutter do not change in the scattering regions, so the pdf of AoA
of the multipaths at BS in macrocells depends on:
1. The eccentricity of the local Eccentro-Scattering disc (and the eccentricity of the distant
Eccentro-Scattering disc if circular disc is not assumed, i.e. eD 6= 0),
73
Distant Scattering
× Gaussian, σMS = 0.5a, Simulated

Gaussian, σMS = 0.5a, Theoretical
• Uniform, Simulated
Local Scattering Uniform, Theoretical
Figure 3.7: Simulated and theoretical pdf of AoA for suburban macrocell with d = 2000m, a =
300m, D = 5000m, aD = 150m, and θD = 15◦ .
2. The ratios E and ED ,
3. The standard deviations of the scattering points around MS and of those around the
center of the distant disc.
In (3.27) and (3.31), θD plays an important role in describing the total angle spread
when both local and distant scatterers are taken into account. θD usually remains constant
and indicates the deviation of the center of the dominant distant scattering cluster from the
center of the mobile service area. If θD is large as compared to the values of θL,max and
θD,max , then the use of two separate directional antennas is a valid recommendation when a
fixed beam antenna system is used at BS, see Fig. 3.6. However a smart antenna can handle
multiple beams, therefore the use of two directional antennas may not be needed. Thus,
74
(3.27) and (3.31) are helpful in designing antenna systems in flat rural, suburban, urban,
and hilly areas. If θD − θD,max is less than θL,max then the pdf of AoA would be the addition
(L) (D)
of pΘ (θ) and pΘ (θ) for the overlapping region. In Fig. 3.7, simulation and theoretical
results, using (3.27) and (3.31), are presented for typical macrocell environments both for
uniform and Gaussian distributions. Keeping the assumptions itemized in section 2.2 of
the previous chapter in view, we use re-radiating scattering structures in our simulations.
We exploit mainly reflection and diffraction phenomena over 2 GHz frequency, i.e. 15 cm
wavelength. At this wavelength all scattering structures mostly buildings, trees and buses
look like big opaque or quasi-opaque spheres which reflect and diffract radio waves at 2 GHz.
Normalized histograms of the number of scattering occurrences are sketched in the plots using
point markers against their respective angles.
The following additional remarks can be made about (3.27) and (3.31):
1. The work in [27] derived the pdf of AoA of the multipath signals at BS considering
scattering points that are Gaussian distributed around MS only, and within the angular
beamwidth of a directional antenna at BS. But no geometrical shape of the scattering
disc was defined, which is an impractical assumption as discussed earlier. The only
effect of using a directional antenna at BS is to reject the AoAs falling outside the
beamwidth, while it does not alter the distribution of AoA at BS. However, by substi-
tuting θL,max by α, the half beamwidth of the directional antenna, used at BS in (3.28)
we get results obtained in [27].
75
2. If we consider σM S ≥ 3a and σD = 0, (3.31) approaches the results found by Piechocki,
Tsoulos and McGeehan [28] for bounded uniformly distributed scatterers confined in
an elliptical scattering disc. Also if we exclude distant scattering clusters, (3.27) gives
the same results found in [28]
3. The work in [3] for the uniformly distributed local scatterers confined in circular disc
in macrocells can be derived either by putting a = b = R in (3.27) or by substituting
σM S ≥ 3a, σD = 0, and a = b = R in (3.31).
4. In macrocell environments where BS antennas are not at higher altitudes, e.g. in urban
areas as compared to rural areas, the scatterers in the vicinity of BS have significant
impact on the pdf of AoA. Then using JGSM approach, the pdf of AoA of the multipath
signals at BS is,
µ ¶
1 (BS) (M S) (D)
pΘ (θ) = p (θ) + pΘ (θ) + pΘ (θ) (3.32)
3 Θ
(M S) (D)
where the sum pΘ (θ) + pΘ (θ) is given in (3.31) while,
( µ ¶)
(BS) 1 a2BS b2BS
PΘ (θ) = 1 − exp (3.33)
2π 2(a2BS sin2 θ + b2BS cos2 θ)σBS
In (3.33), aBS and bBS are the semi-major and semi-minor axes of the Eccentro-
Scattering disc around BS.
These results illustrate furthermore, the generality of the Eccentro-Scattering model.
As mentioned earlier, the maximum delay of the multipaths (τmax ) has a significant
effect on the pdf of AoA at BS. The pdf of AoA derived using the Eccentro-Scattering model
76
depends on the size of the Eccentro-Scattering disc, which in turn depends on τmax , whereas
this is not the case for the corresponding pdf derived using unbounded scattering [2, 16, 27].
It was mentioned in [2] that, in the Gaussian AoA model, the distribution of arriving waves
in azimuth is assumed to be Gaussian without specific mention of the scatter density required
to produce it. Nevertheless it was stated earlier in [54] that, if the Gaussian model for the
distribution of scatterers round the mobile is assumed then the angular distribution seen
from BS is also Gaussian. In fact, if we consider unbounded scatterers or multipaths from
infinitely far distant scatterers, we get an exactly bell-shaped Gaussian pdf of AoA of the
multipaths at BS. However if we consider Gaussian distributed scatterers confined in some
disc i.e. the Eccentro-Scattering disc, then the pdf of AoA would be Gaussian distributed
with its feet depending on the size of the disc, the distance between BS and MS, and the
standard deviation of the scattering points around MS and BS under the zooming effect, see
Fig. 3.3 and Table 3.1.
An interesting general result in picocell, microcell, and macrocell environments, is the
similarity of the plots of the pdf of AoA for uniform and Gaussian scattering when the
standard deviation of the Gaussian distributed scatterers around MS is greater than twice
the semi-major axis of the respective Eccentro-Scattering discs. In other words, assuming
scattering to be either uniformly or Gaussian distributed for sparsely populated areas gives
almost the same distribution of AoA of multipaths at BS, see Fig. 3.4.
77
Table 3.1: Summary of the Spatial Channel Models
Respective Used Corresponding Most common shape

Environment Models Scatter Eccentro-Scattering of the plot of
Distribution Model substitution AoA distribution
Macrocell Rural Petrus [26], Uniform a = b = R in (3.27)
Ertel [3], and OR
Piechocki [28] σM S = 3a, σBS = σD = 0,
and a = b = R in (3.31)
Urban/ Petrus [26], Uniform a = b = R in (3.27)

Suburban Ertel [3], and OR
and a = b = R in (3.31)
Janaswamy [2] Gaussian a = 4σM S , σBS = σD = 0
and Lotter [27] in (3.31)
Hilly Petrus [26], Uniform a = b = R in (3.27)
Ertel [3], and OR
and a = b = R in (3.31)
d
Microcell Liberti [24] Uniform e= 2a in (3.10)
and Ertel [3] OR
σM S = 3a, σBS = 0 in (3.17)
Janaswamy [2] Gaussian a = 4σM S , σBS = 0
d
Picocell Liberti [24] Uniform e = 2a in (3.10)
and Ertel [3] OR
σM S = 3a, σBS = 0 in (3.17)
Janaswamy [2] Gaussian a = 4σM S , σBS = 0
Effect of Directional Lotter [27] Gaussian z1 = 0, z2 = ∞, and,
Antenna at BS θL,max = α in (3.28),
where α = half of
BW
the beamwidth of
directional antenna
78
3.4 Modeling the Impact of Scattering around BS on the AoA Statistics
Table 3.1 summarizes the results and comparisons of the Eccentro-Scattering model with
previous physical channel models. Also in Table 3.1, the most common shapes of the pdf of
AoA of the multipaths at BS are given for all cellular environments in Gaussian distributed
scattering, which depend on σM S , σBS , e, and E of the Eccentro-Scattering discs. The work
in [2, 3, 16, 24, 27, 28] can easily be extracted from Table 3.1.
3.4 Modeling the Impact of Scattering around BS on
the AoA Statistics
In this section, we address the issue of the impact of heavy scattering around BS on the AoA
statistics of multipath signals at BS. We utilize JGSM proposed in chapter 2, along with a
scattering-free region, which makes it feasible to model the effect of scattering around BS
on the azimuthal distribution. This combination successfully answers the formation of pdf
tails on both sides of the mean angle of arrival in the measurements performed by Pedersen
et al. [4] for an urban macrocell environment. JGSM along with the provision of scattering-
free region, provides simpler modeling as compared to the Gaussian Macrocell Eccentro-
Scattering model in section 3.3.2, where Gaussian distributed scatterers are confined in an
elliptical or circular disc. In Gaussian Macrocell Eccentro-Scattering model, the eccentricity
of the Eccentro-Scattering disc is adjusted to provide appropriate modeling for a specific
cellular environment, which makes it more complicated than the combination of JGSM and
79
y S
rMS
Q
rBS
M
d
2α
Rnull β
θ
B x
Figure 3.8: The proposed scattering model
scattering-free zone.
3.4.1 Low Antenna-Height Urban Macrocell Model
Fig. 3.8 depicts the proposed scattering model where BS and MS are surrounded by Gaussian
distributed scatterers represented by two separate Gaussian functions.
Let Rnull be the radius of circular shaped scattering-free zone around BS as shown in
Fig. 3.8. The value of Rnull depends on the radius of the cell, height of BS antenna, clustering
of the structures around BS, and the terrain of BS location.
A Gaussian model of the spatial pdf of the scatterers around MS and BS is written in
(2.4) and (2.9), respectively.
80
The area bounded by the sector BSQ in Fig. 3.8 is a function of the angle θ, with angles
between β and θmax , where θmax is the maximum angle of arrival and gives the measure
for the maximum angular spread. For indoor picocells, multipath signals arrive at BS from
all directions, i.e. θ ∈ {−180◦ , 180◦ }. For outdoor microcells, a directional antenna of half
beamwidth α = 60◦ is usually used keeping the maximum angular spread as 120◦ . Whereas
in outdoor macrocells, the maximum AoA, θmax , and the minimum AoA, θmin , depend on
the standard deviation of the distribution of the scatterers around MS, the distance between
BS and MS and the angle β, as,
³ 4σ ´ d
MS
θmax,min ≈ β ± arcsin , for σM S ≤ (3.34)
d 4
Considering (2.4) and the geometry in Fig. 3.8, the density of the scattering points
around MS can be described by the bivariate Gaussian distribution,
 ³ 2 2 ´
 rBS exp − rBS + d − 2rBS d cos(θ − β) , for r > 0
(M S) 2 2 BS

0, elsewhere
In a similar way, considering (2.9) and the geometry in Fig. 3.8, the density of the
scattering points around BS can be found as,

 ³ 2 ´
 rBS exp − rBS , for r > 0
(BS) 2 2 BS
pRBS ,Θ (rBS , θ) = 2πσBS 2σM BS (3.36)

0, elsewhere
From (3.35), (3.36) and Fig. 3.8, the Cumulative Distribution Function (CDF) of the
scattering points around MS and BS would be, respectively,
81
Z θ Z ∞ µ 2
¶
(M S) rBS −rBS − d2 + 2rBS d cos(ζ − β)
PΘ (θ) = 2
exp 2
drBS dζ (3.37)
θmin 0 2πσM S 2σM S
and
Z θ Z ∞ µ 2
¶
(BS) rBS −rBS
PΘ (θ) = 2
exp 2
drBS dζ (3.38)
θmin Rnull 2πσBS 2σBS
We assume that the received signal at BS antenna interacts with one single scatterer in
the last reflection/scattering, therefore the AoA of the multipaths from scatterers around BS
and MS are two disjoint events. Hence, the pdf of AoA of the multipaths from all scattering
points, pΘ(θ) , will be basically the addition of the derivatives of (3.37) and (3.38) with respect
to θ, i.e.,
1 ³ (M S) ´ 1µd d (BS)
¶
(BS) (M S)
pΘ (θ) = p (θ) + pΘ (θ) = P (θ) + PΘ (θ) (3.39)
2 Θ 2 dθ Θ dθ
therefore,
Z ∞ µ 2
¶
1 rBS −rBS − d2 + 2rBS d cos(θ − β)
pΘ (θ) = 2
exp 2
drBS
2 0 2πσM S 2σM S
Z ∞ µ 2
¶
1 rBS −rBS
+ 2
exp 2
drBS (3.40)
2 Rnull 2πσBS 2σBS
Simplifying (3.40), we get the pdf of AoA of the multipaths at BS from the Gaussian dis-
tributed scatterers around BS and MS as,

(r µ ¶ µ ¶
Ω π d d2 sin2 (θ − β) d cos(θ − β)
pΘ (θ) = cos(θ − β) exp − 2
erfc − √ 2
4π 2 σM S 2σM S 2σM S
µ ¶ µ ¶ )
d2 R2
+ exp − 2 + exp − null2
(3.41)
2σM S 2σBS
82
R 2π
where Ω is normalizing constant such that 0
pΘ (θ)dθ = 1, and erfc(x) is the complementary
R∞
error function defined as: erfc(x) = √2 exp(−t2 )dt.
π x
3.4.2 Results and Discussion
In outdoor environments, the majority of scattering points are clustered together with their
density decreases as the distance from MS increases. This fact suggests a Gaussian model
for the distribution of scatterers in outdoor environments. Since the distribution of AoA in
outdoor environments strictly depends on the separation between BS and MS and the nature
of scattering structures; so, we use different values for Rnull in our proposed scattering model
to represent rural, suburban/bad urban, and urban environments as listed in Table 3.2.
Equation (3.41) represents the general closed-form expression for the pdf of AoA of the
multipaths at BS using the proposed scattering model. The following remarks can be made
regarding this formula:
1. We observe that the second and third terms in equation (3.41), which do not depend
on the AoA, θ, reveal the fact that the tails of the pdf of the AoA are usually uniform,
and their height can be easily controlled by the appropriate selection of Rnull .
2. If we consider σM S = ∞ and σBS = 0, pdf in (3.41) approaches a uniform distribution
for the AoA of the multipaths at BS. This is a common assumption made usually
to study the Doppler characteristics of the channel and is based on the fact that
83
Measurement Data [Pedersen]

Proposed JGSM with using Rnull
300 GSDM [Janaswamy]
USM(Circular) [Ertel,Petrus]
USM(Uniform Eccentro−Scattering)
250
200
Rate of Occurance
150
100
50
0
−25 −20 −15 −10 −5 0 5 10 15 20 25
Angle of Arrival, Degrees
(a) Linear Scale
Measurement Data [Pedersen]

Proposed JGSM with using Rnull
GSDM [Janaswamy]
10
2 USM(Circular) [Ertel,Petrus]
USM(Uniform Eccentro−Scattering)
1
10
Rate of Occurance
0
10
−1
10
−2
10
−25 −20 −15 −10
Angle of Arrival, Degrees
(b) Logarithmic Scale
Figure 3.9: pdf of AoA at BS in urban macrocell mobile environments
84
multipaths are arriving from all directions with equal probabilities.
3. If we consider either σBS = 0 or Rnull > 4σBS , pdf in (3.41) approaches the result
found by Janaswamy [2] for Gaussian distributed scatterers located around MS only.
4. The work in [27] derived the pdf of AoA of the multipaths at BS considering scatter-
ing points that are Gaussian distributed and situated around MS within the angular
beamwidth, 2α, of a directional antenna at BS. The only effect of using directional an-
tenna at BS is the rejection of the AoAs falling outside the beamwidth of the antenna,
with no alteration of the distribution of AoA at BS. However, by substituting θmax by
α in (3.41), we get the same results obtained in [27].
5. In [2, 4], a Gaussian function with standard deviation of 6◦ was suggested to fit field
measurements performed in Aarhus, Denmark, which provided a good match except
for the tails of the histogram. It was observed that the tails were actually heavier
than predicted by Gaussian distribution. In fact, these tails are formed due to the
multipath signals reflected by the scatterers situated around BS. The height of these
tails depends on the standard deviation of the Gaussian distributed scatterers around
BS, σBS , and the value of Rnull . A lower value of Rnull produces heavier tails and vice
versa. Equation (3.41) explains the formation of the tails, which is due to the AoA
bouncing from the scatterers in the vicinity of BS.
In Fig. 3.9(a) and 3.9(b), plots for the rate of occurrence of AoA of the multipaths at BS
have been presented using Uniform Scattering Model [3, 28], the Gaussian Scatter Density
85
3.5 Conclusion
Table 3.2: Values of Rnull for different macrocell environments
Environment Proposed Range of Rnull
Outdoor Macrocellular Rural Area Rnull > 4σBS
Sub Urban/Bad Urban 3 σBS < Rnull ≤ 4 σBS
Urban 1 σBS < Rnull ≤ 3 σBS
Model (GSDM) [2] and the proposed scattering model besides measurement data [4]. The
first conclusion we can make from these plots is that Gaussian assumption for the distribution
of scatterers in outdoor environment is more appropriate as compared to uniform assumption.
Fig. 3.9(b), enlightens the tails of the distribution of AoA, on logarithmic scale. Here, we
can observe the effect of scattering around BS, which is successfully overcome by adjusting
the value of Rnull in the proposed scattering Model.
3.5 Conclusion
In this chapter, we have studied the spatial characteristics of cellular mobile channel for
picocell, microcell, and macrocell environments assuming uniform and Gaussian distribution
for the scatterers. We have utilized the Eccentro-Scattering Model proposed in the previous
chapter and derived general expressions for the pdf of AoA of the multipath signals at BS
applicable to picocell, microcell, and macrocell environments assuming uniform and Gaussian
scatter density. The same methodology can be further used to derive the spatial statistics
86
3.5 Conclusion
of the mobile channel for a 3-D environment. The derived results show that the previous
spatial models can easily be extracted from our proposed model with appropriate selection
of parameters. We have thoroughly discussed the results and compared them with those of
all existing models.
Theoretical results when compared with some available measurements both in indoor
and outdoor environments, show good proximity with the realistic situations. We can thus
visualize our proposed model to be useful in simulating several propagation scenarios for
wireless communications systems. The derived results, in closed form, can also be used
in further research work to model Doppler characteristics and tracking properties of time-
varying fading channels.
We have also addressed the issue of the impact of local-to-BS scattering on the spa-
tial characteristics and implemented JGSM for low antenna-height urban environment by
introducing an adjustable scattering-free region around BS. The adjustable scattering-free
region around BS models the extent of scattering in the vicinity of BS and thus can easily
be used as a control-valve for the inclusion/exclusion of scattering objects in the vicinity of
BS according to their anticipated effect on the angular distribution of the cellular mobile
channel. We have found that the JGSM along with the provision of scattering-free region
provides good fitness to the field measurements when compared with all existing Gaussian
scattering models that consider only one Gaussian function for the distribution of scatterers
around MS.
87
Chapter 4
Characterization of Angle Spread
Section 4.1 of this chapter presents an overview of the significance of angle spread in MIMO
channel modeling and itemizes the contributions made in the chapter. Section 4.2 discusses
various definitions of angle spread, which are already being used in the literature. In section
4.3, a novel method for quantifying angle spread of multipath power is proposed and its
relations with the previous definitions are discussed. Section 4.4 presents the effect of AoA
distribution truncation on the angle spread, while section 4.5 itemizes various factors that
cause the truncation of AoA distributions. Section 4.6 compares different azimuthal models
with the measurement campaigns on the basis of angle spread and section 4.7 concludes the
chapter and presents final remarks.

4.1 Overview
4.1 Overview
Use of adaptive antenna arrays in MIMO communication systems has generated significant
interest in recent years. MIMO systems utilizing advanced diversity and spatial filtering
schemes [55], offer very high information-theoretic capacities [18, 37]. The optimum antenna
array topology and combining algorithm are strongly related to the azimuth dispersion of
the mobile radio channel [4], especially during small-scale fading. Hence, in order to make
realistic evaluations of the capabilities of diversity and spatial filtering schemes in MIMO
systems, deep understanding of the statistics of the azimuth angle spread is required.
Angle spread is the key factor of the second order statistics (or the correlation statis-
tics) of fading processes in wireless communications [22, 23, 33, 49]. The concept of isotropic
scattering or omni-directional azimuthal propagation modeling [23] does not cope with the
realistic situations anymore, therefore a lot of work has been done on non-isotropic spatial
channel models [30, 38, 49, 56] to simulate the angular energy distribution at BS antenna (or
antenna array). Use of these angular energy distributions in the mathematical calculations
for spatial fading correlations, opened a new debate on finding the exact definition of angle
spread in the research community. Various definitions of the angle spread have been used in
the literature so far, but unfortunately none of them is capable of establishing a generalized
relationships among different parameters of the angle spread and multipath fading. This
motivated us to focus on the spatial statistics of the cellular mobile channel and introduce
a generalized definition of the angle spread for its diverse use in correlation calculations.
89
4.1 Overview
Main contributions of this chapter are thus summarized as follows:
1. We discuss all previous definitions of the angle spread and propose a novel generalized
method of quantifying the angle spread of the multipath power. Our method provides
almost all parameters of the dispersion of multipath power in space, which can be
further used for calculating more accurate spatial correlations and other statistics of
multipath fading channels. These proposed parameters are also helpful in finding the
exact standard deviation of truncated or distorted angular distributions as well as of the
angular data acquired in measurement campaigns, which can lead to the computation
of the exact separation distances among array elements needed for diversity antennas.
2. Keeping the recent use of truncated Gaussian and Laplacian functions as the distribu-
tions of AoA in view, we indicate the severity of the effects of distribution truncation
on the angle spread.
3. We itemize the factors which cause such truncations in the AoA distributions and
provide analytical solution to compensate for their effects.
4. We propose the angle spread as the goodness-of-fit measure in measurement campaigns
and show the comparisons of some notable azimuthal models with the measurement
results.
90
4.2 Angle Spread
4.2 Angle Spread
arrive at the receiver antenna from different azimuth directions about the horizon [22, 23, 49].
The distribution of these multipath components in the azimuth is conveniently described by
the function, p(θ), where θ is the azimuthal angle of arrival (AoA) [32].
Depending on their choices and requirements, the researchers are currently using several
definitions of quantifying angle spread in multipath fading channels. These definitions involve
total angular span [29], beamwidth, the rms value of the angular data, standard deviation
of Gaussian [30] or Laplacian distribution [31] and shape factor defined in [32, 33]. However,
no clear and unanimous definition has yet appeared in the literature.
Use of Beamwidth and rms value as the definition of angle spread are often ill suited for
general application to periodic functions such as angle of arrival (AoA) distributions [33].
Standard deviation (SD), σθ , of the Gaussian angular energy distribution heavily depends on
the total angular span, θspan , and the degree of truncation of the distribution (if there is any
truncation in the distribution). So the utilization of SD of an exact bell-shaped Gaussian
function as the true SD of the angular data which represents a truncated Gaussian, will
certainly lead to wrongful results.
The shape factor Λ defined in [32, 33] (we are keeping the same notation) used to denote
91
4.3 The Proposed Method to Quantify Angle Spread
the angle spread in the range from zero to one, can be considered as the most favorable one,
since it is invariant under the changes in transmitted power, under any series of rotational
or reflective transformation of p(θ) [33]. However, this definition has the disadvantage of
not providing true physical information about the angle spread, i.e. information either in
degrees or radians. This disadvantage is the main cause of its unpopularity in the research
community. We will show in sequel that one of our parameter in the proposed angle spread
modeling resembles this definition and gives the same results.
Our method of quantifying angle spread is based on Fourier coefficients of p(θ), and is analo-
gous to the method used in [32]. However, we utilize trigonometric moments for calculating
the statistics of directional data in [57] instead of exponential moments. It can easily be
shown that both methods generate the same Fourier coefficients. Since the use of trigono-
metric moments is advantageous in manipulating discrete data, we will later extend our
method from continuous distribution to the discrete data obtained in measurements.
Let R̄n = C¯n + j S¯n be defined as the nth complex trigonometric moment of the angular
R 2π
energy distribution p(θ) whose total power is equal to P0 = 0
p(θ)dθ. The trigonometric
parameters, C¯n and S¯n for the angular energy distribution p(θ) are defined as
92
Z 2π
1
C̄n = p(θ) cos(nθ)dθ (4.1)
P0 0
and
Z 2π
1
S̄n = p(θ) sin(nθ)dθ (4.2)
P0 0
In case of discrete measured or observed data, the definitions for the trigonometric
parameters, C¯n and S¯n can be modified as
N
1 X
C̄n = fi cos(nθi ) (4.3)
F0 i=1
and
N
1 X
S̄n = fi sin(nθi ) (4.4)
F0 i=1
PN
where F0 = i=1 fi for i = 1, · · · , N and fi is the number of occurrences for the AoA θi in
the ith bin of histogram.
R̄n can also be written as,
R̄n = |R̄n | ej θ̄n (4.5)
p
where |R̄n | = C̄n2 + S̄n2 is the mean resultant of the nth trigonometric moment and θ̄n =
tan−1 ( C̄S̄nn ) is its direction [57]. If the first moment is considered, θ̄1 gives the mean angle
of the distribution, p(θ), i.e. the mean AoA in our case. From now onward, we will denote
θ̄1 as θ̄, the mean AoA. Furthermore, we will utilize only first and second moments in our
characterization of angle spread.
93
Now we define our basic measure of angular dispersion, the circular variance, S0 as,
S0 = 1 − |R̄1 | (4.6)
where |R̄1 | is the magnitude of the first trigonometric moment of p(θ). S0 is invariant under
the changes in transmitted power, under any series of rotational or reflective transformation
of p(θ). We see that 0 ≤ S0 ≤ 1, which means that the observed AoAs are tightly clustered
about the mean direction θ̄, then |R̄1 | will be closer to 1 and S0 will be nearly zero. On the
other hand, if the AoAs are widely dispersed then |R̄1 | will be small and S0 will be nearly 1.
S0 and |R̄1 | can easily be transformed to the shape factor Λ discussed in [32] and the
conventional definition of standard deviation, as
q
Λ = 1 − |R̄1 |2
q
= 2S0 − S02 (4.7)
and
p
σθ = −2 ln(1 − S0 )
q
= −2 ln(|R̄1 |) (4.8)
where σθ is the conventional standard deviation of the angular energy distribution in ra-
dians. It gives the true physical information about the angular dispersion of the data. It
is also invariant under the circumstances discussed for S0 . Since it provides true physical
information about the dispersion of the multipath signals in space, it can also stand as the
major candidate for the unanimous definition of angle spread of multipath signals. If σθ is
94
4.4 Effect of Distribution Truncation on the Angle Spread
already known, then the circular variance S0 and the mean resultant of the first moment
|R̄1 | can also be deduced as,

σθ2
S0 = 1 − e− 2 (4.9)
σθ2
|R̄1 | = e− 2 (4.10)
The shape factor Λ and σθ are also inter-related as,
p
σθ = − ln(1 − Λ2 ) (4.11)
p 2
Λ= 1 − e−σθ (4.12)
In [33], Durgin et al. have also proposed two more shape factors angular constriction γ
and orientation parameter θMF in their study. γ is a measure of how multipath concentrates
about two azimuthal directions and θMF provides the azimuthal direction of maximum fading.
These two parameters can also be related to the complex trigonometric moments as,
|R̄2 − R̄12 |
γ= (4.13)
1 − |R̄1 |2
and
1
θM F = phase{R̄2 − R̄12 } (4.14)
2
It has been observed in [54] that if the bell-shaped spatial Gaussian model for the distribution
of scatterers around the mobile is assumed, then the angular distribution as seen from the BS
95
0.016
Truncated Gaussian pdf

0.014 Untruncated Gaussian pdf
0.012
0.01
pdf
0.008
0.006
0.004
0.002
0
−135 −90 −45 0 45 90 135 180 225
angle in degrees
Figure 4.1: Comparison between truncated (θspan = 90◦ ) and untruncated (θspan = 360◦ ) Gaussian
density functions, σg = 30◦ , θ̄ = 45◦
is also Gaussian. Based on the measurements, Pedersen et al. [4] found that the scattering
structures in urban environments usually give rise to the Gaussian distribution in azimuthal
angle of arrival and the Laplacian distribution in power azimuthal spectrum (PAS). Laplacian
function was also considered the best candidate for the distribution in angle of arrival in
indoor environments [1]. However, many situations have been observed when neither the
distribution of the angle of arrival matches exactly a true bell-shaped Gaussian nor the
PAS fits an exact double exponential Laplacian function, then a truncated Gaussian or
Laplacian function is usually considered as an approximate solution. Such situations usually
emerge in outdoor environment when Gaussian distributed scatterers do not produce exact
Gaussian distributions in AoA at BS, but instead they form some truncated, distorted, cut
96
or constant-added versions of Gaussian distributions. Measurements in [4, 56] show that
in addition to bell-shaped Gaussian distribution in AoA, there is always some additional
part, which distorts the Gaussian function. This additional part certainly disturbs the angle
spread measure. That is why the SD of the exact bell-shaped Gaussian function, σg , can not
be trusted any more to be used in calculating the spatial fading correlations among antenna
array elements. If it is still used, the spatial fading correlation calculations will certainly
result in wrong separations for diversity antennas.
The true SD, σθ , of the angular energy distribution or the pdf of the AoA as seen at
BS, depends on the peak/s of the distribution, as well as on the angular span of the angular
data. SD of the exact bell-shaped Gaussian function, σg , remains equal to the true SD, σθ ,
of the Gaussian azimuthal distribution as long as the total angular span, θspan , remains more
than 8σg (or twice of 4σg ). Angular span, θspan , can be written as,
θspan = θmax − θmin (4.15)
where θmax and θmin are the maximum and minimum angles of arrival, respectively.
As soon as the angular span lowers 8σg (in case of a truncated Gaussian), the true SD
of the angular energy starts decreasing, and it decreases very sharply for the smaller values
of the span.
A truncated Gaussian function is defined in terms of angular span as,
Cg ³ (θ − θ̄)2 ´
(G)
p̃Θ (θ) =√ exp − (4.16)
2πσg 2σg2
97
60
Gaussian, σ =60°
g
55 Gaussian, σ =45°
g
°
Gaussian, σg=30
50 Gaussian, σ =15°
g
True Standard Deviation of AoA data, σθ, Degrees

Gaussian, σg=7.5°
45 Uniform
40
35
30
25
20
15
10
0
360 320 280 240 200 160 120 80 40 0
Angular Span, 2θ0, Degrees
Figure 4.2: Effect of Gaussian distribution truncation on the angle spread
(G)
where θ̄ is the mean AoA and Cg is the normalizing constant chosen to make p̃Θ (θ) a density
function. In Fig. 4.1, a truncated Gaussian density function is plotted along with a full-span
or untruncated Gaussian density function. Cg is given as,
1
Cg = (4.17)
erf( 2θ√span
2σ
)
g
(G)
In equation (4.16), σg defines the probability density function (pdf), p̃Θ (θ), but does
not provide the correct information about the deviation of the data from the mean AoA, θ̄.
σg is in fact the SD of the untruncated or full-span Gaussian with θspan = 8σg . For such
untruncated case, Cg will be equal to 1, otherwise it will always be greater than 1.
After some mathematical manipulations given in Appendix 4-A, the true SD, σθ , for the
98
0.03
Truncated Laplacian pdf

Untruncated Laplacian pdf
0.025
0.02
pdf
0.015
0.01
0.005
0
−135 −90 −45 0 45 90 135 180 225
angle in degrees
Figure 4.3: Comparison between truncated (θspan = 90◦ ) and untruncated (θspan = 360◦ ) Laplacian
density functions, σl = 30◦ , θ̄ = 45◦
(G)
truncated Gaussian function, p̃Θ (θ), can easily be related to the SD of full-span Gaussian
function, σg , as,
Cg θspan σg ³ θ2 ´
span
σθ2 = σg2 − √ exp − (4.18)
2π 8σg2
Fig. 4.2 shows the effect of truncation of the Gaussian function on the angle spread
measured as the true SD, σθ , using various values of σg . Fig. 4.2 also provides the corre-
sponding effect for the case of uniform distribution in AoA. We observe that the effect of
decreasing angular span, θspan , on the angle spread σθ in case of uniform distribution is linear
√
as expected from the definition, σθ = θspan /(2 3). On the other hand, the decreasing θspan
clearly affects σθ for large values of σg in case of Gaussian distribution, but does not affect
99
σθ for small values of σg . One more interesting result that can be observed is the curve
of σθ versus θspan for Gaussian distribution follows the linear curve of uniform distribution,
indicating the turning of Gaussian distribution into uniform distribution in case of severe
truncation.
Similarly Laplacian assumption of the angular energy distribution can also be ap-
proached using the same analogy. A truncated Laplacian function is defined in terms of
angular span as,

³ √
(L) Cl 2|θ − θ̄| ´
p̃Θ (θ) =√ exp − (4.19)
2σl σl
where σl is the SD of untruncated or full-span Laplacian function. Cl is the normalizing

(L)
constant chosen to make p̃Θ (θ) a density function. In Fig. 4.3, a truncated Laplacian
density function is plotted along with a full-span or untruncated Laplacian density function.
Cl is given as
1
Cl = ¡ θspan ¢
(4.20)
1 − exp − √
2σl
As mentioned earlier in the Gaussian case, in equation (4.19), σl also defines the pdf,
(L)
p̃Θ (θ), but does not provide the correct information about the deviation of the data from
(L)
the mean AoA, θ̄. If p̃Θ (θ) represent an untruncated Laplacian, Cl becomes unity and σθ
equals σl , otherwise Cl always remains greater than 1 and σθ is less than σl .
Appendix 4-B provides the derivation of σθ of truncated Laplacian pdf in terms of σl
100
60
Laplacian, σl=60°
55 Laplacian, σl=45°
Laplacian, σl=30°
50 Laplacian, σl=15°
True Standard Deviation of AoA data, σθ, Degrees

Laplacian, σl=7.5°
45 Uniform
40
35
30
25
20
15
10
0
360 320 280 240 200 160 120 80 40 0
Angular Span, 2θ0, Degrees
Figure 4.4: Effect of Laplacian distribution truncation on the angle spread
and θspan . σθ can be written as,

¡ θspan ¢ µ ¶
Cl exp − √
2σl
√
σθ2 = Cl σl2 − exp 4σl2 + 2
θspan + 2 2σl θspan (4.21)
4
Fig. 4.4 shows the effect of truncation of the Laplacian function on the true SD, σθ ,
using various values of σl . Fig. 4.4 provides almost similar results as discussed in the case of
Fig. 4.2. However, one interesting result which can be deduced by comparing both figures is
that the effect of truncation on the angle spread is more gradual in case of Laplacian than
that in Gaussian, where sudden decline in σθ occurs for low values of θspan .
Keeping the above discussion in view, we can easily conclude that any calculations or
simulations carried out for spatial correlations by using truncated Gaussian or Laplacian
101
4.5 Factors which Cause Gaussian Distribution Truncation
models for the distribution of AoA will result in wrongful outcomes. Hence, the use of σg
of an untruncated Gaussian in [58] (where a truncated Gaussian is used to represent the
angular energy distribution for calculating correlations among antenna elements in order to
compare BER performance of a uniform circular array with that of a uniform linear array)
would certainly lead to unexpected and wrongful results.
4.5 Factors which Cause Gaussian Distribution Trun-
cation
The major factors which give rise to the situations where a Gaussian azimuthal distribution
is subjected to truncate, can be categorized as follows:
Factor 1:
With the use of directional antenna at the base station in Gaussian scattering environment
[27] one can avoid the scatterers falling out of the beam-range of antenna. This gives rise
to a truncated Gaussian distribution of the AoA at the base station. In chapter 3, we have
explained the effect of directional antenna at BS on the AoA distribution. In this case,
the angular span θspan of the AoA data will be equal to the beam width of the directional
antenna, 2α, as shown in Fig. 4.5(a). This sort of situations usually belongs to the space
division multiple access (SDMA) systems which rely on the use of adaptive narrow-beam
102
antennas and the nonhomogeneous distribution of users in a cellular system to increase
system capacity [27].
Factor 2:
The streets crowded with the automobile traffic represent an elliptical part of a Gaussian
scattering region centered at MS. 3D angular investigations at a BS site [14] also show street
canyon dominated propagation. This sort of situations have been modeled in [44]. Such
elliptical scattering discs are referred to as the Eccentro-Scattering discs in chapter 2 of
our thesis, due to the fact that the eccentricity of the disc can be altered according to the
scattering structures in the street. Fig. 4.5(b) shows a scenario where a mobile is surrounded
by Gaussian distributed scatterers in an elliptical disc in a crowded urban street. Usually
high-rise buildings define the boundaries of scattering disc. Here the angular span depends
on the major axis a and the minor axis b of the scattering disc and can be written as,
µ µ ³ a ´¶¶
−1 b −1
θspan = 2 tan tan sin (4.22)
a d
where d is the distance between BS and MS.
Factor 3:
In urban macrocell and microcell environment, the height of BS antenna is usually kept low,
so the high-rise buildings near BS also play an important role in the dispersion of angular
103
MS
BS θspan=2 α MS
θspan
BS
(a) Use of Directional Antenna [27], (see (b) Eccentro-Scattering scenario, (see sec-
section 3.3) tion 3.2 and 3.3)
Distant Scattering Disc

Local Scattering Disc
CD
MS
MS
θspan
BS
BS
(c) Local-to-BS scattering scenario, (see (d) Far Scattering scenario [38], (see sec-
section 3.4) tion 3.3)
Figure 4.5: The factors which cause truncation of the Gaussian distribution in AoA
104
energy as seen at BS. In the measurements campaigns of both indoor [1] and outdoor [4, 56]
environments, it was observed that in addition to the double exponential Laplacian (indoor
case) and bell-shaped Gaussian (outdoor case), there are always very uniform tails on both
sides of the mean AoA in the angular domain. The formation of these uniform tails in the
distribution of AoA is in fact the aftermath of the reflections/scattering of the radio signal
from the scatterers that surround BS, see Fig. 4.5(c). A detailed discussion on the issue
of scattering around BS can be found in section 3.4 of this thesis, where JGSM has been
proposed to model the effect of such scattering. The uniform tails cause deformation in
the shape of Gaussian or Laplacian, and hence alter the measure of angle spread, i.e. the
standard deviation. The angular span here mostly depend on the scatter spread around BS
or in other words, the span of uniform tails. Uniform tails usually span the whole antenna
beamwidth if a directional antenna of certain beamwidth (mostly 120◦ ) is used at the base
station.
Factor 4:
It has been found that the scatterer locations (or equivalently, the azimuth angles of the mul-
tipath components) are not distributed uniformly over space but tend to be concentrated
in certain regions [38]. Far scatterers like high-rise buildings (in urban environments) or
mountains (in rural environments) significantly contribute to the multipath scattering phe-
nomenon in addition to the scattering structures located near the mobile stations [14, 25, 38].
105
Hence, they give rise to non-symmetric and non-isotropic scattering. This kind of scattering
induces a non-uniform and non-symmetric distribution in the angle of arrival (AoA) as seen
at the base-station (BS). The measurement campaign conducted in [59] reveals that in some
urban areas, Power Azimuthal Spectrum (PAS) exhibits a narrow peak symmetrically cen-
tered at 0◦ described by a truncated Laplacian curve, but also embodies an additional part
centered around 8◦ . This additional part which breaks the symmetry of the PAS around 0◦ ,
can be considered as the aftermath of the remote or far scattering.
Far scattering gives rise to multi-modal distribution of AoA at BS, where every peak
in the distribution represents a concentration center of the scattering cluster. So in such
cases, the true standard deviation of the angular data differs significantly from the standard
deviation of single Gaussian approximated as the distribution of AoA. We have presented a
detailed study of the effect of far or distant scatterers on the azimuthal angular distribution
of multipaths in chapter 3. Fig. 4.5(d) shows the situation of far scattering which gives rise
to the formation of multi-modal distribution in the AoA. Here the angular span is given as,
³ 4 σ (L) ´ ³ (D) ´
−1 g −1 4 σg
θspan ≈ sin + sin + |θD − β| (4.23)
d dD
where
• d is the distance between BS and MS
• dD is the distance between BS and the center of the far scattering cluster, CD
• β and θD are the angles of d and dD with some reference axis (either the axis of antenna
106
4.6 Angle Spread as the Goodness-of-Fit Measure in Measurement Campaigns
array or the axis of antenna broad side), and
(L) (D)
• σg and σg are the standard deviations of local Gaussian scattering and far Gaussian
scattering regions.
4.6 Angle Spread as the Goodness-of-Fit Measure in
Measurement Campaigns
Many measurement campaigns [1, 4, 48, 60] have been undertaken to characterize the actual
azimuthal distribution or power azimuth spectrum (PAS) in the test environment. Several
analytical models have also been proposed to fit the results of such measurement campaigns
in closed-form formulas, either by using Gaussian functions [4], Laplacian functions [1, 4],
trigonometric functions [61] or the combinations of Gaussian and trigonometric functions [2],
(also Eccentro-Scattering and JGSM models, see chapter 3). No such measure of the angular
dispersion has been developed so far, on which the proximity of these models could be tested
with the measurement results. Usually standard deviations, σg of the Gaussian and σl of
the Laplacian functions are taken as the measures of angle spread to fit some specific model
to measurement data, but they do not provide adequate information, especially in the cases
of truncated Gaussian and Laplacian functions. On the other hand, equation (4.8) offers
substantial help in this regard. It gives almost all necessary information about the angle
spread, no matter what functions and how much truncation are used. We will thus exploit
107
°
Meas. [Spencer], σθ=36
−1 Simulations, σ =31.4°
10 θ
°
Eccentro−Scattering, σ =35.7
θ
°
Uniform Scattering, σθ=29.5
Gaussian, σ =24.5°, σ =24.5°
g θ
° °
Gaussian, σg=30 , σθ=30
−2
10
Rate of Occurrences
−3
10
−4
10
−200 −150 −100 −50 0 50 100 150 200

Angle of Arrival, θ, Degrees
Figure 4.6: Comparison of the distribution in AoA for the candidate models (Eccentro-Scattering
Model [section 3.2], Gaussian [2] and Uniform Elliptical Scattering Model [3]) with the measure-
ments [1] and simulations in indoor environments
this measure as our goodness-of-fit criterion on good-fitting models for the two renowned
measurement results presented in [1] (Indoor) and [4] (Outdoor). One very important point,
which should be kept in mind while comparing the goodness-of-fit for various candidate
models to the measurement data, is that the shape of the PAS or exact distribution is not so
important rather the variance of the angle spread is important [62]. The same methodology
can also be extended to other models assumed to fit the measurement results of various
campaigns.
108
−1
10
−2
10
Rate of Occurrences
−3
10
°
Measurements [Spencer], σθ=36
° °
Laplacian, σl=36 , σθ=34.2
° °
Laplacian, σl=25.5 , σθ=25
Eccentro−Scattering Model, σ =35.7°
θ
−4
10 Raised Laplacian, σl=25.5°, σθ=35.9°
−200 −150 −100 −50 0 50 100 150 200

Figure 4.7: Comparison of the distribution in AoA for the candidate models (Eccentro-Scattering
Model [section 3.2], Laplacian [1] and Raised-Laplacian) with the measurements [1] in indoor en-
vironments
4.6.1 Indoor Environments
In indoor environments, the scatterers are usually bounded by some rigid structures, forming
the boundary of the scattering region. For example, the furniture and the people in a
hall represent the scattering objects while walls indicate clear boundaries to those objects.
Previous research work [53] shows that in indoor environment the delay spread is usually
in the range of 0.1 − 0.2 µsec, which specifies the dimensions of the scattering area. So,
exact bell-shaped Gaussian assumption either for the distribution of scatterers [2] or for the
distribution of the AoA in indoor environments is usually not considered suitable (see chapter
109
°
Measurements [Pedersen], σ =7.1
θ
Gaussian (GSDM), σ =6°, σ =6°
300 g θ
° °
JGSM or Raised Gaussian, σ =6 , σ =7.2
g θ
250
200
Rate of Occurrences
150
100
50
0
−25 −20 −15 −10 −5 0 5 10 15 20 25
Figure 4.8: Comparison of the distribution in AoA for the candidate models (GSDM [2] and JGSM
[section 3.4]) with the measurements [4] in outdoor environments
3), while the three other candidates such as the elliptical uniform scattering model [3], the
Gaussian Eccentro-Scattering model and the Laplacian azimuthal distribution [1] are usually
believed to be more appropriate. Using the definition of the angle spread given in (4.8),
(4.18) and (4.21), we can compare all these candidate models for goodness-of-fit with the
measurement data [1].
In Fig. 4.6 and 4.7, all candidate models are plotted along with their respective angle
spread measures. In Fig. 4.6, we observe that the simulations (which are in fact the normal-
ized histograms of the arrival angles as seen from the BS) and the measurements tend to form
almost identical curves with a difference of 4.5◦ in their angle spreads. This difference seems
110
to be due to the irregular shape of simulations at the tails. Gaussian Eccentro-Scattering
model (section 3.2) with σθ = 35.7◦ and uniform scattering model [3] with σθ = 29.5◦ are
observed as the best fits of first and second choices. Gaussian curves with σθ = 24.5◦ [2] and
with σθ = 30◦ are far from goodness-of-fit in both shape and angle spread.
In Fig. 4.7, the same measurement data has been plotted along with different Laplacian
functions. Laplacian functions are usually considered the best fits to model the pdf of AoA
in indoor environments [1] and the PAS in outdoor environments [4]. We observe that the
Laplacian function alone fails in modeling the measurements on the basis of both shape and
angle spread. If σl is increased to 36◦ , the Laplacian deviates from the measurements in the
narrow stem of the measurements pdf. On the other hand, if some constant value based
on the height of tails of the measurements pdf, is added to the Laplacian with σl = 25.5◦ ,
the resultant raised Laplacian function shows good proximity to the measurements, both in
shape and angle spread. Even the raised Laplacian surpasses the Eccentro-Scattering model.
Actually the additional constant compensates for the effects caused by the scattering local
to BS. These effects have been discussed in section 3.4 extensively.
4.6.2 Outdoor Environments
In outdoor environments, the majority of scattering points are clustered together with their
density decreasing as the distance from MS increases. This fact suggests a Gaussian model
for the distribution of scatterers in outdoor environments [25, 27]. Measurement results for
111
4.7 Conclusion
the azimuthal distribution in outdoor urban environment [4] also showed a strong tendency
towards Gaussian distribution. In [2], a Gaussian function with σg = 6◦ was suggested to
fit these measurements. Using (4.8), we see that σθ of measurements is 7.1◦ , which shows
1.1◦ discrepancy with the Gaussian. Fig. 4.8 shows that this discrepancy lies in the tails
of the measurement histogram where the Gaussian function alone is not able to model the
measurements. In section 3.4, formation of these tails has been discussed in the light of
the aftermath of the scattering phenomenon in the vicinity of BS and Jointly Gaussian
Scattering Model (JGSM) or raised Gaussian function has been suggested. In Fig. 4.8, a
raised Gaussian is shown with σθ = 7.2◦ , where a constant value based on the height of the
tails of the measurement histogram is added to the Gaussian function with σg = 6◦ . We see
that the resultant raised Gaussian function shows good agreement with the measurements
both in shape and angle spread.
4.7 Conclusion
In this chapter, we have proposed a novel generalized method of quantifying the angle spread
of the multipath power distribution. The proposed method provides almost all parameters
about the angle spread, which can be further used for calculating more accurate spatial
correlations of the multipath fading channels. The proposed parameters are also useful in
finding the exact standard deviation of the truncated angular distributions and the angular
data acquired in measurement campaigns. The degree of accuracy in correlation calculations
112
4.8 Appendix 4-A
can lead to the computation of exact separations among array elements needed for diversity
antennas. Recent use of truncated Gaussian or Laplacian functions as azimuthal distribution
of multipath signals has simplified the calculation of correlations in MIMO channels. We
have indicated that the use of standard deviation of full-span functions as the standard
deviation of the truncated function causes severe effects on the angle spread, which in turn
distorts the accuracy of correlation figures in MIMO channels. Due to the importance of
angle spread in the fading statistics, we have proposed its use as the goodness-of-fit measure
in measurement campaigns. The proposed method of quantifying angle spread can thus be
used in finding the accurate separations among array elements in outdoor MIMO systems
where measurement campaigns provide basis for channel models.
4.8 Appendix 4-A
Since the conventional definition of the standard deviation, σθ , and the one given in (4.8)
have the same meanings, we can use the conventional method to find the relationship between
(G)
true SD, σθ , of the truncated Gaussian, p̃Θ (θ), and the SD, σg , of untruncated or full-span
(G)
Gaussian function. The variance of the random variable, θ, with the function, p̃Θ (θ), is
defined as,
Z θmax
(G)
σθ2 = (θ − θ̄)2 p̃Θ (θ)dθ (4.24)
θmin
113
4.9 Appendix 4-B
where θ̄ is the mean AoA defined in section 4.3. It can also be defined conventionally as,
Z θmax
(G)
θ̄ = θp̃Θ (θ)dθ (4.25)
θmin
(G)
Using p̃Θ (θ) given in (4.16), and integrating with the help of tables of integrals [63], we get,
" ½ ¾
−Cg σg √ (θmax − θ̄)2 √ (θ̄ − θmax )2
σθ2 = √ 2 (θmax − θ̄) exp − + 2(θ̄ − θmin ) exp −
2 π 2σg2 2σg2
½ ³ #
√ θmax − θ̄ ´ ³ θ̄ − θ ´¾
min
−σg π erf √ − erf √ (4.26)
2σg 2σg
(G)
Let the distribution, p̃Θ (θ), be symmetrical around mean AoA, θ̄ (also see Fig. 4.1), then
the maximum and minimum values of the angular distribution can be written as,
θspan
θmax , θmin = θ̄ ± (4.27)
2
(G)
where θspan is the angular span of the distribution, p̃Θ (θ). Using these minimum and
maximum values in (4.26) and simplifying, we get the true variance of the truncated Gaussian
(G)
distribution, p̃Θ (θ), as given in (4.18). It is also evident from (4.18) that the true SD does
not depend on θ̄.
4.9 Appendix 4-B
A method similar to the one used in Appendix 4-A, can be exploited to derive the relationship
(L)
between true SD, σθ , of the truncated Laplacian, p̃Θ (θ), and the SD, σl , of untruncated or
full-span Laplacian function.
114
4.9 Appendix 4-B
(L)
The variance of the random variable, θ, with the function, p̃Θ (θ), is defined as,
Z θmax
(L)
σθ2 = (θ − θ̄)2 p̃Θ (θ)dθ (4.28)
θmin
where θ̄ is the mean AoA defined in section 4.3 and equation (4.25). Since the truncated
(L)
Laplacian p̃Θ (θ) given in (4.19) is symmetrical (also see Fig. 4.3), the true variance from
θmin to θ̄ is the same as that from θ̄ to θmax . Thus we can write (4.28) as
Z θ̄
(L)
σθ2 =2 (θ − θ̄)2 p̃Θ (θ)dθ (4.29)
θmin
(L)
Using p̃Θ (θ) given in (4.19), and integrating with the help of tables of integrals [63], we get,
" ½
−Cl 2 ³ √2 ´¾
σθ2 = σl 1 − exp − (θ̄ − θmin )
2 σl
#
½
√ ³ √2 ´¾
− (θ̄ − θmin )(θ̄ − θmin − 2σl ) exp − (θ̄ − θmin ) (4.30)
σl
(L)
Assuming p̃Θ (θ) as symmetrical around θ̄ and substituting the value of θmin from (4.27), we
get σθ2 in terms of θspan as given in (4.21). From (4.21), it is again evident that the true SD,
σθ , does not depend on θ̄, but it strictly depends on θspan .
115
Chapter 5
Modeling Temporal Characteristics of
Mobile Channel
This chapter gives a brief overview of temporal channel modeling in its section 5.1. In section
5.2, the description of the proposed temporal modeling approach is presented. Section 5.3
and 5.4 present the derivations of the closed-form formulas for the pdf of time of arrival for
pico/micro and macrocell environments. In the end, section 5.5 gives the summary and final
remarks of the chapter.

5.1 Overview
5.1 Overview
5.1.1 Background
Modern wireless communication systems aim to provide reliable services to as many users as
possible regardless of their locations and mobility. This goal is seriously impeded by three
major channel impairments, i.e. multipath, delay spread (DS), and co-channel interference
(CCI). Multipath occurs in wireless communications when various incoming radio waves
reach their destination (the receiver) from different directions and/or with different time
delays [64]. The differences in propagation delays among these multiple propagation paths
cause delay spread, which in turn induces inter-symbol interference (ISI). The time-dispersive
nature of the channel determines the maximum data rate that may be transmitted without
requiring equalization and also determines the accuracy of navigational services such as vehi-
cle location [65]. Large values of DS (> 10% of the symbol duration), result in considerable
degradation in system performance in terms of the attainable data rates [64]. Therefore,
characterizing temporal dispersion of the channel is essential for wideband systems [66], e.g.
wideband code division multiple access (WCDMA), due to strong association between DS
and the inverse of signal bandwidth. Thus, it provides a measure of frequency selectivity
of the channel [67]. Accurate and, if possible, simple propagation models would lead to an
effective design and evaluation of modern communication systems. Such models are low-cost
and handy means to accurately predict radio wave propagation behavior.
117
5.1 Overview
5.1.2 Problem Formulation
Early models for the multipath arrival time in urban areas date back to [68–70]. It was
suggested in [68] that the sequence of path delays follows a Poisson process. But, this as-
sumption did not fit the empirical data particularly for early arrivals and, therefore, Turin et
al. [68] recommended a modified Poisson function to resolve this inconsistency (see footnote
6 in [68]). The modified Poisson function provided a better fit for the measured data as
compared to the Poisson sequence [70] but it is still not good enough. In [4], the exponential
decaying function has been reported as the best fit for estimated delays in outdoor environ-
ments. Indeed, the exponential function is an excellent fit to empirical data but it lacks a
theoretical explanation. In [2], excellent derivation and analysis has been presented for the
Gaussian Scatter Density Model (GSDM). Yet, analytical evaluation of the integral related
to the probability density function (pdf) of Time of Arrival (ToA) is not permissible due
to model complexity [2]. Thus, no closed-form formula for the pdf of ToA is provided for
the GSDM model. In [3], a closed-form formula for the pdf of ToA has been presented for
the Geometrically Based Single Bounce Macrocell (GBSBM) model, but it is valid only for
a circular scattering disc. This was a limitation because in some macrocell locations, e.g.
street canyon propagation [71], an elliptical scattering disc produces a more realistic propa-
gation model, the Eccentro-Scattering model. In macrocell environments, distant scatterers
far from Base Station (BS) and Mobile Station (MS) have considerable effect on the received
signal. However, few scattering models acquiesced the effect of distant scatterers on the
118
5.1 Overview
multipath signal. The model in [72] predicts the power of reflected paths from distant reflec-
tors with no consideration of spatial or temporal aspects of the received signal. In chapter
3, a model for the spatial statistics of cellular environments including the effect of distant
scatterers has been presented. So far, no scattering model incorporated the effect of distant
scatterers on the temporal dispersion of the multipath signal in macrocell environments.
5.1.3 Contributions
In this chapter, we derive simplified closed-form expressions for the ToA distributions for
picocell/microcell and macrocell environments by considering the scatterers confined in
Eccentro-Scattering discs. This is a more general approach from which the results of the
previous models, such as GBSBM [3], can be deduced as its special cases. For macrocell
environments, we also include the effect of dominant distant scatterers on the temporal dis-
persion of the multipath signals. Objects such as hills, mountains, and skyscrapers act as
clustered scatterers/reflectors when they have line of sight (LoS) to both BS and MS [16, 25].
The derived formulas can be used to simulate temporal dispersion of wireless signal in several
propagation conditions. The pdf of ToA is very important for finding the covariance of two
multipath signals as a function of their frequency separation. Covariance of two multipath
signals is in fact the characteristic function of the pdf of time delays [22]. Besides the handy
use of the pdf of ToA in determining the coherence bandwidth of a particular system, it also
emerges as a basic characteristic of the system capacity along with the pdf of AoA [22]. The
119
5.2 Model Description
pdf of AoA has been discussed in detail in chapter 3. Almost complete description of the
wireless system can be achieved if the pdfs of time and angle of arrival are known.
In this chapter, we also present formulas for the areas of intersection between two
elliptical regions and between a circle and an ellipse. These formulas are necessary for the
derivation of the pdf of ToA and can also be used in other mathematical applications.
arrive at the receiver antenna with identical or different delays [22, 23, 49]. The distribution
of these multipath components in time-delay is conveniently described by the function p(τ ),
where τ is the delay of the multipath component in time domain. Thus, a specific signal
delay τ defines a set of scatterers bounded by an elliptical region, with foci BS and MS and
distinct major and minor axes. This ellipse will be referred to as the ‘bounding ellipse’ in our
work. Such a bounding ellipse has been shown in Fig. 5.1. The scatterers on the boundary
of the bounding ellipse give rise to single bounce multipath components [24, 51]. Therefore,
different delays correspond to different confocal ellipses [13], with BS and MS as common
focal points.
Based on the assumptions given in chapter 2, an nth bounding ellipse defines a specific
120
S bτ n
y
rMS
M
rBS d
C
aτ n
β
B x
Figure 5.1: Geometry of the proposed temporal channel model
Distant
Scatterers Medium Distant
Scatterers
BS
MS
MS
Scatterers local to BS τn+2 Scatterers

τn+1
local to MS
τn
BS τ0 MS
Figure 5.2: Temporal model for a typical multipath fading environment
121
set of scatterers that gives rise to multipath components arriving between time delays τn and
τn+1 = τn + δτ . This phenomenon is explained in Fig. 5.2, where different sets of scatterers
are giving rise to the formation of different confocal bounding ellipses.
Consider an nth bounding ellipse centered at C with foci B (BS) and M (MS) separated
by a distance d and with semi-major and semi-minor axes aτn and bτn respectively, as shown
in Fig. 5.1. rBS and rM S satisfy,
rBS + rM S = 2aτn (5.1)
Eccentricity of this bounding ellipse is defined as,
q
e τn = 1 − κ2τn (5.2)
where κτn is the aspect ratio of the bounding ellipse and equals bτn /aτn . Eccentricity of the
bounding ellipse can also be defined in terms of the MS-BS separation and the major axis
as,
d
eτn = (5.3)
2aτn
Using radio signal propagation theory, the relationship among time delay τn , path trav-
elled by the multipath signal, rBS + rM S , and speed of light c can be written as,
rBS + rM S = cτn (5.4)
Let a specific time delay τn define a set of scatterers bounded by nth bounding ellipse.
Then the semi-major axis aτn and semi-minor axis bτn of the nth bounding ellipse can be
122
written as,
cτn
aτn = (5.5)
2
1q 2
bτn = 4aτn − d2 (5.6)
2
In our work, we will denote the semi-major axis and eccentricity of the outermost
bounding ellipse by the notations aτmax and eτmax , respectively. Both variables are related
to the maximum delay spread, τmax . Use of aτn should not be confused with the notation
a in chapters 2 and 3, where a denotes the semi-major axis of a fixed Eccentro-Scattering
disc in a spatial channel model. However, aτmax and eτmax are equivalent to a and e of the
Eccentro-Scattering spatial channel model in the case of picocell and microcell environments,
but not equivalent to a and e of the Eccentro-Scattering spatial channel model in macrocell
environment. This is due to the fact that a and e of the Eccentro-Scattering disc in case
of picocell and microcell environments depend on τmax , while being independent of τmax in
macrocell environment, they depend only on the terrain and orientation of the streets, roads
or valleys.
In order to find the pdf of ToA, we will first find the cumulative distribution function
(CDF) of ToA, which may be calculated as the probability of a scatterer being placed inside
the nth bounding ellipse corresponding to time delay τn . Thus,
Aτn (τn )
Pτ (τn ) = (5.7)
A
where Aτn (τn ) is the area of overlap between the nth bounding ellipse corresponding to time
delay τn and the Eccentro-Scattering disc (pico/micro or macro) and A is the area of the
123
5.3 pdf of ToA for Picocells and Microcells
whole Eccentro-Scattering disc. Then, the pdf of ToA would be the derivative of the CDF
with respect to τn . Thus,
d 1 d
pτn (τn ) = Pτn (τn ) = Aτ (τn ) (5.8)
dτn A dτn n
In this section, we derive the pdf of ToA of the multipaths for picocell and microcell envi-
ronments. In such environments, BS and MS are located closer to each other and both have
scatterers around them. The antenna heights are relatively low and multipath scattering
is assumed near both BS and MS. Therefore, BS and MS are located at the focal points
of the Eccentro-Scattering disc. It was explained earlier that BS and MS are the common
focal points for all bounding ellipses. So, the pico/micro Eccentro-Scattering disc encloses
all bounding ellipses such that the bounding ellipse corresponding to the maximum allowed
delay, τmax , coincides the pico/micro Eccentro-Scattering disc. Thus,
2aτmax 2a
τmax = = (5.9)
c c
Fig. 5.3 represents the scattering model for picocell and microcell environments. In
the figure, C is the common center of all bounding ellipses and the pico/micro Eccentro-
Scattering disc; and B (BS) and M (MS) are their common focal points. a and b are the
semi-major and semi-minor axes of the pico/micro Eccentro-Scattering disc.
For the nth bounding ellipse corresponding to the time delay τn in the case of picocells and
124
S b or bτ max
y
M
d
C
a or aτ max
β
B x
Outermost bounding ellipse
Figure 5.3: Proposed temporal channel model for picocells and microcells
microcells, the following parameters are defined,
cτn
aτn = (5.10)
2
1q 2
bτn = 4aτn − d2 (5.11)
2
d
eτn = (5.12)
cτn
The area of the pico/micro Eccentro-Scattering disc in Fig. 5.3, A, is defined as,
A = πab
= πaτmax bτmax (5.13)
Therefore, the area of overlap between the nth bounding ellipse corresponding to time
delay, τn , and the pico/micro Eccentro-Scattering disc will be equal to the whole area
bounded by that bounding ellipse, i.e.,
Aτn (τn ) = πaτn bτn
πcτn p 2 2
= c τn − d2 (5.14)
4
125
Substituting (5.13) and (5.14) into (5.8), we get the pdf of ToA for picocell and microcell
environments as follows [3],

 2 2
 c(2c
√ d2 − 4a2 e2 )
τn −p
, τ0 < τn < τmax
pτ (τn ) = 4a2 1 − e2 c2 τn2 − 4a2 e2 (5.15)

0, elsewhere.
where e is a constant representing the eccentricity of the pico/micro Eccentro-Scattering
disc, defined as e = d/(2a) and τ0 is a constant representing the delay of the first multipath
arrival or usually delay in LoS path, defined as τ0 = d/c. Equation (5.15) can also be written
as, 
 τn (2 − e2τn )
 √ q , τ0 < τn < τmax
2 2 2
pτ (τn ) = τmax 1 − e 1 − eτn (5.16)


0, elsewhere.
where eτn is a variable representing the eccentricity of the nth bounding ellipse, which cor-
responds to the time delay τn and is defined as eτn = d/(cτn ).
Fig. 5.4 illustrates the effect of changing the value of the semi-major axis of the pico/micro
Eccentro-Scattering ellipse, a, or the distance between BS and MS, d, on the pdf of ToA in
picocells and microcells. In Fig. 5.4(a), we keep the distance between BS and MS fixed at the
value, d = 12 m, and change the value of the major axis of the Eccentro-Scattering ellipse, a,
consequently e and τmax change. Since d is fixed, then the delay of the LoS, τ0 , is also fixed.
Therefore, an increase in the value of the semi-major axis of the Eccentro-Scattering ellipse,
a, would result in an increase in the value of the maximum delay, τmax , thus the temporal
spread increases. This is the case when an increase in the number of scatterers influences the
multipath signal. Conversely, small values of the temporal spread correspond to situations
when only small numbers of scatterers influence the multipath signal, those influential scat-
126
0.5
a = 12 m ,e = 0.5
s s
as = 10 m ,es = 0.6
0.45 a = 8 m ,e = 0.75
s s
0.4
0.35
Rate of Occurrence
0.3
0.25
0.2
0.15
0.1
0.05
0
4 4.5 5 5.5 6 6.5 7 7.5 8
−8
Time of Arrival, (seconds) x 10
(a)
0.5
d = 8 m, es = 0.5
d = 9.6 m, e = 0.6
s
0.45 d = 12 m, es = 0.75
0.4
0.35
Rate of Occurrence
0.3
0.25
0.2
0.15
0.1
0.05
0
2.5 3 3.5 4 4.5 5 5.5 6
−8
(b)
Figure 5.4: pdf of ToA for picocell and microcell environments
127
5.4 pdf Of ToA for Macrocells
terers exist around the direction of LoS thus the scattering ellipse (the Eccentro-Scattering
disc) gets higher values for its eccentricity.
Fig. 5.4(b) illustrates the effect of changing the distance between BS and MS, d, on
the temporal spread of the multipath signal. We choose a fixed value for the semi-major
axis of the pico/micro Eccentro-Scattering disc, a = 8 m, and change the distance between
BS and MS, d. Consequently, the eccentricity of the pico/micro Eccentro-Scattering disc, e,
and τ0 change. Since a is fixed, then the maximum delay, τmax , is also fixed. Therefore, an
eccentricity value close to 0 implies that MS is in the neighborhood of BS, so the delay of
the first multipath signal (or the delay in the LoS) is small and, hence, the temporal spread
is large. Whereas, an eccentricity value close to 1, indicates MS is far from BS, so the delay
of the first multipath signal (or the delay in the LoS) is large, i.e. τ0 approaches τmax , and,
thus, the temporal spread would be small. Before e becomes 1, MS should be handed over
to a new BS in order to guarantee uninterrupted coverage. For more discussion on this issue
refer to the section 4 of chapter 6.
In macrocell environments, several local and distant scatterers contribute to the multipath
signal as shown in Fig. 5.2.
Here, BS is usually far away from MS, i.e. the distance between BS and MS is in the
128
order of kilometers and BS is typically positioned higher than the surrounding scatterers,
i.e. more than 30 meters above ground [14]. Therefore, BS is located outside the scattering
ellipse, i.e. no scatterers are assumed around BS, and MS is located at its center. In [73],
a scattering model has been presented in which BS is assumed to be at the center of the
scattering area and MS is somewhere at the edge. This is an impractical assumption based
on numerous empirical and theoretical results. In macrocells, less, or no, scatterers exist in
the vicinity of BS. The scattering maps measured during the trials confirm that in many
cases the significant scattering from a mobile in the field is constrained to an area centered
on the mobile [54].
Due to the far distance separating BS and MS, the multipath signal is also influenced by
objects far from transmitter and receiver. Objects such as hills, mountains, and skyscrapers
act as clustered scatterers when they have LoS to both BS and MS [25]. Therefore, signals
reflected from these structures arrive at the receiver as clusters in space and time. The effect
of distant scatterers is apparent in suburban and hilly areas. In bad urban environment, a
two-cluster model is more appropriate [14], where distant scatterers are usually the high-rise
buildings. Therefore, our model encompasses a distant scattering disc in addition to the
local scattering ellipse in suburban and hilly areas. For simplicity of the derivation, we will
derive the pdf of ToA for macrocells in two parts; first due to local scatterers around MS
and second due to the dominant distant scatterers.
129
a
Bounding
Ellipse
b
α1
B C M
bτ n Lτ n
Ls
Local
Scattering
d Ellipse
aτ n
(a) Local macro Eccentro-Scattering disc along
with bounding ellipse
Distant aD
Scattering
Disc bD
O
(x1, y1) LD
(x2, y2)
Bounding
Ellipse
θD
B C M
bτ n
d
aτ n
(b) Distant macro Eccentro-Scattering disc along
with bounding ellipse
Figure 5.5: Temporal channel model for macrocells
130
5.4.1 Local Scatterers
The proposed scattering model for local scatterers in macrocell environment is depicted
Fig. 5.5(a). In the figure, C, B (BS), and M (MS) are the center and two focal points of
the bounding ellipse corresponding to time delay τn with semi-major and semi-minor axes
aτn and bτn , respectively. The local scattering ellipse is centered at M with semi-major and
semi-minor axes a and b, respectively. The location of a scattering point on the bounding
ellipse corresponding to time delay τn , Lτn , from M is defined by (rτn , φ), whereas the
location of a scattering point on the local scattering ellipse, Ls , from M is defined by (rs , φ).
The bounding ellipse corresponding to τn and the scattering ellipse intersect at Lx , which
is located at angle α1 as shown in Fig. 5.5(a). The equation defining the bounding ellipse
corresponding to delay τn in Fig. 5.5(a) is,

³ ´2
rτn cos φ − d/2 rτ2n sin2 φ
+ =1 (5.17)
a2τn b2τn
Solving 5.17, rτn can be found as,
4a2τn − d2 c2 τn2 − d2
rτn = = (5.18)
2(2a2τn − d cos φ) 2(cτn − d cos φ)
The equation defining the Macro Eccentro-Scattering ellipse in Fig. 5.5(a) is,
rs2 cos2 φ rs2 sin2 φ

+ =1 (5.19)
a2 b2
Solving 5.19, rs can be found as,
a2 b2
rs = p (5.20)
a2s sin2 φ + b2 cos2 φ
131
Here, an important part of the derivation is to find the area of overlap between the
bounding and the scattering ellipses. Referring to Fig. 5.5(a), the area of overlap between
the bounding ellipse corresponding to delay τn and the scattering ellipse, A1 (τn ), can be
calculated using the following integral,

Z α1 Z π
A1 (τn ) = rs2 (φ)dφ + rτ2n (φ)dφ (5.21)
0 α1
Since α1 equals φ, when rτn = rs ; equating (5.18) and (5.20) and solving for φ, we get
α1 = cos−1 (f1 ) (5.22)
where
√
4a2 κ2 cτ d + f2 (d2 − c2 τ 2 )
f1 = (5.23)
f3
f2 = e2 (c2 τ 2 − d2 )2 + 4a2 κ2 (d2 − e2 c2 τ 2 ) (5.24)
f3 = e2 (c2 τ 2 − d2 )2 + 4a2 κ2 d2 (5.25)
and κ and e are, respectively, the aspect ratio and eccentricity of the Macro Eccentro-
Scattering disc.
After carrying out the required mathematical manipulations and parameter substitu-
tions in (5.21), we get the area of the overlapping region between the bounding ellipse
corresponding to delay τn and the scattering ellipse in Fig. 5.5(a), A1 (τn ), as follows,
µ ¶
2 −1 tan α1
A1 (τn ) = a κ tan
κ
( µ√ ¶)
c2 τ 2 (e2τn − 1) −π eτn sin α1 2 1 + eτ tan(α/2)
+ p + +p tan−1 √ n
4 1 − e2τn 1 − eτn cos α1 1 − e2τn 1 − eτn
(5.26)
132
Substituting A = πab and A1 (τn ) from (5.26) into (5.8), we get the pdf of ToA for macrocell
environment due to local scattering disc as follows,

  

 2
a κq2
f4
 1 f −  , τ0 < τn < τmax
2 5
pτ (τn ) = πa κ 2 2 2
(1 − e f1 ) 1 − f1 (5.27)



0, elsewhere.
where,
(
1 ³ p ´
f4 = √ 4a2 κ2 cd f2 − 2f2 c2 e2 τn (d2 − c2 τn2 )(d2 − c2 τn2 + 2a2 κ2 )
f3 f2
)
1 ³ 2 2 2 3 2 2 p ´
+ 2 16a κ e c τn d(d − c2 τn2 ) + 4e2 c2 τn (d2 − c2 τn2 )2 f2 (5.28)
f3
( s )
c(2c2 τ 2 − d2 ) π (cτ n + d)(1 − f 1 )
f5 = p n − tan−1
2 c2 τn2 − d2 2 (cτn − d)(1 + f1 )
p
cd2 (cτn f1 − d) 1 − f12 (c2 τn2 − d2 )2 f4
+ + p (5.29)
4(cτn − df1 )2 4(cτn − df1 )2 1 − f12
d + 2a
τmax = (5.30)
c
Assuming circular scattering disc around MS, i.e. a = b = R, as in GBSBM [3] is a special
case of our work. So, a simplified version of the pdf of ToA formula in [3] would be,
 ½ r p ¾
2 2

 c(2cpτn − d2 ) π −1 (cτn + d − 2R) (2R − cτn ) d2 − (cτn − 2R)2

 2πR2 c2 τ 2 − d2 2 − tan + ,
n
(cτn + d + 2R) 2(2c2 τn2 − d2 )
pτ (τn ) =

 τ0 < τn ≤ τmax


0, elsewhere
(5.31)
133
5.4.2 Distant Scatterers
In Fig. 5.5(b), the proposed scattering model for dominant distant scatterers in macrocell
environment is illustrated. In the figure, C, B (BS), and M (MS) are the center and
focal points of the bounding ellipse corresponding to delay τn with semi-major and semi-
minor axes aτn and bτn , respectively. The following distances are defined; dBO = |BO|,
dOM = |OM |, rBL = |BLD |, and rM L = |LD M |. O is the center of the Distant Eccentro-
Scattering disc located at the point (dBO , θD ) with semi-major and semi-minor axes aD and
bD , respectively. The location of any scattering point, LD , on the distant Eccentro-Scattering
disc with respect to B is defined as (rBL , θ) and with respect to M is defined as (rM L , φ).
The distant scattering disc intersects with the bounding ellipse corresponding to delay τn at
(x1 , y1 ) and (x2 , y2 ) as shown in Fig. 5.5(b).
The bounding ellipse corresponding to delay τn in Fig. 5.5(b) is defined in (5.17); we
can re-write it in Cartesian coordinates as,

p s µ ¶
c2 τn2 − d2 c2 τn2 d
y τn = − xτ n − (5.32)
cτn 4 2
The distant Eccentro-Scattering disc in Fig. 5.5(b) is defined as,

q
yD = dBO sin θD − κD a2D − (xD − dBO cos θD )2 (5.33)
Here also, an important part of the derivation is to find the area of overlap between
the bounding ellipse corresponding to delay τn and the distant Eccentro-Scattering disc. To
find the area of intersection, we need to find the abscissas of the two points of intersection,
134
x1 and x2 in Fig. 5.5(b). Therefore, equating (5.32) and (5.33), we obtain the two solutions
for x. Alternatively, to simplify the derivation, we will assume that distant scatterers are
confined within a circular scattering disc with radius RD . The general case of elliptical
distant scattering disc can be handled similarly; although huge expressions will be required.
We adopt the following simplified equation for a circular distant scattering disc,
q
d 2
yD = tan θD − RD − (xD − d/2)2 (5.34)
2
Solving (5.32) and (5.34) simultaneously for x = xτn = xD , we get the following two real
solutions,
1 2
x1,2 = (d ∓ cτn f6 ) (5.35)
2d
where,
r q
2 2
f6 = c2 τn2 + 4RD + (d2 − 2c2 τn2 ) sec2 θD + 2 tan θD (d2 − c2 τn2 )(4RD − c2 τn2 sec2 θD )
(5.36)
From (5.32) and (5.34), the area of intersection, A2 (τn ), would be,
Z x2
A2 (τn ) = (yτn (x) − yD (x)) dx (5.37)
x1
Equation (5.37) can be simplified to,
µq q ¶
cτn f6 p
2 2 2 2 2 2 2 2 2 2
A2 (τn ) = 4RD d − c τn f6 c τn − d d − f6
4d2
µ ¶ µ ¶
cτn p 2 2 −1 f 6 2 −1 cτ n f6 cτn f6
+ c τn − d2 sin + RD sin − tan θD (5.38)
4 d 2RD d 2
2
Substituting, A = πRD , and the value of A2 (τn ) from (5.38) in (5.8), we get the pdf of ToA
135
1.4
ToA from Local and Distant Scatterers
ToA from Local Scatterers only
ToA from Distant Scatterers only
1.2
1
Rate of Occurrence
0.8
0.6
0.4
0.2
0
3 3.5 4 4.5 5 5.5 6 6.5 7
−6
Figure 5.6: pdf of ToA for macrocells with the effect of distant scatterers
due to circular distant scattering disc as,

( µq ¶ q
1 2
pτ (τn ) = 2 2
c(f6 + τn f7 ) 2 2
4RD d − c τn f6 − d tan θD + cτn f7 (c2 τn2 − d2 )(d2 − f62 )
2 2 2
2πRD d
µ q µ ¶¶ )
c(2c2 τn2 − d2 ) f6
+ p f6 d2 − f62 + d2 sin−1 (5.39)
2 c2 τn2 − d2 d
where,
( )
c2 τn tan θD (4R 2
D + (2c2 2
τn − d2
) sec 2
θD )
f7 = 1 − 2 sec2 θD + p (5.40)
f6 2 2 2 2 2 2
(d − c τn )(4RD − c τn sec θD )2
Equation (5.39) is valid for τmin < τn ≤ τmax , where τmax and τmin are defined as,
sµ ¶ µ ¶2
2
2 d d
τmax , τmin = + tan θD ± RD (5.41)
c 2 2
The pdf of ToA due to local scatterers is given in (5.27) and the pdf of ToA due to distant
scatterers is given in (5.39). However, it should be noted that each of these two formulas
136
5.5 Conclusions
has to be applied over the corresponding region and the pdf of ToA is assumed to be zero
outside these regions. Nevertheless, when the two regions overlap, the resultant pdf of ToA
would then be the normalized sum of (5.27) and (5.39) over the common region, see Fig. 5.6.
The following model parameters were used in the simulation: d = 1000 m, a = 500 m,
e = 0.5, θD = 63.4◦ , and RD = 600 m. In several macrocell environments, e.g. hilly and
bad urban, a scattering model should consist of the distant scattering disc in addition to
local scattering disc, in order to imitate the physical scattering behavior more pragmatically.
Hence, Fig. 5.6 signifies the importance of the proposed model as it incorporates the effect
of distant scatterers in addition to that of local scatterers.
5.5 Conclusions
In this chapter, we have discussed the temporal characteristics of cellular mobile channel in
picocell, microcell, and macrocell environments. We have employed the proposed Eccentro-
Scattering model to derive the pdf of ToA of the multipath signal for these cellular environ-
ments. In macrocell environment, our model incorporated the effect of distant scatterers,
far from BS and MS on the temporal dispersion of the multipath signal in addition to that
of local scatterers. A simplified generic closed-form formula for the pdf of ToA due to local
scatterers has been derived from which previous models can be easily reproduced. For distant
scatterers, a circular scattering disc has been adopted; nevertheless the same methodology
can be used for an elliptical scattering disc or an Eccentro-Scattering disc.
137
5.5 Conclusions
The presented formulas can be used to simulate temporal dispersion of the multipath
channel in a variety of propagation conditions. Furthermore, these formulas are also help-
ful in designing efficient equalizers to combat intersymbol interference (ISI) for frequency-
selective fading channels.
138
Chapter 6
Effect of Mobile Motion on the
Spatio-Temporal Characteristics
In section 6.1 of this chapter, an overview of the previous work and chapter contributions
are presented. Section 6.2 describes the three MS motion scenarios which affect the spatial
and temporal statistics of the channel. In section 6.3, the important spatial parameters are
described, and the behavior of these parameters under MS motion is discussed in detail.
Section 6.4 presents the description of important temporal parameters and discusses their
behavior under MS motion. Finally, section 6.5, concludes the chapter and presents final
remarks.
6.1 Overview
6.1 Overview
Propagation models are usually considered low-cost and handy means of predicting accurate
radio wave propagation behavior. In order to enhance capacity, efficient exploitation of radio
channel resources [5, 14] is required. Conventional physical channel models can be considered
as stationary models, since they exclude the effects of motion on the channel characteristics.
Recently, very good work has been presented to provide dynamic channel models by consid-
ering the motion of mobile station and/or scatterers [25, 74–79]. In [14], measurements have
been presented for the elevated BS-antenna systems (rural environments), which show incon-
sistent behavior of the angle spread when an increase occurs in MS-BS separation. However,
no explanatory model has been proposed so far to explain the behavior of spatio-temporal
statistical parameters of the multipath fading channel under the effect of the displacement of
MS or BS. This motivated us to analyze the spatial and temporal statistics of mobile radio
channel and investigate their behavior under mobile motion.
The main contributions of this chapter may thus be summarized as follows:
1. We consider a realistic situation of a moving MS for the characterization of multipath
fading channel and investigate how this motion affects the spatio-temporal character-
istics at BS.
2. We present three mobile motion scenarios that are responsible for the alterations in the
spatio-temporal characteristics and plot the behavior of various spatial and temporal
140
6.2 Mobile Motion Scenarios
spread quantifiers during these motion scenarios. We also explain the behavior of angle
spread under the effect of mobile motion, observed in the field measurements [14]. We
show that the model successfully simulates the time-variability of the angle and delay
spreads induced during the course of MS motion.
3. We identify two different cases when the terrain and clutter of MS surroundings have
an additional effect on the temporal spread of the channel during MS motion. These
cases can provide good basis for the performance evaluation in those wireless systems
which employ additional time-delay processing techniques.
As we have already discussed in chapter 2 that the multipath components of a received
signal arrive at the receiver antenna from different azimuth directions about the horizon
with identical or different delays. The distributions of these multipath components in the
azimuth and time are conveniently described by the functions p(θ) and p(τ ), where θ is the
azimuthal AoA and τ is the delay of the multipath component in time domain. The work in
this chapter is based on the general assumptions made in chapter 2. Moreover, we assume
uniformly distributed scatterers in the scattering region. However, the same methodology
can also be extended to the case of Gaussian distributed scatterers.
Let MS moves from position Pi to position Pi+1 with velocity v and angle φv with x-axis
141
Next MS position
in scenario # 2
Next MS position
y Next MS position in scenario # 1
in scenario # 3
Pi+1 φv
δd
di+1 Pi
δβ
di
βi β i+1
x
BS
Figure 6.1: Illustration of motion scenarios
and traverses a small distance δd such that δd → 0, in an infinitesimal small time interval δt
in the direction of v, see Fig. 6.1. It can be shown that
δd = vδt (6.1)
This change in the displacement causes a change in the whole scattering environment around
MS. If we use subscripts i and i + 1 to represent the values of various scattering variables
at positions Pi and Pi+1 , respectively, then we can write the general expression for the next
state of MS-BS separation in terms of previous state as
q
di+1 = d2i + 2δd di cos(β − φv ) + δd2 (6.2)
The above change in the MS-BS separation strictly depends on v and φv , because v and φv
collectively define MS motion. Especially φv forces the MS-BS separation either to undergo
speedy change or keep constant. We thus consider three possible trajectories of MS with
142
respect to BS position as shown in Fig. 6.1. The motion scenarios which give rise to these
trajectories are as follows:
Scenario 1: MS is moving along the direction of LoS towards or away from BS. In this
case φv equals β and both do not change during MS motion. Obviously, the distance
between BS and MS will change linearly.
Scenario 2: MS is moving in a circular path around BS. In this case, both φv and β change
at each instant of time and φv is equal to β + δβ + π/2 during MS motion, where δβ is
an infinitesimally small interval such that δβ → 0, Fig. 6.1. Due to circular motion of
MS around BS, the distance between BS and MS remains constant.
Scenario 3: MS is neither moving along the direction of LoS nor in a circular path with
respect to BS. This is the general case. Here, φv is assumed to be constant during MS
motion for a certain specified period of time depending on the rate of data processing.
β and the distance between BS and MS change at each instant of time, see Fig. 6.1.
Note, the change in β is infinitesimally small.
143
6.3 Behavior of Spatial Characteristics under Mobile Motion
6.3 Behavior of Spatial Characteristics under Mobile
Motion
6.3.1 Important Spatial Parameters
In this section, we revisit some of the important spatial parameters already defined and
explained in chapter 4.
Angular span, θspan : This is the total width of the AoA distribution, p(θ), on the angular
domain. It can be written as,
θspan = θmax − θmin (6.3)
where θmax and θmin are the maximum and minimum angles of arrival, respectively.
Circular Variance, S0 : It is the basic measure of the dispersion of the multipath signals
in space. It is defined as,
S0 = 1 − |R̄1 | (6.4)
where R̄1 = C̄1 + j S̄1 is the first complex trigonometric moment of p(θ). The trigono-
metric parameters C̄1 and S̄1 are defined as follows,

Z 2π
1
C̄1 = p(θ) cos(θ)dθ (6.5)
P0 0
and
Z 2π
1
S̄1 = p(θ) sin(θ)dθ (6.6)
P0 0
144
R 2π
where P0 = 0
p(θ)dθ.
As we have discussed in detail in section 4.3 that S0 is invariant under the changes
in transmitted power, and any series of rotational or reflective transformation of p(θ).
We see that 0 ≤ S0 ≤ 1, which means that the observed AoAs are tightly clustered
about the mean direction θ̄; then, |R̄1 | will be closer to 1 and S0 will be nearly zero.
On the other hand, if the AoAs are widely dispersed then |R̄1 | will be small and S0
will be nearly 1.
Mean Angle of Arrival, θ̄: It can also be referred to as the mean direction of the first
complex trigonometric moment, R̄1 , and is written as,
µ ¶
−1 S̄1
θ̄ = tan (6.7)
C̄1
Angle Spread Shape Factor, Λ: It is another measure of the azimuthal dispersion of the
multipath signals, used by Durgin et al. in [32]. It is closely related to the circular
variance and the first trigonometric moment proposed in in section 4.3 as,
q
Λ = 1 − |R̄1 |2
q
= 2S0 − S02 (6.8)
Standard Deviation of the AoA, σθ : Conventionally it is the square root of the second
central moment of p(θ).

s
Z 2π
σθ = (θ − θ̄)2 p(θ)dθ (6.9)
0
145
S
rBS
rMS
θ
2b
BS C MS
d
2a
(a) Typical scattering model for microcell and picocell environments
Ci MS at Pi
BS φv
Ci+1
MS at Pi+1
Direction of motion with velocity, v
(b) Movement of scattering disc due to MS motion in scenario 3
Figure 6.2: Microcell and picocell environments
It can easily be written in terms of S0 and R̄1 as,
p
σθ = −2 ln(1 − S0 )
q
= −2 ln(|R̄1 |) (6.10)
It is the most important spatial parameter and provides the true physical information
about the angular dispersion of the data in radians.
146
6.3.2 Effect of Mobile Motion on the Spatial Characteristics of
the Channel
6.3.2.1 Picocell and Microcell Environments
In microcell and picocell environments, BS and MS are located close to each other and
both have scatterers in their vicinity. The antenna heights are relatively low and multipath
scattering is assumed near BS as well as near MS. So, according to the Eccentro-Scattering
model, BS and MS are located at the focal points of the Eccentro-Scattering disc as shown
in Fig. 6.2(a). Good examples of this model are the shopping malls, street crossings and
wireless Local Area Networks (WLANs).
In picocell and microcell environments, multipath signals arrive at BS antenna from all
directions as shown in Fig. 6.2(a). Therefore, the azimuthal angular span, θspan , can always
be considered as 360◦ .
The pdf of AoA of the multipaths at BS from all scattering points within the Eccentro-
Scattering disc, pΘ (θ), in equation (3.10), can also be written as,
(1 − e2 )3/2
pΘ (θ) = (6.11)
2π (1 − e cos(θ − β))2
Considering the three motion scenarios and equation (6.2), change in eccentricity, e, of
the Eccentro-Scattering disc can be written as,
di+1 − di
ei+1 − ei = (6.12)
2a
147
−3
x 10
9
0.035
8
0.03
0.025
6
0.02 5
pdf
pdf
4
0.015
3
0.01
0.005
1
0 0
−150 −120 −90 −60 −30 0 30 60 90 120 150 180 210 −200 −150 −100 −50 0 50 100 150 200 250
AoA, degrees AoA, degrees
(a) Micro- and picocells, Scenario 1 (b) Micro- and picocells, Scenario 2
0.018 0.07
0.016
0.06
0.014
0.05
0.012
0.04
0.01
pdf
pdf
0.008 0.03
0.006
0.02
0.004
0.01
0.002
0 0
−200 −150 −100 −50 0 50 100 150 200 250 −5 0 5 10 15 20 25 30 35 40 45 50 55 60 65
AoA, degrees Angle of Arrival, degrees
(c) Micro- and picocells, Scenario 3 (d) Macrocells, Situation 1

0.035 0.035
0.03 0.03
0.025 0.025
0.02 0.02
pdf
pdf
0.015 0.015
0.01 0.01
0.005 0.005
0 0
−20 −10 0 10 20 30 40 50 60 70 80 −60 −40 −20 0 20 40 60
Angle of Arrival, degrees Angle of Arrival, degrees
(e) Macrocells, Situation 2 (f) Macrocells, Situation 3
Figure 6.3: Effect of MS motion on the pdf of AoA, various line styles show the plots at three
different MS positions
148
Scenario 1 Scenario 2 Scenario 3
360 360 360
300 300 300 Angular Span, θspan

Angle Spread, σθ
Mean AoA
240 240 240

Angle in degrees
180 180 180
120 120 120
60 60 60
0 0 0
10 20 30 40 50 0 10 20 30 10 20 30 40
Distance travelled in meters
Figure 6.4: Behavior of θspan , θ̄ and σθ under the effect of MS motion in microcell and picocell
environments
where di and di+1 represent the MS-BS separations at positions Pi and Pi+1 , respectively,
see Fig. 6.1 and 6.2(b).
Fig. 6.3(a), 6.3(b) and 6.3(c) show the effect of MS motion on the pdf of AoA of the
multipaths at BS for microcell and picocell environments in MS motion scenarios 1, 2, and
3, respectively. The behavior of various spatial spread parameters under the effect of MS
motion is also plotted in Fig. 6.4 and 6.5.
We see that,
• During motion scenario 1, the mean value of the pdf of AoA, θ̄, remains constant while
a considerable change occurs in its SD, σθ . As MS moves away from BS, SD decreases
149
Scenario 1 Scenario 2 Scenario 3

1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7

Λ (between 0 and 1)
S0 (between 0 and 1)
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
10 20 30 40 50 0 10 20 30 10 20 30 40 50
Figure 6.5: Behavior of S0 and Λ under the effect of MS motion in microcell and picocell environ-
ments
and vice versa. Since, in microcell and picocell environments, multipath signals arrive
at BS from all directions, angular span, θspan , remains 360◦ . Similar behavior of θspan
can also be seen in other scenarios.
In this scenario, since the arrival angles tend to cluster more and more tightly about
θ̄ at each instant of time, S0 and Λ decrease sharply as shown in Fig. 6.5.
• During motion scenario 2, the mean value, θ̄, of the pdf of AoA changes at each instant
of time while no change occurs in its SD, σθ . Since shape of the pdf does not disturb,
S0 and Λ remain constant.
• During motion scenario 3, both mean value, θ̄, of the pdf of AoA and SD, σθ , change
150
at each instant of time. Such varying effect can also be seen in S0 and Λ. That is why
this scenario is referred to as the general case and scenario 1 and scenario 2 are its
special cases with φv = β and φv = β + δβ + π/2, respectively.
6.3.2.2 Macrocell Environments
In macrocell environment, the BS is usually far away from MS, i.e. the distance between BS
and MS is in the order of kilometers [14]. Rural, urban, suburban, and hilly areas are the
best examples of macrocell environments.
In macrocell environments, the multipath signals arriving at BS antenna result mainly from
two-fold phenomena, viz.
1. Local scattering
2. Distant scattering
These phenomena have been explained in detail in chapter 1 and 3.
Fig. 6.6(a) shows the Eccentro-Scattering model for typical macrocell environment where
MS is located at the center of the local scattering disc at distance d from BS while the center
of the dominant scattering disc is at distance dBD and dM D from BS and MS, respectively.
θD is the angle between the line joining the center of the dominant scattering disc to BS and
the x-axis whereas β is the angle between LoS and x-axis.
151
Q Distant Scattering
Y
Disc
Local Scattering Disc
CD
D,max rMCD b
a
rBCD
M
d
D

L,max
B X
(a) Typical scattering model for macrocell environment
Y
Q rMC D,1 rMCD,2
CD
Pi+1
rBCD
φv
θ di+1
di Pi
θD
X
B
β θL,max,i
β+δβ θL,max,i+1
(b) Movement of scattering disc due to MS motion according to scenario 3
Figure 6.6: Macrocell environment
152
The azimuthal angular spans due to local and distant scattering discs, θL,max and θD,max ,
are defined in (3.18) and (3.19), respectively.
From the geometry of Fig. 6.6(a), the total azimuthal angular span, θspan , is defined as,
θspan = θL,max + θD,max + θD − β (6.13)
The eccentricities of the local and distant Eccentro-Scattering discs, e and eD , do not
depend on d or dD but they instead depend on the terrain and clutter of the scattering
regions and can be considered as fixed values between 0 and 1. For rural areas both are
usually considered as 0s, i.e. circular scattering discs are assumed.
The pdf of AoA of the multipaths at BS defined in (3.27) can also be written as,
 q

 2
2E(1 − e ) cos(θ − β) 1 − e2 cos2 (θ − β) − E 2 sin2 (θ − β)



 ¡ ¢2 ,

 π 1 − e2 cos2 (θ − β)



 β − θL,max ≤ θ ≤ β + θL,max










 0, β + θL,max < θ < θD − θD,max
pΘ (θ) =


 ¡ ¢q

 2E cos(θ − θ ) − e2
cos θ cos θ 1 − e2D cos2 θ − ED
2
sin2 (θ − θD )


D D D
¡
D
¢2 ,

 2 2

 π 1 − e cos θ


D








 0, elsewhere
(6.14)
where E = d/a and ED = dBD /aD are important ratios, which help in designing directional
antennas for local and dominant distant scattering regions especially in rural areas.
153
Situation 1 Situation 2 Situation 3

70 80 120
70
60 100
Angular Span, θspan
Angle Spread, σθ
60
Mean AoA
50 80
50
Angle in degrees
40 60
40
30 40
30
20 20
20
10 10 0
40 60 80 100 40 60 80 100 40 60 80 100
Figure 6.7: Behavior of θspan , θ̄ and σθ under the effect of MS motion in macrocell environment
Let MS moves from position Pi to position Pi+1 traversing a small distance δd as given
in (6.1); (also see Fig. 6.6(b)). The distance between BS and MS changes from di to di+1
according to the general expression given in (6.2). Consequently, E in (6.14) changes from
Ei to Ei+1 as,
2
Ei+1 = Ei2 + 2δE Ei cos(β − φv ) + δE2 (6.15)
where δE = δd /a = vδt /a. Since distant scattering clusters are usually the sets of stationary
structures like hills and high-rise buildings in towns and villages, dBD remains constant dur-
ing the course of MS motion. Therefore, change occurs only in local scattering parameters.
However, the dominant distant scattering disc has an immanent influence on the angular
spread of the multipath signals at BS in macrocell environment. Therefore, the effect of
154
mobile motion on the angular spread due to dominant distant scattering must be considered
in addition to the aftermath of the motion scenarios given in section 6.2. We can classify
the behavior of spatial characteristics of mobile channel under the combined effect of mobile
motion and distant scattering disc into the following three situations.
Situation 1:
In this situation, the mobile moves towards (or away from) a dominant distant scattering
cluster (like a town, a village or a hill) in the direction of line of sight (LoS) with BS. This
situation usually develops during motion scenario 1 given in section 6.2. This situation is
depicted in Fig. 6.8(a).
In this situation, d changes with respect to v, but almost no change occurs in β. On
the other hand, angle spread, σθ , decreases (or increases) slightly but angular span, θspan ,
and mean AoA, θ̄, remain constant.
Situation 2:
In this situation, the mobile moves in a circular or quasi-circular path with respect to BS,
and passes ahead or behind a dominant distant scattering cluster. This situation is usually
developed during motion scenario 2 given in section 6.2. This situation is elaborated in
Fig. 6.8(b).
155
CD
BS MS
(a) Situation 1
BS CD
MS
(b) Situation 2
MS
CD
BS
(c) Situation 3
Figure 6.8: Collective effect of distant scattering cluster and MS motion on the spatial spread of
cellular mobile channel in macrocell environment
156
Situation 1 Situation 2 Situation 3

0.35 0.35 0.7
Λ (between 0 and 1)
S0 (between 0 and 1)
0.3 0.3 0.6
0.25 0.25 0.5
0.2 0.2 0.4
0.15 0.15 0.3
0.1 0.1 0.2
0.05 0.05 0.1
0 0 0
40 60 80 100 40 60 80 100 40 60 80 100
Figure 6.9: Behavior of S0 and Λ under the effect of MS motion in macrocell environment
In this situation, d does not change (or if it changes, the change is usually very minor),
but major change occurs in the value of β, proportional to the distance travelled by MS.
Thus in this situation, angle spread, σθ , and angular span, θspan , first decrease as MS comes
closer to the distant scattering cluster, stay fixed for some time proportional to the radius
of cluster and then increase again. Mean AoA, θ̄, also changes continuously.
Situation 3:
In this situation, neither MS travels in the direction of LoS nor in a circular path with
respect to BS. It follows the motion scenario 3 given in section 6.2 and passes alongside the
157
dominant distant scattering cluster. This situation is illustrated in Fig. 6.8(c).
In this situation, d and β both change at each instant of time in proportional to the
velocity, v, of the MS. On the other hand, angle spread, σθ , and angular span, θspan , both
decrease as MS comes closer to the distant scattering cluster as seen from BS, but they
increase when the distance between MS and distant scattering cluster increases. The effect
of increasing or decreasing d on σθ is mainly dependent on the closeness of the distant
scattering cluster from the MS. Mean AoA, θ̄, also does not remain constant and alters with
respect to the direction of motion of MS.
This situation is observed very often in rural and hilly areas. Thus the smart antenna
system employed at BS must be designed accordingly in order to guarantee the reliability of
service to the desired users.
The behavior of various spatial spread parameters for a moving mobile under the above
situations is shown in Fig. 6.7 and 6.9. The effect of mobile motion on the pdf of AoA under
these situations is also shown in Fig. 6.3(d), 6.3(e) and 6.3(f).
Behavior of angle spread in rural areas, plotted in Fig. 3.2.5 of COST 259 [14] can easily
be explained on the basis of above-mentioned situations. This behavior strictly depends on
the angle of MS motion, φv , and the location of distant scattering cluster. Therefore, it is
always inconsistent with the increase or decrease in MS-BS separation.
158
6.4 Behavior of Temporal Characteristics under Mobile Motion
6.4 Behavior of Temporal Characteristics under Mo-
bile Motion
6.4.1 Temporal Channel Model
Based on the assumptions given in section 2.2 of chapter 2, an nth bounding ellipse defines a
specific set of scatterers that give rise to multipath components arriving between time delays
τn and τn+1 = τn + δτ . A temporal model along with all employed parameters has been
explained in detail in section 5.2 of chapter 5, (also see Fig 5.1).
Let a specific time delay τn define a set of scatterers bounded by nth bounding ellipse.
Then the semi-major axis aτn and semi-minor axis bτn of the nth bounding ellipse can be
written as,
cτn
aτn = (6.16)
2
1q 2
bτn = 4aτn − d2 (6.17)
2
As we mentioned earlier in chapter 5, we denote the semi-major axis and eccentricity of
the outermost bounding ellipse by the notations aτmax and eτmax , respectively. Both variables
are related to the maximum delay spread, τmax . Use of aτn should not be confused with the
notation a in section 6.3, where a denotes the semi-major axis of a fixed Eccentro-Scattering
disc in a spatial channel model. However, aτmax and eτmax are equivalent to a and e of the
Eccentro-Scattering spatial channel model in the case of picocell and microcell environments,
159
Received
Signal Power
Threshold
0 max
Excess Delay, sec
Figure 6.10: A general power delay profile
but not equivalent to a and e of the Eccentro-Scattering spatial channel model in macrocell
environment. This is due to the fact that a and e of the Eccentro-Scattering disc in case
of picocell and microcell environments depend on τmax , while being independent of τmax in
macrocell environment, they depend only on the terrain and orientation of the streets, roads
or valleys.
160
6.4.2 Some Important Temporal Parameters
Delay in LoS, τ0 : This is the time at which signal travels between BS and MS in LoS,
whether LoS exists or not. That is,
d
τ0 = (6.18)
c
Mean Excess Delay, τ̄ : This is the first moment of the power-delay profile, P (τ ), with
respect to delay in LoS, τ0 . That is [80],

Z
τ̄ = (τn − τ0 )P (τn )dτn (6.19)
Maximum Delay, τmax : This can be specified as the time delay for which the power of
the multipath component, P (τn ), falls below a specific power level, referred to as the
threshold value. When the signal level is lower than the threshold, it is passed as
noise [24].
Maximum Excess Delay, τe : This is the time difference between the maximum delay,
τmax , and the delay in LoS, τ0 , [24]. That is,
τe = τmax − τ0 (6.20)
rms Delay Spread, στ : This is the square root of the second central moment of the power-
delay profile, P (τn ). This can also be referred to as the standard deviation of P (τn )
about the mean excess delay, τ̄ . That is, [8]

sZ
στ = (τn − τ0 − τ̄ )2 P (τn )dτn (6.21)
161
Pi+1
B Ci Ci+1 Pi
I.
2 aτ max
2 aτ max
(a) Case 1: aτmax remains constant and eτmax changes
B Ci Ci+1 Pi Pi+1
I.
2 aτ max , i
(
2 aτ max , i +1 = 2 aτ max , i + δ a )
(b) Case 2: aτmax changes and eτmax remains constant
Figure 6.11: Illustration of the lifetime of scatterers
The effect of temporal dispersion on the performance of a digital receiver can be reliably
related to the rms delay spread, independently of the shape of power-delay profile [80].
These time-delay spread parameters are illustrated in Fig. 6.10.
6.4.3 Lifetime of Scatterers
As MS travels, the scattering environment changes such that some scatterers move out of the
scattering disc and some others move in; as a result new multipath components contribute
to the received signal. This phenomenon is referred to as multipath generation [81]. The
162
distance between BS and MS, d, and maximum delay, τmax , collectively form a bounding
ellipse with parameters eτmax and aτmax as given in (5.3) and (6.16). This bounding ellipse
defines the outer boundary of the scattering points associated with time delays up to τmax .
Thus, MS motion gives rise to the phenomenon of multipath generation as well as to an
alteration in the shape of the bounding ellipse.
According to terrain and clutter of the surroundings, MS motion affects the bounding
ellipse in one of the following two cases:
6.4.3.1 Case 1
This case defines a situation where τmax is kept fixed during the course of MS motion, see
Fig. 6.11(a). Namely, as MS moves it loses some scatterers but does not attain new influential
ones that can affect the temporal spread of the received signal. In other words, it can be said
that the lifetime of multipath components has finished for the present cell and MS should
be handed over, immediately, to a new BS in order to guarantee continuous reception. In
such a case, τmax and consequently aτmax do not change for the current cell while eτmax and
bτmax change due to change in d. Also, τ0 and στ incur a change in their values. Considering
three motion scenarios and (6.2), change in eτmax can be written as,
di+1 − di
eτmax , i+1 − eτmax , i = (6.22)
c τmax
In this case, eccentricity of the outer bounding ellipse, eτmax , takes a value in the range
163
Received Signal
Power
e,2
e,1
0,1 0,2 Fixed max
Excess Delay, sec
(a) Case 1
Received Signal
Power
Fixed e
Fixed e
Threshold
0,1 0,2 max,1 max,2 ex

Excess Delay, sec
(b) Case 2
Figure 6.12: Behavior of power delay profile under the effect of MS motion
164
between 0 and 1 during the course of MS motion. Hence, this value increases as the distance
between BS and MS increases such that eτmax approaches 1 at the edge of the cell where the
signal power falls lower than the threshold value. Before eτmax approaches 1, MS should be
handed over to the new cell and it comes under the influence of the new cell’s BS. Therefore,
in this case handover depends either on the power threshold or on the situation that τ0
approaches τmax (i.e. eτmax approaches 1 due to the terrain and clutter of the surroundings).
Effect of MS motion on the pdf of ToA and other statistical parameters has been illustrated
in Fig. 6.12(a) for this case.
6.4.3.2 Case 2
This case defines a situation where στ or τe is kept fixed during the course of MS motion as
long as MS remains in the same cell. In this situation, both τ0 and τmax change according
to the change in d, but the eccentricity of the outer bounding ellipse, eτmax , does not incur
change in its value. Instead, the semi-major axis of the outer bounding ellipse, aτmax , which
is a measure of τmax , changes as shown in Fig. 6.11(b). Considering equation (6.2) and the
three motion scenarios, change in aτmax can be written as,
di+1 − di
aτmax , i+1 − aτmax , i = (6.23)
2 eτmax
In this case, MS attains more and more scatterers from the surroundings, in addition to the
existing ones at each instant of time. Here, aτmax gets a change in its value between R0 and
cτex
2
; where, τex is the extreme maximum delay at which the signal power drops below the
165
M
Pi
B Ci φv
Ci+1
Pi+1
(a)
M R
B Ci
Pi φv
Ci+1
Pi+1
(b)
Figure 6.13: Movement of scattering discs due to MS motion in (a) picocells and microcells (b)
macrocells
threshold value and MS is then handed over to a new BS; and, R0 is an arbitrary constant
value depending on the close-in-reference distance which is the minimum practical separation
between BS and MS. Effect of MS motion on the pdf of ToA and other statistical parameters
has been illustrated in Fig. 6.12(b) for this case.
166
−7
x 10
σ , case 2
τ
τ , case 2
e
6 τ , case 2
0
Mean ToA, case 2
τ , case 2
max
στ, case 1
5 τe, case 1
τ0, case 1
Mean ToA, case 1
τmax, case 1
time−delay in seconds
0
100 102 104 106 108 110 112 114 116 118
distance in meters
Figure 6.14: Behavior of στ , τe , τ0 , τ̄ and τmax in picocells and microcells under the effect of MS
motion when lifetime of scatterers is based on case 1 and case 2
6.4.4 Effect of Mobile Motion on ToA Statistics
6.4.4.1 Picocell and Microcell Environments
In picocell and microcell environments, the Eccentro-Scattering disc covers the whole area
bounded by the outer bounding ellipse, as shown in Fig. 6.13(a). The pdf of ToA of the
multipaths at BS from all scattering points within the outer bounding ellipse or Eccentro-
Scattering disc of picocell/microcell environments, p(τn ), in (5.16) can be re-written as,
c(2c2 τn2 − d2 )
pτ (τn ) = p p (6.24)
4a2τmax 1 − e2τmax c2 τn2 − d2
167
(a) Mico/Pico, Case 1 (b) Micro/Pico, Case 2

200 30
pdf of ToA at P
1
25 pdf of ToA at P
2
150 pdf of ToA at P3
20
pdf
pdf
100 15
10
50
5
0 0
0.3 0.4 0.5 0.6 0.7 0.3 0.4 0.5 0.6 0.7
Time−delay in µ sec Time−delay in µ sec
(c) Macro, Case 1 (d) Macro, Case 2

40 20
30 15
pdf
pdf
20 10
10 5
0 0
3.2 3.4 3.6 3.8 4 3 3.5 4 4.5
Time−delay in µ sec Time−delay in µ sec
Figure 6.15: Effect of MS motion on the pdf of ToA; various line styles show the plots at different
MS positions
Here, the effect of motion on the temporal statistics of the channel can be explained on the
basis of the lifetime of scatterers discussed earlier.
Case 1: In this case, as mentioned earlier, τmax is kept fixed during the course of MS motion
while the eccentricity of the outer bounding ellipse, eτmax , changes due to the change in
d according to (6.2) and (6.22). Therefore, the semi-major axis of the outer bounding
ellipse, aτmax , remains constant. Since τmax is kept fixed, so with changing d, the values
of τ0 , τe , and στ change. Fig. 6.15(a) represents the plots of the pdf of ToA at various
MS positions.
168
Case 2: In this case, as mentioned earlier, στ , or τe is kept fixed during the course of
MS motion while both τ0 and τmax change. Therefore, the eccentricity of the outer
bounding ellipse, eτmax , also remains constant and a change occurs in aτmax according
to (6.23), due to the change in d according to (6.2). In other words, the shape of the
pdf of ToA does not alter as a result of MS motion. Fig. 6.15(b) represents the plots
of the pdf of ToA at various MS positions for this case.
Fig. 6.14 shows the behavior of various temporal spread parameters under the effect of
motion in picocell and microcell environments.
6.4.4.2 Macrocell Environments
In macrocell environments, the outer bounding ellipse intersects some part of the scattering
disc around MS as shown in Fig. 6.13(b). As MS moves from position Pi to position Pi+1 ,
it re-acquires almost the same scattering environment at the new position without having
any change in its shape. However, since change occurs in MS-BS separation, d, temporal
spread parameters are significantly affected. Let R be the effective radius of the scattering
disc around MS and eτmax be the eccentricity of the outer bounding ellipse. The pdf of ToA
of the multipaths at BS from all scattering points within the outer bounding ellipse, p(τn ),
in (5.31) for macrocell environment can also be written as,

" s p #
c(2r2 − e2τmax ) π e τ + (r − m) (r − m) e 2 − (r − m)2
pτ (τn ) = p − tan−1 max
− ¡ τmax ¢
πmR r2 − e2τmax 2 eτmax − (r − m) 2 2r2 − e2τmax
(6.25)
169
−6
x 10
4.5
3.5
3 σ , case 1
τ
τ , case 1
time−delay in seconds
e
τ , case 1
0
2.5
Mean ToA, case 1
τ , case 1
max
στ, case 2
2
τe, case 2
τ0, case 2
1.5 Mean ToA, case 2
τmax, case 2
0.5
0
1000 1050 1100 1150
distance in meters
Figure 6.16: Behavior of στ , τe , τ0 , τ̄ and τmax in macrocells under the effect of MS motion when
lifetime of scatterers is based on case 1 and case 2
where r = τn /τmax and m = R/aτmax .
The effect of change in MS-BS separation on temporal spread parameters in macrocell
environment can also be discussed on the basis of the lifetime of scatterers. In case 1, τmax
is kept fixed during the course of MS motion while the eccentricity of the outer bounding
ellipse, eτmax , changes due to the change in d. Likewise, the values of τ0 , τe , and στ also
change as shown in Fig. 6.16. Also, Fig. 6.15(c) represents the plots of the pdf of ToA at
various MS positions.
On the other hand, in case 2, στ , or τe is kept fixed during the course of MS motion
while both τ0 and τmax are allowed to change. Thus, the shape of the pdf of ToA does not
170
6.5 Conclusion
alter as a result of MS motion. Fig. 6.15(d) represents the plots of the pdf of ToA at various
MS positions for case 2. Fig. 6.16 shows the behavior of various temporal spread parameters
under the effect of motion in macrocell environments.
Comparing Fig. 6.14 and Fig. 6.16, we observe that the values of τe and στ in macrocell
environments are very small as compared to those of τ0 and τmax . This is because of large
BS-MS separations and concentrations of scatterers in the vicinity of MS. It means that
the approach of keeping στ or τe fixed (i.e. case 2) is the most appropriate for macrocells.
Otherwise, temporal array processing receivers like Rake receivers cannot offer outstanding
performance. On the other hand, in picocell and microcell environments, both approaches
(case 1 and case 2) are equally applicable and either of them can be used, according to the
scattering situation (terrain) in the vicinity of BS and MS.
6.5 Conclusion
In time-varying fading channels, a direct representation of the distribution of AoA, being a
prerequisite in calculating correlation coefficients, is needed to assess the potential of beam-
oriented algorithms such as Space Division Multiple Access (SDMA) [23, 82]. On the other
hand, besides the handy use of the delay spread in determining the coherence bandwidth
of a particular system [22], it is the main cause of frequency selective fading. It has also
attracted the attention of many researchers in order to design efficient receivers in wireless
171
6.5 Conclusion
mobile communications. Since mobile motion is responsible for the time variability of cellular
mobile channel, it is necessary to investigate its effect on the spatial and temporal statistics
of the cellular mobile channel. In this chapter, we have discussed several motion scenarios
that are responsible for such an effect.
We have formulated the changes in the AoA and ToA distributions of the multipath
signals at BS during the course of MS motion for different cellular environments. We have
plotted the behavior of all important spatial and temporal statistical parameters under the
effect of mobile motion. The proposed theoretical results in spatial characteristics can be
extended to characterizing and tracking transient behavior of Doppler spread in time-varying
fast fading channel; likewise the proposed theoretical results in temporal characteristics can
be utilized in designing efficient equalizers for combating inter-symbol interference (ISI) in
time-varying frequency-selective fading channels.
172
Chapter 7
Fast Fading Channel Modeling for
Single-Carrier DS-CDMA Systems
In this chapter, section 7.1 presents a comprehensive overview of adaptive multiuser detection
in single-carrier DS-CDMA system. In section 7.2, a state-space model is developed on the
basis of the statistics of time-varying multipath fading channel. Section 7.3 presents the
Kalman filter based adaptive detection scheme for DS-CDMA system, which utilizes an
autoregressive (AR) model for signature sequence estimation. Section 7.4 summarizes the
chapter and presents concluding remarks.

7.1 Overview
7.1 Overview
7.1.1 History of DS-CDMA Detectors
In a code division multiple access spread spectrum (CDMA) system a number of users
simultaneously transmit information over a common channel using different code sequences
referred to as signatures. The receiver usually knows the assigned signatures and correlates
them with the received signal. If the assigned signatures are orthogonal, then a bank of
decoupled single-user detectors consisting of matched filters followed by threshold detectors
can achieve optimum demodulation [83, 84].
In most of the CDMA systems of practical importance, transmitters send information
independently. Therefore, signals from different users arrive asynchronously at the receiver.
Since their relative time delays are arbitrary, the cross-correlation between the received
signals coming from different users is non-zero. To achieve a low level of interference the
assigned signatures need to have low cross-correlations for all relative time delays. A low
cross-correlation between signatures is obtained by designing a set of mutually orthogonal
signatures [83]. In practice, however, it is difficult to keep the signatures orthogonal in
synchronous systems due to many impairments of the channel.
The presence of multiple access interference (MAI) has been considered as the major
limiting factor in CDMA systems. Many techniques have been proposed to decode users
successfully in the presence of MAI [84–90], while addressing CDMA demodulation problem.
174
7.1 Overview
The matched filter (MF) receiver demodulates the received signal with the use of a single-
user detector which consists of a matched filter followed by a threshold detector [84]. In
MF receivers, the MAI is modeled as additive white Gaussian noise (AWGN), ignoring
its cyclo-stationary character [84] (especially in case of short-code CDMA). An alternative
receiver structure is a maximum likelihood multiuser receiver (MLMR) for synchronous and
asynchronous transmission. The MLMR consists of a bank of matched filters followed by a
Viterbi maximum likelihood detector [91]. The computational complexity of this optimum
demodulator increases exponentially with the number of users [92].
The concept of interference suppression by exploiting its cyclo-stationary nature has
been used elsewhere, notably in [93], with applications to digital subscriber lines. MAI
cancellation has been used properly in systems where MAI is demodulated and subtracted
from the received signal prior to the user of interest signal detection. This method of MAI
detection and cancellation is combined with an adaptive antenna array in [94]. The concept
is critically dependent on successful MAI demodulation [85]. A significant improvement in
exploiting the cyclo-stationary nature of MAI which led to signature sequence adaptation
is obtained in a number of adaptive receiver structures which are very effective in MAI
cancellation and are moderately complex [85]. An adaptive filter is necessary to handle
time varying system parameters. But it is shown that the adaptive linear receiver is able
to completely remove the effect of multipath propagation provided the multipath parameter
variations are slower than the adaptive algorithm convergence speed.
175
7.1 Overview
A general approach to the design of multiuser systems is based on receiver optimization.
Due to the exponential complexity of MLMR, suboptimum multiuser detection has been an
active area of research. As an alternative to receiver optimization for synchronous multiuser
channels, transmitter precoding has been recently proposed [86, 95–97]. So considering the
good behavior of linear adaptive receiver in MAI cancellation, an improvement was made in
the same technique by shifting the bulk of processing to the transmitter and making it jointly
adaptive with the receiver [86]. With this joint adaptation of transmitter and receiver, better
MAI suppression is achieved and consequently a larger number of independent users were
accommodated per unit of bandwidth relative to the systems with adaptive receivers only.
The concept of the joint adaptive transmitter-receiver is based on feedback information from
the corresponding receiver. The information obtained from the receiver is used to calculate
the optimum signature for the respective transmitter. The signatures are adaptively adjusted
according to the MSE criterion of optimality during the training period as well as during
the data transmission [86].
The joint transmitter-receiver optimization as well as transmitter precoding relies on
channel knowledge. When the system operates in the time-duplex mode, the channels’
responses can be estimated at the base station. Otherwise, channel estimates obtained by
mobile receivers have to be transmitted to the base via feedback channels. It is sufficient
to transmit channel estimates once per packet. The crucial assumptions for this scheme to
work are that channel characteristics remain constant over the block of precoded bits and
the transmitter has knowledge (at least sufficiently accurate) about all multipath profiles.
176
7.1 Overview
Selection of an ensemble of signature sequences that minimize total interference power with
matched filter (MF) receivers is considered in [98, 99].
Transmitter precoding assumes that the symbols of the users on the downlink are al-
ready modulated with some linearly independent signature sequences, and essentially undoes
the cross correlation introduced by the non-orthogonality of these signature sequences. A
drawback associated with transmitter adaptation in general is the feedback bandwidth re-
quired, which increases with the number of transmitter coefficients to be estimated [90].
It means that the feedback transmission of filter coefficients to the transmitter must not
require excessive bandwidth. These concerns need to be addressed for signature optimiza-
tion to be useful in practical systems. An improvement to tackle this problem is presented
in [90]. A ‘reduced-rank’ transmitter adaptation scheme, in which each signature sequence
is constrained to lie in a lower dimensional subspace, spanned by some orthogonal basis, is
presented in [90]. The weights for the basis are then selected to optimize the performance cri-
terion, the output signal-to-interference-plus noise ratio (SINR). Different orthogonal bases
are assigned to different users. Selection of the subspace dimension allows a tradeoff between
the number of parameters to be estimated and steady-state performance. Their results show
that adaptive receivers based on a least squares performance criterion generally perform
worse than the non-adaptive RAKE, or MF receiver, since the adaptive receiver introduces
tracking error when the signatures are time-varying [90].
As far as the type of signature waveform or ’code’ is concerned, it also affects the design
177
7.1 Overview
and performance of DS-CDMA detector. There are two types of codes commonly used in
DS-CDMA. They are:
1. Short or Repeated code
2. Long or Non-repeated code
The use of short code in DS-CDMA systems gives rise to cyclo-stationary MAI which can be
exploited in adaptive detection as discussed earlier; whereas long code is usually difficult to
estimate with gradient-based algorithms over time-varying channels. However, MAI in the
case of long code, is usually assumed as Gaussian, which simplifies the receiver structure.
The use of short or repeated signature sequences in DS-CDMA is shown to be of great
advantage for avoiding interference in adaptive techniques for multiuser detection [85, 86, 90].
The receiver structure in [85] uses short codes and requires no knowledge of the transmitted
signature sequences. Instead, it requires training bits and symbol timing of the desired
user. The minimization of the external information requirement, whether in form of training
sequence or knowledge of signature waveforms, has been a hot area of research in recent years.
In the adaptive operations, some external information, (sometimes in the form of training
sequences) must be sent not only during the startup period but also after sudden changes
in the channel response, or when a new active user appears [88, 89]. Especially in highly
time-varying channels or fast fading environment the need to retransmit training sequence is
very necessary otherwise performance of the multiuser detector degrades significantly. This
178
7.1 Overview
additional burden exploits a large part of the bandwidth resources with no beneficial use.
Thus a large effort has been made in eliminating the need for training sequences by proposing
blind adaptive multiuser detection [88, 89, 100].
7.1.2 Time-Varying Nature of the DS-CDMA Channel
The classical approaches [85, 89] used so far for signature adaptation suppose that either
multipath multiuser channels are fixed (whether known or unknown) or are slowly time-
varying. But in real life mobile channels are highly time-varying and their time-varying
nature can be interpreted as a dynamic system with uncertainties in its coefficients [35]. As
we know the mobile communication channel introduces a variety of transmission impairments
including multipath fading, noise and interference. Many of the signal processing algorithms
used to mitigate the multipath and fading aspects of the mobile channel require accurate
knowledge or estimates of the channel state [87].
In DS-CDMA systems, accurate detection of transmitted symbols depends mainly on
the accurate estimation of received code waveforms. The performance of adaptive multiuser
detector degrades significantly if signature estimates are not accurate. This lack of accuracy
can be due to high level of interference caused by non-ideal cross-correlation properties of
time-varying signatures of different users. At high vehicle speeds (> 100 Km/h), the channel
conditions change significantly. So the signatures of different users in the system get highly
correlated and the performance of the system degrades. To combat this dynamic nature of
179
7.1 Overview
time-varying channels and other fading effects, state-space approach has been proposed as a
good technique [34, 101].
7.1.3 Contributions of the Chapter
Autoregressive (AR) model is usually considered the simplest way to capture the dynamic
nature of time-varying flat fading channel [102–107]. We thus develop an AR model of
time-varying flat fading channel by using its second order fading statistics and utilize it fur-
ther to establish a linear state-space equation pair for signature sequence estimation in direct
sequence code division multiple access (DS-CDMA) system. We then exploit the Kalman fil-
tering approach to incorporate our proposed state-space equation pair in an algorithm, meant
for estimating channel-distorted received signature sequences. It is an established fact [34]
that the Kalman filter is a good optimal linear minimum mean squared error (MMSE) de-
tector if a first order linear state-space model is applied to DS-CDMA system. As mentioned
earlier, we also use the Kalman filter as the MMSE solution for signature distortions caused
by the time-varying fading channel. However, different from the model used in [34], where the
unknown transmitted symbol vector has been used as state-vector, we use channel-distorted
received signature vector as the state-vector in our model. This approach is based on the fact
that the time-varying channel behaves as the AR model depending on its past values [23, 35].
In our proposed receiver structure, the need for training sequence is bound to the startup
period only. Later on, the receiver adapts itself to the changes of the channel during data
180
7.2 State-Space Approach in Multipath Fading Channel Modeling
transmission depending on previous decisions. Simulation results show that being based on
the Kalman filter and of non-gradient nature, our proposed algorithm combats effectively
the impairments and fading effects caused by time-varying multipath channel.
7.2 State-Space Approach in Multipath Fading Chan-
nel Modeling
The performance of a wireless communication system strictly depends on the extent of the
prior knowledge of the time-varying fading channels [20], which in turn depends on the fad-
ing statistics utilized in the channel estimator. There is a lot of work on designing optimum
and sub-optimum channel estimators like the Kalman trackers [106, 107] or Wiener Least
Mean Square (WLMS) predictors [108], that utilize the fading statistics of cellular mobile
channel. State-space model is the best way to develop a channel estimator based on the
valid assumption that time-varying fading channel is Markovian in nature. Therefore, the
varying part of the channel is modeled either Autoregressive (AR) [102, 104], or Autoregres-
sive Moving-Average (ARMA) [109], based on Jakes-Reudink autocorrelation function [23].
There is also a need for a highly time-selective fast fading channel model, which can become
a basis for optimum channel estimators for highly spatio-resolution systems like multi-input
multi-output (MIMO) and space division multiplexing (SDMA) systems. Our motivation
to present this section is thus to develop a state-space model based on the statistics of
181
time-varying fast fading channels, for both the dynamic Rayleigh and Ricean processes.
7.2.1 Characterization of Multipath Fading Channels
Consider the equivalent low-pass signal received over a time-variant fading multipath channel
[67]
X
r (t) = αl (t) e−2 j π fc τl (t) s (t − τl (t)) + n(t) (7.1)
l
where s(t) is the equivalent low-pass transmitted signal, fc is the carrier frequency, αl is the
attenuation factor for the signal received on lth path, τl (t) is the propagation delay for the
lth path and n(t) is the additive white Gaussian noise with zero mean and σn2 variance.
Therefore, time-variant impulse response h(τ ; t) of the fading multipath channel comprised
of L + 1 paths can be written as,

L
X
h(τ ; t) = αl (t)e−2 j π fc τl (t) δ(τ − τl (t)) (7.2)
l=0
h(τ ; t) is a complex-valued Gaussian random process in the t variable, and the amplitude
variations in the received signal (7.1) are due to the time-variant multipath characteristics
of h(τ ; t). When the impulse response h(τ ; t) is modeled as a zero-mean complex-valued
Gaussian process, the envelope |h(τ ; t)| at any instant t is Rayleigh-distributed. In the event
that there are fixed scatterers or signal reflectors in the medium, in addition to randomly
moving scatterers, |h(τ ; t)| has a Rice distribution [67]. Furthermore, if the differential path
delays τi − τj are small compared to the duration of a modulated symbol; then the received
signal can be considered to exhibit flat-fading and channel can be written as independent of
182
τ . Writing the channel h(τ ; t) as the combination of fixed and time-varying parts [110], we
get,
r L
Kt 1 X
h(t) = α0 (t)e−2 j π fc τ0 (t) + √ lim αl (t)e−2 j π fc τl (t)
Kt + 1 Kt + 1 M →∞ l=1
r L
Kt j(2πfD cos(ψ0 )t+ϕ0 ) 1 1 X j(2πfD cos(ψl )t+ϕl )
= e +√ lim √ ρl e (7.3)
Kt + 1 Kt + 1 L→∞ L
| {z } | l=1
{z }
h (t)c
ht (t)
where,
• hc (t) and ht (t) are the line of sight (LoS) and diffuse components of the time-varying
channel h(t).
• Kt = E{|hc (t)|2 /|ht (t)|2 } is the ratio of the LoS component’s power to that of the
diffuse component and is known as the Ricean factor [110].
• fD is the maximum Doppler spread of the channel and can be defined as fD = v/λ,
where v is the velocity of the mobile, and λ is the wavelength of the signal.
• ψi , i = 1, ..., L is the angle between the ith incoming multipath signal and the mobile
direction; the subscript i = 0 denotes the LoS component.
• ϕi , i = 1, ..., L is the phase angle due to the ith incoming multipath signal and is equal
to −2πfc τi (t). In the absence of LoS component, i.e. in a Rayleigh channel, the phases
ϕi are uniformly distributed on [−π, π].
L in Equation (7.3) can be made arbitrarily large enough to ensure ht (t) to be a complex-
valued Gaussian process as discussed earlier. According to Bello’s WSSUS model [107, 111],
183
y Scattering Point
Direction of
MS Motion
ψ
Incoming
Multipath
φv
θ φ
Line of Sight x
BS MS
Figure 7.1: Typical multipath fading environment
(i)
all the channel multipaths are independent; therefore, each term ht , i = 1, ..., L, in the
summation of time-varying part ht in (7.3) is a zero-mean, wide-sense-stationary complex

(l6=i)
Gaussian process, uncorrelated with any other term ht . So the autocorrelation function
R(∆t) of the channel h for the time-lag ∆t can be written as,
R(∆t) = E{h(t)h∗ (t + ∆t)} (7.4)
where E{·} represents the expectation and x∗ is the complex conjugate of a complex-valued
variable x.
Using (7.3), we get [110]

Z π
Kt −j2πfD cos(ψ0 )∆t 1
R(∆t) = e + p(ψ)e−j2πfD cos(ψ)∆t dψ (7.5)
Kt + 1 Kt + 1 −π
where p(ψ) is the distribution of angle of arrival of incoming multipaths with respect to the
direction of MS motion, φv , as shown in Fig. 7.1. It is evident from Equation (7.5) that
R(∆t) depends on p(ψ). Let φ be the angle of arrival (AoA) of the multipath signal at MS
184
Direction of
MS Motion
S ψ
φ
θ
Q 2b
B C M
d
2a
Figure 7.2: Eccentro-Scattering model for picocell and microcell environments
with respect to the x-axis as shown in Fig. 7.1. Then the distribution of the angle of arrival
at MS, p(φ), can be defined for various cellular environments as follows:
Picocell and Microcell Environments: As discussed in chapter 2, the Eccentro-Scattering
model for picocell and microcell environments consists of an elliptical scattering disc
with BS and MS on its foci, as shown in Fig. 7.2. In these environments, scatterers are
usually uniformly distributed. Due to the symmetry of the focal points of the ellipse
with respect to its center, the distribution of AoA at MS will be the same as that at
BS given in (3.9). Therefore,
(4a2 − d2 )3/2
pΦ (φ) = (7.6)
4πa(2a − d cos φ)2
where a is the semi-major axis of the pico/microcell Eccentro-Scattering disc and
d is the MS-BS separation. In these environments, the LoS component or specular
component, hc (t), usually exists.
Macrocell Environments In macrocell environments, the MS is usually surrounded by
185
ψ Direction of
MS Motion
S
φ φv
BS d MS a
b
Figure 7.3: Eccentro-Scattering model for macrocell environments
uniformly or Gaussian distributed scatterers in an Eccentro-Scattering disc as shown
in Fig. 7.3. The BS in macrocells is usually free from local scatterers [13]. Considering
the geometry of the scattering disc, the distribution of AoA at MS is given as,
ab
pΦ (φ) = 2 (7.7)
2π(a2 sin φ + b2 cos2 φ)
where a and b are the semi-major and semi-minor axes of the macrocell Eccentro-
Scattering disc. In flat-rural macrocell areas, the Eccentro-Scattering disc becomes
circular or quasi-circular and the distribution of AoA at MS becomes uniform. Hence,
1
pΦ (φ) = (7.8)
2π
Since in picocell and microcell environments the velocity of MS is very low, there is no
considerable Doppler effect on the carrier frequency. Hence, channel can be considered as
time-invariant for the duration of at least one data frame. Therefore, there is no need of
high-complexity channel estimation algorithms, which are usually meant for rapidly time-
varying channels. An ordinary channel estimator utilizing low-complexity LMS algorithm
186
can be employed at either of two wireless link ends. The main goal of our research work in
this section is to model a fast fading channel estimator that combats the rapid variations of
the channel. Thus, we will put emphasis on the macrocellular case where the velocity of MS
is usually very high (i.e., > 120 km/h).
In macrocellular environments, the LoS component, hc (t), of the channel usually does
not exist and the channel behaves completely as ‘Rayleigh’ fading process [13]; therefore,
Kt = 0. Using (7.8), we can also write for p(ψ),
1
pΨ (ψ) = (7.9)
2π
Therefore, R(∆t) in (7.5) can be finally written as,
R(∆t) = σh2 J0 (2πfD ∆t) (7.10)
where J0 (·) is the zeroth order Bessel function of the first kind and σh2 = R(0) is the variance
of time-varying fading channel. Similar result for the autocorrelation function was first
reported in [23] and thus is known as ‘Jakes’ Model. Without loss of generality, we can
assume σh2 = 1 in order to simplify our results.
To simulate the structured variations of time-selective wireless fading channel, three
types of linear models are usually considered [103, 104, 109, 112, 113], which are:
1. Autoregressive (AR) or ‘All-pole’ model
2. Moving average (MA) or ‘All-zero’ model
187
3. Autoregressive moving average (ARMA) model
Out of these three, the AR model is the most frequently used model because of its simplicity
and ease of designing (i.e., the equations that determine its parameters are linear). The
Yule-Walker equations [112, 113], which are used to determine the parameters of the model,
become the normal equations in case of the AR model. The Levinson-Durbin recurrent algo-
rithm is used to solve these equations. The key of the algorithm is the recursive computation
of the filter coefficients, beginning with the first order and increasing the order recursively,
using the lower order solutions to obtain the solution to the next higher order [103, 112].
Selecting the order is another difficult problem in developing a linear model. In [114],
the information theoretic results show that the first-order AR model provides a sufficiently
accurate model for time-selective fading channels and, therefore, we will adopt it henceforth.
Thus, using discrete-time notations, h(i) varies according to [35]
h(i) = ξh(i − 1) + v(i) (7.11)
where ξ is the autoregressive (AR) coefficient which accounts for the variations in the channel
due to Doppler shift, and v(i) is the zero-mean complex Gaussian noise with covariance σv2
and is statistically independent of h(i − 1).
Using Yule-Walker equations [35, 112, 113], ξ and σv2 can be written as,
ξ = R(∆t) = J0 (2πfD ∆t) (7.12)
σv2 = 1 − |ξ|2 (7.13)
188
7.3 Kalman Filter Based Adaptive Detection for DS-CDMA System
7.2.2 State-Space Model of the Communication System over Fast
Fading Multipath Channels
If we consider Nt tap channel, then we can rewrite equations (7.1) and (7.11) in vector form
as,

h(i) = Ξ h(i − 1) + v(i) 
(7.14)

r(i) = sT (i)h(i) + n(i)
where, Ξ =diag{ξ0 , ξ1 , · · · ξNt −2 , ξNt −1 }, h(i) = [h0 (i) h1 (i) · · · hNt −2 hNt −1 ]T , v(i) =
[v0 (i) v1 (i) · · · vNt −2 vNt −1 ]T and s(i) = [s0 (i) s1 (i) · · · sNt −2 sNt −1 ]T .
The model in (7.14) is the well known State-Space model of a communication system over
time-varying fading channel.
7.3 Kalman Filter Based Adaptive Detection for DS-
CDMA System
In this section we present a first order linear state-space model of a DS-CDMA multiuser
channel. We use the Kalman filter for the estimation of channel-distorted received signature
sequences. In our proposed receiver structure, the need for training sequence is bound to the
startup period only. Later on, receiver adapts itself to the changes of the channel during data
transmission depending on previous decisions. Simulation results show that being based on
the Kalman filter and of non-gradient nature, our proposed algorithm combats, effectively,
189
Estimator Noise
Signature Sequence
v (i)
Estimator, zk (i|i)
AWGN Noise Receiver Filter

n (i) ck (i)
Time-Varying Channel
Data bk (i)
Decision Device
Hk (i) bˆ ( i )
k
Received
Vector, r (i)
Signatures sk (i) Time-Varying Demodulator
MAI
Transmitter CDMA Channel Receiver
Figure 7.4: Discrete-time baseband model for synchronous DS-CDMA system
the time-varying channel impairments and multipath fading effects.
7.3.1 System Model
We assume a synchronous DS-CDMA system with K users and processing gain N . The
N × 1 received vector is given by [87, 90],
K
X
r(i) = Ak [H+ −
k (i)bk (i) + Hk (i − 1)bk (i − 1)]sk + n(i) (7.15)
k=1
where
• bk (i) is the ith symbol transmitted by user k with power E[|bk (i)|2 ] = 1,
• sk is N × 1 signature vector of user k,
190
• Ak is the amplitude of the signal transmitted by user k,
• n(i) is the N × 1 white Gaussian noise vector with σ 2 IN covariance matrix,
• H+
k (i) is N × N time-varying channel matrix for user k representing the contribution
from symbols bk (i), k = 1, · · · , K.
• H−
k (i−1) is N ×N time-varying channel matrix for user k representing the inter-symbol
interference (ISI) from symbols bk (i − 1), k = 1, · · · , K.
7.3.1.1 Channel Model
The spread symbols for user k in (7.15) are passed through the discrete time-varying channel
with impulse response given by the N × 1 vector,
£ ¤T
hk (i) = hk,0 (i) hk,1 (i) hk,2 (i) · · · hk,Lk −1 (i) 0 · · · 0 (7.16)
where Lk represents the number of paths, assumed to be spaced at the chip duration, Tc =
T /N , such that N > Lk , T is the symbol duration and (·)T is the matrix transpose operator.
The time-varying channel matrices in (7.15), H±

k (i) can then be written as [90],
h i
+(N −2) +(N −1)
H+
k (i) = h+0 +1 +2
k (i) hk (i) hk (i) · · · hk (i) hk (i) (7.17)
h i
−(N −1) −(N −3)
H−
k (i) = h−N
k (i) hk (i) hk (i) ··· h−2
k (i) h−1
k (i)
(7.18)
191
where hk±n (i) is hk (i) shifted down (+) or up (-) by n positions, and the vacant positions are
filled up with zeros. More conveniently, we can also write the N × N channel matrices as
 
hk,0 (i) 0 0 ··· 0 0
 hk,1 (i) hk,0 (i) 0 ··· 0 0 
 
 hk,2 (i) hk,1 (i) hk,0 (i) ··· 0 0 
 .. .. .. 
 . . .. .. 
+  . . . . . . 
Hk (i) =   (7.19)
 hk,Lk −1 (i) hk,Lk −2 (i) hk,Lk −3 (i) ··· 0 0 
 
 0 hk,Lk −1 (i) hk,Lk −2 (i) ··· 0 0 
 . .. .. . . . .. .. 
 .. . . . . 
0 0 0 · · · 0 hk,0 (i)
 
0 ··· 0 hk,Lk −1 (i) · · · hk,2 (i) hk,1 (i)
 0 ··· 0 0 ··· hk,3 (i) hk,2 (i) 
 .. . . .. .. .. .. 
 .. 
 . . . . . . . 
 
−  0 ··· 0 0 · · · hk,Lk −1 (i) hk,Lk −2 (i) 
Hk (i) =   (7.20)
 0 ··· 0 0 ··· 0 hk,Lk −1 (i) 
 
 0 ··· 0 0 ··· 0 0 
 .. . . .. .. .. .. .. 
 . . . . . . . 
0 ··· 0 0 ··· 0 0
We see that H−
k (i) is a sparse matrix if Lk << N . In order to simplify the derivation and
formulation of our estimation algorithm, we will assume that the channel delay spreads are
small compared with the symbol duration and, hence, neglect ISI (i.e., H−
k (i) = 0). This
is a viable assumption and has also been made in [90]. We, therefore, use Hk (i) to denote
the channel matrix for user k. The necessary conditions for the optimal signatures pre-
sented here can be extended to asynchronous CDMA with ISI by expanding the observation
window for the received signal. This complicates the derivations while adding little insight
to the synchronous case, so that only synchronous CDMA with negligible ISI is considered
throughout the chapter. Rewriting equation (7.15) for synchronous DS-CDMA system over
192
non-frequency selective fast fading channel, we get
K
X
r(i) = Ak Hk (i)bk (i)sk + n(i) (7.21)
k=1
We observe that the transmitted signature sequence, sk , is distorted by the time-varying
channel fading channel, Hk ; and as a result, channel-distorted version, zk , of signature
sequence is received. We can, therefore, write the received vector as
K
X
r(i) = Ak bk (i)zk (i) + n(i) (7.22)
k=1
The time-varying received signature vector zk (i) is actually the result of convolution between
the transmitted signature waveform sk and the time-varying channel vector hk (i) for user k.
zk (i) = sk ∗ hk (i) (7.23)
zk (i) in (7.23) can also be written as
zk (i) = Hk (i)sk (7.24)
where Hk (i) is the channel matrix defined in (7.19). For coherent reception over an ideal
time-invariant AWGN channel,
zk (i) = sk (7.25)
Fig. 7.4 depicts a synchronous DS-CDMA system [90]. Different from [90], where the
channel coefficients have been assumed to be time-invariant, we consider more realistic fast
fading time-selective channels.
To account for the dynamic variations that arise in the channel due to the relative motion
193
between transmitter and receiver, we use the AR model given in (7.11). Because of linear
processing between the transmitted signature sk with power, sH

k sk = 1 and time-varying
channel hk (i), the transmitted signature sk deforms to time-varying signature, zk (i), after
passing through time-varying channel. Therefore, we can also write our AR channel model
as,
zk (i) = Ξk zk (i − 1) + v(i) (7.26)
where,
£ ¤ £ ¤T
Ξk = diag ξk,0 ξk,1 ξk,2 · · · ξk,N −1 and v(i) = v0 (i) v1 (i) v2 (i) · · · vN −1 (i) .
Using (7.12), the AR coefficient, ξk , for user k can be written as,
∗
ξk = E{zk,l (i)zk,l (i + T )} = J0 (2πfD T ) (7.27)
where T is the symbol duration. σv2 can be found as given in (7.13).
7.3.1.2 Receiver Model
The received signal in (7.21) or (7.22) is the input to a linear filter with coefficients ck . Like
in [90], we also consider two types of filters:
1. Coherent RAKE filter (i.e., maximum ratio combiner) or matched filter (MF), given
by
Hk (i)sk
ck (i) =
kHk (i)sk k
zk (i)
= (7.28)
kzk (i)k
194
2. An MMSE filter given by
ck (i) = M−1 (i)Hk (i)sk
= M−1 (i)zk (i) (7.29)
where
K
X
H
M(i) = A2k Hk (i)sk sH 2
k Hk (i) + σn IN
k=1
XK
= A2k zk (i)zH 2
k (i) + σn IN (7.30)
k=1
is the time-varying received covariance matrix and (·)H denotes the Hermitian trans-
pose.
Thus we can say that our receiver consists of a bank of matched filters as its first part, where
we match the received signal vector with the estimated time-varying signature sequence,
zk (i) of user k. Thus, the output of conventional detector for user k can then be written as,
yk (i) = zH
k (i)r(i)
K
X
= Ak zH
k (i)bk (i)zk (i) + Aj zH
k (i)bj (i)zj (i) + ñ(i) (7.31)
j6=k
PK
where j6=k Aj zH
k (i)bj (i)zj (i) is the time-varying MAI, which depends on the time-varying
channel matrix Hk (i) for user k and thus on the quality of the channel estimates or received
signature estimates.
In order to detect the transmitted symbols, any of the two filters given in (7.28) and
(7.29) can be considered. The former’s output is equivalent to the output of the bank of
195
matched filters in (7.31). If the latter is considered, then the output of the bank of matched
filters in (7.31) is passed to the linear MMSE filter that outputs the following decision for
user k [87],
½ ´ ¾
1 ³ −1 H
b̂k (i) = sgn M (i)zk (i)r(i)
Ak k
(Ã K ! )
³X ´−1
= sgn A2k zk (i)zH 2
k (i) + σn zH
k (i)r(i) (7.32)
k=1 k
where sgn represents the signum function. In this case, we use the time-varying linear
transformation on linear MMSE receiver basis, to eliminate the effects of time-varying MAI
and noise. As we know, the single-user matched filter receiver is optimized to fight the
background white noise, n(i), exclusively, where as the decorrelating detector eliminates
the multiuser access interference (MAI) disregarding the background noise. In contrast the
MMSE linear detector can be seen as a compromise solution that takes into account the
relative importance of each interfering user and the background noise. In fact both the
conventional receiver and the decorrelating receiver are the limiting cases of the MMSE
linear detector [87].
7.3.2 Kalman Filter-Based Signature Sequence Tracking
As mentioned in the previous sections, we base our time-varying signature sequence esti-
mation algorithm on the Kalman filter approach because of its good tracking properties
in a highly time-varying environment and its active minimization of estimation error vari-
ance [115]. Given the noisy measurements up to the time i in (7.22), and using signature
196
state equation in (7.26), we have to find the best estimate of zk (i + 1|i) which minimizes
the error variance. In our case the estimation error vector is equal to the difference between
{True signature vector at time i + 1} and {Estimated signature vector at time i + 1 given
the measurements up to time i},
z̃k (i + 1) = zk (i + 1) − ẑk (i + 1) (7.33)
and the estimation error covariance matrix is,
P̃k (i + 1) = cov(h̃k (i + 1)) (7.34)
Thus our criterion is given as,
Jk (i + 1) = E{z̃H
k (i + 1)z̃k (i + 1)} = tr{P̃k (i + 1)} (7.35)
The implementation of our adaptive algorithm for signature sequence estimation starts with
a training mode that is used to acquire initial zk (i) estimates, after which it reverts to a
decision-directed mode. In the training mode, the receiver knows the transmitted symbols,
whereas in the decision-directed mode, the decoded symbols replace the information symbols.
We will focus on the decision-directed mode and assume that initial signature sequence
estimates are available through startup training.
Selecting zk (i) = [zk,1 (i) zk,2 (i) zk,3 (i) · · · zk,N −1 (i) zk,N (i)]T as the state vector, we can
exploit equation (7.26) as the state update for user k.
197
Measurement Prediction Signature Prediction

r (i |i-1) zk (i | i-1)
Estimator Noise
v (i)
- ek(i) Signature Correction

zk (i| i)
+ Signature Estimator
Receiver Filter
ck (i)
Decision Device
bˆ ( i )
k
Received
Vector, r (i) Demodulator
Receiver
Figure 7.5: Proposed model of the receiver structure for the Kalman filter based adaptive multiuser
detection
198
7.3.2.1 State-Space Model for the Adaptive Multiuser Detection
From the general form of the State-Space model presented in (7.14), we can re-write our
state-space model specifically for the signature sequence estimation case as,
¾
zk (i + 1) = Ξk zk (i) + v(i + 1)
P (7.36)
r(i) = K k=1 Ak bk (i)zk (i) + n(i)
7.3.2.2 Signature Sequence Estimation Algorithm
Our proposed algorithm consists of two parts:
1. Use of training sequence to find optimal signature estimates as a startup
2. To drive algorithm using decisions made in the detection prior to prediction and esti-
mation throughout the process of data transmission.
Various steps involved in our proposed algorithm are elaborated in Fig. 7.5, where the
proposed receiver structure for the Kalman filter based adaptive detection of DS-CDMA
signals is shown. We can present our signature sequence estimation algorithm based on the
state-space model presented in (7.36) in the following steps:
1. Initialize the algorithm with zk (0|0) and P̃k (0|0).
2. Decode b̂k (i) by using (7.32).
199
3. Obtain Predictions zk (i + 1|i) and P̃k (i + 1|i) as,
zk (i + 1|i) = Ξk zk (i|i) + σv v(i) (7.37)
P̃k (i + 1|i) = Ξk P̃k (i|i)ΞH 2 H

k + v(i)σv v (i) (7.38)
4. As an innovation step, find N ×1 observation error vector and observation error variance
as,
ek (i + 1) = r(i) − Ak b̂k (i)zk (i + 1|i) (7.39)
Re,k (i + 1) = FH 2
k (i)P̃k (i + 1|i)Fk (i) + σv IN (7.40)
where Fk (i) = Ak b̂k (i) and b̂k (i) is found in step 2.
5. Find the Kalman gain as
Gk (i + 1) = P̃k (i + 1|i)Fk (i)R−1

e,k (i + 1) (7.41)
6. Correct the estimates made in step 4 as,
zk (i + 1) = zk (i + 1|i + 1)
= zk (i + 1|i) + Gk (i + 1)e∗k (i + 1) (7.42)
P̃k (i + 1) = P̃k (i + 1|i + 1)
= (IN − Gk (i + 1)FH
k (i))P̃k (i + 1|i) (7.43)
7. Use real-time corrected zk (i + 1) for the detection of b̂k (i + 1) and repeat steps 3-7 to
estimate and correct the received signature sequence vector at i+2 time-instant.
200
7.3.3 Modes of Operation
Training sequence is usually employed to provide information of the channel to the receiver
so that it can get the accurate initial estimates of the signature sequences. Training sequence
helps the receiver to quickly adapt itself to the channel conditions. The length of the training
sequence is kept as small as possible as no useful information is being conveyed during
the training mode. This length is usually equal to the number of iterations required by
the receiver to converge to its optimum value. During simulations, the proposed adaptive
multiuser detector has been found to converge in about 45 symbols. Thus a training sequence
of length 50 is used for subsequent simulations. It is envisaged that this receiver can operate
in following four different modes.
7.3.3.1 Decision Directed (DD) mode
In this mode, the receiver operates with Decision feedback only. The initial channel coeffi-
cients are assumed arbitrarily. The algorithm finds measurement error vector in (7.39) by
using the decision made in the previous iteration. There exists a chance of error propagation,
as subsequent decisions are heavily dependent upon the decisions made in earlier iterations.
201
7.3.3.2 Training Directed (TD) and DD mode
For overcoming the possibility of error propagation, training can be employed during the
convergence time of the algorithm. In this mode, the receiver operates in Training Directed
(TD) mode for 50 iterations and then switches to DD mode for the rest of the data trans-
mission.
7.3.3.3 TD and Non-Estimation (NE) mode
In this mode, the receiver operates in TD mode for 50 iterations and then switches to NE
mode in which the estimated channel response is kept constant at the last optimized value
achieved at the end of TD mode. This mode is similar in analogy to the linear adaptive
MMSE receiver, which assumes time-invariant channel.
7.3.3.4 Repeated TD and NE mode
In this mode, the receiver is provided training after short intervals of actual data transmis-
sion. The receiver operates in TD mode for 50 iterations and then switches to NE mode for
about one frame or block of data symbols. This pattern is repeated till the end of the data
transmission. This mode outperforms TD and NE mode but performs worse than TD and
DD together and DD alone modes.
202
Performance at 60 Hz Performance at 220 Hz

0 0
10 10
Single User Bound Single User Bound
Proposed Model Proposed Model
RLS RLS
LMS LMS
−1 −1
10 10
Average BER
Average BER
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
6 8 10 12 6 8 10 12
SNR in dB SNR in dB
Figure 7.6: BER versus SNR with N = 16 and K = 4
7.3.4 Simulation Results
In this section we present numerical results, which illustrate the relative performance of
our model. Numerical results are obtained using Monte-Carlo simulation. In each trial
of simulation, 5,000 data bits were generated for each user. All users have equal power
and undergo independent multipath fading with different multipath gains in their channel
impulse responses. The fading coefficients are regenerated with each simulation trial using
Jakes’ model [23]. The carrier frequency is 2.4 GHz and the mobile speed is considered
27 Km/h and 99 Km/h taking two different scenarios of Doppler spread (60 and 220 Hz).
203
0.35
0.3
0.25
0.2
J (i)
k
0.15
0.1
0.05
0
0 50 100 150 200 250 300 350 400 450 500
Iteration, i
Figure 7.7: The criterion, Jk (i)=tr{P̃k (i)}, for arbitrary user, k, at iteration, i
Each user generates bit streams of data with p(−1) = p(+1) = 1/2. Random sequences of
length 16 are used to spread each data bit of K users. The Signal to Noise ratios (Eb /N0 )
are set in ascending order from 0 dB to 15 dB.
In Fig. 7.6, Bit Error Rate (BER) is evaluated in each value of Eb /N0 for a population
of 4 users, with equal powers and random signature sequences in a fading environment of
Doppler spread 60 and 220 Hz, separately. We have considered two multipaths Lk = 2,
k = 1, 2, · · · K for each user k. The channel coefficients (elements of hk , k = 1, 2, · · · , K)
are taken independently and the channel vectors hk are normalized so that E{khk k2 } = 1.
The results show that Doppler spread and multipath fading present a major impeding effect
204
on the performance of high-speed digital transmission when we don’t consider compensation
for the channel. The results also show that successful estimation of the received signature
waveforms in such a time varying environment leads us to relatively good performance. The
results show that our proposed model outperforms when it is compared to
1. Adaptive MMSE with LMS algorithm [85]
2. RLS
3. Single user bound
all in different Doppler conditions.
Fig. 7.7 shows that convergence of the Kalman filter-based signature sequence estimation
algorithm is very fast compared to other gradient based algorithms (almost two times faster
than LMS based algorithm given in [85]). It also shows that our criterion, Jk (i) = trP̃k (i),
discussed in section 7.3.2 is minimized properly which indicates that our estimation process
is running successfully. In our simulations, we have taken initial value of P̃k (i) as 0.05IN
for each user with different multipath gains obtained in training. Jazwinski [116] has shown
that the initial statistics, P̃k (0|0), is forgotten as more data are processed. These results
follow from the fact that as long as the filter is stable and the system (state-space model in
(7.36)) is completely controllable and observable then
lim P̃k (i|i) = P̃k {very small and constant value} as i −→ ∞ (7.44)
205
Thus the estimation error approaching zero implies that the state estimate converges to the
true value given that there exist enough data. The initial estimates will affect the transient
performance of the algorithm, since a large initial P̃k (0|0) gives a large the Kalman gain,
therefore heavily weighting the initial measurements and ignoring the model.
The different modes of operation of linear adaptive CDMA receiver are evaluated over
fast fading channel, where the channel conditions are assumed to change in every iteration.
The BER performance results for all operation modes are shown in Fig. 7.8. The following
observations are made from these results:
1. The BER performance achieved by ‘TD and DD’ mode is the best among all of four
modes.
2. In ‘DD alone’ mode, the receiver suffers from error propagation phenomenon, which
degrades its performance compared to ‘TD and DD’ mode.
3. In ‘TD and NE’ mode, the receiver performs worst, which means that ordinary CDMA
receiver can never cope with the variations in the time-varying channel.
4. The performance improves marginally when the combination of TD and NE mode is
repeated after short intervals of iterations, especially before each data block/frame
transmission.
The above-mentioned observations indicate that the changing behavior of the channel affects
the receiver performance severely whenever channel estimation and compensation is not
206
0
10
DD
TD+DD
TD+NE
Repeated (TD+NE)
−1
10
Average BER
−2
10
−3
10
−4
10
0 2 4 6 8 10 12 14 16 18
SNR in dBs
Figure 7.8: BER performance comparison of different operation modes of the Kalman filter-based
adaptive CDMA detector, N = 31, K = 4
performed. This situation improves when the training is repeated after short intervals. This
means that the channel information provided to the receiver by the training sequence in a
highly time varying environment is of great help since the channel response is continuously
changing. It implies that the receiver continuously needs to perform channel estimation and it
cannot just rely on the converged channel response achieved at the end of training. Moreover,
when training is not employed at the startup and the algorithm is running in decision directed
mode only, there are chances of error propagation. Therefore, it is sometimes required to
train the filter before it is run in decision directed mode.
207
7.4 Conclusions
7.4 Conclusions
In this chapter, an autoregressive (AR) model was developed for the time-varying flat fading
channel on the basis of its second order fading statistics. The model was then utilized further
to establish a linear state-space equation pair for signature sequence adaptation in direct
sequence code division multiple access (DS-CDMA) system. Moreover, a decision directed
signature sequence estimation algorithm was proposed which worked on an iterative basis
during the course of data transmission in a rapidly time-varying environment. The algorithm
was based on the proposed state-space model and the Kalman filter so that it could estimate
and track the variations in the signature sequence over multipath and time varying fading
channels.
The performance of the proposed algorithm was compared with the other adaptive algo-
rithms over time-varying channels and it was found that the proposed algorithm outperforms
all other gradient-based algorithms like LMS and RLS, in tracking the rapid changes of the
channel. A linear minimum mean squared error (MMSE) detector was used to detect the
symbols, which were then exploited in updating the received signature waveforms, distorted
by the time-varying channel. Simulation results were presented which showed that the per-
formance of a linear adaptive receiver could be improved significantly with signature tracking
on high Doppler spreads in DS-CDMA system. The proposed model can also be extended
to eliminate other channel impairments such as CFO (Carrier Frequency Offset) in OFDM
systems.
208
Chapter 8
Conclusions and Future Work
This chapter first gives a brief summary of the thesis in section 8.1 and then discusses final
conclusions and future research work based on the results of this dissertation in section 8.2.
8.1 Summary of the Thesis
In Chapter 2, we have addressed the issue of physical channel modeling for the cellular
mobile communication system. We have extensively studied the previous approaches used
for modeling cellular mobile channel in picocell, microcell, and macrocell environments. We
have developed necessary channel modeling parameters and proposed a generalized physical
channel model, referred to as the ‘Eccentro-Scattering Model’. This model can be applied to
any type of cellular environment with appropriate choice of eccentricity, semi-major axis, and
distribution of scatterers around MS and/or BS. We have also introduced a more practical
scattering model, the Jointly Gaussian Scattering Model (JGSM), which consists of two
Gaussian functions each for the distribution of scatterers around BS and MS. The same
methodology can also be used to develop a generalized spatial channel model for a 3-D
environment.
In Chapter 3, we have studied the spatial characteristics of cellular mobile channel for
picocell, microcell, and macrocell environments assuming uniform and Gaussian distribu-
tion for the scatterers. Utilizing the Eccentro-Scattering Model and JGSM, we have derived
general expressions for the pdf of AoA of the multipath signals at BS. The derived results
show that the previous spatial models can easily be extracted from our proposed model with
appropriate selection of parameters. We have thoroughly discussed the results and compared
them with those of all existing models. The theoretical results are also compared with some
available measurements both in indoor and outdoor environments. The resulting compar-
210
isons show good agreement with the realistic situations. We can thus assume our proposed
model to be useful in simulating several propagation scenarios for wireless communications
systems. The derived results, in closed form, can also be used in further research work to
model Doppler characteristics and tracking properties of time-varying fading channels.
In Chapter 3, we have also addressed the issue of the impact of local-to-BS scattering on
the spatial characteristics and implemented JGSM for low antenna-height urban environment
by introducing an adjustable scattering-free region around BS. The adjustable scattering-free
region around BS models the extent of scattering in the vicinity of BS and thus can easily
be used as a control-valve for the inclusion/exclusion of scattering objects in the vicinity of
BS according to their anticipated effect on the angular distribution of the cellular mobile
channel. We have found that the JGSM along with the provision of scattering-free region
provides good fitness to the field measurements when compared with all existing Gaussian
scattering models that consider only one Gaussian function for the distribution of scatterers
around MS.
In Chapter 4, we have proposed a novel generalized method of quantifying the angle
spread of the multipath power distribution. The proposed method provides almost all pa-
rameters associated to the angular spread, which can be further used for calculating more
accurate spatial correlations of the multipath fading channels. The proposed parameters are
also useful in finding the exact standard deviation of the truncated angular distributions and
the angular data acquired in measurement campaigns. The degree of accuracy in correlation
211
calculations can lead to the computation of exact separations among array elements needed
for diversity antennas. Currently, truncated Gaussian or Laplacian functions are often used
to represent the azimuthal distribution of multipath signals. Such functions usually simplify
the calculations of correlation in MIMO channels. We have indicated that the use of standard
deviation of full-span functions as the standard deviation of the truncated function causes
severe effects on the angle spread, which in turn distorts the accuracy of correlation figures
in MIMO channels. Due to the importance of angle spread in the fading statistics, we have
proposed its use as the goodness-of-fit measure in measurement campaigns. The proposed
method of quantifying angle spread can thus be used in finding the accurate separations
among array elements in outdoor MIMO systems where measurement campaigns provide
basis for channel models.
In Chapter 5, we have discussed the temporal characteristics of cellular mobile chan-
nel in picocell, microcell, and macrocell environments. We have employed the proposed
Eccentro-Scattering model to derive the pdf of ToA of the multipath signal for these cellular
environments. In macrocell environment, our model incorporated the effect of distant scat-
terers, far from BS and MS on the temporal dispersion of the multipath signal in addition
to that of local scatterers. A simplified generic closed-form formula for the pdf of ToA due
to local scatterers has been derived, from which previous models can be easily reproduced.
The presented formulas can be used to simulate temporal dispersion of the multipath chan-
nel in a variety of propagation conditions. Furthermore, these formulas are also helpful in
designing efficient equalizers to combat intersymbol interference (ISI) for frequency-selective
212
fading channels.
In Chapter 6, we have investigated the effects of mobile motion on the spatial and
temporal statistics of the cellular mobile channel. We have discussed several motion scenarios
that are responsible for such effects. We have formulated the changes in the AoA and ToA
distributions of the multipath signals at BS during the course of MS motion for different
cellular environments. We have plotted the behavior of all important spatial and temporal
statistical parameters under the effect of mobile motion. The proposed theoretical results in
spatial characteristics can be extended to characterizing and tracking transient behavior of
Doppler spread in time-varying fast fading channels; likewise the proposed theoretical results
in temporal characteristics can be utilized in designing efficient equalizers for combating
inter-symbol interference (ISI) in time-varying frequency-selective fading channels.
In Chapter 7, an autoregressive (AR) model was developed for the time-varying flat
fading channel on the basis of its second order fading statistics. The model was then utilized
further to establish a linear state-space equation pair for signature sequence adaptation in
DS-CDMA systems. Moreover, a decision directed signature sequence estimation algorithm
was proposed which worked on iterative basis during the course of data transmission in
a rapidly time-varying environment. The algorithm was based on the proposed state-space
model and the Kalman filter so that it could estimate and track the variations in the signature
sequence over multipath and time varying fading channel.
The performance of the proposed algorithm was compared with the other adaptive algo-
213
8.2 Conclusions and Future Work
rithms over time-varying channels and it was found that the proposed algorithm outperforms
all other gradient-based algorithms like LMS and RLS, in tracking the rapid changes of the
channel. A linear minimum mean squared error (MMSE) detector was used to detect the
symbols, which were then exploited in updating the received signature waveforms, distorted
by the time-varying channel. Simulation results were presented which showed that the per-
formance of a linear adaptive receiver could be improved significantly with signature tracking
on high Doppler spreads in DS-CDMA system.
This research has studied spatial channel models for cellular mobile systems and their use in
the characterization of multipath fading channels. The results were presented mainly in five
parts; a) modeling of scattering mechanisms, b) derivation of the closed-form expressions for
the spatio-temporal characteristics, c) generalization of the quantitative measure of angular
spread, d) investigation of the effect of mobile motion on the spatio-temporal characteristics,
and e) characterization of fast fading channel and its use in signature sequence adaptation
for DS-CDMA system.
The results presented in this work have implications to cellular wireless system design,
depicted in Fig. 1.5 of Chapter 1. The analytical results for the distributions of the angle
and time of arrival of multipath signals can be useful for the purposes of system design
214
and performance evaluation in cellular systems. The generalized quantitative definition of
the angular spread proposed in the work will be used for calculating more accurate spatial
correlations and thus the exact separation distances between array elements required for
maximizing capacity in MIMO systems or diversity antennas. For the first time, angular
spread was proposed as the goodness-of-fit measure in angular measurement campaigns based
on its key role in the fading processes. The behavior of some important spatial and temporal
statistical parameters was investigated under the effect of mobile motion. The results, in
closed mathematical form in this regard, can be useful in studying the behavior of time-
varying frequency-selective fading channels. In the last part of the thesis, a Kalman filter
based adaptive DS-CDMA receiver structure was proposed, where the simulation results
showed that the performance of a linear adaptive receiver could be improved significantly
with signature tracking on high Doppler spreads in DS-CDMA systems.
Besides the use of Fig. 1.5 of Chapter 1 in proposing the methodology of our dissertation,
it is also helpful in introducing the strategy of our future research work. This strategy can
be grouped according to the following three plans:
Plan 1
The general formulas derived for the pdf of AoA of the multipath signals at BS in (3.10),
(3.17), (3.27), (3.31), and (3.41) provide almost all information needed for calculating the
spatial correlations among the antenna elements of MIMO arrays. We are thus planning to
215
use these formulas along with the angular spread quantifiers, discussed in Chapter 4, to find:
1. the spatial correlations among antenna elements of MIMO arrays which are responsible
to decrease MIMO system capacity,
2. the exact separations between the antenna elements required to maximize MIMO ca-
pacity,
3. the design parameters, needed for diversity antenna to overcome deep fades at BS
antennas.
Plan 2
The general formulas derived for the pdf of ToA in Chapter 5, can be further extended to
finding coherence bandwidth, which is an important parameter in the designing mechanism
of channel equalizers. ToA distributions of Chapter 5 and temporal spread parameters of
Chapter 6 are also helpful in designing RAKE receiver fingers for CDMA and W-CDMA
multiuser detectors. If we apply the RAKE finger number decision rule, given in [117, 118]
and finger assignment algorithm, discussed in [119] on the general expressions of the pdf of
ToA in (5.15), (5.27) and (5.39), we will certainly approach a generalized RAKE receiver
finger number selection criteria for all cellular environments, according to their local terrain
and clutter.
216
Plan 3
In Chapter 7, a state-space model was developed on the basis of the second order fading
statistics of the time-varying channel. The model was flexible to be used for any type of
cellular environment by simply changing the parameters of the AoA distribution in (7.6) and
(7.7). However, for the simplicity of theoretical results, an isotropic scattering environment
was considered. As we have discussed in detail in Chapter 1 that in modern communication
systems, antenna arrays are usually used on both link ends, which adds directionality to the
MS. In such cases, the state-space model can be modified by using the effects of induced
directionality on the fading statistics. Such effects are studied in [26, 120–122]. Spatial
correlations in non-isotropic scenarios are also discussed extensively in [123, 124]. However,
to the best of our knowledge, there is no inclusive time-varying channel estimation model
available in the literature which can be able to accommodate the directionality induced
by MS antennas. Therefore, the state-space modeling approach discussed in Section 7.2 of
Chapter 7, can easily be extended to non-isotropic scattering scenarios.
217
Bibliography
[1] Q. H. Spencer, B. D. Jeffs, M. A. Jensen, and A. L. Swindlehurst, “Modeling the
statistical time and angle of arrival characteristics of an indoor multipath channel,”
IEEE J. Select. Areas Commun., vol. 18, no. 3, pp. 347–360, 2000.
[2] R. Janaswamy, “Angle and time of arrival statistics for the gaussian scatter density
model,” IEEE Trans. Wireless Commun., vol. 1, no. 3, pp. 488–497, July 2002.
[3] R. B. Ertel and J. H. Reed, “Angle and time of arrival statistics for circular and
elliptical scattering models,” IEEE J. Select. Areas Commun., vol. 17, no. 11, pp.
1829–1840, November 1999.
[4] K. I. Pedersen, P. E. Mogensen, and B. H. Fleury, “A stochastic model of the tempo-
ral and azimuthal dispersion seen at the base station in outdoor propagation environ-
ments,” IEEE Trans. Veh. Technol., vol. 49, no. 2, pp. 437–447, March 2000.
[5] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless Communi-
cations. Trumpington Street, Cambridge, UK: Cambridge University Press, 2003.

BIBLIOGRAPHY
[6] V. H. MacDonald, “The cellular concept,” Bell Sys. Tech. J., vol. 58, no. 1, pp. 15–41,
January 1979.
[7] N. J. Boucher, Cellular Radio Handbook, 4th ed. New York, USA: John Wiley, 2001.
[8] T. S. Rappaport, Wireless Communications, Principles and Practice, 2nd ed. Upper
Saddle River, New Jersey, USA: Prentice Hall, 2002.
[9] G. Foschini, “Layered space-time architecture for wireless communication in a fading
environment when using multi-element antennas,” Bell Labs Tech. J., vol. 1, pp. 41–59,
Autumn 1996.
[10] G. D. Golden and G. J. Foschini, “Detection algorithm and initial laboratory results
using V-BLAST space-time communication architecture,” IEE Electron. Lett., vol. 35,
no. 1, pp. 14–16, January 1999.
[11] A. F. Molisch, Wireless Communications. West Sussex, England: John Wiley & Sons,
2005.
[12] P. Rapajic, “Information capacity of the space division multiple access mobile commu-
nication system,” Wireless Personal Commun., vol. 11, no. 1, pp. 131–159, October
1999.
[13] G. L. Stüber, Principles of Mobile Communication, 2nd ed. Norwell, Massachusetts,
USA: Kluwer Academic Publishers, 2001.
219
BIBLIOGRAPHY
[14] L. M. Correia, Wireless flexible personalised Communications: COST-259, European
Co-operation in Mobile Radio Research. Chichester, England: John Wiley, 2001.
[15] J. S. Blogh and L. Hanzo, Third-generation systems and intelligent wireless net-
working: Smart Antennas and Adaptive Modulation. New York, USA: Halsted Press,
2002.
[16] P. Van Rooyen, M. P. Lotter, and D. Van Wyk, Space-Time Processing for CDMA
Mobile Communications. Kluwer Academic Publishers, 2000.
[17] C. Shannon, “A mathematical theory of communication,” Bell Sys. Tech. J., vol. 27,
pp. 79–423, 1948.
[18] G. J. Foschini and M. J. Gans, “On limits of wireless communications in a fading
environment when using multiple antennas,” Wireless Personal Commun., vol. 6, pp.
311–335, 1998.
[19] D. Chizhik, F. Rashid-Farrokhi, J. Ling, and A. Lozano, “Effect of antenna separation
on the capacity of BLAST in correlated channels,” IEEE Commun. Lett., vol. 4, no. 11,
pp. 337–339, November 2000.
[20] M. K. Simon and M.-S. Alouini, Digital Communication over Fading Channels. Third
Avenue, New York, USA: John Wiley & Sons, 2000.
[21] R. S. Blum, J. H. Winters, and N. R. Sollenberger, “On the capacity of cellular systems
with MIMO,” IEEE Commun. Lett., vol. 6, no. 6, pp. 242–244, June 2002.
220
BIBLIOGRAPHY
[22] R. H. Clarke, “A statistical theory of mobile-radio reception,” Bell Sys. Tech. J.,
vol. 47, pp. 957–1000, July-August 1968.
[23] W. C. Jakes, Microwave Mobile Communications. New York, USA: John Wiley &
Sons, 1974.
[24] J. C. Liberti, Jr. and T. S. Rappaport, “A geometrically based model for line-of-sight
multipath radio channels,” in Proc. 46th IEEE Veh. Technol. Conf., April/May 1996,
pp. 844–848.
[25] J. Fuhl, A. F. Molisch, and E. Bonek, “Unified channel model for mobile radio systems
with smart antennas,” IEE Proc. Radar Sonar Navigation, vol. 145, no. 1, pp. 32–41,
February 1998.
[26] P. Petrus, J. H. Reed, and T. S. Rappaport, “Geometrical-based statistical macrocell
channel model for mobile environments,” IEEE Trans. Commun., vol. 50, no. 3, pp.
495–502, March 2002.
[27] M. P. Lotter and P. Van Rooyen, “Modeling spatial aspects of cellular CDMA/SDMA
systems,” IEEE Commun. Lett., vol. 3, no. 5, pp. 128–131, May 1999.
[28] R. J. Piechocki, G. V. Tsoulos, and J. P. McGeehan, “Simple general formula for pdf
of angle of arrival in large cell operational environments,” IEE Electron. Lett., vol. 34,
no. 18, pp. 1784–1785, September 1998.
221
BIBLIOGRAPHY
[29] T. L. Fulghum, K. J. Molnar, and A. Duel-Hallen, “The Jakes fading model for antenna
arrays incorporating azimuth spread,” IEEE Trans. Veh. Technol., vol. 51, no. 5, pp.
968–977, September 2002.
[30] R. M. Buehrer, “The impact of angular energy distribution on spatial correlation,” in
Proc. 56th IEEE Veh. Technol. Conf., vol. 2, September 2002, pp. 1173–1177.
[31] J. Tsai, R. M. Buehrer, and B. D. Woerner, “Spatial fading correlation function of
circular antenna arrays with laplacian energy distribution,” IEEE Commun. Lett.,
vol. 6, no. 5, pp. 178–180, May 2002.
[32] G. D. Durgin and T. S. Rappaport, “Basic relationship between multipath angular
spread and narrowband fading in wireless channels,” IEE Electron. Lett., vol. 34, no. 25,
pp. 2431–2432, December 1998.
[33] ——, “Theory of multipath shape factors for small-scale fading wireless channels,”
IEEE Trans. Antennas Propagat., vol. 48, no. 5, pp. 682–693, May 2000.
[34] T. J. Lim and Y. Ma, “The Kalman filter as the optimal linear minimum mean squared
error multiuser CDMA detector,” IEEE Trans. Info. Theory, vol. 46, no. 7, pp. 2561–
2566, July 2000.
[35] Z. Liu, X. Ma, and G. B. Giannakis, “Space-timing coding and kalman filtering for
time-selective fading channels,” IEEE Trans. Commun., vol. 50, no. 2, pp. 183–186,
February 2002.
222
BIBLIOGRAPHY
[36] A. M. Rao and D. L. Jones, “Efficient detection with arrays in the presence of angular
spreading,” IEEE Trans. Signal Processing, vol. 51, no. 2, pp. 301–312, February 2003.
[37] J. H. Winters, “On the capacity of radio communications systems with diversity in
rayleigh fading environments,” IEEE J. Select. Areas Commun., vol. 5, no. 5, pp.
871–878, June 1987.
[38] A. F. Molisch, “Effect of far scatterer clusters in MIMO outdoor channel models,” in
Proc. 57th IEEE Veh. Technol. Conf. (VTC), vol. 1, April 2003-Spring, pp. 534–538.
[39] W. C. Y. Lee, “Effects on correlation between two mobile radio base-station antennas,”
IEEE Trans. Commun., vol. COM-21, no. 11, pp. 1214–1224, November 1973.
[40] K. Yu and B. Ottersten, “Models for MIMO propagation channels: A review,” Wiley
J. Wireless Commun. Mobile Computing, vol. 2, no. 7, pp. 653–666, November 2002.
[41] A. F. Molisch, “A generic model for mimo wireless propagation channels in macro-
and microcells,” IEEE Trans. Signal Processing, vol. 52, no. 1, 2004.
[42] “3GPP TR 25.996: Spatial channel model for multiple input multiple output (MIMO)
simulations (Release 6),” 3rd Generation Partnership Project (3GPP), Tech. Rep.,
2003.
[43] D. D. N. Bevan, V. T. Ermolayev, A. G. Flaksman, and I. M. Averin, “Gaussian chan-
nel model for mobile multipath environment,” EURASIP J. Applied Signal Processing,
vol. 2004, pp. 1321–1329, August 2004.
223
BIBLIOGRAPHY
[44] A. F. Molisch, A. Kuchar, J. Laurila, K. Hugl, and R. Schmalenberger, “Geometry-
based directional model for mobile radio channels: principles and implementation,”
Eur. Trans. Telecommun., vol. 14, pp. 351–359, July/August 2003.
[45] J. Litva, A. Ghaforian, and V. Kezys, “High-resolution measurements of AOA and
time-delay for characterizing indoor propagation environments,” in Proc. Antennas
Propagat. Society International Symp., vol. 2 AP-S Digest, July 1996, pp. 1490–1493.
[46] R. D. Tingley and K. Pahlavan, “Space-time measurement of indoor radio propaga-
tion,” IEEE Trans. Instrument. Measurement, vol. 50, no. 1, pp. 22–31, February
2001.
[47] A. S. Y. Poon and M. Ho, “Indoor multiple-antenna channel characterization from 2
to 8 GHz,” in Proc. IEEE International Conf. Commun. (ICC’03), vol. 5, May 2003,
pp. 3519–3523.
[48] J. Fuhl, J.-P. Rossi, and E. Bonek, “High-resolution 3-D direction-of-arrival determi-
nation for urban mobile radio,” IEEE Trans. Antennas Propagat., vol. 45, no. 4, pp.
672–682, April 1997.
[49] R. Vaughan and J. B. Andersen, Channels, Propagation and Antennas for Mobile
Communications. IEE Press, 2003.
[50] Y. Chen and V. K. Dubey, “Accuracy of geometric channel-modeling methods,” IEEE
Trans. Veh. Technol., vol. 53, no. 1, pp. 82–93, January 2004.
224
BIBLIOGRAPHY
[51] J. D. Parsons, Mobile Radio Propagation Channel, 2nd ed. West Sussex, England:
John Wiley & Sons, 2000.
[52] A. Popoulis, Probability, Random Variables, and Stochastic Processes, 3rd ed. USA:
McGraw-Hill, 1991.
[53] A. J. Paulraj and C. B. Papadias, “Space-time processing for wireless communica-
tions,” IEEE Sig. Proc. Magn., vol. 14, no. 6, pp. 49–83, November 1997.
[54] C. Ward, M. Smith, A. Jeffries, D. Adams, and J. Hudson, “Characterising the radio
propagation channel for smart antenna systems,” Electron. and Commun. Eng. J.,
vol. 8, no. 4, pp. 191–200, August 1996.
[55] A. F. Naguib, A. Paulraj, and T. Kailath, “Capacity improvement with base-station
antenna arrays in cellular CDMA,” IEEE Trans. Veh. Technol., vol. 43, no. 3, pp.
691–698, August 1994.
[56] J. B. Andersen and K. I. Pedersen, “Angle-of-arrival statistics for low resolution an-
tennas,” IEEE Trans. Antennas Propagat., vol. 50, no. 3, pp. 391–395, March 2002.
[57] K. V. Mardia, Statistics of Directional Data. Orlando, USA: Academic Press, 1972.
[58] J. Tsai, R. M. Buehrer, and B. D. Woerner, “BER performance of a uniform circular
array versus a uniform linear array in a mobile radio environment,” IEEE Trans.
Wireless Commun., vol. 3, no. 3, pp. 695–700, May 2004.
225
BIBLIOGRAPHY
[59] K. I. Pedersen, P. E. Mogensen, and B. H. Fleury, “Power azimuth spectrum in outdoor
environments,” IEE Electron. Lett., vol. 33, no. 18, pp. 1583–1584, August 1997.
[60] P. E. Mogensen, F. Frederiksen, H. Dam, K. Olesen, and S. L. Larsen, “TSUNAMI-II
stand alone test bed,” in Proc. ACTS Mobile Telecommunication Summit, Granada,
Spain, November 1996, pp. 517–527.
[61] S. A. Zekavat and C. R. Nassar, “Power-azimuth-spectrum modeling for antenna array
systems: A geometric-based approach,” IEEE Trans. Antennas Propagat., vol. 51,
no. 12, pp. 3292–3294, December 2003.
[62] D. Asztely and B. Ottersten, “Modified array manifold for signal waveform estimation
in wireless communications,” in Proc. 30th Asilomar Conf. Signals, Systems, Comp.,
vol. 1, November 1996, pp. 738–741.
[63] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products. Fifth
Avenue,New York, USA: Academic Press, 1980.
[64] S. Haykin, Communication Systems. New York: John Wiley & Sons, 2001.
[65] J. B. Andersen, T. S. Rappaport, and S. Yoshida, “Propagation measurements and
models for wireless communications channels,” IEEE Commun. Magn., vol. 33, no. 1,
p. 4249, January 1995.
226
BIBLIOGRAPHY
[66] J. B. Andersen and P. Eggers, “A heuristic model of power delay profiles in land-
mobile communications,” in Proc. URSI Int. Symp. Electromagnetic Theory, Sydney,
Australia, 1992.
[67] J. G. Proakis, Digital Communications, 4th ed. New York: McGraw-Hill, 2001.
[68] G. Turin, F. Clapp, T. Johnston, S. Fine, and D. Lavry, “A statistical model of urban
multipath propagation,” IEEE Trans. Veh. Technol., vol. VT-21, pp. 1–9, February
1972.
[69] D. C. Cox and R. P. Leck, “Distributions of multipath delay spread and average excess
delay for 910 M Hz urban mobile radio paths,” IEEE Trans. Antennas and Propagat.,
vol. AP-23, no. 5, pp. 206–213, March 1975.
[70] H. Suzuki, “A statistical model for urban radio propagation,” IEEE Trans. Commun.,
vol. COM-25, pp. 673–680, July 1977.
[71] J. Laurila, K. Kalliola, M. Toeltsch, K. Hugl, P. Vainikainen, and E. Bonek, “Wideband
3D characterization of mobile radio channels in urban environment,” IEEE Trans.
Antennas and Propagat., vol. 50, no. 2, pp. 233–243, February 2002.
[72] H. Hashemi, “Simulation of the urban radio propagation channel in the presence of
distant reflectors,” in Proc. IEEE Global Telecommun. Conf. (GLOBECOM’91), vol. 3,
December 1991, pp. 1940–1946.
227
BIBLIOGRAPHY
[73] L. Jiang and S. Y. Tan, “Geometrically-based channel model for mobile-communication
systems,” Microwave and Optical Technology Lett., vol. 45, no. 6, pp. 522–528, June
2005.
[74] S. P. Stapleton, X. Carbo, and T. McKeen, “Tracking and diversity for a mobile
communications base station array antenna,” in Proc. 46th IEEE Veh. Technol. Conf.,
vol. 3, April/May 1996, pp. 1695–1699.
[75] J. G. Markoulidakis, G. L. Lyberopoulos, D. F. Tsirkas, and E. D. Sykas, “Mobil-
ity modeling in third-generation mobile telecommunication systems,” IEEE Personal
Commun., pp. 41–56, August 1997.
[76] S. Hu, H. Hallen, and A. Duel-Hallen, “Physical channel modeling, adaptive prediction
and transmitter diversity for flat fading mobile channel,” in Proc. 2nd IEEE Workshop
Signal Processing Advances Wireless Commun., 1999, pp. 387–390.
[77] C. B. Margarita, V. M. Josep, and M. M. Olga, “Time evolutionary wideband propa-
gation channel model for user positioning in UMTS,” in Proc. 54th IEEE Veh. Technol.
Conf., October 2001, pp. 1442–1446.
[78] H. Omote and T. Fujii, “Time-space path model in wide-band mobile communica-
tions,” in Proc. 57th IEEE Veh. Technol. Conf., vol. 3, April 2003, pp. 1659–1663.
[79] P. I. Bratanov and E. Bonek, “Mobility model of vehicle-borne terminals in urban
cellular systems,” IEEE Trans. Veh. Technol., vol. 52, no. 4, pp. 947–952, July 2003.
228
BIBLIOGRAPHY
[80] T. K. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Palma, “A survey of vari-
ous propagation models for mobile communication,” IEEE Antennas Propagat. Mag.,
vol. 45, no. 3, pp. 51–82, June 2003.
[81] R. J. Piechocki, G. V. Tsoulos, and J. P. McGeehan, “A new stochastic spatio-temporal
propagation model (SSTPM) for mobile communications with antenna arrays,” IEEE
Trans. Commun., vol. 49, no. 5, pp. 855–862, May 2001.
[82] O. Norklit, P. Eggers, P. Zetterberg, and J. B. Andersen, “The angular aspect of
wideband modelling and measurments,” in Proc. 4th IEEE International Symp. Spread
Spectrum Techniques Applications (ISSSTA), vol. 1, September 1996, pp. 73–78.
[83] D. V. Sarwate and M. B. Pursley, “Cross-correlation properties of pseudorandom and
related sequences,” Proc. IEEE, vol. 68, pp. 593–619, May 1980.
[84] M. B. Pursley, D. V. Sarwate, and W. E. Stark, “Error probability for direct-sequence
spread-spectrum multiple-access communications, Part I: Upper and lower bounds,”
IEEE Trans. Commun., vol. COM-30, pp. 975–984, May 1982.
[85] P. Rapajic and B. Vucetic, “Adaptive receiver structures for asynchronous CDMA
systems,” IEEE J. Select. Areas Commun., vol. 12, no. 4, pp. 685–697, May 1994.
[86] ——, “Linear adaptive transmitter-receiver structures for asynchronous CDMA sys-
tems,” Eur. Trans. Telecommun., vol. 6, no. 1, pp. 21–27, January/February 1995.
[87] S. Verdu, Multiuser Detection. Cambridge, U. K.: Cambridge University Press, 1998.
229
BIBLIOGRAPHY
[88] C. Antón-Haro, J. A. R. Fonollosa, Z. Zvonar, and J. R. Fonollosa, “Probabilistic
algorithms for blind adaptive multiuser detection,” IEEE Trans. Signal Proc., vol. 46,
no. 11, pp. 2953–2966, November 1998.
[89] G. Woodward and B. S. Vucetic, “Adaptive detection for DS-CDMA,” Proc. of the
IEEE, vol. 86, no. 7, pp. 1413–1434, July 1998.
[90] G. S. Rajappan and M. L. Honig, “Signature sequence adaptation for DS-CDMA with
multipath,” IEEE J. select. areas Commun., vol. 20, no. 2, pp. 384–395, February
2002.
[91] S. Verdu, “Minimum probability of error for asynchronous gaussian multiple-access
channels,” IEEE Trans. Inform. Theory, vol. IT-32, pp. 85–96, January 1986.
[92] A. Duel-Hallen, J. Holtzman, and Z. Zvonar, “Multiuser detection for CDMA systems,”
IEEE Personal Commun., vol. 2, no. 2, pp. 46–58, April 1995.
[93] B. R. Petersen and D. D. Falconer, “Minimum mean square equalization in cyclosta-
tionary and stationary interference analysis and subscriber line calculations,” IEEE J.
Select. Areas Commun., vol. 9, no. 6, pp. 931–940, August 1991.
[94] R. Kohno, H. Imai, M. Hatori, and S. Pasupathy, “Combination of an adaptive array
antenna and a canceller of interference for direct-sequence spread-spectrum multiple-
access system,” IEEE J. Select. Areas Commun., vol. SAC-8, no. 4, pp. 691–699, May
1990.
230
BIBLIOGRAPHY
[95] B. R. Vojčić and W. M. Jang, “Transmitter precoding in synchronous multiuser com-
munications,” IEEE Trans. Commun., vol. 46, no. 10, pp. 1346–1355, October 1998.
[96] Z. Tang and S. Cheng, “Interference cancellation for DS-CDMA systems over flat
fading channels through pre-decorrelating,” in Proc. IEEE PIMRC’94, vol. 2, Hague,
Netherlands, September 1994, pp. 435–438.
[97] W. M. Jang, B. R. Vojčić, and R. L. Pickholtz, “Joint transmitter-receiver optimiza-
tion in synchronous multiuser communications over multipath channels,” IEEE Trans.
Commun., vol. 46, no. 2, pp. 269–278, February 1998.
[98] S. Ulukus and R. D. Yates, “Iterative construction of optimum signature sequence
sets in synchronous CDMA systems,” IEEE Trans. Inform. Theory, vol. 47, no. 5, pp.
1989–1998, July 2001.
[99] S. Ulukus and A. Yener, “Iterative transmitter and receiver optimization for CDMA
networks,” IEEE Trans. Wireless Commun., vol. 3, no. 6, pp. 1879–1884, November
2004.
[100] M. Honig, U. Madhow, and S. Verdu, “Blind adaptive multiuser detection,” IEEE
Trans. Info. Theory, vol. 41, no. 4, pp. 944–960, July 1995.
[101] A. H. Sayed and T. Kailath, “State-space approach to adaptive RLS filtering,” IEEE
signal proc. magazine, vol. 11, no. 3, pp. 18–60, July 1994.
231
BIBLIOGRAPHY
[102] S. J. Howard and K. Pahlavan, “Autoregressive modelling of wide-band indoor radio
propagation,” IEEE Trans. Commun., vol. 40, no. 9, pp. 1540–1552, September 1992.
[103] O. A. Dobre and O. Panfilov, “Linear modeling of the flat fading channels,” in Proc.
IEEE Int. Conf. Telecommun. (ICT), Bucharest, Romania, June 2001.
[104] K. E. Baddour and N. C. Beaulieu, “Autoregressive models for fading channel simula-
tion,” in Proc. IEEE Global Telecommun. Conf., vol. 2, 2001, pp. 1187–1192.
[105] G. B. Giannakis, Y. Hua, P. Stoica, and L. Tong, Signal Processing Advances in Wire-
less and Mobile Communications. Upper Saddle River, New Jersey, USA: Prentice
Hall, 2001, vol. 1.
[106] Z. Liu, X. Ma, and G. B. Giannakis, “Spacetime coding and kalman filtering for
time-selective fading channels,” IEEE Trans. Commun., vol. 50, no. 2, pp. 183–186,
February 2002.
[107] C. Komninakis, C. Fragouli, A. H. Sayed, and R. D. Wesel, “Multi-input multi-output
fading channel tracking and equalization using kalman estimation,” IEEE Trans. Signal
Proc., vol. 50, no. 5, pp. 1065–1076, May 2002.
[108] L. Lindbom, A. Ahlén, M. Sternad, and M. Falkenström, “Tracking of time-varying
mobile radio channels - part II: A case study,” IEEE Trans. Commun., vol. 50, no. 1,
pp. 156–167, January 2002.
232
BIBLIOGRAPHY
[109] G. W. K. Colman, S. D. Blostein, and N. C. Beaulieu, “An ARMA multipath fading
simulator,” in Proc. 7th Virginia Tech / MPRG Symp. on Wireless Personal Commun.,
1997, pp. 5.1–5.12.
[110] C. Tepedelenlioğlu, A. Abdi, G. B. Giannakis, and M. Kaveh, “Estimation of Doppler
spread and signal strength in mobile communications with applications to handoff and
adaptive transmission,” Wireless Commun. Mob. Comput., no. 1, pp. 221–242, 2001.
[111] P. A. Bello, “Characterization of randomly time-variant linear channels,” IEEE Trans.
Commun. Syst., vol. CS-11, pp. 360–393, December 1963.
[112] J. G. Proakis, C. M. Rader, F. Ling, and C. L. Nikias, Advanced Digital Signal Pro-
cessing. New York: MacMillan, 1992.
[113] C. W. Therrien, Discrete Random Signals and Statistical Signal Processing. Engle-
wood Cliffs, New Jersey, USA: Prentice Hall, 1992.
[114] H. Wang and P. Chang, “On verifying the first-order Markovian assumption for a
Rayleigh fading channel model,” IEEE Trans. Veh. Technol., vol. 45, pp. 353–357,
May 1996.
[115] J. V. Candy, Signal processing: The model-based approach. New York, USA: McGraw-
Hill, 1986.
[116] A. H. Jazwinski, Stochastic processes and filtering theory. New York, USA: Academic
Press, 1976.
233
BIBLIOGRAPHY
[117] M. Z. Win, G. Chrisikos, and N. R. Sollenberger, “Performance of RAKE reception in
dense multipath channels: Implications of spreading bandwidth and selection diversity
order,” IEEE J. Select. Areas Commun., vol. 18, no. 8, pp. 1516–1525, August 2000.
[118] Chi-Min Li and Hsueh-Jyh Li, “A novel RAKE receiver finger number decision rule,”
IEEE Antennas Wireless Propagat. Lett., vol. 2, no. 1, pp. 277–280, 2003.
[119] M. Abou-Khousa, M. El-Tarhuni, and A. Ghrayeb, “A novel finger assignment algo-
rithm for RAKE receivers in CDMA systems,” in Proc. IEEE Int. Conf. Commun.,
vol. 5, Paris, France, June 2004, pp. 2516–2520.
[120] P. Petrus, J. H. Reed, and T. S. Rappaport, “Effects of directional antennas at the
base station on the Doppler spectrum,” IEEE Commun. Lett., vol. 1, no. 2, pp. 40–42,
March 1997.
[121] Wee Teck Ng and V. K. Dubey, “Comments on ‘On the doppler spectrum at the
mobile unit employing a directional antenna’,” IEEE Commun. Lett., vol. 6, no. 11,
pp. 472–474, November 2002.
[122] ——, “Effect of employing directional antennas on mobile OFDM system with time-
varying channel,” IEEE Commun. Lett., vol. 7, no. 4, pp. 165–167, April 2003.
[123] P. D. Teal, T. D. Abhayapala, and R. A. Kennedy, “Spatial correlation for general
distributions of scatterers,” IEEE Signal Processing Lett., vol. 9, no. 10, pp. 305–308,
October 2002.
234
BIBLIOGRAPHY
[124] D. B. Smith and T. D. Abhayapala, “Generalised space-time model for Rayleigh fading
channels with non-isotropic scatterer distribution,” Electron. Lett., vol. 39, no. 21, pp.
1541–1543, October 2003.
235

Whole PDF

Uploaded by

Copyright:

Available Formats

Whole PDF

Uploaded by

Document Information

Original Title

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

Whole PDF

Uploaded by

Copyright:

Available Formats

MODELING AND CHARACTERIZATION OF

MULTIPATH FADING CHANNELS IN CELLULAR

MOBILE COMMUNICATION SYSTEMS

A DISSERTATION SUBMITTED IN FULFILMENT OF THE REQUIREMENTS

Noor Muhammad Khan

SCHOOL OF ELECTRICAL ENGINEERING AND TELECOMMUNICATIONS

This thesis is based mainly on the following publications:

[J01] Noor M. Khan, Mohammed T. Simsim and Predrag B. Rapajic, “A Generalized

2006, Paper ID: VT-2005-00203. Paper Status: In press

Paper Status: With Associate Editor, Prof. Ali Abdi

Scattering Distribution”, International J. on Info. Technol. & Telecommun. (IJIT),

vol. 2/2006, pp. 97-105, June 2006.

to Local and Distant Scatterers in Cellular Environments”, Submitted to IEE Proc.

on Microwaves, Antennas and Propagat. in March 2006, Paper ID: MAP-2006-0066,

Paper Status: Under Review

Kalman Filter Estimation in Channel Modeling for Multiuser Detection in Time-

Varying Environment”, in Proc. International Workshop on Ultra Wideband Systems

(IWUWS’03), Oulu, Finland, June 2003, Paper No. 1052

Based Channel Estimation”, in Proc. Australian Telecommun. Networks and Applica-

based Channel Estimation in OFDM Multi-user Single-Input-Multi-Output System”,

in Proc. IEEE International Symp. Spread Spectrum Techniques and Applications

(ISSSTA’04), Sydney, Australia, August 2004, pp. 802-806

Commun. (SympoTIC’04), Bratislava, Slovak Republic, October 2004, pp. 33-38

Aspects of Mobile Channel in Uniformly Distributed Scatters”, in Proc. 8th IEEE

International Conf. on Telecommun. (ConTel’05), Zagreb, Croatia, vol. 1, June 2005,

Conf. (VTC’05-Fall), Dallas, USA, vol. 4, September 2005, pp. 2201-2205

Asia-Pacific Conf. on Commun. (APCC’05), Perth, Australia, October 2005, pp.

Research Symp. (PIERS’06), Cambridge, USA, March 2006

Channel Tracking”, Accepted for publication in Proc. the Progress in Electromagnetic

Research Symp. (PIERS’06), Cambridge, USA, March 2006

Research Symp. (PIERS’06), Cambridge, USA, March 2006

Melbourne, Australia, May 2006

“Time of Arrival Statistics in Cellular Environments”, Accepted for publication in

Velencia, Spain, September 2006

List of Tables xviii

List of Figures xix

1.1 History of the Capacity Demands in Wireless Communication Systems . . . 1

1.2 Multiantenna Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2.1 Smart Antenna Systems . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.3 Channel Fading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3.1 Mobile Radio Propagation Environment . . . . . . . . . . . . . . . . 6

1.3.2 Small-Scale Multipath Fading . . . . . . . . . . . . . . . . . . . . . . 8

1.3.3 Types of Small-Scale Fading . . . . . . . . . . . . . . . . . . . . . . . 11

1.3.3.1 Based on Multipath Time-Delay Spread . . . . . . . . . . . 11

1.3.3.2 Based on Doppler Spread . . . . . . . . . . . . . . . . . . . 13

1.4 Cellular Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.4.1 Picocell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.4.2 Microcell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.4.3 Macrocell Environment . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.5 Capacity of MIMO Systems in Fading Channels . . . . . . . . . . . . . . . . 18

1.6 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.7 Scope of the Thesis and Proposed Methodology . . . . . . . . . . . . . . . . 23

1.8 Contributions of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2 A Generalized Spatial Channel Model 30

2.1.1 Overview of Physical Spatial Channel Modeling . . . . . . . . . . . . 31

2.1.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.1.3 Contributions in the Chapter . . . . . . . . . . . . . . . . . . . . . . 35

2.3 General Channel Modeling Parameters . . . . . . . . . . . . . . . . . . . . . 39

2.3.1 Shape of the Scattering Region . . . . . . . . . . . . . . . . . . . . . 39