Channel Simulation Models for Mobile

Broadband Communication Systems

Carlos A. Gutirrez

Channel Simulation Models for Mobile

Broadband Communication Systems

Doctoral Dissertation

University of Agder
Faculty of Engineering and Science
2009

Doctoral Dissertations at the University of Agder 16

ISBN : 978-82-7117-658-7
ISSN: 1504-9272
Carlos Adrin Gutirrez Daz de Len, 2009
Printed by the Printing Office, University of Agder
Kristiansand

To my wife and child, Andrea and Pablo

and to my parents, Imelda and Carlos

Contents
Summary

xi

Acknowledgments

xiii

List of Figures

xix

List of Tables

xxi

Acronyms

xxiii

1 Introduction
1.1 Mobile Broadband Wireless Communication Systems . . .
1.2 Channel Simulation Models . . . . . . . . . . . . . . . . .
1.2.1 The Need for Channel Simulation Models . . . . .
1.2.2 The Sum-of-Cisoids Channel Simulation Approach
1.3 Goals of the Doctoral Project . . . . . . . . . . . . . . . .
1.4 Organization of the Manuscript . . . . . . . . . . . . . . .
2 The
2.1
2.2
2.3

2.4

2.5

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

Reference Model
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The Narrowband Rician Fading Channel Model . . . . . . . . . .
Statistical Characterization of the Reference Channel Model . . .
2.3.1 ACFs . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2 PSDs . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.3 Average Doppler Shift and Doppler Spread . . . . . . . .
Statistical Characterization of the Signal Fading . . . . . . . . .
2.4.1 PDFs of the Channels Envelope and Phase . . . . . . . .
2.4.1.1 First-Order PDF of the Envelope . . . . . . . .
2.4.1.2 First-Order PDF of the Phase . . . . . . . . . .
2.4.2 LCR and ADF . . . . . . . . . . . . . . . . . . . . . . . .
2.4.2.1 LCR . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.2.2 ADF . . . . . . . . . . . . . . . . . . . . . . . .
2.4.3 ACF of the Channels Squared Envelope . . . . . . . . . .
Distributions of the AOA and Particular Cases . . . . . . . . . .
2.5.1 The von Mises PDF of the AOA and the Associated ACF
and DPSD of the Channels Diffuse Component . . . . . .
vii

.
.
.
.
.
.

1
1
2
2
4
5
6

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
7
7
9
9
11
11
13
13
13
14
15
16
17
17
18

. 18

viii

Channel Simulation Models for Mobile Broadband Communications

2.5.2
2.6

The Laplacian PDF of the AOA and the Associated ACF

and DPSD of the Diffuse Component . . . . . . . . . . . . . 20
Chapter Summary and Conclusions . . . . . . . . . . . . . . . . . . 23

3 Narrowband SOC-Based Simulation Models

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 SOC-Based Channel Simulators: General Structure and Classes . .
3.3 The Stochastic SOC-Based Simulation Model . . . . . . . . . . . .
3.3.1 Structure and Considerations . . . . . . . . . . . . . . . . .
3.3.2 Mean Value, Variance, and Average Power . . . . . . . . . .
3.3.3 Correlation Properties and Spectral Characteristics . . . . .
3.3.3.1 ACFs . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.3.2 DPSDs . . . . . . . . . . . . . . . . . . . . . . . .
3.3.3.3 Average Doppler Shift and Doppler Spread . . . .
3.3.4 PDFs of the Stochastic SOC Models Envelope and Phase .
3.3.4.1 First-Order PDF of the Envelope . . . . . . . . .
3.3.4.2 First-Order PDF of the Phase . . . . . . . . . . .
3.3.5 A Note on the LCR and ADF of the Stochastic SOC Models Envelope . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.6 ACF of the Stochastic SOC Models Squared Envelope . . .
3.4 The Deterministic SOC-Based Simulation Model . . . . . . . . . .
3.4.1 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.2 Mean Value and Average Power . . . . . . . . . . . . . . . .
3.4.3 Correlation Properties and Spectral Characteristics . . . . .
3.4.3.1 Time-averaged ACFs . . . . . . . . . . . . . . . .
3.4.3.2 DPSDs . . . . . . . . . . . . . . . . . . . . . . . .
3.4.3.3 Average Doppler Shift and Doppler Spread . . . .
3.4.4 ACF of the Deterministic Squared Envelope . . . . . . . . .
3.5 On the Ergodicity of the Stochastic SOC-Based Simulation Model
3.6 Chapter Summary and Conclusions . . . . . . . . . . . . . . . . . .
4 Parameter Computation Methods
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Overview of the Existing Parameter Computation Methods for
SOC-Based Channel Simulators . . . . . . . . . . . . . . . . . . .
4.2.1 The EMEDS . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.2 The LPNM . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 The GMEA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Description . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.2 Differences Between the GMEA and Other Versions of the
MEA . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . .
4.3.3.1 von Mises Distributed AOAs . . . . . . . . . . .
4.3.3.2 Laplacian Distributed AOAs . . . . . . . . . . .
4.4 The RSAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.1 Description . . . . . . . . . . . . . . . . . . . . . . . . . .

25
25
26
27
27
28
28
28
30
30
31
32
32
34
35
35
35
36
37
37
38
38
39
40
42

43
. 43
.
.
.
.
.

44
44
45
46
46

.
.
.
.
.
.

48
49
49
51
53
53

Contents (Continuation)

4.5

4.6

4.4.1.1 Basic Approach . . . . . . . . . . . . . . . . . .

4.4.1.2 Problems of the Basic Approach . . . . . . . . .
4.4.1.3 Improved Approach . . . . . . . . . . . . . . . .
4.4.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . .
4.4.2.1 von Mises Distributed AOAs . . . . . . . . . . .
4.4.2.2 Laplacian Distributed AOAs . . . . . . . . . . .
Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . .
4.5.1 Emulation of the Correlation and Spectral Characteristics
of the Reference Models Random Component . . . . . . .
4.5.1.1 ACF . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.1.2 Average Doppler Shift and Doppler Spread . . .
4.5.2 Emulation of the Fading Statistics of the Reference Model
4.5.2.1 PDF of the Envelope . . . . . . . . . . . . . . .
4.5.2.2 PDF of the Phase . . . . . . . . . . . . . . . . .
4.5.2.3 ACF of the Squared Envelope . . . . . . . . . .
Chapter Summary and Conclusions . . . . . . . . . . . . . . . . .

ix
.
.
.
.
.
.
.

53
54
56
57
57
58
59

.
.
.
.
.
.
.
.

61
61
67
70
70
74
79
90

5 Simulation of Mobile MIMO Fading Channels Under SingleBounce Non-Isotropic Scattering Conditions
91
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.2 Reference Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2.1 The Geometrical One-Ring Scattering Model . . . . . . . . 92
5.2.2 Narrowband MIMO Rayleigh Fading Channel Model . . . . 94
5.2.3 Correlation Properties of the Reference Model . . . . . . . 95
5.2.3.1 ACF and DPSD of the MIMO Channel Gains . . 95
5.2.3.2 Space-Time CCF of the MIMO Channel Gains . . 96
5.3 The Simulation Model . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.3.1 The Stochastic SOC Simulation Model . . . . . . . . . . . . 96
5.3.2 Correlation Properties of the SOC Simulation Model . . . . 97
5.4 Parameter Computation Methods . . . . . . . . . . . . . . . . . . . 97
5.4.1 The GMEA . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.4.2 The RSAM . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.5 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.5.1 Considerations . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.5.2 Emulation of the ACF . . . . . . . . . . . . . . . . . . . . . 101
5.5.3 Emulation of the SCCF . . . . . . . . . . . . . . . . . . . . 101
5.6 Simulation of Wideband MIMO Fading Channels . . . . . . . . . . 106
5.6.1 Modeling and Statistical Characterization of Wideband
Mobile MIMO Fading Channels . . . . . . . . . . . . . . . . 106
5.6.2 Simulation Approach . . . . . . . . . . . . . . . . . . . . . . 107
5.7 Chapter Summary and Conclusions . . . . . . . . . . . . . . . . . . 109
6 Design of TDL Models for WSSUS Channels with Diffuse Impulse Responses
111
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.2 About the Validity of SSTDL Models . . . . . . . . . . . . . . . . 113

Channel Simulation Models for Mobile Broadband Communications

6.3

6.4

6.5
6.6

6.2.1 The SSTDL Modeling Approach . . . . . . . . . . . . .

6.2.2 The Validity of SSTDL Models . . . . . . . . . . . . . .
Statistical Behavior of SSTDL Models . . . . . . . . . . . . . .
6.3.1 Statistical Description of the Reference Channel Model
6.3.2 Statistical Description of -Spaced TDL Models . . . .
Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . .
6.4.1 The Truncated-Exponential-Decay PDP . . . . . . . . .
6.4.2 Effects of Aliasing on the FCF . . . . . . . . . . . . . .
6.4.3 The Influence of the FCF on the System Performance .
Avoiding the Problems Caused by SSTDL Models . . . . . . .
Chapter Summary and Conclusions . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

113
114
116
116
117
119
119
120
120
123
125

7 Summary of Contributions and Open Problems

129
7.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.2 Open Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
A Derivation of the Squared Envelope ACF of the Reference
Model

131

B Derivation of the Squared Envelope ACF of the Stochastic

SOC-Based Simulation Model

135

C Derivation of the Squared Envelope ACF of the Deterministic

SOC-Based Simulation Model
137
D Closed-Form Expression for the Squared Envelope ACF of the
Sample Functions of the Stochastic Homogeneous SOC Model
with Uncorrelated IQ Components
141
E Theorems About the Design of SOC-Based Simulation Models
with Uncorrelated IQ Components
145
F Validation of the Expressions Obtained for the Squared Envelope
ACF of the Deterministic SOC-Based Simulation Model
149

Summary
Mobile broadband wireless communication systems (MBWCS) are emerging as
a solution to provide broadband services to users on the move. These systems
are expected to operate in a wide variety of propagation scenarios, at different
mobile speeds, and at various frequency bands. Under such a variety of requirements, flexible and efficient channel simulation models will prove fundamental
for the laboratory analysis of MBWCS. Currently, most of the existing channel
simulation models are either too complex as to allow for an efficient performance
investigation of MBWCS, or they cannot be applied to the simulation of some relevant classes of mobile fading channels. To overcome these limitations, we present
in this doctoral a flexible and efficient methodology for the design of channel simulation models for MBWCS. Such a methodology is based on the sum-of-cisoids
(SOC) approach, an approach that is closely in line with the electromagnetic
plane-wave propagation model.
We build our channel simulators upon a class of ergodic SOC simulation
models. For the computation of the SOC model parameters, we introduce two
simple methods that enable the design of simulation models for mobile fading
channels characterized by any type of Doppler power spectral densities (DPSDs).
The proposed methods are well-suited for the simulation of both single-input
single-output (SISO) and multiple-input multiple-output (MIMO) channels. We
evaluate the methods performance with respect to their accuracy for emulating
important statistical functions of the channel, such as the autocorrelation function
(ACF), the envelope probability density function (PDF), and the ACF of the
squared envelope. In the case of MIMO channels, we evaluate the methods
performance in terms of the approximation of the channel temporal ACF and
spatial cross-correlation function (SCCF). The obtained results demonstrate the
excellent performance of the proposed methods.
This dissertation is also intended to provide a comprehensive treatise of the
theory behind the design of SOC simulation models for mobile fading channels.
In this respect, the statistical properties of SOC channel simulators are thoroughly analyzed. Important contribution are given concerning the correlation
properties of the square envelope of SOC simulators. Such contributions include
the derivation of closed-form expressions for the squared envelope ACF of the
xi

xii

Channel Simulation Models for Mobile Broadband Communications

SOC simulation model, and the analysis of the ergodicity properties of the SOC
models squared envelope.
We also revisit here the concept of the symbol-spaced tapped line model
(SSTDL) for WSSUS channels. In this regard, we present a discussion on the
problems of SSTDL models, and we propose a simple solution to avoid them.
The usefulness of such a solution is exemplary demonstrated by analyzing the bit
error probability of a multi-carrier code division multiple access (MC-CDMA)
system.

Acknowledgments
The completion of this Dissertation was a monumental task that I would not have
been able to accomplish without the help and the support of many individuals
and institutions to whom I want to express my gratitude.
I am very grateful to my advisor, Prof. Matthias P
atzold. I learned a lot
working with him and his guidance was decisive to achieve the goals of my PhD
project. I am also thankful to the members of the evaluation committee, Profs.
Dongwoo Kim, Neji Youssef, and Jose J. Gonzalez. My gratitude goes also to the
Doctoral Fellows from the ICT Department, specially to Ali Chelli, Batool Talha,
Bjrn Olav Hogstad, Dmitry Umansky, Gulzaib Rafiq, Martin Choux, Ram Kumar, Sigurd Eskeland, and Yuanyuan Ma. They made my stay in Grimstad
something pleasant to remember and I benefited so much from their technical
knowledge and expertise. Thanks go to my teachers Prof. Andreas Prinz and
Prof. Vladimir Oleshchuk. I am thankful too to the Project Secretary of the
Mobile Communications Group, Mrs. Katharina P
atzold, and the Coordinator
of the PhD Program in Mobile Communication Systems, Mrs. Trine Tonnessen.
I kindly acknowledge the financial support that the Mexican National Council
of Science and Technology (CONACyT) gave me to pursue the PhD degree in
Norway. I also acknowledge the opportunity given to me by the University of
Agder to make my PhD in a topic I feel so passionate about.
I was lucky to enjoy the friendship of Ahmed A. A. S. AbouGhonim, Fanny
Paris, Mikael Snaprud, and Morten Goodwin Olsen. Thanks go to all of them.
I am thankful too to my friends and teachers from CICESE, specially to Paul
Medina, Prof. Jaime S
anchez, Prof. Roberto Conte, and Prof. Jose R. Gallardo.
My deepest thanks to my parents, Imelda and Carlos; my brothers, Luis
and Aaron; my sisters-in-law, Monica and Eli; my parents-in-law, Guadalupe
and Ramiro; and to my brother-in-law, Ramiro. They sacrificed a lot in order
for me to make my PhD in Norway. There are no words I can use to express
my gratefulness to my wife, Andrea. She gave me the heart to complete this
enterprise. I am also grateful to Pablo, my son. He was my motivation during
the final and decisive stage of my doctoral work.
Last but not least, I thank the Lord. Bringing my doctoral studies to a good
end has been another of the many gifts he has given to me.
Carlos A. Gutierrez
June 11, 2009
Aguascalientes, Ags., Mexico
xiii

xiv

Channel Simulation Models for Mobile Broadband Communications

List of Figures
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
3.1
3.2

3.3

4.1

4.2

4.3

Channels envelope and its first-order PDF for different Rician

factors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Channels phase and its first-order PDF for different Rician factors.
The von Mises PDF of the AOA. . . . . . . . . . . . . . . . . . . .
ACF of the channels diffuse component by considering the von
Mises PDF of the AOA. . . . . . . . . . . . . . . . . . . . . . . . .
DPSD of the channels diffuse component by considering the von
Mises PDF of the AOA. . . . . . . . . . . . . . . . . . . . . . . . .
The Laplacian PDF of the AOA. . . . . . . . . . . . . . . . . . . .
ACF of the channels diffuse component by considering the Laplacian PDF of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . .
DPSD of the channels diffuse component by considering the Laplacian PDF of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . .

14
15
20
20
21
22
23
23

Block diagram of an SOC-based simulation model for Rician fading

channels. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Comparison between the theoretical envelope PDF of the stochastic SOC-based simulation model and the Rice PDF by considering
different values for both the Rician factor and the number of cisoids. 33
Comparison between the theoretical phase PDF of the stochastic SOC-based simulation model and the phase PDF of reference
model by considering different values for both the Rician factor
and the number of cisoids. . . . . . . . . . . . . . . . . . . . . . . . 34
Comparison between the ACF of the channels diffuse component
and the ACF of the SOC-based simulation models random component by considering the von Mises PDF of the AOA and the
generalization of the MEA proposed by other authors. . . . . . . . 50
Comparison between the DPSD of the channels diffuse component and the DPSD of the SOC-based simulation models random
component by considering the von Mises PDF of the AOA and the
GMEA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Comparison between the DPSD of the channels diffuse component and the DPSD of the SOC-based simulation models random
component by considering the Laplacian PDF of the AOA and the
GMEA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
xv

xvi

Channel Simulation Models for Broadband Wireless Communications

4.4

Relationship between the gains and AOAs of the SOC-based simulation models random component by applying the BRSAM to the
von Mises PDF of the AOA. . . . . . . . . . . . . . . . . . . . . . . 55

4.5

Comparison between the Rayleigh PDF and the first-order PDF

of the envelope of the simulation models random component by
applying the BRSAM to the von Mises PDF of the AOA. . . . . . 55

4.6

Comparison between the DPSD of the channels diffuse component and the DPSD of the SOC-based simulation models random
component by considering the von Mises PDF of the AOA and the
RSAM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.7

Comparison between the DPSD of the channels diffuse component and the DPSD of the SOC-based simulation models random
component by considering the Laplacian PDF of the AOA and the
RSAM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4.8

Comparison among the GMEA, RSAM, and LPNM I in terms

of the emulation of the ACF of channels diffuse component by
considering the von Mises PDF of the AOA. . . . . . . . . . . . . . 62

4.9

Comparison among the GMEA, RSAM, and LPNM II in terms

of the emulation of the ACF of channels diffuse component by
considering the von Mises PDF of the AOA. . . . . . . . . . . . . . 63

4.10 Comparison among the GMEA, RSAM, and LPNM III in terms
of the emulation of the ACF of channels diffuse component by
considering the von Mises PDF of the AOA. . . . . . . . . . . . . . 64
4.11 Root mean square error between the ACF of the channels diffuse
component and the ACF of the SOC-based simulation models
random component by considering the von Mises PDF of the AOA
with different pairs of parameters. . . . . . . . . . . . . . . . . . . 66
4.12 Absolute error between the average Doppler shift of the channels diffuse component and the average Doppler shift of the SOCbased simulation models random component by considering the
von Mises PDF of the AOA and the GMEA, RSAM, and LPNM. . 68
4.13 Relative error between the Doppler spread of the channels diffuse
component and the Doppler spread of the SOC-based simulation
models random component by considering the von Mises PDF of
the AOA and the GMEA, RSAM, and LPNM. . . . . . . . . . . . 69
4.14 Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOC-based
simulation model by applying the GMEA to the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.15 Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOC-based
simulation model by applying the LPNM I to the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

Figures (Continuation)

xvii

4.16 Comparison between the first-order PDF of the reference models

envelope and the first-order PDF of the envelope of the SOC-based
simulation model by applying the LPNM II to the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.17 Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOC-based
simulation model by applying the LPNM III to the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.18 Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOC-based
simulation model by applying the RSAM to the von Mises PDF of
the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.19 Root mean square error between the envelope distributions of the
channels diffuse component and SOC-based simulation models
random component by considering the von Mises PDF of the AOA
with different pairs of parameters. . . . . . . . . . . . . . . . . . . 75
4.20 Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based simulation model by applying the GMEA to the von Mises PDF of
the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.21 Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based simulation model by applying the LPNM I to the von Mises PDF of
the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.22 Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based simulation model by applying the LPNM II to the von Mises PDF of
the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.23 Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based simulation model by applying the LPNM III to the von Mises PDF of
the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.24 Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based simulation model by applying the RSAM to the von Mises PDF of the
AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.25 Comparison among the GMEA, RSAM, and LPNM I in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to zero and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.26 Comparison among the GMEA, RSAM, and LPNM II in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to zero and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

xviii

Channel Simulation Models for Broadband Wireless Communications

4.27 Comparison among the GMEA, RSAM, and LPNM III in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to zero and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.28 Comparison among the GMEA, RSAM, and LPNM I in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to two and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.29 Comparison among the GMEA, RSAM, and LPNM II in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to two and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.30 Comparison among the GMEA, RSAM, and LPNM III in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to two and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.31 Comparison among the GMEA, RSAM, and LPNM I in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to four and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.32 Comparison among the GMEA, RSAM, and LPNM II in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to four and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.33 Comparison among the GMEA, RSAM, and LPNM III in terms of
the emulation of the ACF of the reference models squared envelope
by considering a Rician factor equal to four and the von Mises PDF
of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1
5.2

5.3
5.4

5.5

6.1

The geometrical one-ring scattering propagation model . . . . . .

Comparison between the ACF of the MIMO channel gains and
the ACF of the simulation model by applying the GMEA and the
RSAM to the von Mises PDF of the AOA. . . . . . . . . . . . . .
Absolute value of the SCCF of the MIMO channel model by considering the von Mises PDF of the AOA. . . . . . . . . . . . . . .
Error between the SCCF of the MIMO channel gains and the SCCF
of the simulation model by applying the GMEA to the von Mises
PDF of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Error between the SCCF of the MIMO channel gains and the SCCF
of the simulation model by applying the RSAM to the von Mises
PDF of the AOA. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. 83

. 84

. 85

. 86

. 87

. 88

. 89
. 93

. 102
. 103

. 104

. 105

Comparison between the absolute value of the FCF of the reference model and the absolute value of the FCF of the SSTDL by
considering the TED-PDP. . . . . . . . . . . . . . . . . . . . . . . 121

Figures (Continuation)
6.2

6.3

6.4

xix

Comparison between the theoretical and the empirical BEPs of a

MUI-free downlink MC-CDMA system with MRC by considering
the SSTDL model and the TED-PDP with a falling factor = 30
ns. In the figure, SF stands for the spreading factor. . . . . . . . . 124
Comparison between the absolute value of the FCF of the reference model and the absolute value of the FCF of the HSSTDL by
considering the TED-PDP. . . . . . . . . . . . . . . . . . . . . . . 126
Comparison between the theoretical and the empirical BEPs of a
MUI-free downlink MC-CDMA system with MRC by considering
the HSSTDL model and the TED-PDP with a falling factor = 30
ns. In the figure, SF stands for the spreading factor. . . . . . . . . 127

F.1 Comparison between the theoretical and the empirical squared envelope ACFs of the SOC-based simulation models sample functions by considering a Rician factor equal to zero and applying the
RSAM to the von Mises PDF of the AOA. . . . . . . . . . . . . . . 150
F.2 Comparison between the theoretical and the empirical squared envelope ACFs of the SOC-based simulation models sample functions by considering a Rician factor equal to two and applying the
RSAM to the von Mises PDF of the AOA. . . . . . . . . . . . . . . 151
F.3 Comparison between the theoretical and the empirical squared envelope ACFs of the SOC-based simulation models sample functions by considering a Rician factor equal to four and applying the
RSAM to the von Mises PDF of the AOA. . . . . . . . . . . . . . . 152

xx

Channel Simulation Models for Broadband Wireless Communications

List of Tables
1.1
1.2
2.1
2.2
3.1

Air Interface Specifications of LTE, UMB, and Mobile WiMAX . .

Air Interface Specifications of the standards IEEE 802.16e, IEEE
802.16m, and IEEE 802.20 . . . . . . . . . . . . . . . . . . . . . . .

3
3

Average Doppler shift and Doppler spread of the channels diffuse

component by considering the von Mises PDF of the AOA. . . . . 21
Average Doppler shift and Doppler spread of the channels diffuse
component by considering the Laplacian PDF of the AOA. . . . . 24
Classification of SOC-based simulation models for Rician fading
channels according to the nature of the cisoids parameters. . . . . 27

xxi

xxii

Channel Simulation Models for Broadband Wireless Communications

Acronyms
3G:
3GPP:
3GPP2:
ACF:
ADF:
AOA:
AOD:
AWGN:
BPSK:
BRSAM:
BS:
CCF:
CDF:
CDMA:
CIR:
CNR:
DPSD:
DS-SS:
EMEDS:
ETSI:
FCF:
GMEA:
GSM:
HSSTDL:
i.i.d.:
IP:
IQ:
ISI:
LCR:
LPNM:
LOS:
LTE:
MBWCS:
MC-CDMA:
MEA:

third generation
third generation partnership project
third generation partnership project
autocorrelation function
average duration of fades
angle of arrival
angle of departure
additive white Gaussian noise
binary phase shift keying
basic Riemann sum approximation method
base station
cross-correlation function
cumulative distribution function
code division multiple access
channel impulse response
carrier-to-noise-ratio
Doppler power spectral density
direct sequence spread spectrum
extended method of exact Doppler spread
European Telecommunication Standards Institute
frequency correlation function
generalized method of equal areas
global system for mobile communications
half-symbol spaced tapped delay line
independent and identically distributed
Internet protocol
Inphase and quadrature
inter-symbol interference
level-crossing rate
Lp -norm method
line of sight
long term evolution
mobile broadband wireless communication systems
multicarrier code division multiple access
method of equal areas

xxiii

xxiv

Channel Simulation Models for Broadband Wireless Communications

MEDS:
MIMO:
MRC:
MT:
MUI:
NLOS:
OFDM:
OFDMA:
PDF:
PDP:
PSD:
QAM:
RSAM:
SC-FDMA:
SCCF:
SDMA:
SISO:
SOC:
SOS:
SSTDL:
STCCF:
SSS:
TACF:
TDL:
TED-PDP:
TF-ACF:
TVLF:
UMB:
UMTS:
US:
VoIP:
WCDMA:
WiFi:
WiMAX:
WSS:
WSSUS:

method of exact Doppler spread

multiple-input multiple-output
maximal ratio combining
mobile terminal
multi-user interference
non-line of sight
orthogonal frequency division multiplexing
orthogonal frequency division multiple access
probability density function
power delay profile
power spectral density
quadrature amplitude modulation
Riemann sum approximation method
single carrier frequency division multiple access
spatial cross-correlation function
spatial division multiple access
single-input single-output
sum-of-cisoids
sum-of-sinusoids
symbol-spaced tapped-delay-line
space-time cross-correlation function
strict-sense stationary
temporal autocorrelation function
tapped-delay-line
truncated-exponential-decay power delay profile
time-frequency autocorrelation function
time-variant linear filter
ultra-mobile broadband
universal mobile telecommunication system
uncorrelated scattering
voice over IP
wideband code division multiple access
wireless fidelity
worldwide interoperability for microwave access
wide-sense stationary
wide-sense stationary uncorrelated scattering

Chapter 1

Introduction
1.1

Mobile Broadband Wireless Communication

Systems

In the opinion of many technology analysts, such as the authors of [Lee07, Orti07],
the next revolution in the wireless communications industry will be triggered
by the emerging mobile broadband wireless communication systems (MBWCS).
These novel systems will combine the best of modern fixed broadband wireless
access networks [Abic06, Stal04] and third generation (3G) cellular telephony systems [Rapp02, Robe06] with the aim of providing broadband multimedia services
to users on the move. MBWCS are expected to deliver high-data-rate services,
such as voice over IP (VoIP) and video on demand, to users moving at speeds as
high as those of fast trains [Li07, Bolt07].
Among the initiatives that are currently ongoing toward the standardization
of MBWCS, we can distinguish those carried out within the third-generation
partnership project (3GPP), the 3GPP 2 (3GPP2), the WiMAX ForumTM , and
the IEEE 802 group. The 3GPP [3Gweb1] is working on the standardization of
an advanced 3G cellular system called Long-Term Evolution (LTE). LTE is an
evolutionary upgrade of the Global System for Mobile Communications (GSM)
and the 3G Wideband Code Division Multiple Access (WCDMA) cellular system
[Robe06]. Its air interface utilizes an orthogonal frequency division multiple access (OFDMA) scheme on the downlink, and a single-carrier frequency division
multiple access (SC-FDMA) scheme on the uplink [Orti07, Schu05]. LTE employs
multiple-input multiple-output (MIMO) transceivers to achieve data rates of up
to 277 Mbps on the downlink and 75 Mbps on the uplink, within a 20 MHz bandwidth. It uses spatial division multiple access (SDMA) techniques to increase cell
capacity [Qua08a].
The 3GPP2 [3Gweb2] recently published the specifications of the Ultra Mobile
Broadband (UMB) standard for beyond-3G cellular systems [3gpUMB]. UMB is
1

Channel Simulation Models for Mobile Broadband Communications

an upgrade of the CDMA2000 standard for 3G cellular networks [Robe06]. The

air interface of UMB is based on an OFDMA scheme, and it incorporates MIMO
transceivers to achieve data transfer rates of up to 288 Mbps on the downlink and
75 Mbps on the uplink, within a 20 MHz bandwidth [3gpUMB]. UMB includes
beamforming and transmission modes basing on SDMA.
The WiMAX Forum, an industry-led nonprofit organization, is developing
the Mobile WiMAX standard. Mobile WiMAX is a trimmed version of the
IEEE 802.16e standard for fixed and mobile broadband wireless access networks
[IEEE06]. It defines an OFDMA-based air interface able to deliver information
at peak data rates of 60 Mbps on the downlink and 23 Mbps on the uplink over a
10 MHz bandwidth [Teo07]. A 2 2 MIMO configuration is required in order to
achieve such high data rates. Even though the Mobile WiMAX standard has not
yet been released, several trial networks basing on Mobile WiMAX have already
been deployed in several countries around the world, such as USA, Pakistan,
Korea, and Colombia [Goza06, Lee07].
The IEEE 802.20 Task Group is developing a new standard for MBWCS which
will specify an efficient packet-based air interface optimized for the transport of
Internet-Protocol- (IP)-based services [Bolt07]. In turn, the IEEE 802.16m Task
Group is working on a high throughput enhancement for licensed cellular bands
of the 802.16e standard.
We summarize the main features of LTE, UMB, and Mobile WiMAX in
Table 1.1, whereas Table 1.2 overviews the target characteristics of the IEEE
802.16m and IEEE 802.20 standards. An overview of the 802.16e standard is
also shown in Table 1.2.

1.2
1.2.1

Channel Simulation Models

The Need for Channel Simulation Models

In order to get insights into the problems that affect the performance of MBWCS,
or any other type of wireless communication systems, one often has to resort
in practice to the use of computer simulators. Computer simulators provide
a powerful, reproducible, and affordable way to assess the system performance.
They can be used as well to verify the correctness of results obtained analytically.
Choosing a proper channel simulation model is fundamental for the laboratory
analysis of MBWCS, as most of the problems affecting the performance of mobile
communication systems, e.g., path loss, shadowing, and signal fading, are caused
by the channel [Skla88]. The importance of channel simulation models is such that
the European Telecommunications Standards Institute (ETSI) has recently issued
a document with specifications for the design of channel simulators for MBWCS
[3gpSCM]. In view of the recommendations in [3gpSCM], and taking into account

Chapter 1 Introduction

Table 1.1: Air Interface Specifications of LTE, UMB, and Mobile WiMAX
Characteristics

LTE

UMB

Mobile WiMAX

Transmission
modes

OFDMA on the
uplink and
SC-FDMA on the
downlink
Licensed bands
below 2.6 GHz

OFDMA and an
alternative mode
based on CDMA

OFDMA

Licensed bands
below 2.6 GHz

Various, ranging
from 1.25 MHz to
20 MHz
MIMO (2 2,
4 4), SDMA , and
beamforming
Scalable from GSM
and WCDMA
cellular networks
Currently ongoing

Various, ranging
from 1.25 MHz to
20 MHz
MIMO (2 2,
4 4), SDMA , and
beamforming
1xEV-DO and
CDMA2000 1X
cellular networks
Released in 2007

Licensed bands
between 2.3 and
3.5 GHz
Various, ranging
from 1.25 MHz to
10 MHz
MIMO (2 2)

Bands

Bandwidth

Smart antenna
techniques
Compatibility
with other
systems
Standard status

Currently ongoing

Sources: [3gpUMB, Goza06, Li07, Qua08a, Teo07].

SDMA and MIMO are complementary to each other. They are not used together.

Table 1.2: Air Interface Specifications of the standards IEEE 802.16e, IEEE
802.16m, and IEEE 802.20
Characteristics
Transmission
modes

Bands

Bandwidth

Smart antenna
techniques

IEEE 802.16e
Three alternatives:
single-carrier,
OFDM, and
OFDMA
Licensed bands
suitable for
mobility below
6 GHz
Various, ranging
from 1.25 MHz to
20 MHz
MIMO (2 2)

Compatibility

Standard status

Released in 2006

IEEE 802.16m

IEEE 802.20

OFDMA

OFDMA

Licensed bands
below 6 GHz

Licensed bands
below 3.5 GHz

Various, ranging
from 5 MHz to
20 MHz
MIMO (2 2,
4 4) and
beamforming
802.16e networks
operating in
OFDMA mode
Currently ongoing

Various, ranging
from 1.25 MHz to
20 MHz
MIMO (different
configurations) and
beamforming
802.11a/g standard

Sources: [Bolt07, Goza06, IEEE06].

Currently ongoing

Channel Simulation Models for Mobile Broadband Communications

the characteristics of standardized channel models for MBWCS [Alme07], such

as the COST 259 [Corr01, Moli06] and the COST 273 [Corr06], we have that a
channel simulation model for MBWCS should meet the following:
They should be able to reproduce the statistical properties of directionselective mobile fading channels.
They should be adequate for the simulation of macrocell and microcell
mobile fading channels.
They should be flexible and simple enough as to allow for the evaluation
of MBWCS operating in environments that exhibit propagation conditions
that change quite often during active sessions.
They should be well-suited for the performance assessment of narrowband
and wideband MIMO systems.
They should be well-suited for the performance assessment of multicarrier
single-input single-output (SISO) and MIMO systems.

1.2.2

The Sum-of-Cisoids Channel Simulation Approach

Several different models exist that allow for a proper simulation of mobile fading
channels, including:
Models based on the Cholesky decomposition algorithm [Erte98, Beau04].
Models based on the IFFT algorithm [Youn00].
Models based on the autorregresive filtering concept[Badd05, Badd04].
Models based on ray tracing [Cora97].
However, simulation models based on the SOS principle introduced by Rice
[Rice44, Rice45] are particularly attractive for the performance evaluation of
MBWCS. In addition to providing an excellent basis for the simulation of temporally correlated narrowband channels [Jake74, Pae02a, Pate05], the SOS principle can easily be applied to the simulation of frequency-selective [Hohe92, Yip95,
Wang07] and space-selective [Han02, Paet01, Pae04a, Pae08a] radio channels.
The design of accurate and efficient SOS-based simulators for mobile fading
channels has been a topic of research of several books and papers, e.g., [Jake74,
p. 70], [Pae02a, ch. 5], and [Pate05, Hohe92, Yip95, Wang07, Han02]. Nevertheless, most of the existing SOS-based simulators have been developed on
the assumption that the channels Doppler power spectral density (DPSD) is
symmetrical with respect to the origin. This poses a serious restriction, since

Chapter 1 Introduction

it has been observed that the DPSD of real-world channels is in general nonsymmetrical [Kott04, Abd02a, Zhao03, Blau06]. Moreover, simulation models
for fading channels having asymmetrical DPSDs are necessary to assess the performance of mobile communication systems under non-isotropic scattering conditions [Pae04a, Miti04]. To close this gap, a new class of SOS channel simulators
that are able to produce complex-valued waveforms with cross-correlated inphase
and quadrature (IQ) componentsas required for the synthesis of channels characterized by asymmetrical DPSDswas introduced in [Paet01]. SOS models
of such a class, which are known as sum-of-cisoids (SOC) models [Paet07], are
closely in line with the plane-wave propagation model [Saun07], and have already
been adopted as the core structure of several space-selective channel simulators
[Paet01, Pae04a, Pae08a]. Unfortunately, the existing parameter computation
methods for SOC fading channel simulators with asymmetrical DPSD rely upon
optimization techniques that make the determination of the model parameters
a complex and time-consuming task. The development of new methods, under
the constraint of simplicity and accuracy, is therefore desirable to facilitate the
performance analysis of MBWCS.

1.3

Goals of the Doctoral Project

In this Doctoral thesis, we aim at developing accurate and efficient SOC channel simulators for MBWCS. The proposed simulators should fulfill the following
requirements:
They should enable the simulation of mobile fading channels characterized
by symmetrical and asymmetrical DPSDs.
They should be adequate for the simulation of fading channels under lineof-sight (LOS) and non-line-of-sight (NLOS) conditions.
They should be suitable for the performance analysis of MBWCS based on
SISO and MIMO technology.
The determination of the model parameters should be done in a simple and
efficient manner.
To accomplish these goals, we propose two simple and effective parameter computation methods for the design of SOC simulation models for mobile fading
channels with arbitrary DPSDs. We will show that the proposed methods emulate the statistical properties of SISO and MIMO channels with high accuracy.
It is also the objective of this dissertation to provide a comprehensive treatise
of the theory behind the design of SOC simulators for mobile fading channels.
In this respect, we present a thorough analysis of the statistical properties of

Channel Simulation Models for Mobile Broadband Communications

SOC channel simulators for mobile Rician fading channels having a time-variant
LOS component. So far, the statistical properties of SOC Rician fading channel
simulators have been studied by considering a time-invariant LOS component.
Particularly, important contributions are given regarding the correlation properties of the squared envelope of SOC simulation models for Rician fading channels.
This includes closed-form solutions for the squared envelope autocorrelation function (ACF) of SOC channel simulators, and a detailed analysis on the ergodicity
properties of the squared envelope of the simulation model.

1.4

Organization of the Manuscript

The outline to the rest of the manuscript is as follows. In Chapter 2, we review the characteristics and the statistical properties of a narrowband mobile
Rician fading channel that we will consider as a reference model to design the parameters computation methods proposed in this dissertation. In addition to the
correlation and spectral properties of the channel model, the review presented in
Chapter 2 includes information about the fading statistics, including the envelope
probability density function (PDF), the level crossing rate (LCR), the average
duration of fades (ADF), and the squared envelope ACF. In Chapter 3, we present
a stochastic SOC simulation model suitable for the simulation of the reference
model described in Chapter 2. Then, in Chapter 4, we introduce our parameter
computation methods. In Chapter 5, we show that such parameter computation
methods can be applied to the simulation of narrowband mobile MIMO fading
channels. We demonstrate in that chapter that the proposed methods produce
excellent results concerning the emulation of the correlation properties of MIMO
channels. The design of simulation models for wideband MIMO channels is also
discussed in Chapter 5. In Chapter 6, we deal with the simulation of wideband
channels. Finally, we summarize the main contributions of this dissertation in
Chapter 7.

Chapter 2

The Reference Model

2.1

Introduction

This chapter describes the characteristics and statistical properties of a narrowband mobile Rician fading channel model that we will consider as reference model
in subsequent chapters to design our channel simulators. The chapter reviews important statistical functions of the model, such as its ACF, DPSD, envelope PDF,
LCR, and ADF. In addition, it reviews the channels average Doppler shift and
Doppler spread, as well as the ACF of the channels squared envelope.
The channel model herein described is based on the assumption that in a microcellular/macrocellular small-scale propagation scenario, the narrowband signal seen by the mobile terminal (MT) is composed of a specular plane wave
and a collection of scattered plane waves that reached the MTs antenna at the
same time but through different paths. Such a modeling approach was originally proposed by Clarke in [Clar68] for characterizing non-frequency-selective
mobile SISO channels, and it was extended afterwards by other researchers to
describe narrowband mobile MIMO channels [Chen00, Shiu00, Abd02b], as well
as to characterize three-dimensional mobile SISO channels [Auli79, Pars00].
The statistical properties of the aforementioned Rician fading channel model
are studied in a number of papers and books, but mostly under the assumption
of isotropic scattering, e.g., [Jake74, Rapp02, Auli79, Pars00]. In contrast, the
review presented here covers both isotropic and non-isotropic scattering scenarios.

2.2

The Narrowband Rician Fading Channel Model

We model the non-frequency-selective mobile fading channel in the equivalent

complex baseband by a complex Gaussian process1 (t), which characterizes a
1
Throughout this dissertation, we will make use of bold symbols and letters to denote random variables and stochastic processes, whereas we will employ normal symbols and letters for
constants and deterministic processes.

Channel Simulation Models for Mobile Broadband Communications

two-dimensional small-scale propagation environment where the scattering is not

necessarily isotropic. We in turn model the process (t) by a superposition of
a stationary zero-mean complex Gaussian process (t) and a time-variant deterministic process m (t), i.e., (t) = (t) + m (t), where the random process (t)
describes the diffuse part of the channel due to scattering and m (t) characterizes
its specular part.
On the basis of the central limit theorem [Papo02, p. 278], and following
Clarkes scattering propagation model [Clar68], we can express the channels
diffuse part in terms of a sum of scattered azimuthal plane waves as follows:
(t) = lim

N
X

n=1




c n exp j 2ff n t + n

(2.1)

where the nth plane wave is characterized by a cisoid with a random gain c n , a
random phase n , and a random Doppler frequency f n . According to Clarkes
model [Clar68], the cisoids phases n are independent and identically distributed
(i.i.d.) random variables, each having a uniform distribution over [, ), while
the gains c n are given such that E{cc2n } = 2 /N for all n = 1, . . . , N , where 2
is the mean power of the channel due to scattering (0 < 2 < ); E{} denotes
statistical expectation. In turn, the Doppler frequencies f n are defined as
n ),
f n , fmax cos(

n = 1, . . . , N

(2.2)

where n is the random AOA of the nth incoming wave, and fmax stands for
the maximum Doppler frequency shift induced on the channels multipath components by the Doppler effect (fmax > 0). The AOAs n introduced above are
assumed to be i.i.d. random variables having some given PDF p (), [, ).
The gains c n , phases n , and AOAs n are furthermore considered as being mutually independent.
On the other hand, we model the specular component m (t) through a single cisoid with a deterministic gain , phase , and Doppler frequency f ,
fmax cos( ), where , [, ), and 0 < .
Under such considerations, the channel model can be written in the form
N
X






c n exp j 2ff n t + n + exp j(2f t + ) .
N
|
{z
}
n=1
{z
}
|
m (t)

(t) = lim

(2.3)

(t)

One can easily verify from the expression above that the mean value m (t) ,
(t)} of (t) is equal to m (t), that is, m (t) = m (t), whereas its variance
E{
Var{ (t)} , E{| (t)m (t)|2 } equals 2 , i.e., Var{ (t)} = 2 . In turn, the power

Chapter 2 The Narrowband Rician Fading Channel

of (t) is given by P (t) , E{| (t)|2 } = 2 , where 2 = 2 + 2 ; | | indicates

complex absolute value. We refer the reader to [Clar68, Jake74] for details on the
physics behind (2.3). There, the suitability of the model for describing real-world
channels is also discussed.

2.3

2.3.1

Statistical Characterization of the Reference

Channel Model
ACFs

The complex Gaussian process (t), which acts in this dissertation as a reference channel model, can be characterized by means of its ACF r (t1 , t2 ) ,
E{ (t1 ) (t2 )}. It is straightforward to verify that the ACF of (t) satisfies
r (t1 , t2 ) = r ( )

(2.4)
2

= r ( ) + exp {j2f }

(2.5)

where = t2 t1 , and r ( ) is the ACF of the channels diffuse component

(t); the notation rxx ( ) stands for the time-origin independent ACF rxx ( ) ,
x (t)x
x(t + )} of a stationary random process x (t), and the operator ()
E{x
indicates complex conjugation. One can easily show for the latter ACF that


) )
r ( ) = 2 E exp(j2fmax cos(

(2.6)

n }N
where is any arbitrary AOA in {
n=1 . Furthermore, since the cisoid exp(j2
) ) is an even function of , we can present (2.6) in the form
fmax cos(
r ( ) = 22

g () exp(j2fmax cos() )d

(2.7)

with g () , [p () + p ()]/2 denoting the even part of the PDF p () of .

Alternatively, we can write
r ( ) = 2

fZmax

fmax

pf (f ) exp{j2f }df

(2.8)

where pf (f ) is the PDF of the random Doppler frequencies f n defined in (2.2).

Such a PDF is found to be equal to
pf (f ) = 2 rect(f )

g (arccos(f /fmax ))
p
.
fmax 1 (f /fmax )2

(2.9)

10

Channel Simulation Models for Mobile Broadband Communications

The function rect(f ) in (2.9) stands for the rectangular function, which we define
as rect(f ) = 1 for |f | < fmax , and rect(f ) = 0 for |f | fmax . It is worth noticing
that the ACF of the diffuse component (t) meets



r I I ( ) + rQ Q ( ) + j rI Q ( ) r QI ( )


= 2 rI I ( ) + jr I Q ( )

r ( ) =

where

rI I ( ) = rQ Q ( ) = 2

(2.10)

g () cos (2fmax cos() ) d

(2.11a)

g () sin (2fmax cos() ) d

(2.11b)

rI Q ( ) = rQ I ( ) =

are the ACFs and cross-correlation functions (CCFs) of the IQ components of

(t), denoted by I (t) and Q (t), respectively. The CCF of any pair of random
x (t)yy (t + )}. It becomes
processes x (t) and y (t) is defined here as rxy ( ) , E{x
evident from (2.10) and (2.11) that if I (t) and Q (t) are uncorrelated, then the
ACF r ( ) of (t) will be an even and a real-valued function; otherwise, it will
be complex-valued and hermitian symmetric.
From the results presented in (2.5) and (2.8), we obtain

r ( ) =

fZmax

fmax



pf (f ) exp j2f df + 2 exp{j2f }.

(2.12)

We recall that a random process is wide-sense stationary (WSS) when its mean
value is constant and its ACF depends only on the time difference [Leon94,
p. 356]. Since the mean value of (t) is equal to m (t) = m (t) = exp{j(2f t+
)} [cf. Sec. 2.2], and r (t1 , t2 ) = r ( ), it follows that the reference model is
WSS process if any of the conditions stated below is fulfilled2 :
Condition 2.1 The received signal is solely composed of scattered waves, so that
(t) = (t), and therefore m (t) = 0.
Condition 2.2 The AOA of the specular component is equal to /2, i.e., =

90 . In such a case, we obtain m (t) = exp j , which is a time-independent
quantity.
Condition 2.1 leads in fact to the well-known Rayleigh fading channel model
[Pae02a, Chap. 3]a model that is widely used for assessing the performance of
2

The reference model will be strict-sense stationary (SSS) indeed, since (t) is a Gaussian
process [Leon94, p. 360].

Chapter 2 The Narrowband Rician Fading Channel

11

mobile communication systems [Hou04, Choi01]. We will pay special attention to

such a particular casewhich we will refer to as the NLOS casebecause of its
relevance. Condition 2.2, on the other hand, is related to very specific scenarios,
and we will not devote further attention to that case.

2.3.2

PSDs

The channel model described by (t) can alternatively be characterized through

its PSD S (f ). It is shown in [Mill04, pp. 373375] that the PSD and the
ACF of a random process form a Fourier transform pair provided that the ACF
depends only on the time difference . Since this is the case for the ACF r ( )
of (t) [see (2.12)], we find by substituting r ( ) from (2.12) into S (f ) =
R
r ( )exp{j2f }d that
S (f ) = 2 pf (f ) + 2 (f f )
g (arccos(f /fmax ))
p
= 22 rect(f )
+ 2 (f f )
fmax 1 (f /fmax )2

(2.13)

(2.14)

where () is the Dirac delta function. This result implies that the PSD of (t)
is confined to a bandwidth of 2fmax , as |f | fmax and pf (f ) = 0 for |f | fmax .
The channels PSD is often referred to as the DPSD due to the relationship
between S (f ) and the distribution of the signal power in the Doppler frequency
domain. We will adopt such a terminology from this point onwards.
For the special case m (t) = 0, S (f ) reduces to the DPSD
S (f ) = 22 rect(f )

g (arccos(f /fmax ))
p
fmax 1 (f /fmax )2

(2.15)

of (t). For this case, we can observe from classic Fourier theory (see [Brig88,
Sec. 3.6]) that if the IQ components of (t) are uncorrelated, meaning that r ( )
is even and real-valued, then S (f ) will be symmetrical with respect to the origin.
On the contrary, if I (t) and Q (t) are mutually correlated, then S (f ) will be
asymmetrical. In this respect, it is worth noticing that measurement campaigns
carried out in different propagation scenarios have shown that the DPSDs of
real-world NLOS channels are in general asymmetrical [Kott04, Zhao03, Blau06].

2.3.3

Average Doppler Shift and Doppler Spread

The first moment and the squared-root second central moment of S (f ) define
the channels average Doppler shift A and Doppler spread D , respectively.
These two quantities play an important role in the investigation of problems
related to the time-variant behavior of the channel, such as speed estimation
[Moha05] and frequency synchronization [Cai06, Niss06]. One can show by taking

12

Channel Simulation Models for Mobile Broadband Communications

account of the Fourier transform relationship between S (f ) and r ( ) that

A

f S (f )df

S (f )df

1 r (0)
j2 r (0)
v
uR

2
u
S (f ) f A df
t
R
,
S (f )df
s

1
r (0) 2 r (0)
=

2
r (0)
r (0)

(2.16)

(2.17)

where the single and double overdots denote the first and second derivatives
with respect to , respectively. Interestingly, (2.16) and (2.17) indicate that the
average Doppler shift and Doppler spread of (t) can unequivocally be determined
from the value, slope, and curvature of r ( ) at the origin ( = 0).
In a similar way, one can show that the average Doppler shift A and Doppler
spread D of the diffuse component (t) are equal to
A ,

f S (f )df

S (f )df

rI Q (0)
2
v
uR

2
u
S (f ) f A df
t
R
,
S (f )df
s 

r

(0)

I
I
=
A2 +
.
2 2 2
=

(2.18)

(2.19)

We took account of the relationships r (0) = 2j r I Q (0) and r (0) = 2

r I I (0)
in writting the previous expressions, where
r I Q (0) =

22 fmax

g () cos()d

(2.20a)

r I I (0) =

2 (2fmax )2

g () cos2 ()d.

(2.20b)

It becomes evident from (2.18) that if the IQ components of (t) are uncorrelated,
meaning that r I Q ( ) = 0, then A = 0. This was to be expected, since the
DPSD of (t) is symmetrical under such conditions.

Chapter 2 The Narrowband Rician Fading Channel

2.4

13

Statistical Characterization of the Signal Fading

The design of robust transceivers for mobile communication systems requires

knowledge on the statistics of the signal fading. In this regard, explicit information about the distributions of the channels envelope and phase, LCR, and
ADF is of utmost interest, as these statistical functions play an important role in
the design of carrier detection/tracking [Brow96], channel codification [Edba89],
and diversity techniques [Rapp02, Sec. 7.10], [Jake74, Sec. 6.3], just to give a few
examples. Information about the ACF of the channels squared envelope is also
relevant, since this function gives insights into the carrier-to-noise-ratio (CNR)
fluctuations produced by noisy fading channels [Pars00, p. 129].

2.4.1

PDFs of the Channels Envelope and Phase

2.4.1.1

First-Order PDF of the Envelope

Concerning the distribution of the reference models envelope (t) , | (t)|2 , one
can show by proceeding as in [Pae02a, Sec. 6.1.1] and [Auli79] that irrespective
of the channels correlation properties, the first-order PDF p (z; t) of (t) equals
2

p (z) =

2z z +
2
I
e
0
2

2z
2

z0

(2.21)

where I0 () is the modified Bessel function of the first kind and order zero. Equation (2.21) can be identified as the Rice distribution [Proa01, p. 46], and it indicates that the channels envelope (t) is a first-order stationary process, since
p (z; t) = p (z). The PDF given above can be rewritten in terms of the ratio
between the powers of the specular and diffuse components as follows
p (z) =





2z(cR + 1)
z 2 (cR + 1)
2z p
exp cR
I0
cR (cR + 1) (2.22)
2
2

for z 0, where 2 = 2 + 2 is the total power of the channel, and cR , 2 /2

is the so-called Rician factor, which is an indicator of the link quality (the higher
the Rician factor, the higher the quality of the wireless link).
For the NLOS case (cR = 0), the PDF in (2.22) reduces to the PDF p (z) of
(t)| of the diffuse component. Such a PDF is given by
the envelope (t) , |
2

p (z) =

2z z2
e
,
2

z 0.

(2.23)

Equation (2.23) can in turn be recognized as the Rayleigh PDF [Proa01, p. 44].
Figure 2.1 shows graphs of the channels envelope obtained by considering
cR {0, 2, 4}, 2 = 1, and fmax = 91 Hz (corresponding to a carrier frequency

14

Channel Simulation Models for Mobile Broadband Communications

of 1.8 GHz and a MT traveling at around 54 km/h). The PDF of (t) defined in
(2.22) is plotted also in that figure for the same values of cR and 2 .
(a)

(b)
1.4

cR = 0

10

cR = 4
0

10

10

fmax = 91 Hz
f = 65 Hz
2 = 1
= 0

10

0.01

0.02

cR = 4

1.2

Envelope PDF, p (z)

Envelope (t) (log scale)

cR = 2

cR = 2
1

cR = 0 (Rayleigh fading)
0.8

2 = 1

0.6

0.4

0.2

0.03

0.04

0.05

0.06

0.07

0
0

0.08

0.5

1.5

2.5

Envelope level, z

time, t (s)

Figure 2.1: (a) Channels envelope (t) (b) and its first-order PDF p (z) for
different Rician factors cR .

2.4.1.2

First-Order PDF of the Phase

In a similar way, one can show that the first-order PDF p (; t) of the channels
phase (t) , arg{ (t)} is given by [Pae02a, Sec. 6.1.1.1], [Auli79]
2

p (; t) =

2 sin2 (2f t )

cos ( 2f t ) e
2
4



erfc
cos( 2f t ) ,

(2.24)

where erfc(x) denotes the complementary error function, which is defined as

erfc(x) ,

Z
x



exp y 2 dy.

(2.25)

Such a PDF can alternatively be presented in terms of the Rician factor as follows
r
ecR
cR
2
p (; t) =
+
cos ( 2f t ) ecR sin (2f t )
2
4

erfc ( cR cos( 2f t )) ,
.
(2.26)
An inspection to (2.26) reveals that the correlation properties of (t) do not
influence the PDF of (t). Besides, (2.26) shows that the channels phase is not

Chapter 2 The Narrowband Rician Fading Channel

15

a first-order stationary process, because the PDF p (; t) of (t) is in general

time-dependent. Nevertheless, the PDF of (t) happens to be time-independent
if either Condition 2.1 or Condition 2.2 [see Sec. 2.3.1] is observed. In particular,
if Condition 2.1 is met, so that cR = 0, then
p (; t) = p ()
1
=
,
2

(2.27)

(2.28)

where p () is the first-order PDF of the diffuse components phase (t) ,

(t)}.
arg {
The phase of the channel and its first-order distribution p (; t) are depicted
in Fig. 2.2 for 2 = 1 and cR {0, 2, 4}. The graphs of p (; t) presented there
correspond to the time instants t = 0 s and t = 0.25 s.
(a)

(b)
1.2

cR = 2

cR = 4

Phase PDF, p (; t)

Phase (t) (radians)

0.01

0.02

0.03

0.04

0.05

time, t (s)

0.06

0.07

t = 0.25s

t = 0s

0.8

0.6

2 = 1

0.4

0.2

fmax = 91 Hz
2 = 1

4
0

cR = 0
cR = 2
cR = 4

cR = 0

0.08

Phase, (radians)

Figure 2.2: (a) Channels phase (t) (b) and its first-order PDF p (; t) for
different Rician factors cR (f = 65 Hz and = 0 ).

2.4.2

LCR and ADF

The first-order PDFs of the envelope and phase of (t) do not provide information
about the coherence or rapidity of the signal fades, as both PDFs are blind to the
channels correlation properties. Such an information, which is fundamental for
the design of bit and symbol interleaving schemes [Proa01, pp. 467470], should
instead be acquired from the LCR and ADF of the channels envelope. The LCR
gives a measure of the number of times the channels envelope crosses a given
level r with positive (or negative) slope within one second, whereas the ADF is
a measure of the average time the envelope remains below that level.

16
2.4.2.1

Channel Simulation Models for Mobile Broadband Communications

LCR

In accordance to Rice [Rice44], the LCR N+ (r) of the envelope (t) is to be

computed for a given signal level r 0 by evaluating
Z

N (r) =

p (r, )d

(2.29)

where p (, ) is the joint PDF of (t) and its derivative (t) = d (t)dt (the
notation () indicates time derivate). The analytical solution of (2.29) was derived
in [Pae02a, Sec. 6.1.1.2] for the envelope of a complex Gaussian process having
essentially the same statistical properties as (t). Based on the result there
presented, we can write

n
r(1 + cR ) 8b0
r 2 (cR + 1) o
exp

N (r) =
R
2
2 3/2

 
Z/2
n c [a sin(y)]2 o
2r p
R 0

cosh
cR (cR + 1) cos(y) exp

cR + 1
0
r

r
cR
cR
+
a0 sin(y) erf
a0 sin(y)
dy
(2.30)
cR + 1
cR + 1

for r 0, where
a0
b0


r I Q (0)
=
f
2
2
=
rI I (0) 2 r2 I Q (0)

2
b0

(2.31)
(2.32)

and erf(x) is the error function, which is defined as

erf(x) ,

Zx
0


exp y 2 dy.

(2.33)

Equation (2.30) shows that N (r) do not depend on the exact shape of the ACF
of (t), but only on its value, slope, and curvature at the origin. For the NLOS
case (cR = 0), N (r) converges to the LCR N of the envelope (t) of (t). By
taking cR = 0 in (2.30), it can be shown that N is equal to
N

b0
p (r),
2

r0

(2.34)

where p (r) is the Rayleigh PDF [see 2.23]. Interestingly, (2.34) shows that the
LCR of the diffuse component is proportional to the Rayleigh PDF.

Chapter 2 The Narrowband Rician Fading Channel

2.4.2.2

17

ADF

On the other hand, the ADF T (r) is to be computed by evaluating the quotient
T (r) =

P (r)
N (r)

(2.35)

Rx
where P (x) , 0 p (y)dy is the cumulative distribution function (CDF) of the
Rice process (t). Such a CDF is equal to [Proa01, Eq. (2.1142)]
P (x) = 1

xe

x2 +( 2r/ )2
2

2r/




I0 x 2
dx.

(2.36)

For the particular case cR = 0, we find that T (r) simplifies to the ADF T (r)
of the envelope (t) of the diffuse component (t). For this latter function, we
find the closed-form solution
s
!
r2
2
2
2

(2.37)
e 1 , r 0.
T (r) =
2r
rI I (0)

2.4.3

ACF of the Channels Squared Envelope

Regarding the ACF r 2 2 (t1 , t2 ) , E{ 2 (t1 ) 2 (t2 )} of the channels squared envelope 2 (t), it is shown in Appendix A that
r 2 2 (t1 , t2 ) = r 2 2 ( )

(2.38)


= r 2 2 ( ) +
rI I ( ) cos (2f )
cR + 1

cR 4 (cR + 2)
+rI Q ( ) sin (2f ) +
(cR + 1)2
42 cR

(2.39)

where
r 2 2 ( ) =


2
4
+ r ( )
2
(cR + 1)

(2.40)

(t)|2 of (t); the correlation funcis the ACF of the squared envelope 2 (t) , |
tions r I I ( ) and r I Q ( ) are defined in (2.11). It is worth mentioning that if
the IQ components of (t) are uncorrelated, so that rI Q ( ) = 0, then (2.39)
and (2.40) reduce to the expressions presented in [Pars00, p. 129] and [Auli79]
for the squared envelopes ACF of isotropic scattering channels.
We observe that the channels squared envelope 2 (t) is a WSS process, since
its ACF 2 (t) depends only on and its mean value m 2 (t) , E{ 2 (t)} can easily
be found to be time-invariant and equal to 2 , i.e., m 2 (t) = 2 .

18

2.5

Channel Simulation Models for Mobile Broadband Communications

Distributions of the AOA and Particular Cases

In order to test and evaluate the performance of the simulation models proposed in this thesis, it is necessary to specify a concrete distribution for the
random AOAs n . In this dissertation, we will consider two widely accepted
AOA distributions, namely, the von Mises distribution [VonM18], and the Laplacian distribution [Kotz01]. We will concisely describe the characteristics of these
distributions in the remaining of this chapter, and we will analyze also the resulting correlation and spectral properties of (t). We restrict our attention to
the diffuse component because the ACF and DPSD of (t) can easily be obtained
once the ACF and DPSD of (t) are knownsee (2.5), (2.14), and (2.15). Besides, we observe that the average Doppler shift A and Doppler spread D of
(t) can be written in terms of the corresponding averaged Doppler shift A and
Doppler spread D of (t) as follows:
A
D

= A + 2 f
q
=
2 f2 + (D )2 + (A )2 (A )2 .

(2.41)
(2.42)

To prevent confusions with the notation, we will make use of the superscripts
and ()LA to indicate whether the statistics of (t) are to be associated
with the von Mises or the Laplacian AOA distributions.

()VM

2.5.1

The von Mises PDF of the AOA and the Associated ACF
and DPSD of the Channels Diffuse Component

The von Mises PDF is a distribution for angular (circular) variates [Wats82] that
was originally introduced by Richard E. von Mises to study deviations of atomic
weights from integer values [VonM18]. This distribution has widely been in use
in the field of directional statistics, and its applications span from geophysics
to vital statistics [Gumb54, Upto73]. It is also known as the circular normal
distributionas it is analogous in many respects to the normal distribution for
linear variates [Gumb53], and it includes other important distributions as special
cases, such as the uniform, cardioid, and wrapped Gaussian [Mard99]. The use
of the von Mises PDF to model the AOA statistics of mobile fading channels
was proposed in [Abd02a]. In that paper, the authors provide evidence of the
suitability of such a PDF to match measured data.
The von Mises PDF and its even part are given by
pVM () ,
gVM () =

exp{ cos( m )}
2I0 ()
exp{ cos() cos(m )}
cosh ( sin() sin(m ))
2I0 ()

(2.43)
(2.44)

Chapter 2 The Narrowband Rician Fading Channel

19

respectively, where [, ), m [, ) denotes the mean AOA, and 0

is a concetration parameter determining the channels angular spread. Following
(2.15), one can readily verify that


p
2
SVM
(f ) = rect (f ) cosh sin(m ) 1 (f /fmax )

2 exp { cos(m )f /fmax }

p
fmax I0 () 1 (f /fmax )2

(2.45)

whereas, by using (2.7), one may demonstrate that

rVM
( ) =

2 I0

p

2 (2fmax )2 + j4fmax cos(m )

I0 ()

(2.46)

It is worth noticing that for = 0, the DPSD shown in (2.45) reduces to the
well-known U-shaped DPSD
S (f ) = rect (f )

fmax

2
1 (f /fmax )2

(2.47)

characterizing isotropic scattering channels [Clar68], whilst the ACF in (2.46)

simplifies to
r ( ) = 2 J0 (2fmax )

(2.48)

where J0 () is the Bessel function of the first kind and order zero.
On the other hand, we find by following (2.18) and (2.19) that the diffuse
VM
components average Doppler shift AVM
and Doppler spread D are given by
I1 ()
AVM
= fmax cos(m )

I0 ()
 2

fmax [I0 () + I2 ()] cos2 (m )
DVM
=

I0 ()
2

1/2
2
+I1 () sin2 (m ) (AVM
)

(2.49)

(2.50)

where I1 () and I2 () denote the first and second order modified Bessel functions,

VM
respectively. We observe that if = 0, then AVM
= 0 and D = fmax / 2.
Figures 2.32.5 show how the von Mises PDF and the resulting ACF and
DPSD of (t) look like for different combinations of the parameters m and
. The graphs presented in those figures were computed by choosing 2 = 1
and fmax = 91 Hz. They are representative of scattering propagation conditions
ranging from isotropic (m = 0 and = 0) to extremely non-isotropic (m = 0
and = 20). We can see in Fig. 2.5 and deduce from (2.45) that the diffuse

20

Channel Simulation Models for Mobile Broadband Communications

1.8

= 10
= 10
= 10
m = 30 m = 45 m = 90

= 20
m = 0

1.6

PDF, p ()

1.4

= 10
m = 0

1.2
1
0.8

=0
m = 0

0.6

=5
m = 0

0.4
0.2
0

180

120

60

60

120

180

AOA, (Degrees)

Figure 2.3: The von Mises PDF of the AOA.

Absolute value of the ACF, |r ()|

1.2

= 10
m = 90

= 10
m = 45

= 10
m = 30

= 20
m = 0

= 10
m = 0

=5
m = 0

=0
m = 0

fmax = 91 Hz

0.8

0.6

0.4

0.2

0
0

Normalized time difference, fmax

Figure 2.4: ACF of the channels diffuse component, r ( ), by considering the

von Mises PDF of the AOA with parameters m and .
components DPSD is symmetrical only when = 0 and/or m = 90 , which
happen to be the only values of m and for which the ACF of (t) is real-valued.
This is in line with the statements made in Section 2.3.1 about the properties of
the DPSD of (t). The corresponding average Doppler shift and Doppler spread
of (t) are presented in Table 2.1.

2.5.2

The Laplacian PDF of the AOA and the Associated ACF

and DPSD of the Diffuse Component

Another relevant model for the AOA statistics of mobile fading channels is given
by the (symmetrical) Laplacian distribution [Pede97, Spen00, Aspl06]. Such a
PDF is used in the field of linear statistics to model data with long tails, and
its applications range from astronomy to biological sciences [Kotz01]. The use

Chapter 2 The Narrowband Rician Fading Channel

21

=0
m = 0

10

= 10
= 10
m = 90 m = 45

=5
m = 0

= 10
m = 0

= 10
m = 30

= 20
m = 0

DPSD, S (f ) (dB)

15
20
25
30
35
40

fmax = 91 Hz
45
50
90

60

30

30

60

90

Doppler frequency, f (Hz)

Figure 2.5: DPSD of the channels diffuse component, S (f ), by considering

the von Mises PDF of the AOA with parameters m and .
VM
Table 2.1: Average Doppler shift AVM
and Doppler spread D of the channels
diffuse component by considering the von Mises PDF of the AOA
(fmax = 91 Hz).
(0 , 0)

(0 , 5)

(0 , 20)

(0 , 10)

(30 , 10)

(Hz):

81.297

88.695

86.322

74.757

(Hz):

64.346

13.857

3.2606

6.6239

15.142

28.027

Parameters (m , ):
VM
A

DVM

(90 , 10)

of this distribution to characterize the AOA statistics of fading channels was

first proposed in [Pede97]. There, the authors demonstrated the goodness of fit
of such a distribution by direct comparison against measured data collected in
outdoor environments. The validity of the Laplacian model has also been tested
and corroborated by other researchers [Spen00].
The Laplacian PDF of the AOA has the form [Pede97, Spen00]
pLA ()

1
exp
cs


|| 2

,
s

[, )

where the parameter s > 0 controls the angular spread, and

 o
n
cs = s 2 1 exp 2
s

(2.51)

(2.52)

R
is a normalization quantity that guarantees pLA ()d = 1. This PDF is
depicted in Fig. 2.6 by considering s {0.3, 0.5, 1, 5}. These values are representative of scattering propagation conditions that vary from moderated (s = 5)
to severe non-isotropic scattering (s = 0.3).
The PDF defined in (2.51) is itself an even function (this is clear, since
LA
p () = pLA ()). Thus, one can show, by substituting pLA () from (2.51)

22

Channel Simulation Models for Mobile Broadband Communications

2.5

s = 0.3
s = 0.5

PDF, p ()

s = 1
s = 5
1.5

0.5

180

120

60

60

120

180

AOA, (Degrees)

Figure 2.6: The Laplacian PDF of the AOA.

into (2.15), that

SLA
(f )


22 exp 2 arccos (f /fmax ) /s
p
= rect(f )
.
cs fmax 1 (f /fmax )2

(2.53)

In contrast to the case of the von Mises PDF, the ACF of (t) cannot be evaluated
analytically when the AOAs of the channels multipath components follow the
Laplacian distribution. Instead, one has to use numerical methods to solve
rLA
( )

fZmax

SLA
(f ) exp{j2f }df.

(2.54)

fmax

Except by the additional work, the numerical evaluation of (2.54) does not pose
any problems, since modern computers and programming languagessuch as
R
MATLAB
can accomplish the task easily.
LA
Regarding the average Doppler shift ALA
and Doppler spread D of (t),
we find by following (2.18) and (2.19) that

ALA

DLA




fmax 8s
2/s
=
1+e
(2.55)
cs (2 + s2 )
(
"
#
)1/2

fmax
s 2
2/s
LA 2
=
1+
(1 e
) (A )
. (2.56)
2
cs (1 + 2s2 )

LA
Figures 2.7 and 2.8 show graphs of |rLA
( )| and S (f ) obtained for the
same values of s as considered for the curves of pLA () depicted in Fig. 2.6.
The results obtained for the corresponding average Doppler shift and Doppler
spread are presented in Table 2.2.

Chapter 2 The Narrowband Rician Fading Channel

23

Absolute value of the ACF, |r ()|

1.2

s
s
s
s

fmax = 91 Hz
1

0.8

= 0.3
= 0.5
=1
=5

0.6

0.4

0.2

0
0

Normalized time difference, fmax

Figure 2.7: ACF of the channels diffuse component, r ( ), by considering the

Laplacian PDF of the AOA with parameter s .
10

DPSD, S (f ) (dB)

15

fmax = 91 Hz

20
25
30
35

s = 0.3
s = 0.5

40

s = 1

45

s = 5
50
90

60

30

30

60

90

Doppler frequency, f (Hz)

Figure 2.8: DPSD of the channels diffuse component, S (f ), by considering

the Laplacian PDF of the AOA with parameter s .

2.6

Chapter Summary and Conclusions

In this chapter, we reviewed the characteristics and statistical properties of the

narrowband Rician fading channel model that we will use as reference model to
design the parameter computation methods proposed in this dissertation.
It was shown in Section 2.3 that the ACF of the complex Gaussian process
(t) characterizing our reference model does not depend on the choice of the
time origin. In spite of this noteworthy characteristic, (t) is in general a nonstationary process, since its mean value is a function of time. However, (t)
proves to be a SSS process under two conditions: (i) Either the LOS component
is absent; (ii) or it is not Doppler shifted. The statistics of (t) have been revised
thoroughly under NLOS conditions.

24

Channel Simulation Models for Mobile Broadband Communications

LA
Table 2.2: Average Doppler shift ALA
and Doppler spread D of the channels
diffuse component by considering the Laplacian PDF of the AOA
(fmax = 91 Hz).
Parameter, s :

0.3

0.5

LA (Hz):
A

87.0814

80.9113

62.1108

16.1574

LA (Hz):
D

8.138

18.8202

40.7789

62.9335

The relevant statistics of the signal fading, such as the distributions of the
channels envelope and phase, ADF, and LCR were studied in Section 2.4. In
that section, it was shown that the envelope (t) of (t) is first-order stationary
and Rician distributed, while its phase (t) has a PDF that depends on time. In
addition, we investigated the ACF of the squared envelope 2 (t) of (t). In accordance to the results presented in Section 2.4.3, the channels squared envelope is
a WSS process. Finally, we described in Section 2.5 the distributions of the AOA
that we will consider throughout this dissertation to assess the performance of
the proposed channel simulation models.

Chapter 3

Narrowband SOC-Based
Simulation Models
3.1

Introduction

We can observe from (2.3) that a hardware/software realization of the mobile

fading channel described by (t) is not possible, since it requires the implementation of a sum of an infinite number of cisoids. Fortunately, most of the statistical
properties of (t) relevant for system performance analysissuch as its correlation properties, spectral characteristics, and the first-order distributions of its
envelope and phasecan satisfactorily be approximated via a simulation model
based on a finite SOC. In this chapter, we present a narrowband stochastic SOCbased simulation model suitable for performing the above mentioned task. Such
a stochastic SOC model constitutes in fact the core structure of the channel simulators proposed in this dissertation. In addition to analyzing the simulation
models correlation and spectral properties, the present chapter reviews the firstorder PDFs of the envelope and phase processes. What is more, the correlation
properties of the resulting squared envelope process are investigated here for the
first time. The ergodicity of the simulation model (in terms of the mean value,
power, and ACF) is also a topic addressed for the first time in this chapter.
We observe that some of the statistical properties of the stochastic SOC model
herein described have previously been analyzed in [Paet07] and [Hogs08]. Specifically, the simulation models ACF, average Doppler shift, and Doppler spread are
studied in [Paet07] for the NLOS case (simulation of Rayleigh fading channels).
The envelope and phase distributions of the simulator are investigated in the
same paper by considering a LOS scenario with a time-invariant specular component (simulation of Rician fading channels), whereas the first-order stationarity
of the envelope process is studied in [Hogs08] upon the NLOS assumption. The
analysis presented throughout this chapter is broader in scope than the one made
25

26

Channel Simulation Models for Mobile Broadband Communications

in the above quoted papers in the sense that we consider a LOS scenario with a
time-variant specular component.

3.2

SOC-Based Channel Simulators: General Structure and Classes

Figure 3.1 shows the general structure of an SOC-based simulation model for the
mobile Rician fading channel model described by the complex Gaussian process
(t). Such a structure consists of two parts. One part is intended for the simulation of the channels diffuse component (t), whereas the other part is aimed
at reproducing the specular component m (t). The simulation of (t) is carried
out upon an SOC model comprising a finite number N Z+ (N < ) of homogeneous cisoids1 , the parameters of whichgains, frequencies, and phasesare
defined either as random variables or deterministic quantities. The symbol Z+
stands for the set of positive integers. Such an SOC model can mathematically
(t) if any of the cisoids parameters
be described by a complex random process
is random, otherwise, it is to be represented by a complex deterministic (pseudorandom) process
(t). The channels specular part m (t) is reproduced by means
of a properly parameterized deterministic cisoid. It goes without saying that the
superposition of
(t) and m (t) (or
(t) and m (t)) is expected to result into
a complex quasi-Gaussian random process (t) (or deterministic process (t))
whose statistical properties resemble those of (t).
Based on the nature (random or deterministic) of the N homogeneous cisoids
underlying the process
(t), we can identify eight fundamental classes of SOCbased simulation models for (t): Seven classes of stochastic models, and one class
of deterministic modelswith the class of deterministic models being a superset
of the other seven classes. The eight classes, which were originally defined in
[Hogs08], are listed in Table 3.1. For the simulation of (t), we will consider
a stochastic simulation model of Class II, meaning that the homogeneous SOC
model
(t) comprises cisoids with random phases, constant gains, and constant
Doppler frequencies. Our motivation to choose an SOC model of such a class is
driven by the fact that only the Class II simulators are autocorrelation ergodic2
(given the restriction N < ). This important property allows for the design
of simulation models able to efficiently approximate the reference models ACF
without the need of averaging over multiple simulation runs. Besides, it will be
shown throughout this chapter that the resulting random process (t) possesses
stationarity characteristics similar as those of the reference model.
1

By homogeneous cisoids we mean a group of cisoids characterized by the same type of

parameters.
2
We point out that the concept of ergodicity does not apply on the Class I simulators, since
the nature of such a class of models is deterministic.

Chapter 3 The Narrowband SOC-Based Simulation Model

27

Block of N homogeneous random/deterministic oscillators

First complexsinusoidal oscillator

of random (deterministic) frequency

Amplifier
of random
(deterministic)
gain

..
.

N th complexsinusoidal oscillator
of random (deterministic) frequency

Random
(deterministic)
phase shift

..
.

..
.

Amplifier
of random
(deterministic)
gain

Random
(deterministic)
phase shift

..
.

Simulated
waveform

(t) (or
(t))

Simulated
channel
waveform
(t)
(or (t))

Deterministic oscillator

Complex-sinusoidal
oscillator of
frequency f

Amplifier
of deterministic
gain

Simulated
specular wave
m (t)

Deterministic
phase shift of
radians

Figure 3.1: Block diagram of an SOC-based simulation model for Rician fading
channels.
Table 3.1: Classification of SOC-based simulation models for Rician fading
channels according to the nature of the cisoids parameters.
Parameters

Gains

Frequencies

Phases

Class I

Deterministic

Deterministic

Deterministic

Class II

Deterministic

Deterministic

Random

Class III

Deterministic

Random

Deterministic

Class IV

Deterministic

Random

Random

Class V

Random

Deterministic

Deterministic

Class VI

Random

Deterministic

Random

Class VII

Random

Random

Deterministic

Class VIII

Random

Random

Random

This attribute refers to the parameters of the SOC model described by

(t) (or
(t)).

3.3
3.3.1

The Stochastic SOC-Based Simulation Model

Structure and Considerations

The stochastic narrowband SOC-based simulation model of Class II considered

in this dissertation is characterized by a random process of the form
(t) =

N
X




cn exp j 2 fn t + n + exp{j(2f t + )} .
|
{z
}
m (t)
|n=1
{z
}

(t)

(3.1)

28

Channel Simulation Models for Mobile Broadband Communications

For the underlying homogeneous random SOC model

(t) =

N
X

n=1




cn exp j 2 fn t + n

(3.2)

we assume the phases n as being mutually independent random variables, each

having a uniform distribution over [, ). In addition, we impose the condition
PN 2
n = 2 on the set of gains {
cn }N
n=1 and define the deterministic Doppler
n=1 c

frequencies fn as
fn , fmax cos(
n ),

n = 1, . . . , N

(3.3)

where
n [, ).

3.3.2

Mean Value, Variance, and Average Power

Under the aforementioned considerations,

(t) proves to be a zero-mean stochas (t)} = 2 . In turn, the composite protic process with constant variance Var{

cess (t) =
(t) + m (t) is found to have a time-variant mean value m (t) ,
E{
(t)} = m (t) and a constant variance Var{
(t)} = 2 . Furthermore, the av2
erage power P (t) , E{|
(t)| } of (t) turns out to be time-invariant and equal
2
2
2
to = + , i.e., P (t) = P = 2 . A comparison of these results with the
ones presented in Section 2.2 for (t) reveals that the simulation model defined
in (3.1) is characterized by exactly the same mean value, variance, and average
power as the reference model.

3.3.3
3.3.3.1

Correlation Properties and Spectral Characteristics

ACFs

(t1 )
(t2 )} of the zero-mean random process

Regarding the ACF r (t1 , t2 ) , E{

(t), one can show without difficulty that [Paet07]

r (t1 , t2 ) = r ( )
=

N
X

n=1



c2n exp j2 fn .

(3.4)
(3.5)

This result shows that the ACF of

(t) depends only on the time difference
= t2 t1 . It can therefore be concluded that
(t) is a WSS process, since
its mean value is constant and its ACF is not influenced by the choice of the
time origin. Other relevant characteristics of r ( ) can readily be noticed by
expressing such a function in terms of the ACFs and CCFs of the IQ components

Chapter 3 The Narrowband SOC-Based Simulation Model

29

(t)} and
(t)} of

I (t) , Re{

Q (t) , Im{

(t) as follows



r I I ( ) + r Q Q ( ) + j r I Q ( ) r Q I ( )


= 2 r I I ( ) + jr I Q ( )

r ( ) =

where

r I I ( ) = r Q Q ( ) =
r I Q ( ) = r Q I ( ) =

N
X
c2

n=1
N
X
c2n
n=1

(3.6)

(3.7a)

sin 2 fn .

(3.7b)

cos 2 fn

One may observe from (3.6) and (3.7) that r ( ) is a complex valued and hermi Q (t) are cross-correlated. On the contrary,
tian symmetric function if
I (t) and
if the IQ components of
(t) are uncorrelated, then r ( ) will be even and real
valued. We recall that the uncorrelation between the processes
I (t) and
Q (t)
is a requisite for the simulation of fading channels characterized by symmetrical DPSDs [Paet98]. Taking account of the expression presented in (3.7b) for
r I Q ( ) and r Q I ( ), we can deduce that a sufficient condition for guaranteeing
Q (t) is the following:
the uncorrelation between
I (t) and
Condition 3.1 The number of cisoids N is even, i.e., N = 2M , where M Z+ ,
and for each pair of parameters (
cn , fn ), n = 1, . . . , N , there exists one and only
one pair (
cm , fm ), n 6= m, such that cn = cm and fn = fm .
Notice that if the above mentioned requirements are met, then
r ( ) = 2

N/2
X

c2n cos(2 fn ).

(3.8)

n=1

It is worth mentioning that the Condition 3.1 becomes a necessary condition for
(t) if the Doppler frequencies are
the uncorrelation of the IQ components of
chosen in such a way that fn 6= 0 n and fn 6= fm , n 6= m. These two latter
considerations are relevant indeed, as they are fundamental for the design of
ergodic channel simulators [see Sec. 3.5].
With respect to the ACF r (t1 , t2 ) , E{
(t1 )
(t2 )} of the simulation model,
it is straightforward to show that
r (t1 , t2 ) = r ( )

(3.9)
2

= r ( ) + exp{j2f }
=

N
X

n=1



c2n exp j2 fn + 2 exp{j2f }.

(3.10)
(3.11)

30

Channel Simulation Models for Mobile Broadband Communications

The equations above indicate that the ACF of (t) possesses the noteworthy
property of being time-shift insensitive. Nonetheless, in contrast to the zeromean process
(t), (t) cannot be claimed to be a WSS process, because its
mean value is in general time-variant. However, this characteristic should not be
considered as a drawback, since the simulation model and the reference model
are similar in that respect. In fact, the wide-sense stationarity property of both
processes (t) and (t) holds under the same conditions: The specular wave m (t)
is absent or not Doppler shifted [cf. Sec. 2.3.1].
3.3.3.2

DPSDs

Concerning the DPSD S (f ) of the simulation model, we obtain on the basis of

[Mill04, pp. 373375] the solution
S (f ) =

N
X

n=1

r ( ) exp{j2f }d


c2n f fn + 2 (f f ).

(3.12)

The DPSD S (f ) of
(t) follows from S (f ) by taking = 0. Consequently,
we can write
S (f ) =

N
X

n=1

c2n f fn .

(3.13)

This latter result can be confirmed by computing the Fourier transform of the
ACF defined in (3.5) [Paet07]. Interestingly, if Condition 3.1 is fulfilled, meaning
that r I Q ( ) = r Q I ( ) = 0, then
S (f ) =

N/2
X

n=1






c2n f fn + f + fn .

(3.14)

Thus, the DPSD of

(t) happens to be an even (symmetrical) function upon
fulfillment of Condition 3.1, as was to be expected.
3.3.3.3

Average Doppler Shift and Doppler Spread

P
From (3.12), and taking into account that N
2n = 2 , one can verify that the
n=1 c
average Doppler shift A and Doppler spread D of (t) are given by:
R
f S (f )df
A , R

S (f )df

Chapter 3 The Narrowband SOC-Based Simulation Model

"
#
N
X
1 2
2
=
f +
cn fn
2
n=1
v
uR

2
u
S (f ) f A df
t
R
,
S (f )df
v "
#
u
N
u1
X
= t 2 2 f2 +
c2n fn2 (A )2 .

31

(3.15)

(3.16)

n=1

What is more, one can deduce by taking = 0 in (3.15) and (3.16) that the
average Doppler shift A and Doppler spread D of
(t) are equal to:


A = A 
=

1
2

=0

N
X

c2n fn

(3.17)

n=1



D = D 
=0
v
u
N
u 1 X

2
t
2 f2 A
=
c

n
n
2

(3.18)

n=1

which are the results presented in [Paet07]. We observe from (3.17) that the
(t) is equal to zero if the Condition 3.1 is met.
average Doppler shift A of
This result is in line with the statement made in Section 2.3.3 regarding the
average Doppler shift of symmetrical DPSDs.
It is worth noticing that the simulation models ACF, DPSD, average Doppler
shift, and Doppler spread are fully specified by the sets of gains {
cn , } and

Doppler frequencies {fn , f }.

3.3.4

PDFs of the Stochastic SOC Models Envelope and Phase

The first-order distributions of the envelope (t)

, |
(t)| and phase
(t)
,
arg{
(t)} of (t) have recently been investigated in [Paet07]. There, analytical
expressions were derived for such statistical functions by assuming the specular
wave m (t) to be not Doppler shifted. In this section, we provide generalized expressions of those formulas covering the more realistic case where m (t) is Doppler
shifted. The generalization of the results presented in [Paet07] is straightforward,
and can systematically be done by following the procedure outlined in [Paet07,
Sec. IV-E]. We will therefore omit details about the derivations and present
only the obtained generalized expressions. The correctness of our results will be
demonstrated by simulations in Chapter 4.

32
3.3.4.1

Channel Simulation Models for Mobile Broadband Communications

First-Order PDF of the Envelope

By proceeding as in [Paet07, Sec. IV-E], one can verify that the first-order PDF
of (t) is equal to
p(z; t) of the envelope (t)
p(z; t) = p(z)
= z(2)2

(3.19)
Z " Y
N

n=1

J0 (2|
cn |x) J0 (2zx)J0 (2||x)x dx,

z 0. (3.20)

is a first-order stationIt can be concluded from the previous equations that (t)
ary process, because its PDF does not change over time. Interestingly, the expression presented in (3.20) proves to be the same as the one obtained in [Paet07,
Eq. (26)]. This means that the time-variant behavior of the specular wave has

We notice that p(z) is fully

no influence on the first-order distribution of (t).
specified by the set of gains {
cn , }. For the special case = 0, the PDF given in
, |
(t)| of
(3.20) reduces to the first-order distribution p(z) of the envelope (t)

as follows

(t). Thus, we can explicitly write the PDF of (t)

p(z) = z(2)2

Z " Y
N

n=1

J0 (2|
cn |x) J0 (2zx)x dx,

z 0.

(3.21)

Investigations in [Paet07] indicate that p(z) is in good agreement with the

Rice distribution for values of N as small as ten if cn = / N n. In fact, it

is shown in the above quoted paper that if cn = / N , then p(z) converges

to the Rice density in the limit N . Figure 3.2 provides an exemplary
demonstration of the veracity of these statements. In this figure, we present a
comparison between the PDF defined in (3.20) and the Rice PDF introduced in

(2.22) by considering cn = / N n, N {10, 50}, cR {0, 2, 4}, and 2 = 1.

We remind that cR = 2 /2 is the Rician factor.

3.3.4.2

First-Order PDF of the Phase

With respect to the first-order PDF p (; t) of the phase

(t)
of (t), it can be
demonstrated that
#
Z Z " Y
N
p (; t) = 2
J0 (2|
cn |x)
0

J0 2x

n=1

z2


2z cos( 2f t ) xz dx dz

(3.22)

Chapter 3 The Narrowband SOC-Based Simulation Model

(a)

(b)

1.4

1.4

Reference model
Simulation model

1.2

cR = 4
cR = 2

0.8

cR = 0

0.6

0.4

0.5

1.5

2.5

Envelope level

cR = 4
cR = 2

0.8

cR = 0

0.6

0.4

N = 10
2 = 1

0.2

Reference model
Simulation model

1.2

Envelope PDF

Envelope PDF

0
0

33

N = 50
2 = 1

0.2

0
0

0.5

1.5

2.5

Envelope level

Figure 3.2: Comparison between the theoretical envelope PDF p(z) of the
stochastic SOC-based simulation model and the Rice PDF p (z)
(reference model) by considering different values for both the Rician
factor cR and the number of cisoids N : (a) N = 10; (b) N = 50.

for [, ). The equation above shows that

(t)
is in general not firstorder stationary, because its distribution p (; t) is not necessarily time-invariant.
However, if m (t) is not Doppler shifted, implying that f = 0, then
p (; t) = p ()
#
Z Z " Y
N
= 2
J0 (2|
cn |x)
0

(3.23)

n=1



q
2
2
J0 2x z + 2z cos( ) xz dx dz,

[, ) (3.24)

which is the result presented in [Paet07, Eq. (28)]. On other hand, if the specular
wave is absent ( = 0), then p (; t) reduces to the first-order distribution p (; t)
(t)} of
of the phase
(t)
, arg{

(t). It can be shown for the latter PDF that

p (; t) = p ()
#
Z Z " Y
N
= 2
J0 (2|
cn |x) J0 (2xz)xz dxdz
0

1
,
2

(3.25)
(3.26)

n=1

[, )

(3.27)

which implies that the first-order distribution of

(t)

is time-invariant and uniform over [, ). Interestingly, the PDF of

(t)

is not influenced by any of the

34

Channel Simulation Models for Mobile Broadband Communications

parameters of
(t) (not even by N ). In view of (3.24) and (3.27), we can claim

that the phase

(t)
of the simulation model is a first-order stationary process if
either f = 0 or = 0.
Figure 3.3 shows a comparison between the PDF given in (3.22) and the
phase PDF of the reference model [see (2.26)]. The graphs presented in that
figure correspond to the time instants t = 0 s and t = 0.25 s, and they were

generated by choosing cn = 1/ N n, N {10, 50}, cR {0, 2, 4}, f = 65 Hz,

and = 0 . One may observe from Fig. 3.3 that the first-order distribution of
the simulation models phase, p (; t), is in excellent agreement with the phase
PDF p (; t) of reference model. Indeed, the graphs of p (; t) are quite close to
the ones of p (; t) even with as few as ten cisoids, as can be seen in Fig. 3.3(a).
(a)

(b)

1.5

1.5

Reference model
Simulation model

Reference model
Simulation model

cR = 4
cR = 2
t = 0.25s t = 0.25s

N = 10
2 = 1
cR = 4
t = 0s
cR = 2
t = 0s

0.5

N = 50
2 = 1

Phase PDF

Phase PDF

cR = 4
t = 0s
cR = 2
t = 0s

0.5

cR = 0

cR = 4
cR = 2
t = 0.25s t = 0.25s

cR = 0

Phase (radians)

Phase (radians)

Figure 3.3: Comparison between the theoretical phase PDF p (; t) of the

stochastic SOC-based simulation model and the phase PDF p (; t)
of reference model by considering different values for both the Rician factor cR and the number of cisoids N : (a) N = 10; (b) N = 50
(f = 65 Hz and = 0 ).
Before we move on to the next topic, we would like to draw the readers
and

attention to the fact that (t)

(t)
exhibit the same first-order stationarity
characteristics as the envelope (t) and phase (t) of the reference model [cf.
Sec. 2.4.1].

3.3.5

A Note on the LCR and ADF of the Stochastic SOC Models Envelope

As of the time of witting of this thesis, exact solutions of the LCR N(r) and
are still lacking. A first attempt to close the gap was made in
ADF T(r) of (t)
[Pae08b]. Closed-form expressions were presented in that paper for both N(r)

Chapter 3 The Narrowband SOC-Based Simulation Model

35

P
and T(r) by considering m (t) = exp{j } and assuming that N
2n fn = 0,
n=1 c
i.e., A = 0. Early investigations showed a reasonably good resemblance between
the analytical solutions provided there and the empirical data obtained from
simulations. However, further analysis demonstrated that the expressions given
in [Pae08b] underestimate the number of crossings at medium levels (r [0.5, 2]).
Such an imprecision could be due to the fact that the cross-correlations at = 0
among the IQ components of
(t) and their time derivatives were neglected when
deriving the formulasit was assumed in [Pae08b] that r I I (0) = r Q Q (0) =
r I Q (0) = r Q I (0) = 0. Those expressions are currently being revisited by the
authors of [Paet07] by relaxing the aforementioned assumption.

3.3.6

ACF of the Stochastic SOC Models Squared Envelope

Concerning the ACF r22 (t1 , t2 ) , E{2 (t1 )2 (t2 )} of the squared envelope
2 (t) , |
(t)|2 of (t), it is shown in Appendix B that
r22 (t1 , t2 ) = r22 ( )

42 cR
= r22 ( ) +
r I I ( ) cos (2f )
cR + 1

cR 4 (cR + 2)
+r I Q ( ) sin (2f ) +
(cR + 1)2

(3.28)

(3.29)

(t)|2 of
where r22 ( ) is the ACF of the squared envelope 2 (t) , |

(t). It is
also shown in Appendix B that
r22 ( ) =

N
X
4
2
+ |r ( )|
c4n .
(cR + 1)2
n=1

(3.30)

One can verify without difficulty that the mean value m2 (t) , E{2 (t)}
of 2 (t) is equal to 2 . Thus, we can conclude that 2 (t) is a WSS process,
since the ACF r22 (t1 , t2 ) depends only on the time difference = t2 t1 , i.e.,
r22 (t1 , t2 ) = r22 ( ), and m2 (t) is constant over time.

3.4
3.4.1

The Deterministic SOC-Based Simulation Model

Structure

In practice, the simulation of (t) is performed by generating sample functions of

(t) and combining the resulting process with the time-variant waveform m (t)
(as depicted in Fig. 3.1). Consequently, the simulators output proves to be a

36

Channel Simulation Models for Mobile Broadband Communications

process of the form

(k) (t) =
(k) (t) + exp{j(2f t + )},

k Z+

(3.31)

where (k) (t) and

(k) (t) denote the kth sample functions of (t) and
(t), re(k)
spectively. The complex process
(t) is given by

(k)

(t) =

N
X

n=1


cn exp j 2 fn t + n(k) ,

k Z+

(3.32)

(k)
(t).
where n is the outcome of n associated to the kth sample function of
(k)
We note that (t) can unequivocally be predicted at any time instant t once
(k)
the values of n are known (those values are to be resolved at random during
the simulation set-up phase and they are kept constant during the simulation
run). The simulators output can therefore be regarded as being a deterministic
(k) (t) can be classified as deterministic SOC-based
process. In fact, (k) (t) and
simulators (Class I simulators).
Information about the time averages3 of the sample functions of (t) is also
relevant for the characterization of the simulation model. Such an information
is fundamental not only to analyze the properties of the generated waveforms
(k) (t), but also to determine whether the simulation model is ergodic or not. In
fact, to investigate the subject of ergodicity, it is necessary to find out under what
conditions the time averages of (t) are equal to the ensemble averages. In that
regard, we present in what follows an analysis of the time-averaged characteristics
of the simulation models sample functions.

3.4.2

Mean Value and Average Power

In the case of the random process

(t), one can easily verify that the mean value

(k) 
(k)
(t)} = 0,
m(k) ,
(t) of
(t) is equal to the ensemble mean, m (t) , E{

if the following condition is fulfilled:

Condition 3.2 The N cisoids underlying
(t) are all Doppler shifted, meaning

that fn 6= 0 n.

(k) 2 
Besides, one may demonstrate that the average power P(k) , |
(t)| of
(k)
2
2
(t)| } = , if the above mentioned

(t) equals that of the ensemble, P = E{|

condition is satisfied and:

Condition 3.3 The N cisoids underlying
(t) are specified in such a way that

fn 6= fm for n 6= m.
3

lim

The time average of a function x(t) is denoted by hx(t)i and defined as hx(t)i ,
RT
1
x(t)dt.
2T T

Chapter 3 The Narrowband SOC-Based Simulation Model

37


On the other hand, we find that the mean value m(k) , (k) (t) of the
sample functions of (t) equals m(k) = 0 k if Condition 3.2 holds and f 6= 0.
If the latter requirement is not satisfied, i.e., if f = 0, then m(k) = exp{j }
k. Since the mean value of the ensemble is in general a function of time m (t) ,
E{
(t)} = exp{j(2f t + )}, we have that m(k) is equal to m (t) only when

(k) 2 
(t)| of (k) (t),
= 0 or f = 0. Regarding the average power P(k) , |
we have P(k) = 2 k if in addition to Conditions 3.2 and 3.3, the following is
observed:
Condition 3.4 The specular wave m (t) and the random cisoids underlying
(t)

are specified in such a way that f 6= fn n.

Notice that the average power of (k) (t) equals the mean power of the ensemble
upon fulfillment of such conditions, since P = E{|
(t)|2 } = 2 [Sec. 3.3.2].

3.4.3

Correlation Properties and Spectral Characteristics

3.4.3.1

Time-averaged ACFs

In order to analyze the correlation properties of the simulation models sample

functions, let us assume that the zero-mean process
(t) is defined in such a way
that Conditions 3.2 and 3.3 are fulfilled, meaning that fn 6= 0 n and fn 6= fm
for n 6= m. Under these considerations, the time-averaged ACF r(k) (k) ( ) ,

(k)
(k)


(t)
(t + ) of the kth sample function of
(t) is found to be equal to
r(k) (k) ( ) =

N
X

n=1

c2n exp{j2 fn },

k.

(3.33)

It follows from (3.33) that the ACF of

(k) (t) is equal to the one of the ensemble,
PN
r ( ), since r ( ) = n=1 cn exp{j2 fn } [Eq. (3.5)]. We observe that the
time-averaged ACF r(k) (k) ( ) can be factorized in terms of the ACFs and CCFs
(k)

(k)

of the IQ components
I (t) and
Q (t) of
(k) (t) in a similar way as described
in Section 3.3.3.1 for the ACF of the ensemble, that is,



r(k) (k) ( ) + r(k) (k) ( ) + j r(k) (k) ( ) r(k) (k) ( )
I
I
Q
Q
I
Q
Q
I


= 2 r(k) (k) ( ) + jr(k) (k) ( )
(3.34)

r(k) (k) ( ) =

where

r(k) (k) ( ) = r(k) (k) ( ) =

I

r(k) (k) ( ) = r(k) (k) ( ) =

I

N
X
c2

cos j2 fn ,

(3.35a)

sin j2 fn ,

k.

(3.35b)

n=1
N
X
c2n
n=1

38

Channel Simulation Models for Mobile Broadband Communications

Let us assume also that the Doppler frequency f of the specular wave m (t)
is defined in such a way that the Condition 3.4 holds, so that f 6= fn n. One


can then show that the time-averaged ACF r(k) (k) ( ) , (k) (t) (k) (t + )
of the kth sample function of (t) meets
r(k) (k) ( ) = r(k) (k) ( ) + 2 exp{j2f }
=

N
X

n=1

c2n exp{j2 fn } + 2 exp{j2f },

(3.36)
k.

(3.37)

Interestingly, a comparison between (3.37) and the result presented in (3.11) for
the ACF r ( ) of the ensemble shows that r(k) (k) ( ) = r ( ) k.
3.4.3.2

DPSDs

For completeness, we will also analyze the spectral characteristics of the simulation models sample functions. To simplify our notation, we will henceforth omit
the superscript {}(k) whenever there is no need to make a distinction among
sample functions, or among their statistical properties.
It is straightforward to show, by computing the Fourier transform of (3.37),
that the DPSD S (f ) of the sample functions of (t) is given by
S (f ) =

N
X

n=1

c2n f fn + 2 (f f ).

(3.38)

Analogously, the DPSD S (f ) of the sample functions of

(t) is found to be
equal to
S (f ) =

N
X

n=1

3.4.3.3

c2n f fn .

(3.39)

Average Doppler Shift and Doppler Spread

By comparing the results presented in (3.38) and (3.12), we realize that the
DPSDs of (t) and (t) are exactly the same, i.e., S (f ) = S (f ). Consequently,
we can deduce that the average Doppler shift A and Doppler spread D of
(t) are equal to those of (t). In the spirit of this reasoning, we can write:


 

1/2
P
P
A = 12 2 f + N
2n fn ; and D = 12 2 f2 + N
2n fn2 (A )2
.
n=1 c
n=1 c

(t) can
Accordingly, the average Doppler shift A and Doppler spread D of
P
2 f and
be presumed to be equal to those of
(t), so that A = 12 N
c

n=1 n n

 1 PN 2 2

1/2
D = 2 n=1 cn fn (A )2
. These results are remarkable indeed, as

they imply that the relevant spectral moments of the simulation model can be
measured from a single sample function (t).

Chapter 3 The Narrowband SOC-Based Simulation Model

3.4.4

39

ACF of the Deterministic Squared Envelope

To close this section, we investigate the time-averaged ACF r2 2 ( ) , 2 (t)


2 (t + ) of the sample functions 2 (t) of 2 (t). A closed-form solution for such

an ACF is derived in Appendix C upon the assumption that the Conditions 3.2,
3.3, and the three conditions stated below are satisfied altogether:

(t) are specified

Condition 3.5 If N 4, then the N Doppler frequencies fn of
in such a way that

l = m = n = k;
fl + fm = fn + fk , iff
or l = n, m = k, l 6= m;

or l = k, m = n, l 6= m.

Condition 3.6 The specular wave m (t) and the random cisoids underlying
(t)

are defined in such a way that |f | =

6 |fn | n.
Condition 3.7 If N 2, then the Doppler frequency f of the specular wave
m (t) and the N Doppler frequencies fn of
(t) are defined in such a way that

f + fl 6= fm + fn , l, m, n.
Under these considerations, it is shown there that

42 cR
r2 2 ( ) = r2 2 ( ) +
rI I ( ) cos (2f )
cR + 1

cR 4 (cR + 2)
+rI Q ( ) sin (2f ) +
(cR + 1)2

(3.40)


where r2 2 ( ) , 2 (t)2 (t + ) is the ACF of the squared envelope 2 (t) ,
|
(t)|2 of
(t). For this latter ACF, it is shown in Appendix C that
r2 2 ( ) =

N
X
4
2
+
|r
(
)|

c4n

(cR + 1)2
n=1

(3.41)

if the Condition 3.5 is met. Notice that the results presented in (3.40) and (3.29)
are equivalent to each other, implying that the ACF of 2 (t) is equal to that of the
ensemble, i.e., r22 ( ) = r2 2 ( ). With respect to the mean value m2 , h2 (t)i
of the sample functions of 2 (t), it can easily be shown that m 2 = 2 . This

result holds provided that the Conditions 3.2 and 3.3 are fulfilled.
We pointed out in Section 3.3.3.1 that if the Doppler frequencies fn of
(t)

satisfy the Conditions 3.2 and 3.3, i.e., if fn 6= 0 n, and fn 6= fm , n 6= m, then the
fulfillment of the Condition 3.1 becomes necessary in order for the IQ components
of
(t) to be uncorrelated. Under such circumstances, the solution given in (3.41)
for r2 2 ( ) is no longer valid, since the Condition 3.1 is not compatible with the

40

Channel Simulation Models for Mobile Broadband Communications

Condition 3.5. This is clear, as the former condition states that for each pair of
parameters (
cn , fn ), n = 1, . . . , N , there exists one and only one pair (
cm , fm ),
n 6= m, such that cn = cm and fn = fm (N is even), whereas the latter condition
establishes that fl + fm = fn + fk iff: l = m = n = k; or l = n, m = k, l 6= m;
or l = k, m = n, l 6= m (N 4). Obviously, if the Condition 3.1 is met, then the
equation fl + fm = fn + fk will have more solutions than the ones specified by
the Condition 3.5.
Since the simulation of fading channels having diffuse components with uncorrelated IQ components is relevant for many practical purposes, e.g., for analyzing
the system performance under isotropic scattering conditions, we derive in Appendix D a solution for r2 2 ( ) by assuming the fulfillment of the Conditions 3.1
3.3. For notational convenience, and without loss of generality, we suppose that
the Doppler frequencies fn are indexed in such a way that fn < fm n < m. The
expression we have obtained for such a particular case is the following
N
X
4
2
r2 2 ( ) =
+ |r ( )|
c4n
(cR + 1)2
n=1
(  N/2

X 2



2


+4 
cm cos 2 fm exp j(m + N m+1 ) 
m=1

N/2

X
k=1

c4k

cos

2 fk



(3.42)

where N is even. Interestingly, the result present above depends on the cisoids
phases n . This is in contrast to the solution given in (3.41), which is not influenced by those parameters. We notice that the result presented in (3.40) for
the time-averaged ACF r2 2 ( ) of the sample functions 2 (t) of 2 (t) does not
change if we remove the Condition 3.5 from our assumptions. However, in this
case the function r2 2 ( ) proves to be different from the ACF of the ensemble
r22 ( ), since r2 2 ( ) 6= r22 ( ) [cf. (3.30) and (3.42)].

3.5

On the Ergodicity of the Stochastic SOC-Based

Simulation Model

On the basis of the statistical properties of

(t) and (t) described so far, we can
analyze the mean, power, and autocorrelation ergodicity of our stochastic SOCbased simulation model. Before we proceed to study the subject, we will define
the concepts of mean ergodicity, power ergodicity, and autocorrelation ergodicity
(cf. [Leon94, Sec. 6.6]).

Chapter 3 The Narrowband SOC-Based Simulation Model

41

x(t)}
Definition 3.1 Let x (t) be a random process whose mean value mx (t) , E{x
is constant over time, i.e., mx (t) = mx . Then, x (t) is said to be mean-ergodic if:
(k)
The mean value mx (t) , hx(k) (t)i of every sample function x(k) (t) of x (t)
(k)
is equal to mx , i.e., mx (t) = mx k.

x (t)|2 } is
Definition 3.2 Let x (t) be a random process whose power Px (t) , E{|x
constant over time, i.e., Px (t) = Px . Then, x (t) is said to be power-ergodic if:
(k)

The time-averaged power Px , h|x(k) (t)|2 i of every sample function x(k) (t)
(k)
of x (t) is equal to Px , meaning that Px = Px k.
x (t1 )
Definition 3.3 Let x (t) be a random process whose ACF rxx (t1 , t2 ) , E{x
x (t2 )} depends only on the time difference = t2 t1 , so that rxx (t1 , t2 ) = rxx ( ).
Then, x (t) is said to be autocorrelation-ergodic if:
The time-averaged ACF rx(k) x(k) ( ) , h(x(k) (t)) x(k) (t+ )i of every sample
function x(k) (t) of x (t) is equal to rxx ( ), that is, rx(k) x(k) ( ) = rxx ( ) k.
We observe that an autocorrelation-ergodic random process is always powerergodic, but the converse is not necessarily true.
In accordance to the previous definitions, and from the results presented in
Sections 3.3 and 3.4, we can conclude that the WSS zero-mean process
(t)
is mean-ergodic, power-ergodic, and autocorrelation-ergodic provided that the
Conditions 3.2 and 3.3 are satisfied. This is clear, since m = m(k) = 0, P =
PN 2
P(k) = 2 , and r ( ) = r(k) (k) ( ) =
n exp{j2 fn } for all k upon
n=1 c
fulfillment of the above mentioned conditions.
On the other hand, if the Conditions 3.23.4 are met, then the simulation
model described by (t) turns out to be power-ergodic and autocorrelationergodic too, as P = P(k) = 2 and r ( ) = r(k) (k) ( ) = r ( ) + 2 exp{j2
f } k. However, the simulation model is in general not mean-ergodic, because its mean value m (t) may vary over time [Sec. 3.3.2]. Nevertheless, if
the Conditions 3.23.4 are satisfied and the specular wave m (t) is not Doppler
shifted, so that f = 0, then (t) proves to be mean-ergodic. Notice that
m (t) = m(k) = 2 exp{j2 } k under such circumstances. Needless to say,
(t) is mean-ergodic if = 0, since (t) reduces to
(t) when = 0.
In a similar spirit, we can conclude from the results presented in Sections 3.3.6
and 3.4.4, that the SOC models squared envelope 2 (t) is a mean-ergodic and
an autocorrelation-ergodic process, since m2 = m2(k) = 2 and r22 ( ) =
r2(k) 2(k) ( ) k. We point out, nonetheless, that these properties are subject to
the fulfillment of the Conditions 3.2, 3.3, and 3.53.7. If any of these conditions is
not met, then the time-averaged ACF r2(k) 2(k) (t) of the kth sample function of
(k)
2 (t) will depend on the set of phases {n , }, as mentioned in Appendix C. In

42

Channel Simulation Models for Mobile Broadband Communications

such a case, the requirements stated in Definition 3.3 will not be satisfied, since
the ACF of the ensemble does not depend on the cisoids phases [see (3.29)].
Moreover, we can conclude on the basis of (3.29), (3.30), (3.40), and (3.42) that
the squared envelope processes 2 (t) and 2 (t) are non-autocorrelation-ergodic if
the simulation model is parameterized in such a way that the Doppler frequencies
fn of
(t) satisfy the Conditions 3.13.3, i.e., if
(t) is autocorrelation-ergodic
and its IQ components are uncorrelated.

3.6

Chapter Summary and Conclusions

In this chapter, we described the characteristics of the stochastic SOC-based simulation model that we have adopted as core structure for the narrowband channel
simulators proposed in this dissertation. Important statistics of the simulation
model, such as its ACF r ( ) and DPSD S (f ), as well as the first-order distri and phase

butions of its envelope (t)

(t),
were revised throughout the chapter.
It was shown in Section 3.3.3 that r ( ) and S (f ) are fully specified by the
cisoids gains and Doppler frequencies. In addition, it was shown in the same section that the random process (t) characterizing our simulation model possesses
similar stationarity characteristics as the complex Gaussian process (t) describing our reference model. We demonstrated in Section 3.3.4 that the envelope
of (t) is a first-order stationary process, whereas its phase

(t)
(t)
is in general not (unless = 0 or f = 0). Similar observations were made in Chapter 2
regarding the first-order stationarity of the envelope (t) and phase (t) of (t).
Interestingly, the numerical results presented in Figs. 3.2 and 3.3 [see Sec. 3.3.4]
indicate that the simulation model is well suited for approximating the first-order
distributions of (t) and (t).
The mean, power, and autocorrelation ergodicity of (t) was also investigated
here. In this regard, it was found that (t) is both power and autocorrelation
ergodic, but its mean ergodicity is restricted to the cases where f = 0 or = 0.
The mean and autocorrelation ergodicity of the SOC models squared envelope
2 (t) was analyzed in this chapter too. The results presented in Sections 3.3.6
and 3.4.4 show that 2 (t) is mean ergodic and autocorrelation ergodic if the
cisoids parameters are computed in conformity with the Conditions 3.2, 3.3, and
3.53.7. We observe, nonetheless, that the squared envelope process 2 (t) is not
autocorrelation ergodic when the simulation models random component
(t) is
autocorrelation ergodic and its IQ components are uncorrelated.

Chapter 4

Parameter Computation
Methods
4.1

Introduction

Once the structure of the stochastic SOC-based simulation model has been defined, the problem consists in finding values for the cisoids parameters that allow
for a proper emulation of the reference model. Basically, the problem lies in specifying the gains and the Doppler frequencies of the simulation models random
component,
(t), in such a way that the statistical properties of
(t) resemble
those of the zero-mean complex Gaussian process (t) characterizing the channels diffuse component. The computation of the phases of
(t) does not require
any special attention, since these parameters have been defined in Section 3.3.1
as i.i.d. random variables uniformly distributed in [, ).
Currently, there exist only two methods suitable for the computation of the
gains and Doppler frequencies of
(t), namely, the extended method of exact
Doppler spread (EMEDS) [Pae08a] and the Lp -norm method (LPNM) [Pae02a,
Sec. 5.1.5]. The EMEDS was introduced in [Hogs05] as an extension of the method
of exact Doppler spread (MEDS) [Paet98] for SOC-based channel simulators. The
LPNM, on the other hand, was proposed in [Paet98] as a parameter computation
method for SOS models, and the approach was later applied in [Paet01] to design
SOC-based channel simulators.
The EMEDS and the LPNM have widely been in use to simulate SISO and
MIMO mobile fading channels, e.g., [Rafi08], [Pae04a]. The performance of both
methods is remarkable indeed, as one may conclude from the analysis presented
in [Pae04a] and [Pae08a]. Unfortunately, the applicability of the EMEDS is restricted to the simulation of isotropic scattering channels, while the numerical
optimization techniques, the LPNM relies upon, significantly increase the complexity of the SOC-based simulator.
43

44

Channel Simulation Models for Mobile Broadband Communications

Aiming at the simulation of both isotropic and non-isotropic scattering channels, and with the purpose of circumventing the computational burden of the
LPNM, we introduce in this chapter two simple and effective methods to address
the parameter computation problem at hand. One of such methods is built upon
a Riemann sum approximation of the ACF of (t), while the other is a generalization of the method of equal areas (MEA) described in [Pae02a, Sec. 5.1.3].
Even though the generalized MEA (GMEA) and the Riemann sum approximation
method (RSAM) are presented here in the context of narrowband SISO systems,
these two fundamental methods can easily be applied to the simulation of narrowband and wideband MIMO systems, as will be demonstrated in Chapters 5
and 6.
We herein analyze the performance of the proposed methods not only with
respect to the emulation of the correlation and spectral characteristics of (t), but
also in terms of the approximation of the envelope and phase distributions of the
Rician fading channel model characterized by the composite process (t) = (t)+
m (t). In addition, we evaluate the accuracy of the methods for approximating
the ACF of the squared envelope of (t). The results produced by the GMEA and
RSAM are compared with results obtained by using the LPNM. For the special
case of the design of simulation models for isotropic scattering channels, it will
be shown that both the GMEA and RSAM simplify to the EMEDS.

4.2

Overview of the Existing Parameter Computation

Methods for SOC-Based Channel Simulators

Before we proceed to introduce the GMEA and the RSAM, we will present in
this section a brief review of the EMEDS and the LPNM.

4.2.1

The EMEDS

In the case on the EMEDS, the parameters cn and fn are defined as [Hogs05]:
cn =

fn = fmax cos

(4.1)




1
2
n
N
4

(4.2)

for n = 1, . . . , N . One can easily verify, by substituting (4.1) and (4.2) into
(3.18), that the Doppler spread of
(t) reduces to the one of isotropic scattering

+
channels for any N Z , meaning that D = D = fmax / 2, N 1. We
(t) and (t), respectively.
recall that D and D denote the Doppler spread of
Thus, as the name suggests, the EMEDS exactly reproduces the Doppler spread
of isotropic scattering channels.

Chapter 4 Parameter Computation Methods

4.2.2

45

The LPNM

The LPNM, as originally introduced in [Paet01] for the design of SOC channel
simulators, also defines the gains of
(t) as in (4.1), but it requires the Doppler

frequencies fn to be computed in a way that minimizes the Lp -norm

(p)
r

1
max

Zmax
0

p



r ( ) r ( ) d

)1/p

p Z+

(4.3)

where max > 0 determines the length of the interval [0, max ] inside of which the
approximation r ( ) r ( ) is of interest. We bring back to memory that
(t), in that order.
r ( ) and r ( ) denote the ACFs of (t) and
There exist other versions of the LPNM, which are surveyed in [Pae02a,
Sec. 5.1.5]. In this chapter, we will consider two of them for the performance
comparison with the parameter computation methods herein proposed. For the
first variant under consideration, the gains and the Doppler frequencies of
(t)
are to be computed in such a way that the cost function
(p)
r , p

(p)
, W1 (p)
r + W2 p

(4.4)

is minimized, where W1 and W2 are weighting factors and

(p)
,
p

( Z
0


p


p (z) p(z) dz

)1/p

p Z+ .

(4.5)

We remember that p (z) and p(z) stand for the first-order PDFs of the envelopes
(p)
of (t) and
(t), respectively. The second variant aims to minimize r , p too,
but it considers only N 1 pairs of parameters (
cn , fn ) for that purpose. To
guarantee that the boundary conditions r (0) = r (0) and r (0) = r (0) are
fulfilled, the remaining parameters cN and fN are obtained by following:
v
u
N
1
u
X
t
2
cN =

c2n
(4.6)
n=1

fN

v
u
N 1
r (0) X  2
1 u
t

cn fn .
cN
4 2

(4.7)

n=1

The minimums of the cost function in (4.4) can be found by applying the optimization algorithm proposed in [Laga98], the which is implemented as fminsearch
R
function in MATLAB
. We will refer to the original version of the LPNM as
the LPNM I, while we will call the first and second variants described above the

46

Channel Simulation Models for Mobile Broadband Communications

LPNM II and LPNM III, respectively. It is worth mentioning that, in accordance

to [Pae02a, Sec. 5.1.5], the LPNM III is the most efficient variant of the LPNM.

4.3
4.3.1

The GMEA
Description

The numerical results presented in Section 3.3.4 indicate that the envelope distribution of the simulation models random component is closely in line with the
Rayleigh distribution [2.23] when the gains cn are given as in (4.1) and N 10
(see Fig. 3.1(a) for the case cR = 0). In fact, it is shown in [Paet07] that the rms
error function
(2)
=
p

( Z
0

h
i2
p (z) p(z) dz

)1/2

(4.8)

is around 0.02 if N = 10, and smaller than 0.01 if N > 20. Motivated by these
results, we choose the same gains cn as in (4.1) for the GMEA.
Concerning the computation of the Doppler frequencies fn , we recall that the
ACF of (t) is given in terms of the PDF p () of the random AOAs n , or
more specifically, in terms of the even part g () of p () [see (2.7)(2.9)]. On
(t) is completely specified by the parameters cn ,
the other hand, the ACF of
fn , and N [see (3.5)]. In view of this, it is reasonable to think that in order for
r ( ) to resemble the ACF of the channels diffuse component for a given value
of N , the gains and Doppler frequencies of
(t) should provide information about
the function g (). In the spirit of this reasoning, and since the gains in (4.1)
are blind to the AOA statistics, we will compute the Doppler frequencies of
(t)
such that the underlying deterministic AOAs
n satisfy the equation
Z n

g ()d =

1
,
2N

n = 2, . . . , N

(4.9)

n1

with
n (0, ). We recall that the uncorrelation between the IQ components
of
(t) is a fundamental feature for the design of simulation models for fading channels having diffuse components characterized by symmetrical DPSDs
[Sec. 3.3.3.1]. We show in Appendix E that if the DPSD of (t) is symmetrical,
and the deterministic AOAs
n satisfy (4.9), then the IQ components of
(t) are
mutually uncorrelated if and only if
Z 1
0

g ()d =

1
.
4N

(4.10)

Chapter 4 Parameter Computation Methods

47

We will consider (4.10) as an initial condition for computing the AOAs

n to
ensure that the parameter computation method presented in this section is well
suited for the simulation of fading channels characterized by symmetrical DPSDs.
On the other hand, it was pointed out in Section 3.4.2 that the simulation
model described by
(t) is mean-ergodic and autocorrelation-ergodic if:
fn =
6 0, n
6 fm , n =
6 m.
fn =

Condition 3.2 :
Condition 3.3 :

(4.11)
(4.12)

An important implication of the requirement stated in (4.9) is that the AOAs

n
will satisfy
n 6=
m for n 6= m, and
n > 0 n. Thus, by demanding the AOAs

n to meet (4.9), we implicitly guarantee the fulfillment of the Condition 3.3, as

the equality fn = fm holds for some n 6= m if and only if
n =
m . Furthermore,
the Condition 3.2 is unsatisfied if and only if
n = /2 for some n. This situation
is rather unlikely when the DPSD of (t), S (f ), is asymmetrical, and it is never
the case when S (f ) is symmetrical and the number of cisoids N is even.
On the basis of (4.9) and (4.10), we can compute the AOAs
n by employing
numerical root-finding techniques to solve
Z n

g ()d =



1
1
n
,
2N
2

n = 1, . . . , N.

(4.13)

The function g () may itself be regarded as being a PDF, the CDF of which is
R
n results into
given by G () , g (x)dx. The evaluation of G () for


1
1
G (
n ) = 2N n + N 2 . Hence, if a closed-form solution exist for the inverse
function G1 of G , then the AOAs
n can be computed by evaluating

n =

G1



1
1
n+N
,
2N
2

n = 1, . . . , N.

(4.14)

Once the AOAs

n are known, the Doppler frequencies fn can easily be obn ) [Eq. (3.3)]. Often, however,
tained by using the transformation fn = fmax cos(
the channels ACF and/or DPSD are introduced without giving any explicit information about the distribution of the random AOAs n , such as in [Blau06].
For those cases, we observe that
Z n

g ()d =

fZmax

fn

1
2

g (arccos(f /fmax ))
p
df
fmax 1 (f /fmax )2

fZmax

pf (f )df.

fn

(4.15)

48

Channel Simulation Models for Mobile Broadband Communications

Taking into account the result presented in (4.13), and given that the DPSD of
(t) is equal to S (f ) = 2 pf (f ) [cf. Eqs. (2.9) and (2.15)], we have
Zfn

S (f )df

2
N

fmax

1
N n+
2

n = 1, . . . , N

(4.16)

where fn (fmax , fmax ). The integral in (4.16) describes a sort of cumulative

Rf
power function that is proportional to the CDF Pf (f ) , pf (x) dx of the
random Doppler frequencies of (t); notice that
Pf (f ) =

1
2

Zf

S (y)dy.

(4.17)

Thus, for the special case where the inverse Pf1 of Pf exists, the Doppler frequencies of
(t) can be computed by evaluating
fn =

Pf1



1
1
N n+
,
N
2

n = 1, . . . , N.

(4.18)

Otherwise, the deterministic Doppler frequencies fn are to be computed by solving (4.16) with the aid of numerical root-finding techniques.

4.3.2

Differences Between the GMEA and Other Versions of the

MEA

The parameter computation method presented in this section establishes the socalled GMEA. This method is basically an extension of the MEA described in
[Pae02a, Sec. 5.1.3] with respect to the simulation of Rayleigh fading channels
characterized by asymmetrical DPSDs. Such an extension was necessary, since
the original MEA is specific for the SOS-based simulation of fading channels
with symmetrical DPSDs. In fact, the deterministic AOAs
n obtained by using

the MEA are confined within (0, 2 ), an interval that does not convey enough
information about the distribution of the random AOAs n as to allow for an
adequate simulation of channels having asymmetrical DPSDs.
We notice that a previous attempt to generalize the MEA was made in
[Pae02c, Sec. 4-B]. The generalization proposed in that paper was designed by
assuming that p () is symmetrical, implying that p () = g (). Under this
consideration, the parameters
n are computed there by solving
Z n
0

p ()d =

1
,
N

n (0, )

(4.19)

Chapter 4 Parameter Computation Methods

49

for n = 1, . . . , N/2. The remaining N/2 AOAs

n are specified as mirrored versions of the solutions of (4.19), i.e.,
n =
N/2+1n , n = 1, . . . , N/2. Such a parameter computation approach is obviously inadequate when p () is not an even
N/2+1n , n = 1, . . . , N/2 produces dufunction. What is more, taking
n =
plicated Doppler frequencies, since cos(
n ) = cos(
N/2+1n ), n = 1, . . . , N/2.
This characteristic not only results into a non-autocorrelation-ergodic simulator,
but also reduces the efficiency of the simulation model. Another problem of the
method in [Pae02c, Sec. 4-B] is that the solutions of (4.19) do not guarantee the
(t) when the DPSD of the channels
uncorrelation between the IQ components of
diffuse component is symmetrical. For this reason, the generalized MEA there
presented proves to be unsuitable for the simulation of channels characterized by
such a type of DPSDs.
To illustrate the above mentioned problems, we present in Fig. 4.1 a comparison between the ACFs of (t) and
(t) by considering the method proposed in
[Pae02c, Sec. 4-B] with N = 20 and the von Mises distribution of the AOA with
pairs of parameters: m = 30 , = 10; and m = 90 , = 10. The former
pair of parameters is representative of the case p () 6= p (), while the latter
pair exemplifies the case where the IQ components of (t) are uncorrelated. It
can be observed from Fig. 4.1 that the generalization of the MEA described in
[Pae02c] produces a poor approximation to the ACF of (t). In contrast, it will
be shown in Section 4.5.1.1 that the GMEA proposed in this chapter provides a
good approximation to rVM
( ) in both cases.

4.3.3

Implementation

In what follows, we will apply the GMEA to the von Mises and Laplacian AOA
distributions with the purpose of demonstrating the methods implementation.

4.3.3.1

von Mises Distributed AOAs

In the case of the von Mises PDF, pVM () [Eq. 2.43], neither the CDF of the even
part of such a distribution, nor the cumulative power function of the resulting
DPSD of (t) [Eq. (2.45)], can be evaluated analytically. The Doppler frequencies
of
(t) shall therefore be computed by numerically finding the values of fn that
solve
1
fmax I0 ()

Zfn


 exp { cos(m )f /f }
p

max
p
cosh sin(m ) 1 (f /fmax )2
df
1 (f /fmax )2
fmax


1
1

(4.20)
N n+
= 0,
|fn | < fmax
N
2

50

Channel Simulation Models for Mobile Broadband Communications

(a)

(b)

1.2

1.2
VM
Reference model, |r
()|

Simulation model, |r ()|

Absolute value of the ACF

Absolute value of the ACF

VM
Reference model, |r
()|

m = 30
= 10

0.8

0.6

0.4

0.2

0
0

m = 90
= 10

0.8

0.6

0.4

0.2

0
0

Simulation model, |r ()|

Normalized time difference, fmax

(c)
VM
Re{r
()}

VM
Reference model, Re{r
()}

Simulation model, Re{r ()}

Simulation model, Re{r ()}

m = 30
= 10

0.5

Real part of the ACF

Real part of the ACF

(d)

Reference model,
1

Normalized time difference, fmax

0.5

m = 90
= 10

0.8

0.6

0.4

0.2

1
0

0
0

Normalized time difference, fmax

(e)

VM
Reference model, Im{r
()}

0.6

Simulation model, Im{r ()}

Imaginary part of the ACF

Imaginary part of the ACF

(f)

VM
Reference model, Im{r
()}

Normalized time difference, fmax

m = 30
= 10

0.5

0.5

Simulation model, Im{r ()}

0.5

m = 90
= 10

0.4
0.3
0.2
0.1
0

1
0

Normalized time difference, fmax

0.1
0

Normalized time difference, fmax

Figure 4.1: Comparison between the ACF of the channels diffuse component
and the ACF of the SOC-based simulation models random component by considering the von Mises PDF of the AOA and the
generalized version of the MEA proposed in [Pae02c] with N = 20.

Chapter 4 Parameter Computation Methods

51

for n = 1, . . . , N . Alternatively, one may compute the underlying AOAs

n of

(t) by solving
1
2I0 ()

Z n
0

exp{ cos() cos(m )} cosh sin() sin(m ) d


1
1
n
= 0,

2N
2

n (0, )

(4.21)

for n = 1, . . . , N . This latter approach is indeed the one we will follow in this dissertation. We recall that the von Mises PDF reduces to the uniform distribution
when = 0. In that case, both (4.20) and (4.21) have the analytical solution
fn = fmax cos




1
n
,
N
2

n = 1, . . . , N.

(4.22)

Without going into details, we observe that (4.22) and (4.2) produce the same set
of Doppler frequencies {fn }N
n=1 . Therefore, we can state that the GMEA contains
the EMEDS as a special case for the simulation of isotropic scattering channels.
4.3.3.2

Laplacian Distributed AOAs

In the case of the Laplacian distribution, pLA () [see (2.51], the Doppler frequencies fn are to be computed by finding the roots of
2
cs fmax

Zfn

fmax





exp 2 arccos (f /fmax ) /s
1
1
p
df
N n+
N
2
1 (f /fmax )2
= 0,

|fn | < fmax

(4.23)

n (0, )

(4.24)

for n = 1, . . . , N . One could also proceed by solving

1
cs

Z n
0

exp




1
1
1

2|| d
n
= 0,
s
2N
2

for n = 1, . . . , N . In either case, we obtain the closed-form solution





s
cs
1

fn = fmax cos
n
ln 1
2
2
2N s

(4.25)

for n = 1, . . . , N , where ln{} is the natural logarithm.

Figures 4.2 and 4.3 show a comparison between the DPSDs of the simulation
model and the reference model by applying the GMEA to the von Mises (Fig. 4.2)
and Laplacian (Fig. 4.3) AOA distributions. Such figures graphically demonstrate

52

Channel Simulation Models for Mobile Broadband Communications

(a)

(b)

m = 0
=0

0
5
10

DPSD (dB)

DPSD (dB)

m = 0
=5

10

15

20

15
20
25
30
35
40

25
45
30
90

60

30

30

60

50
30

90

Doppler frequency, f (Hz)

30

(c)
5

m = 30
= 10

0
5

10

10

15

15

DPSD (dB)

DPSD (dB)

20
25
30

25
30
35

40

40

45

45
0

30

60

Doppler frequency, f (Hz)

90

m = 90
= 10

20

35

50
30

90

(d)

5
0

60

Doppler frequency, f (Hz)

50
90

60

30

30

60

90

Doppler frequency, f (Hz)

Figure 4.2: Comparison between the DPSD of the channels diffuse component
(blue-solid lines) and the DPSD of the random component of the
SOC-based simulation model (black-dotted lines with markers)
by considering the von Mises PDF of the AOA and the GMEA
(fmax = 91 Hz, 2 = 1, and N = 10).
that despite the fact that the GMEA assigns the same amount of power to each
of the cisoids in
(t), the method specifies the Doppler frequencies fn in such a
way that the power of
(t) is more densely concentrated within the zones where
the DPSD of (t) is higher. This characteristic enables the method to provide a
good approximation to the correlation and spectral properties of (t), as will be
demonstrated in Section 4.5.

Chapter 4 Parameter Computation Methods

53

(a)

(b)

0
5

0
5

s = 0.3

10
15

15

DPSD (dB)

DPSD (dB)

s = 0.5

10

20
25
30

20
25
30

35

35

40

40

45
50
30

30

60

45
90

90

60

Doppler frequency, f (Hz)

30

30

60

90

60

90

Doppler frequency, f (Hz)

(c)

(d)

s = 5
5

s = 1

10

DPSD (dB)

DPSD (dB)

10

15

15

20

20
25
25

30
90

30

60

30

30

Doppler frequency, f (Hz)

60

90

35
90

60

30

30

Doppler frequency, f (Hz)

Figure 4.3: Comparison between the DPSD of the channels diffuse component
(t) (blue-solid lines) and the DPSD of the random component

(t) of the SOC-based simulation model (black-dotted lines with

markers) by considering the Laplacian PDF of the AOA and the
GMEA (fmax = 91 Hz, 2 = 1, and N = 10).

4.4

The RSAM

4.4.1

Description

4.4.1.1

Basic Approach

For the second parameter computation method, the RSAM, we will assume that
the PDF of the random AOAs n contains no singularities, so that one can regard
the integral underlying the ACF of the channels diffuse component [see (2.7)] as
being a proper integral. Under such an assumption, the ACF of (t) can be

54

Channel Simulation Models for Mobile Broadband Communications

written as a midpoint Riemann sum [Edwa98, Sec. 5.4] of the form



N
h
h
X
22
1 i
1 i
r ( ) =
lim
g
n
exp j2fmax cos
n
.
N N n=1
N
2
N
2
If we drop the limit from the equation above, then we may presume that


N
h
h
22 X
1 i
1 i
r ( )
g
n
exp j2fmax cos
n
. (4.26)
N n=1
N
2
N
2
A comparison of (4.26) and the expression given in (3.5) for the ACF of
(t)
suggests that r ( ) will render a good approximation to r ( ) if we choose:

1
n
Ns 2
g (
n )
= PN
m )
m=1 g (

n =

(4.27)

cn

(4.28)

for n = 1, . . . , N . The methodology given by (4.27) and (4.28) establishes a parameter computation method that we will refer to as the basic RSAM (BRSAM).
It is worth mentioning that the idea behind the BRSAM has recently been applied
in [Gut07a] to simulate mobile MIMO Rayleigh fading channels, yielding remarkable results concerning the emulation of the spatial CCF and temporal ACF of



the channel. Interestingly, the Doppler frequencies fn = fmax cos N

n 12 obtained by using this method always satisfy the Condition 3.3, and the fulfillment
of the Condition 3.2 is guaranteed too provided that [0, ) is partitioned into an
even number of segments, implying that N = 2M , M Z+ .
4.4.1.2

Problems of the Basic Approach

From experiments, we have observed that irrespective of the AOA statistics, the
BRSAM produces an excellent
to the ACF of (t) for values of
h approximation
i
N
N
close to the origin, say, 4fmax , 4fmax . However, our experiments have also
revealed that this method performs poorly regarding the emulation of the envelope distribution of (t) if the range of g () is huge. Under such circumstances,
there is a large difference between the gains cn of the cisoids. While this characteristic does not entail any problems for the emulation of r ( ), it does affect the
ability of
(t) for approximating the PDF of the envelope of (t), as the envelope
distribution of the simulation models random component is heavily influenced by
the gains cn [see (3.21)]. Figures 4.4 and 4.5 illustrate the problem by considering
the von Mises PDF pVM () with parameters m = 0 and {10, 20}. Such
parameters characterize the AOA statistics of propagation scenarios that exhibit

Chapter 4 Parameter Computation Methods

55

(a)

(b)

10

20

c2
c3

10

c4

20

Gains, cn (dB)

30

= 10

V M ()

g
P
N

n=1

c2n

c6

c4

c7

n=1

c2n

80

c6
c7
c8

140

c9

80

N = 10
2 = 1

c5

60

120

c8

= 20

V M ()

g
P
N

100

70

c9

160

c10

90
100

c3

40

40

60

c2

20

N = 10
2 = 1

c5

50

c1

Gains, cn (dB)

c1

c10

180

1
0

10

200

1
0

Value of the deterministic AOAs,

n

10

Value of the deterministic AOAs,

n

Figure 4.4: Relationship between the gains cn and AOAs

n of the SOCbased simulation models random component,
(t), by applying
the BRSAM to the von Mises PDF of the AOA with parameters
m = 0 and {10, 20}.
(a)

(b)

p(z)

= 10

p(z)

1.8

= 20

Envelope PDF

Envelope PDF

1.6
0.8

N = 10
2 = 1

0.6

0.4

p (z)

1.4

N = 10
2 = 1

1.2
1
0.8
0.6

p (z)

0.4

0.2

0.2
0
0

0.5

1.5

Envelope level

2.5

0
0

0.5

1.5

2.5

Envelope level

Figure 4.5: Comparison between the Rayleigh PDF (p (z)) and the first-order
PDF of the envelope of the simulation models random component
(p(z)) by applying the BRSAM to the von Mises PDF of the AOA
with parameters m = 0 and {10, 20}.
severe non-isotropic scattering conditions. Figure 4.4 shows graphs of the relationship between the gains cn and AOAs
n obtained by applying the BRSAM
with N = 10. Plots of the Rayleigh distribution [Eq. (2.23)] and the PDF of the
envelope of
(t) [Eq. (3.21)] are presented in Fig. 4.5.

56
4.4.1.3

Channel Simulation Models for Mobile Broadband Communications

Improved Approach

To overcome the aforementioned problem, we will present in what follows an

improved version of the BRSAM. To that end, we will assume that the PDF of
the AOA is defined in such a way that its even part has at most one maximum in


[0, ). Thereby, for any given threshold 0, sup{g ()}[0,) , where sup{}
denotes the supremum, we can identify one and only one subinterval IU in [0, )
satisfying
g () > ,

IU

(4.29)

implying that g () is above the threshold only along IU . If the value of is

chosen small, so that g () 0
/ IU , then one can state that
Z
2
r ( )
g () exp{j2fmax cos() }d.
(4.30)
IU

In such a case, it makes sense to compute the gains and AOAs of the stochastic
SOC model described by
(t) by taking into account only the subinterval IU .
Following this reasoning, we will redefine the AOAs
n as follows


u
1

n = +
n
, u >
(4.31)
N
2
for n = 1, . . . , N , where and u designate the lower and the upper boundaries of IU . The methodology given by (4.28) and (4.31) constitutes the RSAM.
We observe that irrespective of the value of , the Doppler frequencies fn =



fmax cos + uN
n 12 obtained by applying the RSAM always meet the
Condition 3.3, such as in the case of the BRSAM. Nevertheless, in contrast to
the BRSAM, the RSAM does not ensure the fulfillment of the Condition 3.2.
Choosing a proper value for the threshold is clearly the critical step in the
RSAM. In fact, when setting the threshold, one has to be aware that the method
will be affected by the same problems of the BRSAM if is too small. On the
other hand, if is given a large value, then the RSAM will become more precise
regarding the approximation of the envelope distribution of (t), but it will loose
accuracy with respect to the approximation of r ( ). While determining an
optimal value for is not a trivial issue, we have found from practice that setting
= 1 103 results in case of the von Mises and Laplacian AOA distributions
in a good approximation to r ( ) and p (z), as will be shown in Section 4.5.
The parameters
n and cn can alternatively be computed by applying the
RSAM directly on the DPSD of (t). We can claim that if there exist only
one subinterval IU in [0, ) for which (4.29) holds, then there exist only one

Chapter 4 Parameter Computation Methods

57

subinterval VU = (f , fu ] in (fmax , fmax ] satisfying

S (f )

2 2
p
,
fmax 1 (f /fmax )2

f VU .

(4.32)

The transformation = arccos (f /fmax ) establishes a one-to-one relation between

the points in VU and IU . Thus, by means of such a transformation, we can rewrite
(4.28) and (4.31) as follows:

cn

v
u

q
u
2

S arccos(fn /fmax ) fmax

fn2
u
q
= t P


N
2
2

fmax
fm
m=1 S arccos(fm /fmax )




1
1
fu
1
arccos
+
n
=

fmax
N
2
q

2
2
f fu +
f2 fmax
fu2
fmax

arccos
2
fmax

(4.33)

(4.34)

for n = 1, . . . , N .
The RSAM has been presented here upon the implicit assumption that there
exists only one cluster of scatterers in the proximities of the receiver. However, it
has been reported from measurements [Spen00] that several clusters of scatterers
may surround the mobile terminal. Consequently, the function g () could have
two or more maximums in [0, ), so that various disjoint subintervals IU,m might
be identified in [0, ) satisfying g () > IU,m , m = 1, . . . , M , 2 M < .
For the simulation of such a kind of channels, we refer the reader to [Gut09a],
where the RSAM has been extended with respect to multiple clusters of scatters.

4.4.2

Implementation

In order to demonstrate the implementation of the RSAM, let us consider again

the von Mises and the Laplacian AOAs distributions. The parameters cn and
n
can be computed by applying the method either on the PDF of the AOA or on
the DPSD of
(t). In this dissertation, we will focus our attention on the former
approach.
4.4.2.1

von Mises Distributed AOAs

In the case of the von Mises PDF1 , pVM (), the boundaries of the subinterval
IU = [ , u ) are to be found by identifying the points in [0, ) at which the
function gVM () crosses the threshold from up to down (corresponding to u )
1

We note that the von Mises PDF pVM () and its even part gVM () have only one maximum
in [0, ), so that at most one subinterval IU can be identified in [0, ) satisfying (4.29).

58

Channel Simulation Models for Mobile Broadband Communications

and/or from down to up (corresponding to ). This has to be done in general

by solving gVM () = 0 with the aid of numerical root-finding techniques.
However, if m {, 0, } and > 0, then the von Mises PDF proves to
be an even function that decreases (for m = 0) or increases (for m = )
monotonically in [0, ). Thus, for such values of m and , there is at most one
real solution to gVM () = 0, (0, ), which if happens to exist is then
given by


ln (2Io ())
+
= arccos
.
(4.35)
cos(m )
Consequently, the boundaries of IU can be found by following

+
8VM +

, if
( g ( ) > 0
=
if g8VM (+ ) 0

0,

or gVM () 6= , [0, )
+
8VM +

k , if
( g ( ) < 0
=
if g8VM (+ ) 0

or gVM () 6= , [0, )

(4.36)

(4.37)

for > 0, where + is given by (4.35) and

dgVM ()
d 

=
sin( m ) exp{ cos( m )}
2I0 ()

+ sin( + m ) exp{ cos( + m )} .

g8VM () =

(4.38)

(4.39)

On the other hand, if = 0, then the boundaries of IU are equal to = 0




and u = , and it can be shown that fn = fmax cos N

n 12 and cn =
p
1/N , n = 1, . . . , N . Under such circumstances, the RSAM and the GMEA
are equivalent to each other, and the EMEDS happens to be a special case also of
the RSAM for the design of simulation models for isotropic scattering channels. In
fact, the RSAM and the EMEDS are developed upon similar approaches. This
is evident, as the latter method has essentially been derived from a Riemann
sum approximation of the ACF r ( ) = J0 (2fmax ) characterizing isotropic
scattering channels [Pae02a, Sec. 5.1.6].
4.4.2.2

Laplacian Distributed AOAs

In the case of the Laplacian distribution, pLA (), the lower boundary of IU is in all
cases equal to = 0, since the even part of pLA () is a monotonically decreasing

Chapter 4 Parameter Computation Methods

59

function in [0, ) and therefore no up-crossings occur within that interval. In

turn, the upper boundary of IU is found to be given as
s
u = ln(cs )
2

(4.40)

which is the solution of gLA () = 0. If the equation above has no solution in

[0, ), then u is set equal to .
(t) and (t) by applying the RSAM
A comparison between the DPSDs of
to the von Mises and Laplacian PDFs is given in Figs. 4.6 and 4.7, respectively.
One may notice from these figures that the DPSD of
(t) exhibits a trend similar
as the one described by the DPSD of (t). Such a characteristic is reflected by
the power of the cisoids of
(t).

4.5

Performance Analysis

In this section, we will analyze the performance of the proposed parameter computation methods with respect to the emulation of the correlation and spectral
characteristics of (t), as well as in terms of the approximation of the fading statistics of the reference model described by the composite process (t) = (t)+m (t).
In addition, the performance of such methods will be compared with that of the
LPNM I, LPNM II, and LPNM III. For the aforementioned purposes, we will
assume that the channels AOA statistics follow the von Mises distribution. We
focus our attention on this distribution since a wide variety of propagation scenarios can be studied from it, spanning from isotropic to non-isotropic scattering
scenarios. However, results obtained by considering the Laplacian PDF can be
found in2 [Gut09b]. The conclusions drawn in that paper about the methods
performance are the same as the ones presented in this chapter.
We carry out our simulations by choosing fmax = 91 Hz, 2 = 1, and taking
into account six different propagation scenarios characterized by the von Mises
PDF with pairs of parameters: (m = 0 , = 0); (m = 0 , = 5); (m =
0 , = 20); (m = 0 , = 10); (m = 30 , = 10); (m = 90 , = 10). The
first and the last of such pairs of parameters are related to channels having diffuse
components characterized by symmetrical DPSDs, whereas the other four pairs
are associated to channels with asymmetrical DPSDs, as illustrated in Fig. 2.5.
Regarding the methods configuration, we set = 1 103 for the RSAM,
whilst we consider p = 2 and max = N/(4fmax ) for the three versions of the
LPNM. In the case of the LPNM I, we use the Doppler frequencies produced
(p)
by the GMEA as initial values to minimize the Lp -norm r defined in (4.3).
On the other hand, for the LPNM II and LPNM III, we employ the gains and
2

We point out, nonetheless, that due to the space limitation, the results obtained for the
LPNM II and LPNM III are not presented in [Gut09b].

60

Channel Simulation Models for Mobile Broadband Communications

(a)

(b)

m = 0
=0

0
5
10

DPSD (dB)

DPSD (dB)

m = 0
=5

10

15

20

15
20
25
30
35
40

25
45
30
90

60

30

30

60

50
30

90

Doppler frequency, f (Hz)

30

(c)
5

m = 30
= 10

0
5

10

10

15

15

DPSD (dB)

DPSD (dB)

20
25
30

25
30
35

40

40

45

45
0

30

60

Doppler frequency, f (Hz)

90

m = 90
= 10

20

35

50
30

90

(d)

5
0

60

Doppler frequency, f (Hz)

50
90

60

30

30

60

90

Doppler frequency, f (Hz)

Figure 4.6: Comparison between the DPSD of the channels diffuse component
(blue-solid lines) and the DPSD of the random component of the
SOC-based simulation model (black-dotted lines with markers)
by considering the von Mises PDF of the AOA and the RSAM
(fmax = 91 Hz, 2 = 1, = 1 103 , and N = 10).
Doppler frequencies produced by the RSAM as initial conditions to minimize the
(p)
cost function r ,p introduced in (4.4). Following the experiments performed in
(p)

[Pae02a, Sec. 5.1.5], we set the underlying weighting factors W1 and W2 of r ,p

to W1 = 1/4 and W2 = 3/4.

Chapter 4 Parameter Computation Methods

61

(a)

(b)

0
5

0
5

s = 0.3

10
15

15

DPSD (dB)

DPSD (dB)

s = 0.5

10

20
25
30

20
25
30

35

35

40

40

45
50
30

30

60

45
90

90

60

Doppler frequency, f (Hz)

30

(c)

30

60

90

60

90

(d)

Doppler frequency, f (Hz)

s = 5

s = 1

10

DPSD (dB)

DPSD (dB)

10

15

15

20

20
25
25

30
90

30

60

30

30

Doppler frequency, f (Hz)

60

90

35
90

60

30

30

Doppler frequency, f (Hz)

Figure 4.7: Comparison between the DPSD of the channels diffuse component
(blue-solid lines) and the DPSD of the random component of the
SOC-based simulation model (black-dotted lines with markers)
by considering the Laplacian PDF of the AOA and the RSAM
(fmax = 91 Hz, 2 = 1, = 1 103 , and N = 10).

4.5.1

4.5.1.1

Emulation of the Correlation and Spectral Characteristics

of the Reference Models Random Component
ACF

Figures 4.84.10 show a comparison between the absolute value of the ACF of
the channels diffuse component [see (2.46)] and the absolute value of the ACF of

(t) by applying the parameter computation methods under consideration with

N = 20. For the sake of clarity, we present the results obtained for the three
version of the LPNM in different figures. In addition to theoretical curves of

62

Channel Simulation Models for Mobile Broadband Communications

(a)

m = 0

VM
Reference model, r
()
GMEA, theory
GMEA, realization
LPNM I, theory
LPNM I, realization
RSAM, theory
RSAM, realization

= 20

Absolute value of the ACF

=5

0.8

0.6

=0

0.4

max =

0.2

N
4fmax

Normalized time difference, fmax

(b)

m = 0

VM
Reference model, r
()
GMEA, theory
GMEA, realization
LPNM I, theory
LPNM I, realization
RSAM, theory
RSAM, realization

= 20

Absolute value of the ACF

=5

0.8

0.6

=0

0.4

max =

0.2

N
4fmax

Normalized time difference, fmax

Figure 4.8: Comparison among the GMEA, RSAM, and LPNM I in terms of
the emulation of the ACF of channels diffuse component (reference
model) by considering the von Mises PDF of the AOA with different
pairs of parameters m and (fmax = 91 Hz, 2 = 1, N = 20,
= 1 103 , p = 2, and max = N/(4fmax )).

Chapter 4 Parameter Computation Methods

63

(a)

m = 0

VM
Reference model, r
()
GMEA, theory
GMEA, realization
LPNM II, theory
LPNM II, realization
RSAM, theory
RSAM, realization

= 20

Absolute value of the ACF

=5

0.8

0.6

=0

0.4

max =

0.2

N
4fmax

Normalized time difference, fmax

(b)
VM
Reference model, r
()
GMEA, theory
GMEA, realization
LPNM II, theory
LPNM II, realization
RSAM, theory
RSAM, realization

= 10

Absolute value of the ACF

m = 0

0.8

m = 30

0.6

0.4

m = 90

max =

N
4fmax

0.2

Normalized time difference, fmax

Figure 4.9: Comparison among the GMEA, RSAM, and LPNM II in terms of
the emulation of the ACF of channels diffuse component (reference
model) by considering the von Mises PDF of the AOA with different
pairs of parameters m and (fmax = 91 Hz, 2 = 1, N = 20,
= 1 103 , p = 2, and max = N/(4fmax )).

64

Channel Simulation Models for Mobile Broadband Communications

(a)

m = 0

VM
Reference model, r
()
GMEA, theory
GMEA, realization
LPNM III, theory
LPNM III, realization
RSAM, theory
RSAM, realization

= 20

Absolute value of the ACF

=5

0.8

0.6

=0

0.4

max =

0.2

N
4fmax

Normalized time difference, fmax

(b)
VM
Reference model, r
()
GMEA, theory
GMEA, realization
LPNM III, theory
LPNM III, realization
RSAM, theory
RSAM, realization

= 10

Absolute value of the ACF

m = 0

0.8

m = 30

0.6

0.4

m = 90

max =

N
4fmax

0.2

Normalized time difference, fmax

Figure 4.10: Comparison among the GMEA, RSAM, and LPNM III in terms
of the emulation of the ACF of channels diffuse component (reference model) by considering the von Mises PDF of the AOA with
different pairs of parameters m and (fmax = 91 Hz, 2 = 1,
N = 20, = 1 103 , p = 2, and max = N/(4fmax )).

Chapter 4 Parameter Computation Methods

65

|r ( )| [see (3.5)], Figs. 4.84.10 present empirical graphs of the absolute value
of the time-averaged ACF r ( ) of the sample functions of
(t). Such graphs
were generated by considering a single realization of
(t). It can be seen in the
figures that the graphs of |r ( )| match perfectly the ones of |r ( )|, meaning
that the autocorrelation-ergodicity of
(t) is preserved by all methods.
Concerning the methods performance, one may observe from the results presented in the three figures that the RSAM outperforms by far the LPNM and the
GMEA. In fact, in the case of the RSAM, no differences between |rVM
( )| and
|r ( )| are visible within the interval [0, 5/fmax ]. The six curves of |r ( )| corresponding to such a method can be distinguished from those of |rVM
( )| only with
the aid of the markers and . Even though the performance of the LPNM and
GMEA is not as impressive as that of the RSAM, these two methods produced
a quite good approximation to rVM
( ). Interestingly, the LPNM II provided
very good results for [1/fmax , 5/fmax ], although the quality of the relationship |r ( )| |rVM
( )| is rather poor around the origin. Indeed, Fig. 4.8 clearly
shows that the LPNM II does not satisfy the boundary condition r (0) = r (0),
P
meaning that N
2n 6= 2 .
n=1 c
Without surprise, we can see in Figs. 4.84.10 that the LPNM performs better
than the GMEA. Nonetheless, the graphs of |r ( )| associated to the GMEA are
in general closer to those of |rVM
( )| at the vicinities of the origin than the graphs
obtained by using the LPNM I and the LPNM II. This remarkable characteristic
suggests that the GMEA is better suited than those versions of the LPNM to
approximate the statistics of (t) that depend only on the value, slope, and
curvature of r ( ) at = 0 (e.g., the average Doppler shift, Doppler spread,
LCR, and ADF).
(2)
The resulting rms error, r , between r ( ) and rVM
( ) [see (4.3)] is shown
(2)

in Fig. 4.11. The error r is plotted there by considering N {10, 20, 30, 40, 50}.
To enable a fair comparison with the LPNM, we set max to N/(4fmax ). The
obtained results confirm that the RSAM is better suited than the other methods
to approximate rVM
( ). The results also show that the LPNM produces a smaller
error than the GMEA, as was to be expected. Among the three versions under
analysis, the LPNM II performed the best, although in some cases it registered a
larger error than the LPNM I. Interestingly, the LPNM III produced in general
worse results than the other two versions of the LPNM, specially when the number
of cisoids was small (N 30). This is due to the fact that the LPNM III has
to sacrifice some accuracy in the approximation of r ( ) to ensure that the
boundary conditions r ( ) = r ( ) and r ( ) = r ( ) are fulfilled. However,
such a sacrifice pays off when it comes to the emulation of the average Doppler
shift and Doppler spread of (t), as will be shown next.

66

Channel Simulation Models for Mobile Broadband Communications

(a)

(b)

10

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
=0

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
= 10

10

(2)

rms error, r

(2)

rms error, r

10

10

10

10

10

10
6

10

10

10

15

20

25

30

35

40

45

10

50

10

15

20

Number of cisoids, N

25

(c)

35

40

45

50

(d)

10

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
=5

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 30
= 10

10

(2)

(2)

rms error, r

10

rms error, r

30

Number of cisoids, N

10

10

10

10

10

10

15

20

25

30

35

40

45

10

50

10

15

Number of cisoids, N

20

25

30

35

40

45

50

Number of cisoids, N

(e)

(f)

10

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
= 20

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 90
= 10
1

(2)

rms error, r

(2)

rms error, r

10
2

10

10

10

10

10

10

15

20

25

30

35

Number of cisoids, N

40

45

50

10

10

15

20

25

30

35

40

45

50

Number of cisoids, N

Figure 4.11: Root mean square error function between the ACF of the channels diffuse component and the ACF of the SOC-based simulation
models random component by considering the von Mises PDF of
the AOA and the GMEA, RSAM, and LPNM (max = N/(4fmax )).

Chapter 4 Parameter Computation Methods

4.5.1.2

67

Average Doppler Shift and Doppler Spread

With respect to the emulation of the spectral characteristics of (t), we plot in

Fig. 4.12 the absolute error
A = |A A |

(4.41)

between the average Doppler shift of (t) [Eq. (2.49)] and the average Doppler
shift of
(t) [Eq. 3.17]. In addition, results obtained for the relative error
D =

|D D |
D

(4.42)

between the Doppler spreads of (t) [Eq. 2.50] and

(t) [Eq. 3.18] are shown in
Fig. 4.13. Again, we have considered N {10, 20, 30, 40, 50} to evaluate these
error functions.
The results presented in Figs. 4.12 and 4.13 indicate that the RSAM performs
better than the LPNM and GMEA regarding the emulation of A and D .
Furthermore, the obtained results confirm that GMEA is in general better suited
than the LPNM I and LPNM II to approximate the average Doppler shift and the
Doppler spread of (t). Interestingly, the performance of the RSAM is essentially
the same irrespective of the value of N , meaning that the number of cisoids exerts
little influence on the slope and curvature of r (0) when the parameters of
(t)
are computed by applying such a method. On the contrary, the results produced
by the LPNM and GMEA reveal that the slope and curvature of r (0) are highly
sensitive to the value of N when applying these methods.
We mentioned in Section 2.3.3 that the average Doppler shift of (t) is equal
to zero when the DPSD characterizing such a process is symmetrical with respect
to the origin. In this respect, it can be shown that the average Doppler shift of the
simulation models random component is equal to zero for both the RSAM and the
GMEA if S (f ) is symmetricalthe proof of this statement is a consequence of
the theorems presented in Appendix E. The results presented in Fig. 4.12 for the
two propagation scenarios associated with symmetrical DPSDs (cases: m = 0 ,
= 0; and m = 90 , = 10) provide evidence of the veracity of such a
(t)
claim. Since the observation of A = 0 indicates that the IQ components of
are uncorrelated [see Sec. 3.3.3.1 and 3.3.3.3], we can conclude that the GMEA
and RSAM are suitable methods for the design of SOC-based simulation models
for fading channel having diffuse components with uncorrelated IQ components.
Such a conclusion cannot be extrapolated to the case of the LPNM, because
this method produces in general an average Doppler shift that is different from
zero, as one can observe in Figs. 4.12(a) and 4.12(f). Not even the LPNM III,
which satisfies the boundary conditions r (0) = r (0) and r (0) = r (0), is

68

Channel Simulation Models for Mobile Broadband Communications

(a)

(b)
1

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
=0

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
= 10

Absolute error, A

Absolute error, A

10

10

10

10

10

10

10

15

10

10

15

20

25

30

35

40

45

10

50

10

15

Number of cisoids, N

20

25

(c)

40

45

50

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

10

10

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 30
= 10
Absolute error, A

m = 0
=5
Absolute error, A

35

(d)

10

10

30

Number of cisoids, N

10

10

15

20

25

30

35

40

45

10

50

10

15

Number of cisoids, N

20

25

30

35

40

45

50

Number of cisoids, N

(e)

(f)

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
= 20
0

10

10

Absolute error, A

Absolute error, A

10

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 90
= 10

10

10

10

10

15

10
3

10

10

15

20

25

30

35

Number of cisoids, N

40

45

50

10

15

20

25

30

35

40

45

50

Number of cisoids, N

Figure 4.12: Absolute error A between the average Doppler shift of the channels diffuse component and the average Doppler shift of the SOCbased simulation models random component by considering the
von Mises PDF of the AOA and the GMEA, RSAM, and LPNM.

Chapter 4 Parameter Computation Methods

69

(a)

(b)

10

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
=0

10

Relative error, D

Relative error, D

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
= 10

10

10

10

10

10

10

10

12

10

14

10

16

10

10

15

20

25

30

35

40

45

10

50

10

15

Number of cisoids, N

20

25

(c)

45

50

10

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 30
= 10
Relative error, D

Relative error, D

40

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
=5

10

10

15

20

25

30

35

40

45

10

50

10

15

Number of cisoids, N

20

25

30

35

40

45

50

Number of cisoids, N

(e)

(f)

10

10

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
= 20
1

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 90
= 10
1

Relative error, D

10

Relative error, D

35

(d)

10

10

30

Number of cisoids, N

10

10

10

10

10

10

10

10

15

20

25

30

35

Number of cisoids, N

40

45

50

10

10

15

20

25

30

35

40

45

50

Number of cisoids, N

Figure 4.13: Relative error D between the Doppler spread of the channels
diffuse component and the Doppler spread of the SOC-based simulation models random component by considering the von Mises
PDF of the AOA and the GMEA, RSAM, and LPNM.

70

Channel Simulation Models for Mobile Broadband Communications

able to guarantee the uncorrelation of the IQ components of (t) when S (f ) is

symmetrical. Actually, the LPNM III is the method that registered the highest
error A in the propagation scenarios that are related to fading channels having
uncorrelated IQ components.

4.5.2

Emulation of the Fading Statistics of the Reference Model

In the remaining of this section, we will evaluate the methods performance in

terms of the emulation of the fading statistics of the mobile Rician fading channel
model characterized by the random process (t) = (t) + m (t). To that end,
we will assume that the specular wave m (t) is parameterized in such a way that
the channels power equals 2 = 1, and cR {0, 2, 4}, where cR = 2 /2 is the
Rician factor. Besides, we will suppose that the Doppler frequency and the phase
of m (t) are equal to f = 65 Hz and = 0 , respectively.
4.5.2.1

PDF of the Envelope

Figures 4.144.18 show a comparison between the first-order PDF p (z) of the
envelope of (t) and the first-order PDF p(z) of the envelope of the SOC-based
simulation model. Curves of p(z) corresponding to the GMEA and LPNM I are
presented in Figs. 4.14 and 4.15, respectively, whilst those corresponding to the
LPNM II, LPNM III, and RSAM are shown in Figs. 4.164.18, in that order.
Such curves were generated by considering N = 10.
The above mentioned figures present both analytical and empirical graphs of
p(z). The analytical graphs were generated by numerically evaluating (3.20) with
R
the aid of the MATLAB
trapz function. For the particular case of the LPNM I
and the GMEA, we notice that (3.20) results always in the same theoretical PDF
p(z) regardless of the simulated propagation scenario. This is due to the fact
that (3.20) is solely influenced by the set of gains {
cn , }, and these two methods
define a unique set of gains cn for each value of N irrespective of the channel
statistics [cf. (4.1)]. In contrast, the solution of (3.20) varies in the case of the
LPNM II, LPNM III, and RSAM from one propagation scenario to another, as
these methods specify the gains cn by taking into account the distribution of the
AOA or the ACF of (t).
On the other hand, the empirical graphs of p(z) were obtained by measuring
= |
the distribution of 50 106 samples of (t)
(t)|2 . Such samples were collected
at the same time instant t = ti , where the value of ti was chosen at random
over a time interval of two hours. We can see in the figures that the analytical
graphs of p(z) are in excellent agreement with the empirical ones, so that one
can presume the correctness of (3.20). In fact, the thesis implied by (3.20) about
p(z) being a function that depends only on the set of gains {
cn , } is supported
by the empirical data obtained for the LPNM I and GMEA. Despite the fact

Chapter 4 Parameter Computation Methods

cR = 4

1.2

cR = 2

Envelope PDF

cR = 0
0.8

71
Reference model, p (z)
Simulation, theory: All scenarios
Simulation, measured: m = 0 , = 0
Simulation, measured: m = 0 , = 5
Simulation, measured: m = 0 , = 20
Simulation, measured: m = 0 , = 10
Simulation, measured: m = 30 , = 10
Simulation, measured: m = 90 , = 10

0.6

0.4

GMEA

0.2

0
0

0.5

1.5

2.5

Envelope level

Figure 4.14: Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOCbased simulation model by applying the GMEA to the von Mises
PDF of the AOA with different pairs of parameters m and
(fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 , and N = 10.
cR = 4

1.2

cR = 2

Envelope PDF

cR = 0
0.8

Reference model, p (z)

Simulation, theory: All scenarios
Simulation, measured: m = 0 , = 0
Simulation, measured: m = 0 , = 5
Simulation, measured: m = 0 , = 20
Simulation, measured: m = 0 , = 10
Simulation, measured: m = 30 , = 10
Simulation, measured: m = 90 , = 10

0.6

0.4

LPNM I:
p=2
max = 4fN
max

0.2

0
0

0.5

1.5

2.5

Envelope level

Figure 4.15: Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOCbased simulation model by applying the LPNM I to the von Mises
PDF of the AOA with different pairs of parameters m and
(fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 , and N = 10).
that these methods produce different sets of Doppler frequencies {fn }N
n=1 for

each scenario, the resulting empirical PDFs of (t) converged in all cases to the
same curve of p(z). What is more, the empirical graphs of p(z) obtained by
using the LPNM I and GMEA are identical to each other, as one can observe by
comparing Figs. 4.15 and 4.14. This was to be expected, since these two methods

72

Channel Simulation Models for Mobile Broadband Communications

(a)

1.2

m = 0

cR = 2

Envelope PDF

Reference model, p (z)

Simulation, theory: = 0
Simulation, measured: = 0
Simulation, theory: = 5
Simulation, measured: = 5
Simulation, theory: = 20
Simulation, measured: = 20

cR = 4

cR = 0
0.8

0.6

LPNM II:
p=2
max = 4fN
max
W1 = 1/4
W2 = 3/4

0.4

0.2

0
0

0.5

1.5

2.5

Envelope level

(b)

1.2

= 10

cR = 2

Envelope PDF

Reference model, p (z)

Simulation, theory: m = 0
Simulation, measured: m = 0
Simulation, theory: m = 30
Simulation, measured: m = 30
Simulation, theory: m = 90
Simulation, measured: m = 90

cR = 4

cR = 0
0.8

0.6

LPNM II:
p=2
max = 4fN
max
W1 = 1/4
W2 = 3/4

0.4

0.2

0
0

0.5

1.5

2.5

Envelope level

Figure 4.16: Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOCbased simulation model by applying the LPNM II to the von Mises
PDF of the AOA with different pairs of parameters m and
(fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 , and N = 10).
specify the gains cn in the same way.
The graphs depicted in Figs. 4.144.18 show a good concordance between
the envelope distributions of the reference model and simulation model. It is
worth mentioning that the curves of p(z) plotted in Fig. 4.18 for the RSAM,
specifically those corresponding to the case cR = 0, m = 0 , and {10, 20}, are
significantly better than the graphs shown in Fig. 4.5 for the BRSAM. However,
despite of such an improvement, a close inspection of Figs. 4.144.18 reveals
that the RSAM is less accurate than the LPNM and the GMEA regarding the
emulation of p (z).

Chapter 4 Parameter Computation Methods

73

(a)

1.2

m = 0

cR = 2

Envelope PDF

Reference model, p (z)

Simulation, theory: = 0
Simulation, measured: = 0
Simulation, theory: = 5
Simulation, measured: = 5
Simulation, theory: = 20
Simulation, measured: = 20

cR = 4

cR = 0
0.8

0.6

LPNM III:
p=2
max = 4fN
max
W1 = 1/4
W2 = 3/4

0.4

0.2

0
0

0.5

1.5

2.5

Envelope level

(b)

1.2

= 10

cR = 2

Envelope PDF

Reference model, p (z)

Simulation, theory: m = 0
Simulation, measured: m = 0
Simulation, theory: m = 30
Simulation, measured: m = 30
Simulation, theory: m = 90
Simulation, measured: m = 90

cR = 4

cR = 0
0.8

0.6

LPNM III:
p=2
max = 4fN
max
W1 = 1/4
W2 = 3/4

0.4

0.2

0
0

0.5

1.5

2.5

Envelope level

Figure 4.17: Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOC-based
simulation model by applying the LPNM III to the von Mises
PDF of the AOA with different pairs of parameters m and
(fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 , and N = 10).

In order to quantitatively measure the methods performance, we present in

Fig. 4.19 the resulting rms error p between p (z) and p(z) [see (4.8)], computed
for each method by considering cR = 0 and N = {10, 20, 30, 40, 50}. The results
presented in Fig. 4.19 leave no doubt that the LPNM I and GMEA are better
suited than the other methods to approximate the envelope distribution of the
reference model when the scattering conditions are non-isotropic. In turn, the
results show that the LPNM II performs better than the rest of the methods
under isotropic scattering conditions. On the other hand, the figure shows that
the highest error was produced by the RSAM. Even though the rms error p

74

Channel Simulation Models for Mobile Broadband Communications

(a)

1.2

m = 0

cR = 2

Envelope PDF

Reference model, p (z)

Simulation, theory: = 0
Simulation, measured: = 0
Simulation, theory: = 5
Simulation, measured: = 5
Simulation, theory: = 20
Simulation, measured: = 20

cR = 4

cR = 0
0.8

0.6

RSAM:
= 1 103

0.4

0.2

0
0

0.5

1.5

2.5

Envelope level

(b)

1.2

= 10

cR = 2

Envelope PDF

Reference model, p (z)

Simulation, theory: m = 0
Simulation, measured: m = 0
Simulation, theory: m = 30
Simulation, measured: m = 30
Simulation, theory: m = 90
Simulation, measured: m = 90

cR = 4

cR = 0
0.8

0.6

RSAM:
= 1 103

0.4

0.2

0
0

0.5

1.5

2.5

Envelope level

Figure 4.18: Comparison between the first-order PDF of the reference models
envelope and the first-order PDF of the envelope of the SOCbased simulation model by applying the RSAM to the von Mises
PDF of the AOA with different pairs of parameters m and
(fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 , and N = 10.
registered by such a method is rather small, it is around two to three times higher
than the one produced by the GMEA/LPNM I. With respect to the LPNM II
and LPNM III, it can be observed that only the former method provides a real
advantage over the RSAM, since the LPNM III and the RSAM perform basically
the same when the number of cisoids is higher than thirty.
4.5.2.2

PDF of the Phase

Concerning the emulation of the first-order PDF of the reference models phase,
p (; t), we present in Figs. 4.204.24 a comparison between p (; t) and the

Chapter 4 Parameter Computation Methods

75

(a)
2.5

(b)
LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
=0

(2)

(2)

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
= 10

rms error, p 102

rms error, p 102

1.5

0.5

0
10

15

20

25

30

35

40

45

0
10

50

15

Number of cisoids, N

20

25

(c)

40

4.5

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
=5

45

50

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 30
= 10

rms error, p 102

rms error, p 102

3.5

(2)

(2)

35

(d)

30

Number of cisoids, N

2.5
2

1.5

1
1

0
10

0.5

15

20

25

30

35

40

45

0
10

50

15

Number of cisoids, N

20

25

(e)

40

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 0
= 20

45

50

rms error, p 102

rms error, p 102

LPNM I
LPNM II
LPNM III
GMEA
RSAM

m = 90
= 10

(2)

(2)

0
10

35

(f)

30

Number of cisoids, N

15

20

25

30

35

40

Number of cisoids, N

45

50

0
10

15

20

25

30

35

40

45

50

Number of cisoids, N

Figure 4.19: Root mean square error between the envelope distributions of the
channels diffuse component and SOC-based simulation models
random component by considering the von Mises PDF of the AOA
and the GMEA, RSAM, and LPNM (max = N/(4fmax )).

76

Channel Simulation Models for Mobile Broadband Communications

1.4

Reference model, p (; t)
Simulation, theory: All scenarios
Simulation, measured: m = 0 , = 0
Simulation, measured: m = 0 , = 5
Simulation, measured: m = 0 , = 20
Simulation, measured: m = 0 , = 10
Simulation, measured: m = 30 , = 10
Simulation, measured: m = 90 , = 10

1.2

Phase PDF

0.8

cR = 4
t=0

cR = 4
t = 0.25

cR = 2
t=0

cR = 2
t = 0.25

GMEA

0.6

cR = 0
t {0, 0.25}

0.4

0.2

Phase (radians)

Figure 4.20: Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based
simulation model by applying the GMEA to the von Mises PDF
of the AOA (fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 , t
{0, 0.25}, and N = 10).
1.4

Reference model, p (; t)
Simulation, theory: All scenarios
Simulation, measured: m = 0 , = 0
Simulation, measured: m = 0 , = 5
Simulation, measured: m = 0 , = 20
Simulation, measured: m = 0 , = 10
Simulation, measured: m = 30 , = 10
Simulation, measured: m = 90 , = 10

1.2

Phase PDF

0.8

cR = 4
t=0

cR = 4
t = 0.25

cR = 2
t=0

cR = 2
t = 0.25
LPNM I:
p=2
max = 4fN
max

0.6

cR = 0
t {0, 0.25}

0.4

0.2

Phase (radians)

Figure 4.21: Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based
simulation model by applying the LPNM I to the von Mises PDF
of the AOA (fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 , t
{0, 0.25}, and N = 10).
first-order PDF of the simulation models phase, p (; t). The graphs of p (; t)
presented in those figures were computed both analytically and empirically for
the methods under investigation by considering N = 10 and t {0, 0.25}. Figures 4.20 and 4.21 show results obtained for the GMEA and LPNM I, in that
order, while results obtained for the LPNM II, LPNM III, and RSAM are pre-

Chapter 4 Parameter Computation Methods

77

(a)
1.2

Reference model, p (; t)
Simulation, theory: = 0
Simulation, measured: = 0
Simulation, theory: = 5
Simulation, measured: = 5
Simulation, theory: = 20
Simulation, measured: = 20

Phase PDF

0.8

cR = 4
t=0

cR = 4
t = 0.25

cR = 2
t=0

cR = 2
t = 0.25

m = 0

0.6

LPNM II:
p=2
max = 4fN
max
W1 = 1/4
W2 = 3/4

cR = 0
t {0, 0.25}

0.4

0.2

Phase (radians)

(b)
1.2

Reference model, p (; t)
Simulation, theory: m = 0
Simulation, measured: m = 0
Simulation, theory: m = 30
Simulation, measured: m = 30
Simulation, theory: m = 90
Simulation, measured: m = 90

Phase PDF

0.8

cR = 4
t=0

cR = 4
t = 0.25

cR = 2
t=0

cR = 2
t = 0.25

= 10

0.6

LPNM II:
p=2
max = 4fN
max
W1 = 1/4
W2 = 3/4

cR = 0
t {0, 0.25}

0.4

0.2

Phase (radians)

Figure 4.22: Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based
simulation model by applying the LPNM II to the von Mises PDF
of the AOA (fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 , t
{0, 0.25}, and N = 10).
sented in Figs. 4.224.24, respectively. For the analytical graphs, we evaluated
(3.22) with the aid of the MATLABs trapz function. For the empirical graphs,
we made the same considerations and followed the same procedure as described
in Section 4.5.2.1 concerning the computation of the empirical envelope distribution. The figures graphically demonstrate the correctness of the analytical
solution given in (3.22) for the PDF of the simulation models phase.
The results presented in Figs. 4.204.24 show that the phase of the simulation
model is in general not a first-order stationary process, since p (; 0) 6= p (; 0.25)
for cR {2, 4}. However, p (; t) proves to be time-invariant and equal to p () =

78

Channel Simulation Models for Mobile Broadband Communications

(a)
1.2

Reference model, p (; t)
Simulation, theory: = 0
Simulation, measured: = 0
Simulation, theory: = 5
Simulation, measured: = 5
Simulation, theory: = 20
Simulation, measured: = 20

Phase PDF

0.8

cR = 4
t=0

cR = 4
t = 0.25

cR = 2
t=0

cR = 2
t = 0.25

m = 0

0.6

LPNM III:
max = 4fN
max
p=2
W1 = 1/4
W2 = 3/4

cR = 0
t {0, 0.25}

0.4

0.2

Phase (radians)

(b)
1.2

Reference model, p (; t)
Simulation, theory: m = 0
Simulation, measured: m = 0
Simulation, theory: m = 30
Simulation, measured: m = 30
Simulation, theory: m = 90
Simulation, measured: m = 90

Phase PDF

0.8

cR = 4
t=0

cR = 4
t = 0.25

cR = 2
t=0

cR = 2
t = 0.25

= 10

0.6

LPNM III:
max = 4fN
max
p=2
W1 = 1/4
W2 = 3/4

cR = 0
t {0, 0.25}

0.4

0.2

Phase (radians)

Figure 4.23: Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based
simulation model by applying the LPNM III to the von Mises
PDF of the AOA (fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 ,
t {0, 0.25}, and N = 10).

1/(2) irrespective of the parameter computation method when cR = 0. These

remarks are in line with the statements made in that respect in Section 3.3.4.2.
With regard to the methods performance, one may observe from Figs. 4.21
4.24 that the five methods provide a close approximation to the phase PDF of the
reference model. Indeed, all methods render an exact match of p (; t) for the
Rayleigh fading scenario (cR = 0). However, for the Rician fading case (cR > 0),
the obtained results show clearly that the RSAM and LPNM III are less effective
than the other methods.

Chapter 4 Parameter Computation Methods

79

(a)
1.2

Reference model, p (; t)
Simulation, theory: = 0
Simulation, measured: = 0
Simulation, theory: = 5
Simulation, measured: = 5
Simulation, theory: = 20
Simulation, measured: = 20

Phase PDF

0.8

cR = 4
t=0

cR = 4
t = 0.25

cR = 2
t=0

cR = 2
t = 0.25

m = 0

0.6

RSAM:
= 1 103

cR = 0
t {0, 0.25}

0.4

0.2

Phase (radians)

(b)
1.2

Reference model, p (; t)
Simulation, theory: m = 0
Simulation, measured: m = 0
Simulation, theory: m = 30
Simulation, measured: m = 30
Simulation, theory: m = 90
Simulation, measured: m = 0

Phase PDF

0.8

cR = 4
t=0

cR = 4
t = 0.25

cR = 2
t=0

cR = 2
t = 0.25

= 10

0.6

RSAM:
= 1 103

cR = 0
t {0, 0.25}

0.4

0.2

Phase (radians)

Figure 4.24: Comparison between the first-order PDF of the reference models
phase and the first-order PDF of the phase of the SOC-based
simulation model by applying the RSAM to the von Mises PDF
of the AOA with different pairs of parameters m and (fmax =
91 Hz, 2 = 1, fp = 65 Hz, = 0 , t {0, 0.25}, and N = 10).
4.5.2.3

ACF of the Squared Envelope

To close this section, we present in Figs. 4.254.33 a comparison between the

ACF of the reference models squared envelope, r 2 2 ( ), and the ACF of the
squared envelope of the simulation model, r22 ( ), by considering N = 20 and
cR {0, 2, 4}. To keep the figures readable, we have plotted the results produced
by each versions of the LPNM against results obtained for the GMEA and RSAM
in different figures. The figures show both theoretical and empirical curves of
the ACF of 2 (t) = |
(t)|2 . The theoretical graphs of r22 ( ) were produced by

80

Channel Simulation Models for Mobile Broadband Communications

evaluating (3.29), while the empirical curves were generated by averaging over the
measured ACFs of 60 samples functions of 2 (t). Such an averaging was necessary,
since the process 2 (t) is not always autocorrelation ergodic3 . However, with the
purpose of demonstrating the correctness of the analytical solutions provided
in Section 3.4.4 for the time-averaged ACF r2 2 ( ) of the sample functions of
2 (t), we present results obtained for r2 2 ( ) in Appendix F. We pointed out in
P
Section 4.5.1.1 that the LPNM II does not satisfy the condition N
2n = 2 .
n=1 c
For this reason, the theoretical ACF of 2 (t) was computed for that particular
method by evaluating

c2R 4
42 cR
+
r I I ( ) cos (2f )
(cR + 1)2 cR + 1

N
1X 2
(4.43)
c
+r I Q ( ) sin (2f )
2 n=1 n

r22 ( ) = r2 2 ( ) +

where
2

r2 2 ( ) = |r ( )| +

"

N
X

n=1

c2n

#2

N
X

c4n .

(4.44)

n=1

One can observe from Figs. 4.254.33 that the results predicted by the theory
are in excellent agreement with the ones obtained in practice. This can be taken
as an indicator of the correctness of the analytical expressions introduced in
Section 3.3.6 for r22 ( ).
Figures 4.284.33 show that the three methods under consideration yield a
very good approximation to the graphs of r 2 2 ( ) when cR > 0. This is not
surprising, since under such circumstances the correlation properties of 2 (t) (and
also of 2 (t)) are mainly determined by the time-variant specular wave m (t). The
characteristics of the parameter computation methods exert in consequence little
influence on the quality of the relationship r22 ( ) r 2 2 ( ). However, for the
NLOS scenario (cR = 0), it is clear that the methods characteristics determine
the capability of the simulation model for emulating the function r 2 2 ( ). In
this respect, we can see in Figs. 4.254.27 that the best fitting of the ACF of
2 (t) is produced by the GMEA and the LPNM I, while the LPNM II yields the
worst. The LPNM III and the RSAM perform similarly, although the LPNM III
is slightly more accurate.
An inspection of the results presented in Figs. 4.254.33 reveals that, irrespective of the parameter computation method, there exists an offset between
We pointed out in Section 3.5 that 2 (t) is non-autocorrelation-ergodic when the IQ components of
(t) are uncorrelated and the Doppler frequencies fn satisfy the Conditions 3.2 and 3.3.
Such a situation ensues in the case of the GMEA and RSAM when the DPSD of the channels
diffuse component is symmetrical [Appx. E] (cases: m = 0 , = 0; and m = 90 , = 10).
3

Chapter 4 Parameter Computation Methods

81

(a)

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM I, theory
LPNM I, realization
RSAM, theory
RSAM, realization

m = 0
cR = 0
= 20

1.8

1.6

=5

1.4

max =

N
4fmax

1.2

=0
1

Normalized time difference, fmax

(b)

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM I, theory
LPNM I, realization
RSAM, theory
RSAM, realization

= 10
cR = 0

1.8

m = 0

1.6

max =

N
4fmax

m = 30

1.4

1.2

m = 90
0.8
0

Normalized time difference, fmax

Figure 4.25: Comparison among the GMEA, RSAM, and LPNM I in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 0 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

82

Channel Simulation Models for Mobile Broadband Communications

(a)

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM II, theory
LPNM II, realization
RSAM, theory
RSAM, realization

m = 0
cR = 0
= 20

1.8

1.6

=5

1.4

max =

N
4fmax

1.2

=0
1

Normalized time difference, fmax

(b)

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM II, theory
LPNM II, realization
RSAM, theory
RSAM, realization

= 10
cR = 0

1.8

m = 0

1.6

max =

N
4fmax

m = 30

1.4

1.2

m = 90
0.8
0

Normalized time difference, fmax

Figure 4.26: Comparison among the GMEA, RSAM, and LPNM II in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 0 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

Chapter 4 Parameter Computation Methods

83

(a)

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM III, theory
LPNM III, realization
RSAM, theory
RSAM, realization

m = 0
cR = 0
= 20

1.8

1.6

=5

1.4

max =

N
4fmax

1.2

=0
1

Normalized time difference, fmax

(b)

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM III, theory
LPNM III, realization
RSAM, theory
RSAM, realization

= 10
cR = 0

1.8

m = 0

1.6

max =

N
4fmax

m = 30

1.4

1.2

m = 90
0.8
0

Normalized time difference, fmax

Figure 4.27: Comparison among the GMEA, RSAM, and LPNM III in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 0 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

84

Channel Simulation Models for Mobile Broadband Communications

(a)
1.6

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM I, theory
LPNM I, realization
RSAM, theory
RSAM, realization

m = 0
cR = 2

1.5

ACF of the squared envelope

= 20
1.4

1.3

=5
1.2

1.1

0.9

0.8

=0

max =

N
4fmax

0.7
0

Normalized time difference, fmax

(b)
1.6

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM I, theory
LPNM I, realization
RSAM, theory
RSAM, realization

= 10
cR = 2

ACF of the squared envelope

1.5

1.4

m = 0
1.3

1.2

m = 30

1.1

0.9

0.8

0.7
0

max =

m = 90
1

N
4fmax

Normalized time difference, fmax

Figure 4.28: Comparison among the GMEA, RSAM, and LPNM I in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 2 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

Chapter 4 Parameter Computation Methods

85

(a)
1.6

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM II, theory
LPNM II, realization
RSAM, theory
RSAM, realization

m = 0
cR = 2

1.5

ACF of the squared envelope

= 20
1.4

1.3

=5
1.2

1.1

0.9

0.8

=0

max =

N
4fmax

0.7
0

Normalized time difference, fmax

(b)
1.6

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM II, theory
LPNM II, realization
RSAM, theory
RSAM, realization

= 10
cR = 2

ACF of the squared envelope

1.5

1.4

m = 0
1.3

1.2

m = 30

1.1

0.9

0.8

0.7
0

max =

m = 90
1

N
4fmax

Normalized time difference, fmax

Figure 4.29: Comparison among the GMEA, RSAM, and LPNM II in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 2 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

86

Channel Simulation Models for Mobile Broadband Communications

(a)
1.6

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM III, theory
LPNM III, realization
RSAM, theory
RSAM, realization

m = 0
cR = 2

1.5

ACF of the squared envelope

= 20
1.4

1.3

=5
1.2

1.1

0.9

0.8

=0

max =

N
4fmax

0.7
0

Normalized time difference, fmax

(b)
1.6

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM III, theory
LPNM III, realization
RSAM, theory
RSAM, realization

= 10
cR = 2

ACF of the squared envelope

1.5

1.4

m = 0
1.3

1.2

m = 30

1.1

0.9

0.8

0.7
0

max =

m = 90
1

N
4fmax

Normalized time difference, fmax

Figure 4.30: Comparison among the GMEA, RSAM, and LPNM III in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 2 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

Chapter 4 Parameter Computation Methods

87

(a)

= 20

1.3

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM I, theory
LPNM I, realization
RSAM, theory
RSAM, realization

m = 0
cR = 4

1.4

1.2

=5
1.1

0.9

0.8

=0

max =

0.7
0

N
4fmax

Normalized time difference, fmax

(b)

ACF of the squared envelope

1.3

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM I, theory
LPNM I, realization
RSAM, theory
RSAM, realization

= 10
cR = 4
m = 0

1.2

1.1

m = 30

0.9

0.8

max =

N
4fmax

m = 90

0.7
0

Normalized time difference, fmax

Figure 4.31: Comparison among the GMEA, RSAM, and LPNM I in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 4 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

88

Channel Simulation Models for Mobile Broadband Communications

(a)

= 20

1.3

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM II, theory
LPNM II, realization
RSAM, theory
RSAM, realization

m = 0
cR = 4

1.4

1.2

=5
1.1

0.9

0.8

=0

max =

0.7
0

N
4fmax

Normalized time difference, fmax

(b)

ACF of the squared envelope

1.3

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM II, theory
LPNM II, realization
RSAM, theory
RSAM, realization

= 10
cR = 4
m = 0

1.2

1.1

m = 30

0.9

0.8

max =

N
4fmax

m = 90

0.7
0

Normalized time difference, fmax

Figure 4.32: Comparison among the GMEA, RSAM, and LPNM II in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 4 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

Chapter 4 Parameter Computation Methods

89

(a)

= 20

1.3

ACF of the squared envelope

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM III, theory
LPNM III, realization
RSAM, theory
RSAM, realization

m = 0
cR = 4

1.4

1.2

=5
1.1

0.9

0.8

=0

max =

0.7
0

N
4fmax

Normalized time difference, fmax

(b)

ACF of the squared envelope

1.3

Reference model, r 2 2 ()
GMEA, theory
GMEA, realization
LPNM III, theory
LPNM III, realization
RSAM, theory
RSAM, realization

= 10
cR = 4
m = 0

1.2

1.1

m = 30

0.9

0.8

max =

N
4fmax

m = 90

0.7
0

Normalized time difference, fmax

Figure 4.33: Comparison among the GMEA, RSAM, and LPNM III in terms of
the emulation of the ACF of the reference models squared envelope by considering a Rician factor cR = 4 and the von Mises PDF
of the AOA with different pairs of parameters m and (N = 20,
2 = 1, p = 2, max = 5/fmax , = 1 103 ).

90

Channel Simulation Models for Mobile Broadband Communications

the ACFs of 2 (t) and 2 (t). Such an offset, which can clearly be distinguished
PN 4
at = 0, is caused by the term
n affecting the ACF r22 ( ) of the
n=1 c
2
2
(t)| of the simulation models random component
squared envelope (t) = |
[see (3.30)]. It is evident from Figs. 4.254.27 that the smallest offset is produced
by the GMEA and LPNM I, and the largest by the LPNM II.

4.6

Chapter Summary and Conclusions

In this chapter, we presented two simple and effective parameter computation

method for the design of SOC-based channel simulation models. The proposed
methods, which we have named the GMEA and the RSAM, are intended to
allow for an accurate emulation of the channels correlation properties and fading
statistics. We evaluated the performance of the proposed methods in terms of
the approximation of the ACF, average Doppler shift, and Doppler spread of the
channels diffuse component. In addition, we evaluated the methods performance
with respect to the emulation of the envelope and phase distributions of a Rician
fading channel model, as well as in terms of the approximation of the squared
envelope ACF. Results obtained by applying three different versions of the LPNM
were also reported in this section and compared with those produced by the
GMEA and RSAM.
All in all, the results presented in Section 4.5 show that the GMEA and
the RSAM are suitable methods for the simulation of mobile fading channels
characterized by symmetrical and asymmetrical DPSDs. However, our investigations indicate that the RSAM is better suited than the GMEA, and also than the
LPNM, to emulate the channels correlation and spectral characteristics. In turn,
the GMEA and the LPNM perform better than the RSAM in terms of the emulation of the channels fading statistics. The three version of the LPNM produce
a better approximation to the channel ACF than the GMEA, but the GMEA
proves to be more accurate than the LPNM I and LPNM II in reproducing the
channels average Doppler shift and Doppler spread.
Owing to their flexibility and good performance, the methods herein proposed
can easily be used to design efficient simulation platforms for the software-assisted
analysis of MBWCS, as we will demonstrate in Chapters 5 and 6.

Chapter 5

Simulation of Mobile MIMO

Fading Channels Under
Single-Bounce Non-Isotropic
Scattering Conditions
5.1

Introduction

Simulation models having the ability to reproduce the statistical properties of

mobile MIMO fading channels in non-isotropic scattering environments are highly
desirable for the laboratory analysis of modern wireless communication systems.
They are important, for example, to study the channel capacity [Abd02a] and
the system bit error rate performance [Miti04] under directional propagation
conditions.
It has been shown in a number of papers, e.g., [Gut07a, Han02, Pae04a,
Paet05], that the simulation of MIMO fading channels can efficiently be performed by means of a SOC model. In fact, investigation carried out in [Mayy07]
and [Pae04a] demonstrate, respectively, that a SOC channel simulator is a very
useful tool for the performance assessment of multi-element antenna OFDM systems and for the analysis of the second order statistics of the channel capacity.
However, as mentioned in the preceding chapter, the LPNM is currently the
only method available for the design of SOC simulators for fading channels under
non-isotropic scattering conditions. The development of new methods is therefore
necessary to bypass the computational burden of the LPNM and facilitate the
analysis of MIMO communication systems operating in non-isotropic scattering
environments.
In this chapter, we show that the GMEA and the RSAM introduced in Chapter 4 can be applied, after some simple modifications, to the simulation of MIMO
91

92

Channel Simulation Models for Mobile Broadband Communications

channels with specified space-time correlation properties. Specifically, we address here the design of SOC simulation models for narrowband mobile MIMO
Rayleigh fading channels in single-bounce scattering environments. It is important to notice that the GMEA and the RSAM cannot directly be applied as they
were presented in Chapter 4 to the case of MIMO channels, since an accurate approximation of the channel spatial CCF (SCCF) requires special considerations.
The outline to the rest of the chapter is as follows. In Section 5.2, we describe
the characteristics of a geometrical narrowband single-bounce scattering mobile
MIMO channel model that will act as a reference model throughout this chapter.
We also provide in that section a brief review of the autocorrelation, spectral,
and spatial cross-correlation properties of such a channel model. In Section 5.3,
we give a concise description of the stochastic SOC model that we will employ as
a basis for the simulation of the reference model. The extension of the GMEA
and the RSAM to the case of MIMO channels is presented in Section 5.4. The
methods performance is evaluated in Section 5.5 with respect to the emulation
of the ACF and the SCCF of the reference model. An application of the GMEA
and the RSAM to the simulation of wideband MIMO channels is discussed in
Section 5.6. Finally, Section 5.7 summarizes the main points of the chapter.

5.2
5.2.1

Reference Model
The Geometrical One-Ring Scattering Model

The frequency-non-selective mobile MIMO channel model considered in this chapter is an extension of the SISO channel model described in Chapter 2 with respect to an elementary 22 multi-element antenna system and a two-dimensional
single-bounce scattering environment. We assume that the transmitter (BS) is
fixed and unobstructed by local scatterers, whereas the receiver (MT) moves at
a speed v and is surrounded by a ring of local scatterers. The geometrical configuration of such a propagation scenario is depicted in Fig. 5.1. The symbols
introduced in that figure stand for:
Tm : the mth transmitter antenna;
Rk : the kth receiver antenna;
Sn : the nth scatterer;
OT : the center of the transmitter antenna array;
OR : the center of the receiver antenna array;
D: the distance from OT to OR .
R: the radius of the ring of scatterers;
T : the distance between the transmitter antennas;

Chapter 5 SOC Simulation of MIMO Channels

93
y

T1
b

d T1,n
T,n

T
T,n

Sn

dR
n,1

R1

R,n

T,max

OT

~v

OR

dR
n,2
d T2,n

R
b

R2

T2

Figure 5.1: The geometrical one-ring scattering propagation model

R : the distance between the receiver antennas;

d Tm,n : the distance from the mth transmitter antenna to the nth scatter;
dR
n,k : the distance from the nth scatter to the kth receiver antenna;
T,n : the distance from OT to the nth scatter;
R,n : the AOA of the nth scattered wave;
T,n : the angle of departure (AOD) of the transmitted wave impinging the
nth scatterer;
T,max : one half of the maximum angle of spread seen at the BS;
T : the orientation angle of the transmitter antenna array;
R : the orientation angle of the receiver antenna array;
: the angle of the vector ~v pointing at the direction of motion of the receiver.
The geometrical one-ring scattering model shown in 5.1 was originally proposed
in [Jake74, Sec. 1.6.1] to study the channel spatial correlation at the base station,
and it was extended in [Shiu00] to the case of MIMO channels. The model has
widely been in use as a basis for assessing the performance of multi-element
antenna systems, e.g., [Chen00, Mayy07].
We will assume that the AOAs R,n are i.i.d. random variables characterized
T
by a circular PDF pR (), whereas T,n , d R
n,k , d m,n , and T,n can be defined as
functions of R,n . Specifically, T,n and T,n can be determined by solving the
following equations in the triangle OT Sn OR [Abd02a, Chen00]
T,n
D
R
=
=
R,n T,n )
T,n )
R,n )
sin(
sin(
sin(

(5.1)

94

Channel Simulation Models for Mobile Broadband Communications

T
while the distances d R
n,k and d m,n are defined as

d Tm,n =
dR
n,k =

T
+ 2T,n + (1)m T T,n cos(T T,n )
4
R
R,n R ).
+ R2 + (1)m R R cos(
4

(5.2)
(5.3)

Notice that the dependence of d Tm,n on R,n is implicit in the random parameters
T,n and T,n . We will suppose that the far-field assumption holds, meaning that
D R sup{T , R } and max,T is small. Thereby, d Tm,n and d R
n,k can be
rewritten as [Abd02a, Chen00]

5.2.2

(1)m T
R,n )
[cos(T ) + T,max sin(T ) sin(
2
(1)m R
R,n R ).
= R+
cos(
2

d Tm,n = T,n +

(5.4)

dR
n,k

(5.5)

Narrowband MIMO Rayleigh Fading Channel Model

Following the modeling approach described in Section 2.2, we characterize the

channel gain associated to the link between the mth transmitter antenna and the
kth receiver antenna in the equivalent complex baseband by a stationary complex Gaussian process km (t) with mean zero and variance 2 km . For simplicity,
we assume that 2 km = 2 m, k, where 0 < 2 < . The random process
km (t) can be represented, in the context of the single-bounce scattering model
in Fig. 5.1, by means of a sum of an infinite number of azimuthal plane waves as
follows [Abd02a, Pae04a]
km (t) =

lim

N
X

n=1

am,n bk,n cn exp {j (2ff n t + n )}

(5.6)

R,n ) and b k,n , bk (

R,n ), with
where a m,n , am (



j(1)m+1 T
R,n ) = exp
R,n )
am (
[cos(T ) + T,max sin(T ) sin(
(5.7)



j(1)k+1 R
R,n ) = exp
R,n R ) .
bk (
cos(
(5.8)

In (5.7) and (5.8), designates the wavelength of the transmitted signal. In

conformity with the considerations made in Section 2.2, we assume that the
random gains cn , the random Doppler frequencies f n , and the random phases n
are statistically independent. The phases n are i.i.d. random variables uniformly
distributed in [, ), and the gains c n satisfy E{|ccn |2 } = 2 /N . On the other

Chapter 5 SOC Simulation of MIMO Channels

95

hand, the Doppler frequencies f n are defined as

f n , fmax cos R,n ,

n.

(5.9)

The equation above is a consequence of the fact that the MT (receiver) can move
in any arbitrary direction over the reference Cartesian plane. This is in contrast
to the case of the mobile SISO channel model described in Chapter 2, where it
is assumed that the MT is moving along the x-axis of the Cartesian plane, and
thus {0, } [Clar68]. The angle is an important parameter of the MIMO
channel model. Indeed, exerts a strong influence on the autocorrelation and
spectral properties of km (t), as will be shown next.

5.2.3
5.2.3.1

Correlation Properties of the Reference Model

ACF and DPSD of the MIMO Channel Gains

Under the above considerations, one can easily verify that the ACF r km km ( ) ,
km (t + )} of the Gaussian process km (t) describing the complex
km (t)
E{
channel gain between Tm and Rk is given by
rkm km ( ) =



pR () exp j2fmax cos( ) d.

(5.10)

For the analysis of the spectral characteristics of km (t), it is convenient to perform the change of variable x = in (5.10). By doing so, and taking into
account that pR is a circular PDF, we obtain the expression
rkmkm ( ) =



pR (x + ) exp j2fmax cos(x) dx

(5.11)



g R ( + ) exp j2fmax cos() d

(5.12)

which can be rewritten as follows

r km km ( ) =

22

Z
0

where g R () = [pR () + p R ()]/2 is the even part of p R (). From (5.12), it

R

is straightforward to show that the DPSD S km km (f ) , r km km ( ) exp

j2f d of km (t) is equal to
Skm km (f ) = 22 rect(f )

g R (arccos(f /fmax ) + )
p
.
fmax 1 (f /fmax )2

(5.13)

96

Channel Simulation Models for Mobile Broadband Communications

It follows from (5.12) and (5.13) that the ACF and DPSD of km (t) are completely specified by the even part of pR () and the parameters , fmax , and 2 .
We observe that Skmkm (f ) is symmetrical if the IQ components I,km(t) and
Q,km(t) of km (t) are uncorrelated. On the contrary, Skm km (f ) is asymmetrical
if I,km (t) and Q,km(t) are cross-correlated. In this respect, it is worth mentioning that the shape of Skm km (f ) depends not only on g R (), but also on the angle . In fact, S km km (f ) is symmetrical if and only if g R (+) = g R (+)
for [0, ). This condition is more general than the one that applies on
the DPSD S (f ) of the SISO channel model described in Chapter 2, where
S (f ) = S (f ) if and only if g () = g ( ) for [0, ).
5.2.3.2

Space-Time CCF of the MIMO Channel Gains

Information about the cross-correlation properties of the complex channel gains

is also relevant for the characterization of the MIMO channel model. Such an
information is important for the analysis of space-time block coding schemes
and space diversity systems [Gold05, Ch. 10], just to give a few examples. The
correlation between km (t) and ql (t) can be studied on the basis of the spacekm (t) ql (t + )}. This
time CCF function (STCCF) rkm ql (T , R , ) , E{
function gives a measure of the correlation between km (t) and ql (t) in terms
of the time difference and the size of the transmitter and the receiver antenna
arrays. From the expression given in (5.6) for km (t), one can show that
rkmql (T , R , ) =

pR () exp {j2fmax cos( ) }

am (, T ) al (, T ) bk (, R ) bq (, R )d.

(5.14)

km (t)
ql (t)} between km (t) and ql (t) is
The SCCF kmql (T , R ) , E{
important too. This function is relevant for the capacity analysis of MIMO fading
channels [Shiu00, Abd02a]. By taking = 0 in (5.14), we obtain the result
km ql (T , R ) =

pR () am (, T ) al (, T )

bk (, R ) bq (, R )d.

5.3
5.3.1

(5.15)

The Simulation Model

The Stochastic SOC Simulation Model

Following the simulation approach presented in Chapter 3, we can approximate

the statistical properties of the complex Gaussian processes km (t) by means of

Chapter 5 SOC Simulation of MIMO Channels

97

a stochastic SOC simulation model of the form

km (t) =

N
X

n=1




am (
n ) bk (
n ) cn exp j 2 fn t + n

(5.16)

where
fn = fmax cos(
n )

(5.17)

and
n [, ). We assume that the phases n are i.i.d. random variables
P
cn |2 = 2 . The
uniformly distributed in [, ) and the gains cn satisfy N
n=1 |
functions am () and bk () are defined in (5.7) and (5.8), respectively.

5.3.2

Correlation Properties of the SOC Simulation Model

It is easy to show that

km (t) is a zero mean WSS random process with variance
2
km (t)

km (t + )} given by
equal to and ACF r km km ( ) , E{
r km km ( ) =

N
X

n=1

c2n exp{j2fmax cos(

n ) }.

(5.18)

km (t)
In addition, one can show that the STCCF r km ql (T , R , ) , E{

km (t) and
ql (t) is equal to

ql (t + )} between
r km ql (T , R , ) =

N
X

n=1

c2n exp{j2fmax cos(

n ) }

am (
n , T ) al (
n , T ) bk (
n , R ) bq (
n , R ). (5.19)
km (t)
ql (t)} between
With respect to the SCCF km ql (T , R ) , E{

km (t)
and
ql (t), we have
km ql (T , R ) =

N
X

c2n am (
n , T ) al (
n , T ) bk (
n , R ) bq (
n , R ). (5.20)

n=1

We observe that if fn 6= 0 n and fn 6= fm n 6= m, then the processes

km (t)
are ergodic with respect to the mean value, the ACF, and the STCCF.

5.4

Parameter Computation Methods

For the simulation of the MIMO channel model, the problem consists in finding
proper values for the gains cn and the AOAs
n such that the autocorrelation
and cross-correlation properties of the random processes
km (t) resemble those

98

Channel Simulation Models for Mobile Broadband Communications

of the Gaussian processes km (t). In particular, we are interested in emulating

the ACF r km ql ( ) and the SCCF km ql (T , R ) of the reference model. In
what follows, we will explain how to apply the GMEA and the RSAM to solve
this parameter computation problem.

5.4.1

The GMEA

The results presented in Chapter 4 indicate that the ACF of km (t) can satis
factorily be approximated by defining the gains cn as in (4.1), i.e., cn = 2 / N ,
and computing the Doppler frequencies fn by solving the equation
Zfn

Skm km (f )df

fmax

2
N

1
N n+
2

(5.21)

for n = 1, . . . , N . Since fn = fmax cos(

n ) [see (5.17)], one can alternatively

compute the Doppler frequencies fn by finding the AOAs

n that solve

Zn

gR ( + )d =

1
2N

1
2

(5.22)

We notice, nonetheless, that the solutions of (5.21) and (5.22) provide information
only about the even part of pR (). This is sufficient to emulate the autocorrelation and spectral characteristics of km (t), but the emulation of the SCCF
between km (t) and ql (t) requires information also about the odd part of pR ()
[see (5.15)]. In order to take the odd part of pR () into account, we propose to
compute the AOAs
n by solving

Zn

pR ( + )d =

1
N

1
n
4

(5.23)

In the equation above, the factor of 1/(4N ) has been introduced to guarantee the
uncorrelation between the IQ components of
km (t) and to ensure that fn 6= fm
n 6= m when the DPSD Skmkm (f ) of the reference model is symmetrical. Such
a factor also ensures that fn 6= 0 n when Skm km (f ) = S km km (f ) and the
number of cisoids is even, i.e., when N = 2M , where M Z+ . We emphasize
that the number of cisoids shall be even in order for this latter characteristic to
hold. Otherwise, if N is odd and S km km (f ) is symmetrical, then the solutions
of (5.23) will produce a Doppler frequency fn equal to zero. The inequalities
fn 6= fm n 6= m and fn 6= 0 n are in general fulfilled when Skm km (f ) is
asymmetrical, irrespective of the value of N .

Chapter 5 SOC Simulation of MIMO Channels

5.4.2

99

The RSAM

In the case of the RSAM, we observe that the STCCF between km (t) and ql (t)
can be expressed as
r km ql (T , R , ) =

N
X
22
p (n ) exp {j2fmax cos(n ) }
lim
N N n=1 R

am (n , T ) al (n , T ) bk (n , R ) bq (n , R ) (5.24)
where n = n2/N 0 , and 0 is a real number in [0, 2/N ]. By removing the
limit in the previous equation, we can write
rkmql (T , R , )

N
22 X
p (n ) exp {j2fmax cos(n ) }
N n=1 R

am (n , T ) al (n , T ) bk (n , R ) bq (n , R ). (5.25)
Based on the above result, and taking account of the expression given in (5.19)
for the STCCF r km ql (T , R , ) of the simulation model, we can state that
r km ql (T , R , ) rkm ql (T , R , ) if
n2
0
Ns
p (
n )
= PN
.
m )
m=1 p (

n =

(5.26)

cn

(5.27)

Even though the value of 0 can arbitrarily be chosen within [0, 2/N ], we recommend to set 0 = /(2N ) to ensure that fn 6= fm n 6= m and to guarantee the
uncorrelation between the IQ components of
km (t) when the DPSD S km km (f )
of the reference model is symmetrical. Furthermore, choosing 0 = /(2N ) also
assures that fn 6= 0 n when S km km (f ) = Skm km (f ) and N = 2M , M Z+ .
This latter condition does not hold if N is odd, as in the case of the GMEA.
We have demonstrated in [Gut07a] that the methodology in (5.26) and (5.27)
provides excellent results concerning the emulation of the ACF rkmkm ( ) and
the SCCF kmql (T , R ). However, to properly approximate the envelope
distribution of the complex Gaussian processes km (t), it is important to impose
a constraint in the range of values that the gains cn may take. With this in mind,
we will assume that pR () has only one maximum in [, ). Then, following
the idea behind the RSAM described in Section 4.4.1.3, we redefine


u
1

n = +
n
,
u > .
(5.28)
N
4
In (5.28), u and designate, in that order, the lower and the upper boundaries

100

Channel Simulation Models for Mobile Broadband Communications

of the subinterval IU = [ , u ] within the which the PDF pR () is above a

given threshold 0, sup{p R ()}[,) .

5.5
5.5.1

Performance Analysis
Considerations

In this section, we will present some simulation examples demonstrating the good
performance of the GMEA and the RSAM concerning the emulation of the ACF
r km km ( ) and the SCCF km ql (T , R ) of the reference model. To that end,
we will assume that the random AOAs R are von Mises distributed, so that:
pR ( + ) =
gR ( + ) =

exp{ cos( m + )}
(5.29)
2I0 ()




exp cos() cos(m )
cosh sin() sin(m ) (5.30)
2I0 ()

where [, ). It is easy to show, by substituting (5.30) into (5.13), that the

DPSD of the complex Gaussian processes km (t) is given by


p
2
SVM
(f
)
=
rect(f
)

cosh

sin(m

)
1

(f
/f
)

max

km km


2 exp cos(m ) f /fmax
p

.
fmax I0 () 1 (f /fmax )2

(5.31)

In turn, by following (5.12), one can show that the ACF of km (t) is equal to
rVM
( ) =
km km

2 I0

p

(2fmax

)2


+ j4fmax cos(m )

I0 ()

. (5.32)

The STCCF rVM

(T , R , ) between km (t) and ql (t) can also be expressed
km ql
in closed form by substituting (5.29) into (5.14). Particularly, for the MIMO
channel gains 11 (t) and 22 (t), we have [Abd02a]
rVM
(T , R , ) =
11 22

T cos(T )}
2 exp{j2
I0 ()
n
h


2R +
T max,T sin(T ) 2
I0 2 4 2 2 +

R cos(R ) 2
T max,T sin(T )
2

i
R sin(R )
sin()
h
R cos(m R )
+j4 cos(m )
io1/2 
T max,T sin(T ) sin()

.
(5.33)

Chapter 5 SOC Simulation of MIMO Channels

101

T = T /, and
R = R /. One can deduce from (5.33)
where = fmax ,
that the SCCF VM
(T , R ) = rVM
(T , R , 0) is equal to
km ql
km ql
VM
(T , R ) =
11 22

T cos(T )}
2 exp{j2
I0 ()
n
h


2
2R +
T max,T sin(T ) 2 + 2
T
R
I0 4 2
i
h
R cos(m R )
max,T sin(T ) sin(R ) j4
io1/2 
T max,T sin(T ) sin()
+
.
(5.34)

For the simulations, we will consider the following triplets of parameters for
the von Mises PDF in (5.29): ( = 0, m = 0 , = 0 ); ( = 5, m = 0 , = 0 );
( = 20, m = 0 , = 0 ); ( = 10, m = 0 , = 0 ); ( = 10, m = 20 , =
10 ); ( = 10, m = 40 , = 50 ). We notice that the first and the last
(f ), while the other four
of such triplets produce a symmetrical DPSD SVM
km km
triplets result in an asymmetrical DPSD. We choose T = R = 90 , max,T = 2 ,
fmax = 91 Hz, 2 = 1, and N = 24. Except by the number of cisoids, these
parameters are the same as the ones considered in [Pae04a] for evaluating the
performance of SOC simulators designed by applying the LPNM. In that paper,
the number of cisoids was equal to N = 25. Here, we have chosen N = 24
to ensure that the GMEA and the RSAM satisfy the inequality fn 6= 0 n.
Regarding the threshold of the RSAM, we set = 1 103 .

5.5.2

Emulation of the ACF

Figure 5.2 shows a comparison between the ACF rVM

( ) of the reference
km km
model [see (5.32)] and the ACF r km km ( ) of the simulation model [see (5.18)]
by applying the GMEA and the RSAM. We can observe from Fig. 5.2 that both
methods produce a very good approximation to the ACF of the reference model.
This indicates that the modifications introduced in this Chapter to the GMEA
and the RSAM do not affect the capability of these methods to emulate the
autocorrelation properties of the complex Gaussian processes km (t).

5.5.3

Emulation of the SCCF

To evaluate the accuracy of the GMEA and the RSAM for approximating the
SCCF km ql (T , R ) of the reference model, we define the following absolute
error function
1122 (T , R ) , | 1122 (T , R ) 11 22 (T , R )|.

(5.35)

102

Channel Simulation Models for Mobile Broadband Communications

(a)
Reference model
Simulation model, GMEA

m = 0
Absolute value of the ACF

= 20, = 0

Simulation model, RSAM

0.8

= 5, = 0
0.6

= 0, = 0

0.4

=
0.2

N
4fmax

10

Normalized time difference, fmax

(b)
m = 40 , = 50

Absolute value of the ACF

= 10

Reference model
Simulation model, GMEA
Simulation model, RSAM

m = 0 , = 0
0.8

0.6

m = 20 , = 10

0.4

N
4fmax

0.2

10

Normalized time difference, fmax

Figure 5.2: Comparison between the ACF r km km ( ) of the MIMO channel

gains (reference model) and the ACF r km km ( ) of the simulation
model by applying the GMEA and the RSAM with N = 24 to the
von Mises PDF of the AOA with different triplets of parameters ,
m , and , (fmax = 91 Hz and 2 = 1).
The absolute value of 11 22 (T , R ) is depicted in Fig. 5.3 for T [0, 20/]
and R [0, 3/]. The resulting error function 11 22 (T , R ) is plotted in
Figs. 5.4 and 5.5 for the GMEA and RSAM, respectively.
The results presented in Figs. 5.4 and 5.5 demonstrate that the GMEA and
the RSAM provide a close approximation to the SCCF 1122 (T , R ). This
is clear, since the error produced by the GMEA under non-isotropic scattering
conditions is smaller than 8102 , while the one produced by the RSAM is lower
than 6 104 . In turn, the error registered by both methods in the isotropic
scattering scenario is smaller than 1 1011 .

Chapter 5 SOC Simulation of MIMO Channels

103

(a)

(b)

(c)

(d)

(e)

(f)

Figure 5.3: Absolute value of the SCCF 11 22 (T , R ) of the MIMO channel

model by considering the von Mises PDF of the AOA with different
triplets of parameters , m , and (2 = 1).

104

Channel Simulation Models for Mobile Broadband Communications

(a)

(b)

(c)

(d)

(e)

(f)

Figure 5.4: Error function r11 22 (T , R ) between the SCCF 11 22 ( )

of the MIMO channel gains (reference model) and the SCCF
11 22 (T , R ) of the simulation model by applying the GMEA
with N = 24 to the von Mises PDF of the AOA.

Chapter 5 SOC Simulation of MIMO Channels

105

(a)

(b)

(c)

(d)

(e)

(f)

Figure 5.5: Error function r11 22 (T , R ) between the SCCF 11 22 ( )

of the MIMO channel gains (reference model) and the SCCF
11 22 (T , R ) of the simulation model by applying the RSAM
with N = 24 to the von Mises PDF of the AOA.

106

5.6

Channel Simulation Models for Mobile Broadband Communications

Simulation of Wideband MIMO Fading Channels

So far, we have focused our attention on the simulation of narrowband mobile

MIMO fading channels. However, the design of simulation models for wideband
MIMO channels is also an important research topic. It is relevant, for example,
for the analysis of multi-element antenna systems that require a large bandwidth
to operate, e.g., those based on spread-spectrum [Moli05, Ch. 18] and ultrawideband [Roys04] technology. In this section, we will discuss an application
of the GMEA and the RSAM to the design of simulation models for wideband
MIMO fading channels in single-bounce scattering environments.

5.6.1

Modeling and Statistical Characterization of Wideband

Mobile MIMO Fading Channels

From a system-theoretic point of view, the mobile fading channel can be described
as a randomly time-variant linear filter (TVLF) [Bell63, Pars00, Pae02a, Proa01].
In this line of thought, we model the wideband subchannel connecting the mth
transmitter antenna Tm with the kth receiver antenna Rk as a randomly TVLF
with an equivalent complex baseband impulse response
h km (t, ) =

L1
X
=0

km, (t) ( ).

(5.36)

In the equation above, km, (t) denotes the complex attenuation factor due to the
superposition of the transmitted signals replicas that reached the kth receiver
antenna through different paths but with the same propagation delay = . In
line with the narrowband MIMO channel model described in Section 5.2.2, we will
assume that the attenuation factors km, (t) are stationary zero-mean complex
Gaussian processes with corresponding variances equal to 2 , = 0, . . . , L1. In
addition, we will assume that the channel exhibits uncorrelated scattering (US)
hkm (t, 1 ) hkm (t + , 2 )} =
[Bell63], meaning that E{h
6 0 iff 1 = 2 . This implies
that the Gaussian processes km, (t) are mutually uncorrelated. Considering a
two-dimensional single-bounce scattering scenario, we can write
km, (t) =

lim

N
X

n=1

R,n, ) bk (
R,n, ) c n, exp {j (2ff n, t + n, )} (5.37)
am (

for = 0, . . . , L1. The parameters of km, (t) can be defined in a similar manner
as we defined their counterparts introduced in Section 5.2.2. Particularly, we will
P
2
2
suppose that the gains cn, are given such that N
n=1 c n, = , = 0, . . . , L 1,
and the AOAs R,n, are i.i.d. random variables, n = 1, . . . , N and = 1, . . . , L.
It should be noticed that if the received signal is composed of a continuum of

Chapter 5 SOC Simulation of MIMO Channels

107

multipath components, then the summation in (5.36) becomes an integration.

This results in a diffuse impulse response given as [Proa01, p. 802]
h km (t, ) = km (t, ).

(5.38)

The attenuation factor km (t, ) in (5.38) is a stationary zero-mean complex

Gaussian processes with variance 2 km . In this case, the US assumption implies
km (t, 1 ) km (t + , 2 )} =
6 0 iff 1 = 2 . The subchannel between Tm
that E{
and Rk can alternatively be described by means of the transfer function
H km (t, ) ,

h km (t, ) exp{2 }d.

(5.39)

For narrowband systems, H km (t, ) can be seen as a function that fluctuates in

the time domain and remains constant in the frequency domain, at least along
the frequency range of the system bandwidth. For wideband systems, H km (t, )
is to be treated as a function that fluctuates within the system bandwidth.
An exact statistical characterization of the random impulse response in (5.36)
is a formidable task that requires knowledge of the multidimensional joint PDFs
of h km (t, ) (or H km (t, )). However, since the attenuation factors km, (t) are
assumed to be complex Gaussian processes, the statistical properties of the MIMO
channel model are completely specified by the ACFs, PSDs, CCFs, and crossPSDs of the random processes h km (t, ). The characterization of MIMO channels
in terms of correlation functions and PSDs is surveyed somewhere else, e.g.,
[Fleu00] and [Moli05, Secs. 6.7 and 7.4]. For the purposes of this section, it is
sufficient to know that the selectivity of the channel in space, time, and frequency
domains can be studied with respect to the following correlation functions:
hkm (t, ) h ql (t, )}
rhkmh ql (T , R ) , E {h
rh kmh km ( ) , E

rH kmH km () , E

hkm (t, ) h ql (t + , )}
{h
H km (t, ) H km (t, + )} .
{H

(5.40)
(5.41)
(5.42)

The functions defined in (5.40)(5.42) are known as the temporal ACF (TACF),
the SCCF, and the frequency correlation function (FCF), respectively.

5.6.2

Simulation Approach

The simulation of h km (t, ) can be carried out upon a time-variant tapped-delay

line (TDL) model of the form
km (t, ) =
h

L1
X
=0

km, (t) ( ).

(5.43)

108

Channel Simulation Models for Mobile Broadband Communications

For the simulation of wideband MIMO channels in single-bounce scattering environments, we define the complex gains
km, (t) as follows

km, (t) =

N
X

n=1


am (
n, ) bk (
n, ) cn, exp j 2 fn, t + n, .

(5.44)

The parameters of
km, (t) are analogous to those of the narrowband SOC simulation model presented in Section 5.3.1, and they can be defined accordingly. In
particular, we suppose that the phases n, are i.i.d. random variables uniformly
distributed in [, ), for n = 1, . . . , N and = 0, . . . , L 1. This consideration guarantees the wide-sense stationarity and the mutual uncorrelation of the
km (t, ).
processes h
The problem at hand lies in computing the gains cn, and the AOAs n, of

km, (t) such that:

rh

km h ql

(T , R ) rh kmh ql (T , R )

rh

kmh km

( ) rh kmh km ( )

rH kmH km () rH kmH km ()
where rh

(5.46)
(5.47)

( ), and rH kmH km () designate, in that order,

the TACF, the SCCF, and the FCF of
h km (t, ). These functions are given as

kmh ql

(T , R ), rh

(5.45)

kmh km

n
o
(t, ) h
ql (t, )
, E h
km
n
o
(t, ) h
ql (t + , )
rh h ( ) , E h
km
km km
n
o
(t, ) H
km (t, + ) .
rH kmH km () , E H
km

(T , R )
km h ql

rh

(5.48)
(5.49)
(5.50)

A solution to this problem has recently been provided in [Paet06]. In that paper,
the authors approximate the TACF and the SCCF of a wideband MIMO channel
model derived from the geometrical one-ring scattering model by applying the
EMEDS and the LPNM. Regarding the emulation of the FCF rH kmH km () of the
MIMO channel, it was demonstrated in that paper that the TDL model in (5.43)
exactly reproduces such an statistical function if the impulse response h km (t, ) is
P
a discrete function in the delay domain and the gains cn, satisfy N
2n, = 2 .
n=1 c
We will refrain ourselves from providing details about the simulation approach
presented in [Paet06], since it was developed independently by other researchers.
However, we observe that the idea can be implemented in a more efficient manner
by employing either the GMEA or the RSAM. In fact, a third party demonstrated
in [Mayy08] that the simulation of wideband MIMO channels can effectively be
performed by applying the idea described in [Paet06] in connection with the
GMEA. We notice that the authors of [Mayy08] made use of an alternative version

Chapter 5 SOC Simulation of MIMO Channels

109

of the GMEA that we proposed in [Gut07b].

The simulation approach described in [Paet06] is based on the assumption
that the impulse responses of the MIMO subchannels are discrete functions in
the delay-domain, meaning that the processes h km (t, ) are given as in (5.36).
Under this consideration, it is evident that the TDL model defined in (5.43) is
a natural structure for the simulation/representation of h km (t, ). However, the
characterization of h km (t, ) by means of a TDL model is not straightforward
when the impulse responses of the MIMO subchannels are given as in (5.38), i.e.,
when hkm (t, ) is a continuous (diffuse) function in . In the following chapter,
we will revisit a well-known procedure for the design of TDL for WSS complex
Gaussian channels characterized by diffuse impulse responses that satisfy the US
condition. The concepts studied in that chapter can be applied together with
the idea proposed in [Paet06] to design simulation models for wideband mobile
MIMO fading channels with continuous-delay impulse responses.

5.7

Chapter Summary and Conclusions

In this chapter, we demonstrated that the GMEA and the RSAM introduced in
Chapter 4 can be applied to the simulation of narrowband mobile MIMO fading
channels with specified correlation properties. We explained here that a proper
emulation of the SCCF of the MIMO channel requires information about both
the even and the odd parts of the PDF of the AOA. For this reason, some minor
modifications were made in Section 5.4 to the GMEA and the RSAM in order to
incorporate information about the odd part of the AOA distribution.
We evaluated the performance of the proposed parameter computation methods with respect to the approximation of the ACF and SCCF of the MIMO
subchannels. The obtained results indicate that the modifications made to the
GMEA and the RSAM do not affect the methods capability for emulating the
ACF of the MIMO subchannels. Concerning the emulation of the SCCF, the obtained results show that both methods provide a very good approximation to such
a correlation function. Particularly, our investigations indicate that the RSAM is
better suited than the GMEA to emulate the correlation properties of the MIMO
channel. Finally, we discussed in Section 5.6 an application of the methods proposed in this dissertation to the design of simulation models for wideband mobile
MIMO fading channels.

110

Channel Simulation Models for Mobile Broadband Communications

Chapter 6

Design of TDL Models for

WSSUS Channels with Diffuse
Impulse Responses
6.1

Introduction

The wide-sense stationary uncorrelated scattering (WSSUS) model proposed by

Bello [Bell63] is a widely accepted model for small-scale mobile fading channels
[Stei87, Skla88, Jeru00, Proa01, Pae02a, Pae02b]. A variety of standardized channel models for wireless and mobile communication systems has been developed
on the basis of this stochastic model [Cos207, EtsGSM, Etsi99, Med98a, Med98b,
Erce04]. In order to obtain realizable and well specified representations of WSSUS
channels, one often has to resort in practice to the use of simulation models. Different types of models can be used for the simulation of WSSUS channels, e.g.,
those presented in [Yip95], [Chri02]. However, a fundamental requirement for
any of such simulators is the proper emulation of the essential characteristics of
the WSSUS model. This is necessary to allow for a reliable and reproducible
system performance investigation.
Several simulation models basing on the time-variant TDL filter concept have
been proposed for the simulation of WSSUS channels with continuous-delay impulse responses, such as those in [Pae02b] and [Yip95]. Among them, the simplest and perhaps most popular one is a so-called symbol-spaced TDL (SSTDL)
model. Briefly speaking, a SSTDL model can be considered as a sampled version of the time-variant channel impulse response (CIR) obtained by using a
sampling rate equal to the reciprocal of the symbol duration. Such a kind of
TDL models was originally introduced by Kailath [Kail59] and Bello [Bell63] as
canonical channel modelswhich are essentially simulation modelsfor bandlimited time-variant linear systems. SSTDL models are nowadays widely in use
111

112

Channel Simulation Models for Mobile Broadband Communications

for the performance analysis of modern wireless communication systems, e.g.,

[Choi01, Yang02, Hou04].
It might seem that the applicability of SSTDL models is unquestionable, since
they were developed in the framework of the sampling theorem [Bell63]. Nevertheless, the results reported in [Guti05] suggest that the validity of such TDL
models is rather doubtful when the concept is applied to WSSUS channels. In
fact, the answer to the question whether SSTDL models are suitable for modeling
WSSUS channels is still lacking in the literature. What is more, despite the fact
that this modeling approach dates back more than forty years [Bell63], [Kail59],
the statistical properties of the resulting SSTDL are not well known. Most of
the literature on this topic concentrates only on describing the structure of the
SSTDL model [Stei87], [Proa01], and even though a statistical analysis can be
found in [Bell63], [Jeru00] and [Syko00], no information concerning the channel
FCF is provided there. The information about the FCF is very important because
this function influences the performance of many wideband and frequency diversity wireless communication systems, such as multicarrier code division multiple
access (MC-CDMA) systems [Choi01, Yang02, Hou04, Hara97, Schu05].
Closing the above mentioned gaps is necessary not only for a better understanding of channel modeling aspects, but also for carrying out a reliable system
performance investigation. In this chapter, we aim at closing them by analyzing
the validity and statistical behavior of SSTDL models for WSSUS channels. It
is also our objective to discuss the problems associated with these TDL models,
as well as to present a simple and effective strategy to avoid them.
The rest of the chapter is organized as follows. Section 6.2 is devoted to
analyzing the suitability of SSTDL models to simulate WSSUS channels. It is
shown in that section that using SSTDL models to model the CIR of WSSUS
channels results in a violation of the US condition from the perspective of bandlimited systems. In Section 6.3, we provide a concise description of the statistical
behavior of these type of simulation models. We show in that section that SSTDL
models suffer from strong limitations in emulating the FCF of WSSUS channels.
In Section 6.4, we show that SSTDL models lead to an imprecise performance
analysis of wireless communication systems sensitive to the FCF. To cope with
this problem, we discuss a solution in Section 6.5 by doubling the sampling rate
of the channel, resulting in a half-symbol-spaced TDL (HSSTDL) model. The
usefulness of this solution is exemplary demonstrated by analyzing the bit error
probability of a down-link MC-CDMA system. Finally, we present our conclusions
in Section 6.6.

Chapter 6 TDL Models for WSSUS Channels

6.2
6.2.1

113

About the Validity of SSTDL Models

The SSTDL Modeling Approach

Before we proceed to analyze the suitability of SSTDL models for modeling

WSSUS channels, it is convenient to review the approach followed in [Bell63]
to develop such a type of simulation models. Toward that end, consider a wireless communication system characterized by the input-output relationship
Z
y(t) = x(t) h(t, ) = x(t)h(t, )d
(6.1)
where denotes convolution, x(t) is the transmitted signal, y(t) is the received
signal, and h(t, ) is the time-variant CIR, which is continuous in both t and
variables. The CIR h(t, ), called the input delay-spread function by Bello, may
be regarded as being the system response at time t to a unit impulse applied
seconds in the past.
For practical purposes, it is possible to derive a discrete-delay1 representation of the CIR h(t, ) on the basis of the sampling theorem by assuming that
either the transmitted signal or the channel itself is bandwidth restricted. For
the latter case, it was shown in [Bell63, pp. 378-379] that if the corresponding
R
channel transfer function H(t, ) = h(t, ) exp{j2}d is confined to a
bandwidth W (meaning that H(t, ) = 0 for || > W), then, according to the
sampling theorem, the CIR h(t, ) may equivalently be expressed as [Bell63, eq.
(123)]
h(t, )

 n
 h
n i
h t,
sinc W
W
W
n=

(6.2)

where sinc() , sin()/() and W W is an arbitrary sampling rate.

In conformity with the thesis in [Bell63], it turns out that the CIR h(t, ) is
well modeled by a TDL linear filter having a time-variant impulse response
) =
h(t,

n=

h(t, n )( n )

(6.3)

where the tap spacing , 1/W is given such that 1/W. The equivalent
) establishes the so-called SSTDL model
discrete-delay CIR described by h(t,
when W = W = Ws , where Ws = 1/T is the nominal bandwidth of the transmit1
In this chapter, we will be concerned only with the discrete-delay modeling of the CIR
h(t, ). The reader is referred to [Pae02a] for a detailed study on the modeling of h(t, ) in
discrete-time domain

114

Channel Simulation Models for Mobile Broadband Communications

ted symbols2 , and T is usually equal to the symbol duration3 . Throughout the
) either as the SSTDL model or as the T -spaced
chapter, we will refer to h(t,
TDL model to stress the fact that = T (Ws = W = W ), while we will
use the term -spaced TDL model for the more general case where T
(Ws W W ).
The equivalence between the CIR h(t, ) and its discrete-delay version defined

by h(t, ) in (6.3) holds in the sense that the former impulse response can be
reconstructed from the interpolation of the latter one, as indicated by (6.2).

6.2.2

The Validity of SSTDL Models

The SSTDL modeling approach may at first seem to be sound and thorough.
Nevertheless, important problems arise when the random nature of the mobile
fading channel is brought into consideration. The main source of conflict comes
from the fact that the SSTDL model was derived on the basis of the sampling
theorem for deterministic signals. If the CIR h(t, ) is deemed to be a random
process, as it is done usually for wireless communication system, then its discretedelay representation should be obtained by invoking the sampling theorem for
stochastic processes instead [Proa01, p. 71]. Thus, the bandwidth restriction is
not to be associated with the (random) channel transfer function H (t, ), but
with the power density spectrum of h (t, ). Fortunately, the amendment of such
h(t, ) similar as the
an imprecision results in an equivalent discrete-delay CIR h
one defined in (6.3), and the results and conclusions drawn in [Bell63] remain
basically unchanged. In this case, however, the equivalence defined in (6.2) holds
only in the zero-mean-square-error sense [Proa01, p. 71]
(
)

 n
 h
i 2
X

n

E h (t, )
h t,
sinc W
= 0.

W
W
n=

(6.4)

We point out that the sampling theorem for stochastic processes applies only on
stationary processes.
A further problem arises if the random CIR h (t, ) meets per definition the
WSSUS condition. If h(t, ) is a WSSUS random process, then the simulation
approach described in Section 6.2.1 is not applicable, because the US condition is
h(t, ) [Schu05].
incompatible with the bandwidth constraint required to obtain h
2

The nominal bandwidth may be defined as the width of the main lobe of the power spectral
density of a digitally modulated signal.
3
For instance, T equals the duration of the quadrature-amplitude-modulation (QAM) symbols in a conventional single carrier system, whereas T equals the chip duration in direct-sequence
spread spectrum (DS-SS) systems [Schu05]. Multicarrier systems [Schu05], [VanN04] are a special case, where T is equal to the (non-cyclic-extended) symbol duration divided by the total
number of subcarriers. In all cases, Ws = 1/T holds. We note that the term symbol is used
in this chapter for those digital signals that carry the information through the channel.

Chapter 6 TDL Models for WSSUS Channels

115

Such an incompatibility often passes unnoticed in the literature. Take as an

example the conclusions drawn in [Proa01] at the end of Section 14.5.1. There,
h(t, ) is a consequence of
it is stated that the uncorrelation of the tap gains of h
the US condition of h (t, ). This statement is not adequate in the way it was
postulated in [Proa01], because it is based on the implicit assumption that the
CIR h (t, ) satisfies both the WSSUS condition and the bandwitdh constraint.
Notice that the equation (14.54) given in [Proa01, Sec. 14.5.1] holds only if the
channel transfer function is band-limited.
Bello proposed a second TDL model which circumvents the incompatibility between the bandwidth limitation of H (t, ) and the US condition of h (t, )
[Bell63, pp. 379, second column]. In deriving that model, Bello shifted the bandwidth constraint onto the input signal x(t). Thereby, since the bandwidth restriction is external to the CIR h (t, ), one can assume without problems that
h(t, ) is a WSSUS process. In that case, it can be shown that the equivalent
discrete-delay CIR is given by [Syko00]
h(t, ) =
h

Z
X

n=

h (t, )sinc(W [ n ])d ( n ).

(6.5)

h(t, ) is not equivalent to h (t, ) in the sense of (6.2), but

The impulse response h
in what
y (t)

n=

x(t n )

h (t, )sinc([ n ])d.

(6.6)

The previous equation implies that the output of the system remains the same
h(t, ); as y (t) = y(t),
if the CIR h (t, ) is replaced by its discrete-delay version h
h(t, ). Similar observations can be made for the SSTDL
where y(t) = x(t) h
model, or more generally, for the -spaced TDL model, since y (t) y(t)
h(t, ).
sinc (W t), where y = x(t) h
Even though the TDL model described in (6.5) is compatible with WSSUS
channels, it is strictly speaking not a valid simulation model for them, since the
tap gains of the equivalent discrete-delay CIR
h
h(t, ) are mutually correlated.
The cross-correlation among the taps can be neglected, nonetheless, if the autocorrelation function of h (t, ) varies slowly in the delay domain, as pointed out
in [Bell63] and [Syko00]. However, since the US condition of the CIR h (t, ) is in
h(t, ), it can be conclude that this simulation model is
general not reflected by h
h(t, ) belongs
not adequate for WSSUS channels. In fact, the random process h
to the class of non-WSSUS processes [Matz05].

116

Channel Simulation Models for Mobile Broadband Communications

6.3

Statistical Behavior of SSTDL Models

Despite the SSTDL model proves to be incompatible with WSSUS channels from
the perspective of band-limited systems, simulation models of this kind are widely
being in use in connection with the WSSUS model to evaluate the performance
of wireless communication systems, e.g., [Choi01, Yang05, Hou04]. In defense
of such a system performance investigation approach, we can argue that the
SSTDL model might coexist with WSSUS channels if the bandwidth constraint
imposed on the CIR h (t, ) is neglected4 . To find out if this is the case, it is
h(t, ) satisfactorily emulates the statistical properties
necessary to know whether h
of WSSUS channels. It should also be determined whether the resulting SSTDL
model lends itself to system performance investigations. To accomplish the abovementioned tasks, we need a complete description of the statistical behavior of the
h(t, ). In this section, we provide such a
(nonequivalent) discrete-delay CIR h
description by assuming that h (t, ) is a zero-mean WSSUS complex Gaussian
process. We will also assume that the CIR h (t, ) is causal and has finite duration.
Consequently, we can write
h(t, ) =
h

N
X

n=0

h (t, n )( n )

(6.7)

where the number of taps N , W max + 1 is defined with respect to the

sampling rate W and the channels maximum excess delay max [Skla88]. The
operator denotes the nearest integer toward minus infinity.

6.3.1

Statistical Description of the Reference Channel Model

Before we proceed, we will summarize the statistical properties of the WSSUS

complex Gaussian CIR h (t, ) that are relevant for the purposes of this chapter.
For a detailed discussion on the statistical properties of WSSUS channels, we
refer the reader to [Bell63, Proa01, Pae02a, Schu05].
In line with the WSSUS Gaussian model, h(t, ) is characterized by its ACF
h(t1 , 1 )h
h (t2 , 2 )}
rhh (t1 , t2 ; 1 , 2 ) , E {h
= rhh ( ; 1 ) (1 2 )

(6.8)

h(t, )h
h (t + , )} and = t1 t2 . Similarly, h (t, ) can be
where rhh ( ; ) , E{h
4

Note, however, that the equivalences described in Section 6.2 for each of the pairs H (t, ),
h (t, ) and y(t), y(t) will no longer be valid. In any case, the corresponding relationships
between those functions should be established as mere approximations in terms of their statistical
properties.

Chapter 6 TDL Models for WSSUS Channels

117

characterized by means of its time-frequency ACF (TF-ACF)

H (t , )H
H (t2 , 2 )}
RH H (t1 , t2 ; 1 , 2 ) , E {H
Z Z 1 1
=
rhh (t1 , t2 ; 1 , 2 )

exp{j2(1 1 2 2 )} d1 d2

= RH H ( ; )

(6.9)
(6.10)

H (t, ) H (t + , + )} and = 1 2 . Equation (6.10)

where RH H ( ; ) , E{H
results from the substitution of rhh (t1 , t2 ; 1 , 2 ) from (6.8) into (6.9), and states
that the channel is WSS in both the time and the frequency variables5 .
The multipath behavior of h (t, ) is characterized in the frequency domain
by the FCF RH H () , RH H (0; ) and in the delay domain by the power delay
profile (PDP)
h(t, )|2 }
S () , E{|h
Z
=
RH H () exp{j2} d.

(6.11)

We will focus our attention on these two statistical quantities (especially on the
FCF), since a TDL channel model is mostly intended to emulate them.

6.3.2

Statistical Description of -Spaced TDL Models

h(t, ) is a sampled version of a Gaussian process, it is in turn a Gaussian

Since h
process, which is therefore completely characterized by its ACF
h(t1 , 1 )h
(t2 , 2 )}
rh h (t1 , t2 ; 1 , 2 ) = E{h
=

X
1
r
(
;

)
(

)
(1 nT )
1
1
2
hh
W2
n=

(6.12)

where W , 1/ . We recall that W = Ws for the T -spaced TDL model. Equah(t, ) is a WSSUS process, which was to be expected
tion (6.12) shows that h
h(t, ) are samples of a WSS random process that fulfills
since the tap gains of h
h(t, ) is an adequate simuthe US condition. We can therefore conclude that h
lation model for WSSUS channels. This is in line with the conclusion drawn in
[Proa01]. In order to determine how accurate is this simulation model in emulating the statistical behavior of the reference model described by the CIR h (t, ),
h(t, ).
it is convenient to turn our attention to the FCF of h
1 , 1 ) H
(t2 , 2 )} of
We can express the TF-ACF R (t1 , t2 ; 1 , 2 ) , E{H
H(t
HH

Actually, the channel is SSS in both domains, since the underlying process is a Gaussian
process.

118

Channel Simulation Models for Mobile Broadband Communications

R
h(t, ), where H

) = h
h
H(t;
h(t, ) exp{j2}d, as the convolution of the
channel TF-ACF RH H (t1 , t2 ; 1 , 2 ) = RH H ( ; ) and the train of delta functions
P
n= (1 nW )/W . In this way, we obtain
RH H (t1 , t2 ; 1 , 2 ) =

1 X
RH H ( ; nW )
W n=

= RH H ( ; )

(6.13)
(6.14)

) H
(t + , + )}. In turn, we can define the FCF
where RH H ( ; ) , E{H
H(t,
h(t, ) as R () , R (0; ). Thereby, we can write
of h
HH
HH

1 X
RH H () =
RH H (0; nW ).
W n=

(6.15)

An equivalent expression that allows for the analytical evaluation of the FCF
RH H () may be written as follows:
N
1 X
RH H () = 2
S (n/W ) exp{j2n/W }.
W n=0

(6.16)

) is WSS in the time domain and wide-sense periodic

We observe that H
H(t;
in the frequency domain with period W . In mathematical terms, this means
that RH H (t1 , t2 ; 1 , 2 ) = RH H ( ; + kW ), where k is an integer. Notice that
RH H () = RH H ( + kW ). The FCF RH H () is also an Hermitian symmetric
().
function, meaning that RH H () = RH
H

HH

For an exact emulation of the FCF RH H () of the reference model, it is required that RH H () = RH H (). However, this equality can be fulfilled only
within the frequency interval [W/2, W/2], because RH H () 6= RH H ()
for || > W/2 due to the periodicity of RH H (). What is more, the FCF
h(t, ) is a weighted sum of shifted replicas
RH H () of the simulation model h
of RH H (), as one can observe from (6.15). Hence, to ensure that the relation
RH H () = RH H () holds for [W/2, W/2], it is necessary that RH H () = 0
for || > W/2, otherwise the replicas of RH H () will overlap and RH H () will
be affected by aliasing. Unfortunately, the fulfillment of this condition cannot be
h(t, )
guaranteed because the FCF of WSSUS channels is not band-limited, and h
does not include any external bandwidth restriction for RH H ()which is actuh(t, ) and the TDL model h
h(t, ) described in
ally the main difference between h
h(t, ) will therefore be affected by a certain degree of
Section 6.2. The FCF of h
aliasing, which will reduce its accuracy for emulating the FCF of h(t, ). This is
in fact the main drawback of the -spaced TDL model. Owing to the aliasing
h(t, ) provides just an approximation to the FCF
effects, the FCF RH H () of h

Chapter 6 TDL Models for WSSUS Channels

119

RH H () of h (t, ) valid for W/2 W/2. Aliasing is a well-known effect,

but it is generally not recognized in the literature that this effect influences the
statistical properties of the TDL simulation model in (6.3).
On the other hand, a reliable system performance analysis requires an accurate emulation of RH H () along the frequency range of the system bandwidth. Hence, satisfying the relation RH H () RH H () within the interval
[Ws , Ws ] is highly desirable for system evaluation purposes. Unfortunately,
this is not possible when we employ a T -spaced TDL model, because the period W
of the FCF of this simulation model is equal to Ws , i.e., W = Ws , implying that
the approximation RH H () RH H () cannot be satisfied for Ws /2 || Ws .
As will be shown next, this characteristic of T -spaced TDL models affects the
performance evaluation of wireless communication systems sensitive to the FCF.

6.4
6.4.1

Numerical Examples
The Truncated-Exponential-Decay PDP

In what follows, we will provide some examples illustrating the issues discussed
in Section 6.3.2. To this end, let us assume that the CIR h (t, ) of the reference
model has a truncated-exponential-decay PDP (TED-PDP)
S () =

1
c

exp {/} , 0 max

0, else

(6.17)

where > 0 is the falling factor of the PDP, and c = [1 exp{max /}] /. The
TED-PDP has been shown to be an adequate model for characterizing the PDP
of outdoor and indoor wideband channels [Erce99]. Indeed, this type of PDP
has been adopted as a reference by several bodies of standardization for wireless
communication systems [Cos207, EtsGSM, Med98b, Med98a, Corr06, Erce04].
In the case of the TED-PDP in (6.17), the FCF RH H () of the reference
model is given by


1 exp max
(1 j2)

.
RH H () =
(6.18)
(1 j2) (1 exp max
)

From (6.16) and (6.17), we can express the FCF RH H () of the -spaced TDL
h(t, ) as
model h
RH H () =

 1

 N


1 exp W
1 exp W
(1 + j2)
 N

 1

.
1 exp W
1 exp W
(1 + j2)

(6.19)

We normalized (6.16) to obtain the above expression, so that RH H (0) = 1.

120

6.4.2

Channel Simulation Models for Mobile Broadband Communications

Effects of Aliasing on the FCF

In Fig. 6.1, we present a comparison between the absolute values of RH H ()

and RH H () by considering max = 800 ns, W = Ws = 20 MHz, and
{30 ns, 60 ns, 105 ns}. These parameters are representative of the propagation
conditions encountered inside office buildings ( = 30 ns), at large open space
environments ( = 60 ns), and at outdoor environments under non-line of sight
conditions ( = 105 ns) [Med98a] for applications in wireless local area networks
(WLANs), such as HIPERLAN/2, where the system bandwidth Ws is equal to
20 MHz [Etsi99].
From the graphs depicted in Fig. 6.1, it is evident that the SSTDL model
does not provide a good approximation to the FCF of the reference model within
the relevant interval [Ws /2, Ws /2]. Actually, by taking the graphs of |RH H ()|
as a reference, we can observe that |RH H ()| exhibits a correlation offset that
increases monotonically as moves from zero to W/2. The above mentioned
offset is indeed an effect caused by aliasing. We can also observe in Fig. 6.1 that
RH H () and RH H () follow different trends for frequencies in excess of W/2.
From the graphs depicted in that figure, it is evident that |RH H ()| =
6 |RH H ()|
for || > W/2.

6.4.3

The Influence of the FCF on the System Performance

Consider a frequency diversity system described by the following baseband equivalent signal model (it is assumed that the system is free from inter-symbol interference (ISI))
~ ~x + n
~y = C
~

(6.20)

~ CM 1 represent a vector of M transmitted signals, a random

where ~x, ~y , n
vector of M received signals, and a random vector with M complex additive
white Gaussian noise (AWGN) components, respectively. The random channel
~ is defined as C
~ = diag{H
~ }, where H
~ CM 1 and hH
~ in = H (t; n 0 )
matrix C
for all t. The symbol 0 stands for the fundamental frequency of the system.
The notation Cmn designates the set of all m n complex matrices, and hin,m
denotes the entry of a matrix at its mth column and nth row (subindex m is
omitted for vectors). The diag{} operator produces a diagonal matrix from a
vector, where the diagonal elements of the matrix are equal to the elements of the
vector between curly braces. The channel correlation matrix is given as follows
~C
~ }. We define the (n, m)th entry of R
~ C = E{C
~ C as
R
~ C in,m = RH H (W mn /M )
hR

(6.21)

Chapter 6 TDL Models for WSSUS Channels

121

(a)
1.4

Absolute value of the FCF

1.2

FCF of the reference model

FCF of the T -spaced TDL model (W = Ws )

= 30 ns

0.8

0.6

0.4

Ws
Ws /2

0.2

0
30

20

10

10

20

30

Frequency separation, f (MHz)

(b)
1.4

Absolute value of the FCF

1.2

FCF of the reference model

FCF of the T -spaced TDL model (W = Ws )

= 60 ns

0.8

0.6

0.4

Ws
Ws /2

0.2

0
30

20

10

10

20

30

Frequency separation, f (MHz)

(c)
1.4

Absolute value of the FCF

1.2

FCF of the reference model

FCF of the T -spaced TDL model (W = Ws )

= 105 ns

0.8

0.6

0.4

Ws
Ws /2

0.2

0
30

20

10

10

20

30

Frequency separation, f (MHz)

Figure 6.1: Comparison between the absolute value of the FCF of the reference
model and the absolute value of the FCF of the SSTDL by considering the TED-PDP with max = 800 ns, Ws = 20 MHz, and
different values for the decay factor of the PDF.

122

Channel Simulation Models for Mobile Broadband Communications

~ C . By assuming binary
where mn = m n denotes the (m n)th lag of R
phase shift keying (BPSK) modulation, and maximal ratio combining (MRC), we
obtain the following expression for the BEP of this system [Schu05, Sec. 2.4.6]
PBEP

1
=

M 
/2 Y
i=1

Eb
i
1+

N0 M sin2

1

(6.22)

where Eb is the bit energy, N0 is the noise density, and i is the ith eigenvalue
~C .
of the correlation matrix R
When using a T -spaced TDL model for the laboratory analysis of such a
frequency diversity system (or any other -spaced TDL model), the channel
~ in (6.20) is substituted by a random matrix C
~ = diag{H
~ }, where
matrix C
~ CM 1 , and hH
~ in = H
~ is then
n 0 ) for all t. The correlation matrix of C
H
H(t;
~ C
~ }, where
~ = E{C
given by R
C

~ in,m = R (W mn /M ).
hR
C
HH

(6.23)

~ = R
~C .
For an accurate system performance analysis, it is desirable that R
C
However, due to the problems discussed in the preceding section, we have
~ in,m hR
~ C in,m ,
hR
C
~ in,m 6= hR
~ C in,m ,
hR
C

for |mn | M/2

(6.24)

for |mn | > M/2 .

(6.25)

i of R
~ are not the
From (6.24) and (6.25), it follows that the eigenvalues
C
~ C . Consequently, the simulated system
same as those of R
~ ~x + n
~y = C
~

(6.26)

will provide an imprecise picture of the BEP of the actual system defined in (6.20).
To demonstrate the veracity of this remark, let us analyze the BEP of a downlink MC-CDMA system [Guti06] comprising M = 64 orthogonal subcarriers with
0 = 312.5 kHz, all of them used as data subcarriers. The nominal bandwidth
of this system is Ws = 20 MHz (Ws = 64 312.5 kHz). The sampling rate T is
therefore equal to T = 1/Ws = 50 ns. We assume a multiple-user-interference(MUI)-free network scenario and BPSK modulation with MRC at the receiver.
With respect to the spreading factor MSF of the system, we consider MSF
{1, 2, 4, 8, 16, 32, 64}. To compute the BEP for different spreading factors, we use
a slightly modified version of (6.22), which is given by
PBEP SF

SF
/2 M
Y

i=1

(M

i SF
Eb
1+

N0 MSF sin2

!1

(6.27)

Chapter 6 TDL Models for WSSUS Channels

(M

123

~ (MSF )
where i SF is the ith eigenvalue of an arbitrary correlation matrix R
CMSF MSF . To measure the BEP of the reference system for a spreading factor
~ (MSF ) as hR
~ (MSF ) in,m = hR
~ C in,m . AnalMSF , we define the (n, m)th entry of R
~ (MSF ) in,m =
ogously, to compute the BEP of the simulated system, we set hR
~ in,m . We have evaluated (6.27) by considering the TED-PDP [see (6.17)]
hR
C
with max = 800 ns and the same values of the falling factor as considered for
drawing the graphs presented in Fig. 6.1. The obtained results are plotted in
Fig. 6.2.
It can be seen in Figs. 6.2(a), 6.2(b), and 6.2(c) that the BEP of the reference
system is not exactly the same as that of the simulated system when MSF > 1.
In fact, the difference between the BEP of both systems increases with increasing
the value of the spreading factor. This is because the aliasing effects affecting
the FCF of the SSTDL model are more severe as the spreading factor increases.
Interestingly, the results presented in Fig. 6.2(c) show that the aliasing effects
can be neglected if is large. This is not surprising, since the FCF RH H () of
h (t, ) decays fast for large values of [cf. Fig. 6.1]. However, the BEP of ~y is
clearly different from that of ~y when MSF = 64, even for a large value of . The
~ (64) do not meet
reason for such a sharp difference is that half of the elements in R

C
(64)
(64)
~
~
the relation hR im,n hR im,n [see (6.25)].

6.5

Avoiding the Problems Caused by SSTDL Models

SSTDL models lead to an imprecise system performance evaluation because of

two reasons. On the one hand, the quality of the relation RH H () RH H () is
affected by aliasing effects within [Ws /2, Ws /2]. On the other hand, RH H () 6=
RH H () for || > Ws /2 since RH H () behaves different from RH H () when || >
Ws /2. Fortunately, these problems can be solved, or at least sufficiently reduced,
simply by increasing the sampling rate of the channel, i.e., by reducing the tap
spacing of h (t, ).
Reducing will result in less severe aliasing effects provided that RH H ()
0 as (most PDPs, including the TED-PDP in (6.17), meet this condition). In addition, it is obvious from (6.15) that a higher sampling rate will result
in a larger period for RH H () and, consequently, in a larger approximation interval. Indeed, it can be shown that RH H () RH H () as 0. Particularly,
if we choose = T /2, then the period W of RH H () will be conveniently equal
to twice Ws , i.e., W = 2Ws . In this way, the approximation RH H () RH H ()
will hold for the relevant interval [Ws , Ws ]. As a result, we will obtain a
more reliable picture of the systems performance while keeping the complexity
of the TDL model low. For the above mentioned reasons, we suggest to sample
the CIR at a rate equal to twice Ws to avoid the problems caused by T -spaced

124

Channel Simulation Models for Mobile Broadband Communications

(a)
1

Bit error probability, BEP

10

Solid lines: BEP by using the reference model

Dashdotted lines: BEP by using the T -spaced TDL model

10

10

= 30 ns

10

10

10

SF
SF
SF
SF
SF
SF
SF

=
=
=
=
=
=
=

1
2
4
8
16
32
64

10

15

20

25

30

Bit energy vs Noise density, Eb /N0 (dB)

(b)
1

Bit error probability, BEP

10

Solid lines: BEP by using the reference model

Dashdotted lines: BEP by using the T -spaced TDL model

10

10

= 60 ns

10

10

10

SF
SF
SF
SF
SF
SF
SF

=
=
=
=
=
=
=

1
2
4
8
16
32
64

10

15

20

25

30

Bit energy vs Noise density, Eb /N0 (dB)

(c)
1

Bit error probability, BEP

10

Solid lines: BEP by using the reference model

Dashdotted lines: BEP by using the T -spaced TDL model

10

10

= 105 ns

10

10

10

SF
SF
SF
SF
SF
SF
SF

=
=
=
=
=
=
=

1
2
4
8
16
32
64
10

15

20

25

30

Bit energy vs Noise density, Eb /N0 (dB)

Figure 6.2: Comparison between the theoretical and the empirical BEPs of a
MUI-free downlink MC-CDMA system with MRC by considering
the T -spaced TDL model and the TED-PDP with different values
of the decay factor (SF stands for the spreading factor).

Chapter 6 TDL Models for WSSUS Channels

125

TDL models. We will indistinctly refer to the oversampled TDL model having a
tap spacing of = T /2 as the T /2spaced TDL model or as the HSSTDL.
Figure 6.3 illustrates the advantages of using T /2-spaced TDL models by
comparing the absolute value of the FCF of the reference model with that of the
FCFs of the T -spaced and T /2-spaced TDL models. This figure demonstrates
that the aliasing effects are less severe for the T /2-spaced TDL model than for
the T -spaced TDL models.
As a numerical example of the benefits of using a T /2-spaced TDL model
for system evaluation, let us consider again the MC-CDMA system described
in Section 6.4. We have recalculated the BEP of this system under the same
considerations as in Section 6.4, but with the difference that = T /2, meaning
~ C1281 with N = 33. The
that the tap spacing is equal to = 25 ns and H
obtained results are shown in Figs. 6.4.
It is clear from Figs. 6.4(a), 6.4(b), and 6.4(c), that the simulation system

~y matches better the BEP of the reference system ~y when the T /2spaced TDL
model is used. What is more, we can observe from the results obtained for a
spreading factor of 64 that the difference between the BEP of ~y and ~y is considerably smaller than when we use the T spaced TDL model (see Figs. 6.2(a),
6.2(b), and 6.2(c)). The reason for such an impressive improvement is that the
approximation RH H () RH H () is valid for [W s, W s] when we use a
T /2-spaced TDL model. We shall mention, nonetheless, that the T /2-spaced
TDL model has been introduced mainly as a solution to enable the approximation of the FCF RH H () along the relevant range of frequencies of the system
Ws Ws . It does not remove completely the aliasing effects observed
within [W/2, W/2], which explains the slight differences between the BEP
of the reference system and the simulation system when = 30 ns and MSF = 64
[see Fig. 6.4(c)]. Other solutions, more sophisticated than the one discussed here,
are necessary to neutralize the aliasing effects, e.g., see [Pae02b].

6.6

Chapter Summary and Conclusions

In this chapter, we showed that important inconsistencies arise when SSTDL

models, or -spaced TDL models in general, are used within the context of
band-limited systems to model WSSUS channels. Such inconsistencies involve
the violation of the US condition. Nevertheless, a -spaced TDL model is a
valid simulation model for WSSUS channels if it is considered out of the context
of band-limited systems.
We showed that the FCF of a -spaced TDL model is a periodic function
with a period equal to W = 1/ . Because of this characteristic, a -spaced
TDL models provides just an approximation to the FCF of the reference model,

126

Channel Simulation Models for Mobile Broadband Communications

(a)
1.4

Absolute value of the FCF

1.2

FCF of the reference model

FCF of the T -spaced TDL model (W = Ws )
FCF of the T /2-spaced TDL model (W = 2Ws )

= 30 ns

0.8

0.6

0.4

Ws
Ws /2

0.2

0
30

20

10

10

20

30

Frequency separation, f (MHz)

(b)
1.4

Absolute value of the FCF

1.2

FCF of the reference model

FCF of the T -spaced TDL model (W = Ws )
FCF of the T /2-spaced TDL model (W = 2Ws )

= 60 ns

0.8

0.6

0.4

Ws
Ws /2

0.2

0
30

20

10

10

20

30

Frequency separation, f (MHz)

(c)
1.4

Absolute value of the FCF

1.2

FCF of the reference model

FCF of the T -spaced TDL model (W = Ws )
FCF of the T /2-spaced TDL model (W = 2Ws )

= 105 ns

0.8

0.6

0.4

Ws
Ws /2

0.2

0
30

20

10

10

20

30

Frequency separation, f (MHz)

Figure 6.3: Comparison between the absolute value of the FCF of the reference
model and the absolute value of the FCF of the HSSTDL by considering the TED-PDP with max = 800 ns, Ws = 20 MHz, and
different values for the decay factor of the PDF.

Chapter 6 TDL Models for WSSUS Channels

127

(a)
1

Bit error probability, BEP

10

Solid lines: BEP by using the reference model

Dashdotted lines: BEP by using the T /2-spaced TDL model

10

10

= 30 ns

10

10

10

SF
SF
SF
SF
SF
SF
SF

=
=
=
=
=
=
=

1
2
4
8
16
32
64

10

15

20

25

30

Bit energy vs Noise density, Eb /N0 (dB)

(b)
1

Bit error probability, BEP

10

Solid lines: BEP by using the reference model

Dashdotted lines: BEP by using the T /2-spaced TDL model

10

10

= 60 ns

10

10

10

SF
SF
SF
SF
SF
SF
SF

=
=
=
=
=
=
=

1
2
4
8
16
32
64

10

15

20

25

30

Bit energy vs Noise density, Eb /N0 (dB)

(c)
1

Bit error probability, BEP

10

Solid lines: BEP by using the reference model

Dashdotted lines: BEP by using the T /2-spaced TDL model

10

10

= 105 ns

10

10

10

SF
SF
SF
SF
SF
SF
SF

=
=
=
=
=
=
=

1
2
4
8
16
32
64
10

15

20

25

30

Bit energy vs Noise density, Eb /N0 (dB)

Figure 6.4: Comparison between the theoretical and the empirical BEPs of a
MUI-free downlink MC-CDMA system with MRC by considering a
T /2-spaced TDL model and the TED-PDP with different values of
the decay factor (SF stands for the spreading factor).

128

Channel Simulation Models for Mobile Broadband Communications

which is valid only within [W/2, W/2]. We also showed that such an
approximation is affected by aliasing effects.
The relation RH H () RH H () does not hold for || > W/2 when = T .
For this reason a SSTDL model is not able to emulate the FCF of the reference
model within the frequency range of the systems bandwidth. This drawback seriously affects the performance analysis of wireless communication systems sensitive
to the FCF.
To avoid the problems caused by T -spaced TDL models, we proposed the use
of T /2-spaced TDL channel models, which are obtained by sampling the CIR at a
rate equal to 2Ws , i.e., = T /2. A T /2-spaced TDL model is more resilient to
the aliasing effects than a T -spaced TDL model provided that the FCF RH H ()
of the reference model approaches to zero as . Another advantage of
T /2 spaced TDL models is that they enable the approximation of the reference
models FCF along the complete frequency range of the system bandwidth. For
this reason, using T /2-spaced TDL models instead of T -spaced TDL models turns
out to be of great advantage for system evaluation purposes.

Chapter 7

Summary of Contributions and

Open Problems
7.1

Contributions

The contributions of this doctoral project can be summarized as follows:

We presented a thorough analysis of the statistical properties of a stochastic SOC simulation model for mobile Rician fading channels having a timevariant LOS component. We analyzed the correlation and spectral properties of the SOC simulator, as well as its envelope and phase PDFs.
We provided closed-form solutions for the squared envelope ACF of the
simulation model. Closed-form expressions were derived for the squared
envelope ACF of the ensemble and also for the squared envelope ACF of
the sample functions. For this latter function, we provided solutions not
only for the case where the IQ components of the SOC model are crosscorrelated, but also for the case where they are mutually uncorrelated. The
correctness of such expressions was demonstrated by simulations.
We investigated the stationarity properties of the SOC simulation model.
In this respect, we showed that the simulation model is a WSS process
provided that its LOS component is time-invariant or equal to zero.
We analyzed the ergodicity of the stochastic SOC simulation model with
respect to the mean value, the power, and the ACF. We found out that
the simulation model is always power ergodic and autocorrelation ergodic,
but the mean-ergodicity property holds only when the LOS component is
time-invariant or equal to zero.
We also analyzed the mean- and the autocorrelation-ergodicity properties of
the simulation models squared envelope. We showed that the squared en129

130

Channel Simulation Models for Mobile Broadband Communications

velope of the SOC model is a mean-ergodic and an autocorrelation-ergodic
process only when its IQ components are uncorrelated.
We presented a generalized version of the MEA that is well-suited for the
design of SOC simulators for narrowband SISO fading channels characterized by any type of DPSDs. We called such a generalized method the
GMEA.
We proposed the RSAM as a suitable method for the design of SOC simulators for narrowband SISO fading channels with arbitrary DPSDs.
We extended the GMEA and the RSAM with respect to the design of SOC
simulators for single-bounce scattering narrowband mobile MIMO fading
channels.
We revisited the concept of the SSTDL model for WSSUS channels. We
showed that, from the perspective of band-limited systems, the SSTDL
model violates the US condition of the channel. However, the SSTDL model
is compatible with the US condition when the model is considered out of
the context of band-limited systems. It was shown that a SSTDL suffers
from strong limitations in emulating the FCF of WSSUS channels. Such
limitations seriously affect the performance analysis of wireless communications systems sensitive to the FCF. A simple solution to avoid the problems
produced by SSTDL models was presented by reducing the tap spacing by
a factor of two.

7.2

Open Problems

We close this chapter by noticing that closed-form solutions are still lacking for
the LCR and the ADF of SOC channel simulation models. Closing this gap is
important to determine whether or not a SOC model is able to reproduce the
fading coherence of the envelope of mobile fading channels. On the other hand,
in this dissertation we focused our attention on the design of SOC simulation
models for mobile fading channels in two dimensional single-bounce scattering
environments. Further research is necessary to enable the SOC simulation of
fading channels in three dimensional multiple-bounce scattering scenarios. This
is particularly relevant to investigate the effects that the spatial polarization of
the received signal exerts on the system performance.

Appendix A

Derivation of the Squared

Envelope ACF of the
Reference Model
In this appendix, we derive a closed-form expression for the ACF r 2 2 (t1 , t2 ) ,
E{ 2 (t1 ) 2 (t2 )} of the squared envelope 2 (t) , | (t)|2 of the narrowband Rician
fading channel model described by the complex Gaussian process (t) = (t) +
m (t). The obtained solution is valid irrespective of the PDF characterizing the
AOA statistics of the channels multipath components.
(t)} = 0
From the definition of r 2 2 (t1 , t2 ), and taking into account that E{
[see Sec. 2.2], one can easily verify that
r 2 2 (t1 , t2 ) , E{ 2 (t1 ) 2 (t2 )}


(t1 ) + m (t1 )|2 |
(t2 ) + m (t2 )|2
= E |


(t1 )|2 + |m (t1 )|2 + (t1 ) m (t1 ) + (t1 ) m (t1 )
= E |


(t2 )|2 + |m (t2 )|2 + (t2 ) m (t2 ) + (t2 ) m (t2 )
|
= r 2 2 (t1 , t2 ) + 4 + 22 2 + w,m (t1 , t2 )
+u,m (t1 , t2 ) + u ,m (t2 , t1 )

(A.1)

where r 2 2 (t1 , t2 ) , E{ 2 (t1 ) 2 (t2 )} is the ACF of the squared envelope 2 (t) ,
(t)|2 of the channels diffuse component (t), whilst the functions w ,m (t1 , t2 )
|
and u ,m (t1 , t2 ) are equal to
w ,m (t1 , t2 ) = E



(t1 )m (t1 ) + (t1 )m (t1 )



(t2 )m (t2 ) + (t2 )m (t2 )



(t1 )|2 (t2 )m (t2 ) + (t2 )m (t2 ) .
u ,m (t1 , t2 ) = E |
131

(A.2)
(A.3)

132

Channel Simulation Models for Mobile Broadband Communications

A compact expression for r 2 2 (t1 , t2 ) can be obtained by noticing that
r 2 2 (t1 , t2 ) , E{ 2 (t1 ) 2 (t2 ) }

(t2 )|2 }
(t1 )|2 |
= E{ |

I (t1 )2 + Q (t1 )2 ] [
I (t2 )2 + Q (t2 )2 ] }
= E{ [

2I (t2 )} + E{
2Q (t1 )
2Q (t2 )}
2I (t1 )
= E{

2I (t1 )
2Q (t2 )} + E{
2Q (t1 )
2I (t2 )}
+E{

(A.4)

where I (t) and Q (t) are the inphase and quadrature components of (t), respectively. The expectations in (A.4) satisfy the relationships [Auli79]
2I (t1 ) 2I (t2 )} = E{
2I (t1 )} E{
2I (t2 )} + 2E 2 {
I (t1 )
I (t2 )}
E{

2Q (t1 ) 2Q (t2 )}
E{
2I (t1 ) 2Q (t2 )}
E{
2Q (t1 ) 2I (t2 )}
E{

=
=
=

2Q (t1 )} E{
2Q (t2 )} + 2E 2 {
Q (t1 )
Q (t2 )}
E{
2I (t1 )} E{
2Q (t2 )} + 2E 2 {
I (t1 )
Q (t2 )}
E{
2
2
2
Q (t1 )} E{
I (t2 )} + 2E {
Q (t1 )
I (t2 )}.
E{

(A.5a)
(A.5b)
(A.5c)
(A.5d)

Hence, by taking account of the autocorrelation and cross-correlation properties

of I (t) and Q (t) [see (2.11)], we find that
2I (t)} + E 2 {
2Q (t)} + E{
2I (t)}E{
2Q (t)}
r 2 2 (t1 , t2 ) = E 2 {
h
2Q (t)}E{
2I (t)} + 2 E 2 {
I (t1 )
I (t2 )}
+E{

i
Q (t1 )
Q (t2 )} + E 2 {
I (t1 )
Q (t2 )} + E 2 {
Q (t1 )
I (t2 )}
+E 2 {


= 4r2 I I (0) + 2 r2 I I ( ) + r2 Q Q ( ) + r2 I Q ( ) + r2 Q I ( )


= 4 + 4 r2 I I ( ) + r2 I Q ( )
= 4 + |r ( )|2 .

(A.6)

This result indicates that the ACF of 2 (t) is not influenced by the choice of the
time origin, so that r 2 2 (t1 , t2 ) = r 2 2 ( ).
The function w,m (t1 , t2 ), on the other hand, may be rewritten as follows
w,m (t1 , t2 ) = E



(t1 )m (t1 ) + (t1 )m (t1 )



(t2 )m (t2 ) + (t2 )m (t2 )


(t1 )m (t1 )} Re{
(t2 )m (t2 )}
= 4 E Re{


I (t1 ) + sin(2f t1 + )
Q (t1 )
= 42 E cos(2f t1 + )


I (t2 ) + sin(2f t2 + )
Q (t2 ) . (A.7)
cos(2f t2 + )

From (A.7), we obtain after straightforward calculations the result

w ,m (t1 , t2 ) = w ,m ( )

(A.8)

Appendix A Squared Envelope ACF of the Reference Model



= 42 cos(2f )r I I ( ) + sin(2f )r I Q ( ) .

133
(A.9)

In turn, the function u ,m (t1 , t2 ) may be expressed as




(t1 )|2 (t2 )m (t2 ) + (t2 )m (t2 )
u ,m (t1 , t2 ) , E |


(t1 )|2 Re{
(t2 )m (t2 )}
= 2 E |

 2
I (t2 )
I (t1 ) + 2Q (t1 )] cos(2f t2 + )
= 2 E [

Q (t2 )
+ sin(2f t2 + )
n
h 


i
I (t2 ) + E 2Q (t1 )
I (t2 )
= 2 cos(2f t2 + ) E 2I (t1 )
h 


io
Q (t2 ) + E 2Q (t1 )
Q (t2 )
+ sin(2f t2 + ) E 2I (t1 )
.

(A.10)

It can be shown that the four expectations in (A.10) are equal to zero, so that
u,m (t1 , t2 ) = 0.

(A.11)

In order to demonstrate the veracity of this result, let us explicitly evaluate the


I (t2 ) . To this end, we recall that
case E 2I (t1 )


I (t) = Re (t) =

lim

N
X

c n cos(2ff n t + n ).

(A.12)

n=1



I (t2 ) can then be arranged as an infinite series
The expectation E 2I (t1 )


I (t2 )
E 2I (t1 )
=

lim

N X
N X
N
X

n=1 m=1 p=1

hn,m,p (t1 , t2 )} (A.13)

E{ccnc mc p } E{h

where
h n,m,p (t1 , t2 ) =


cos(2ff n t1 ) cos( n ) sin(2ff n t1 ) sin( n )



cos(2ff m t1 ) cos( m ) sin(2ff m t1 ) sin( m )


cos(2ff p t2 ) cos( p ) sin(2ff p t2 ) sin( p ) .

(A.14)

The average value of h n,m,p (t1 , t2 ) with respect to the random phases proves to
be equal to zero, since E{cos( n ) cos( m ) cos( p )} = E{sin( n ) sin( m ) sin( p )} =
E{cos( n ) cos( m ) sin( p )} = E{sin( n ) sin( m ) cos( p )} = 0 for all combinations
of n, m, and p (even for n = m = p). In view of this, we can state that
hn,m,p (t1 , t2 )} = 0,
E{h

n, m, p



which in turn implies that E 2I (t1 ) I (t2 ) = 0.

(A.15)

134

Channel Simulation Models for Mobile Broadband Communications





Analogously, it can be proved that E 2Q (t1 ) I (t2 ) = E 2I (t1 ) Q (t2 ) =


E 2Q (t1 ) Q (t2 ) = 0 (we will omit the details for reasons of brevity). Consequently, u,m (t1 , t2 ) = u ,m (t2 , t1 ) = 0.
By substituting the results presented in (A.6), (A.9), and (A.11)) into (A.1),
we finally obtain the expression

r 2 2 (t1 , t2 ) = r 2 2 ( ) + 4 + 22 2 + 42 rI I ( ) cos (2f )

+rI Q ( ) sin (2f ) .

(A.16)

We notice that r 2 2 (t1 , t2 ) is time-shift insensitive, i.e., r 2 2 (t1 , t2 ) = r 2 2 ( ).

Equation (2.39) follows from (A.16) by rewriting r 2 2 ( ) in terms of the Rician
factor cR = 2 /2 .

Appendix B

Derivation of the Squared

Envelope ACF of the
Stochastic SOC-Based
Simulation Model
In this appendix, we derive a closed-form solution for the ACF r22 (t1 , t2 ) ,


E 2 (t1 )2 (t2 ) of the squared envelope 2 (t) , |
(t)|2 of the stochastic SOCbased simulation model described by the random process (t) =
(t) + m (t).
Starting from the definition of r22 (t1 , t2 ), one can show that [cf. Appx. A]
r22 (t1 , t2 ) = r22 (t1 , t2 ) + 4 + 22 2 + w ,m (t1 , t2 )
+u ,m (t1 , t2 ) + u ,m (t2 , t1 )

(B.1)

where r22 (t1 , t2 ) , E{2 (t1 )2 (t2 )} is the ACF of the squared envelope 2 (t) ,
(t)|2 of the simulation models random component
|

(t), whereas the functions

w ,m (t1 , t2 ) and u ,m (t1 , t2 ) are defined as
w ,m (t1 , t2 ) , E



(t1 )m (t1 ) +
(t1 )m (t1 )




(t2 )m (t2 ) +
(t2 )m (t2 )



(t1 )|2
u ,m (t1 , t2 ) , E |

(t2 )m (t2 ) +
(t2 )m (t2 ) .

(B.2)
(B.3)

One can easily verify, by proceeding as we did in Appendix A to obtain the results
presented in (A.9) and (A.11), that
u ,m (t1 , t2 ) = 0

(B.4)

w ,m (t1 , t2 ) = w ,m ( )


= 42 cos(2f ) r I I ( ) + sin(2f ) r I Q ( ) (B.5)
135

136

Channel Simulation Models for Mobile Broadband Communications

where r I I ( ) and r I Q ( ) are the correlation functions defined in (3.7). In

addition, it is straightforward to show that
r22 (t1 , t2 ) =

N X
N X
N X
N
X
l=1 m=1 n=1 p=1



cl cm cn cp exp j2(fl fm )t1





exp j2(fn fp )t2 E exp j(l m + n p ) . (B.6)

Since the random phases n are mutually independent and uniform over [, )
[Sec. 3.3], the expectation in (B.6) is different from zero only when: l = m = n =
p; l = m, n = p, l 6= n; and l = p, m = n, l 6= m. Consequently, we have
N
X

r22 (t1 , t2 ) =

c4l

Case: l=m=n=p

p=1 q=1
q6=p

N
X

c2m c2n

m=1 n=1
n6=m

|l=1
{z }

N X
N
X

N X
N
X

{z

Case: l=m,n=p,l6=n





c2p c2q exp j2 fp (t2 t1 ) exp j2 fq (t2 t1 )
{z

Case: l=p,m=n,l6=m

c4l

"

N
X

m=1
l=1
N
N
XX
+
c2p c2q
p=1 q=1

c2m

#2

N
X

c4n

n=1

exp{j2 fp } exp{j2 fq }

N
X

c4k .

(B.7)

k=1

It follows from (B.7) that the ACF of 2 (t) is time shift insensitive, that is,
P
r22 (t1 , t2 ) = r22 ( ). Taking account of the property N
2n = 2 [Sec. 3.3.1],
n=1 c
PN 2
and given that r ( ) =
c exp{j2 fn } [Eq. (3.5)], we finally obtain
n=1 n

r22 ( ) = 4 + |r ( )|2

N
X

c4n .

(B.8)

n=1

By substituting the results presented in (B.4), (B.5), and (B.8) into (B.1), we
find that the ACF of the simulation models squared envelope 2 (t) is equal to
r22 (t1 , t2 ) = r22 ( )

= r22 ( ) + 4 + 22 2 + 44 cos(2f ) r I I ( )

+ sin(2f ) r I Q ( )
(B.9)

where r22 ( ) is given by (B.8). The expression presented in (3.29) follows from
(B.9) by rewriting r22 ( ) in terms of the Rician factor cR .

Appendix C

Derivation of the Squared

Envelope ACF of the
Deterministic SOC-Based
Simulation Model
In this appendix, we compute a closed-form solution for the time-averaged ACF
r2 2 ( ) of the squared envelope 2 (t) , |
(t)|2 of the deterministic SOC-based
simulation model (t) =
(t) + m (t) introduced in Section 3.4.1. To that end,
we start by noticing that
r2 2 ( ) ,

|
(t)|2 |
2 (t + )|2

= r2 2 ( ) + 4 + 22 2 + u,m (t, t + ) + u,m (t + , t)

+z,m (t, t + ) + z,m (t + , t) + w,m ( )

(C.1)


where r2 2 ( ) , 2 (t)2 (t + ) is the time averaged ACF of the squared envelope 2 (t) , |
(t)|2 of
(t), and:
u,m (t, t + ) ,
u,m (t + , t) ,
z,m (t, t + ) ,
=
z,m (t + , t) ,
=
w,m ( ) ,



|
(t)|2
(t + ) m (t + ) +
(t + ) m (t + )
(C.2)



|
(t + )|2
(t) m (t) +
(t) m (t)
(C.3)



2

|m (t)|
(t + ) m (t + ) +
(t + ) m (t + )



2
(t + ) m (t + ) +
(t + ) m (t + )
(C.4)



2

|m (t + )|
(t) m (t) +
(t) m (t)



(t) m (t) +
(t) m (t)
(C.5)
2




(t) m (t) +
(t) m (t)



(t + ) m (t) +
(t) m (t + ) .
(C.6)
137

138

Channel Simulation Models for Mobile Broadband Communications

In order to obtain a compact expression for r2 2 ( ), we need first to find a closedform solution for r2 2 ( ) and the time-averaged functions defined in (C.2)(C.6).
For that purpose, we will make the following assumptions regarding the Doppler


PN
(t) =
n exp j(2 fn t + n ) and the
frequencies fn of the SOC model
n c


Doppler frequency f of the specular wave m (t) = exp j(2f t + ) :
6 0, n
fn =
fn =
6 fm , n =
6 m

Condition 3.2 :
Condition 3.3 :

If N 4, then

l = m = n = k;
fl + fm = fn + fk , iff
or l = n, m = k, l 6= m;

or l = k, m = n, l 6= m.
Condition 3.6 :
|f | =
6 |fn |, n
Condition 3.7 :
If N 2, then f + fl 6= fm + fn , l, m, n.
Condition 3.5 :

(C.7)
(C.8)

(C.9)
(C.10)
(C.11)
(C.12)

In the case of the function u,m (t, t + ), we have

u,m (t, t + ) =
=



|
(t)|2
(t + ) m (t + ) +
(t + ) m (t + )

N X
N X
N
X

l=1 m=1 n=1

cl cm cn exp{j2(f fn ) }

exp{j(l m n + )}
ZT
1
lim
exp{j2(fl fm fn + f )t} dt
T 2T
T

N X
N X
N
X
p=1 q=1 k=1

cp cq ck exp{j2(fk f ) }

exp{j(p q + k + )}
ZT
1
lim
exp{j2(fp fq + fk f )t} dt.
T 2T
T

The two integrals above are equal to zero in the limit T if the condition
established in (C.12) is satisfied. Consequently, we can state that
u,m (t, t + ) = 0,

if f + fl 6= fm + fn , l, m, n.

(C.13)

Analogously, one can show that the time average u,m (t + , t) equals
u,m (t + , t) = 0,

if f + fl 6= fm + fn , l, m, n.

(C.14)

Appendix C Squared Envelope ACF of the Deterministic SOC Model

139

Notice that u,m (t, t + ) = u,m (t + , t) upon fulfillment of the Condition 3.7.
However, such a relationship is in general not valid.
On the other hand, for the time-averaged function z,m (t, t + ), we have





z,m (t, t + ) = 2
(t + ) m (t + ) +
(t + ) m (t + )
" N
X
= 3
cl exp{j2(f fl ) } exp{j( l )}
l=1

1
lim
T 2T
+

N
X

m=1

ZT

exp{j2(f fl )t} dt

cm exp{j2(fm f ) } exp{j(m )}

1
lim
T 2T

ZT

exp{j2(fm f )t} dt .

If the Condition 3.6 is met, then the integrals introduced in the previous expression are equal to zero in the limit T , implying that
z,m (t, t + ) = 0,

if |fn | =
6 |f | n.

(C.15)

if |fn | =
6 |f | n.

(C.16)

In a similar way, one can show that

z,m (t + , t) = 0,

It is worth noticing that the results presented in (C.15) and (C.16) hold even
when the absolute value is removed from the Condition 3.6, i.e., if fn 6= f n.
In turn, for the function w,m ( ), it can be shown that if the Condition 3.6
is fulfilled, then


(t) m (t) +
(t) m (t)


(t + ) m (t + )

(t + ) m (t) +
h


= 2 exp{j2f }
(t)
(t + )

i
+ exp{j2f }
(t)
(t + )


= 22 Re exp{j2f } r ( )


= 42 cos(2f ) rI I ( ) + sin(2f ) rI Q ( )

w,m (t, t + ) =

(C.17)

where r ( ), rI I ( ), and rI Q ( ) denote, in that order, the ACF of

(t)
[see (3.33)], the ACF of the inphase component of
(t) [see (3.35a)], and the
CCF between the IQ components of
(t) [see (3.35b)].

140

Channel Simulation Models for Mobile Broadband Communications

Finally, for the time-averaged ACF r2 2 ( ) of 2 (t), we have
r2 2 ( ) =
=

|
(t)|2 |
(t + )|2

N X
N X
N X
N
X

cl cm cn cp

l=1 m=1 n=1 p=1





exp j(l m + n p ) exp j2(fn fp )

1
lim
T 2T

ZT

exp{j2(fl fm + fn fp )t} dt.

(C.18)

If the condition stated in (C.9) is met, then the integral in (C.18) will be different
from zero only when: l = m = n = p; l = n, m = p, l 6= m; l = p, m = n, l 6= n.
In such a case, we obtain [cf. Appx. B]
N
X

r2 2 ( ) =

c4l

Case: l=m=n=p
N X
N
X
p=1 q=1
q6=p

N
X

c2m c2n

m=1 n=1
n6=m

|l=1
{z }

N X
N
X

{z

Case: l=m,n=p,l6=n





c2p c2q exp j2 fp (t2 t1 ) exp j2 fq (t2 t1 )
{z

Case: l=p,m=n,l6=m

c4l +

"

N
X

m=1
l=1
N
N
XX
+
c2p c2q
p=1 q=1

c2m

#2

N
X

c4n

n=1

exp{j2 fp } exp{j2 fq }

= 4 + |r ( )|2

N
X

c4n .

N
X

c4k

k=1

(C.19)

n=1

On the basis of the results presented in (C.1), (C.13)(C.17), and (C.19), we

can conclude that

r2 2 ( ) = r2 2 ( ) + 4 + 22 2 + 44 cos(2f ) rI I ( )

+ sin(2f ) rI Q ( )
(C.20)

if the conditions established in (C.7)(C.12) are fulfilled. Notice that the expression in (C.20) does not depend on the set of phases {n , }. However, without
going into details, we observe that this characteristic does not hold if any of the
conditions in (C.7)(C.12) is not met.

Appendix D

Closed-Form Expression for

the Squared Envelope ACF of
the Sample Functions of the
Stochastic Homogeneous SOC
Model with Uncorrelated IQ
Components
In this appendix, we outline the derivation of the time-averaged ACF r2 2 ( ) ,


2
(t)|2 of the deterministic
(t) 2 (t + ) of the squared envelope 2 (t) , |
process
(t) characterizing the sample functions of the stochastic homogeneous
SOC model
(t). We do so under the assumption that the Conditions 3.13.3 are
fulfilled, meaning that:
Condition 3.1 :

Condition 3.2 :
Condition 3.3 :

The number N of cisoids in

(t) is even, i.e., N = 2M,
+
M Z , and for each pair of parameters (
cn , fn ) there
exist one and only pair (
cm , fm ), m 6= n, such that
cn = cm and fn = fm .

fn =
6 0, n.
fn =
6 fm , n 6= m.

Notice that if these conditions are satisfied, then the IQ components

I (t) and

Q (t) of
(t) are uncorrelated [cf. (3.6) and (3.7)]. For notational convenience,
and without compromising the generality of our results, we will furthermore assume that the cisoids parameters are indexed in such a way that fn < fm n < m.
141

142

Channel Simulation Models for Mobile Broadband Communications

Our starting point is the expression given in (C.18) for r2 2 ( ), the which we
present again for completeness:

r2 2 ( ) =

|
(t)|2 |
(t + )|2

N X
N X
N X
N
X

cl cm cn cp

l=1 m=1 n=1 p=1





exp j(l m + n p ) exp j2(fn fp )

1
lim
T 2T

ZT

exp{j2(fl fm + fn fp )t} dt.

(D.1)

For the case analyzed in Appendix C, the integral in (D.1) is different from zero
in the limit T only when: l = m = n = p; l = n, m = p, l 6= m; and
l = p, m = n, l 6= n. However, for the case at hand, the integral is different
from zero also whenever l 6= m 6= n 6= q and: fl = fn , fm = fp; fl = fn ,
fp = fm; fn = fl , fm = fp ; and fn = fl , fp = fm . Under such
circumstances, we have
r2 2 ( ) = 4 + |r ( )|2
N/2

X
l=1

m=1

N/2

c2n

n=1

N/2
X
p=1

q=1

X
k=1

n=1




c2m exp j m + N m+1 exp j2 fm


exp j n + N n+1

exp j2 fn






c2p exp j p + N p+1 exp j2 fp

( N/2
X
N/2

c4n




c2l exp j l + N l+1 exp j2 fl

( N/2
X
X

N
X






c2q exp j q + N q+1 exp j2 fq


c2k exp j k + N k+1

exp j2 fk

where r ( ) is the ACF defined in (3.8). One can easily verify that
r2 2 ( ) = 4 + |r ( )|2

N
X

n=1

c4n

(D.2)

Appendix D Squared Envelope ACF: Uncorrelated IQ Components

N/2 N/2
X
X

l=1 m=1

143



c2l c2m exp j(l + N l+1 m N m+1 )

h


i

exp j2(fl fm ) + exp j2(fl + fm )
+

N/2 N/2
X
X
n=1 p=1



c2n c2p exp j(n + N n+1 p N p+1 )

h



i
exp j2(fn + fp ) + exp j2(fn fp )

= 4 + |r ( )|2
N/2 N/2

XX

l=1 m=1

N
X

c4n

n=1



c2l c2m exp j(l + N l+1 m N m+1 )




exp j2(fl fm ) + exp j2(fl + fm )



i
+ exp j2(fl + fm ) + exp j2(fl fm )
4

+ |r ( )|
N/2 N/2

+4

XX

l=1 m=1
m6=l

N
X

c4n

n=1



c2l c2m exp j(l + N l+1 m N m+1 )

cos 2 fl cos 2 fm .

Hence, by taking into account the identity

PN
2
m=1 |xm | , we can finally write
r2 2 ( ) =

+ |r ( )|

N
X

(D.3)

PN

n=1

PN

m=1 xn xm
m6=n

P
2


= N
x

n
n=1

c4n

n=1

( N/2

X



2

+4 
c2l exp j(l + N l+1 ) cos 2 fl 

l=1

N/2

m=1

c4m

cos

2 fm



(D.4)

144

Channel Simulation Models for Mobile Broadband Communications

Appendix E

Theorems About the Design of

SOC-Based Simulation Models
with Uncorrelated IQ
Components
Theorem E.1 Let the DPSD of the channels diffuse component, S (f ), be a
symmetrical and continuous function in (fmax , fmax ], and suppose that the gains
n of the SOC model described by the complex random process
cn and the AOAs

(t) are given such that

cn =
Z n

g ()d =

,
N

n = 1, . . . , N

(E.1)

1
,
2N

n = 2, . . . , N.

(E.2)

n1

Then, the IQ component

I (t) and
Q (t) of
(t) are mutually uncorrelated if and
only if
Z 1

g ()d =

1
.
4N

(E.3)

Proof To prove this theorem, we start by noticing from (E.2) that the deterministic AOAs
n fulfill the relationships 0 <
n < n and
n 6=
n n 6= m.
Consequently, the Doppler frequencies fn , fmax cos(
n ) of
(t) satisfy fn 6= fm
for all n 6= m. Now, if fn 6= fm and cn = cm for all n 6= m (as in the present
145

146

Channel Simulation Models for Mobile Broadband Communications

case), then I (t) and Q (t) are mutually uncorrelated if and only if [see (3.7b)]
fn = fN n+1 ,

n = 1, . . . , N.

(E.4)

For notational convenience, and without loss of generality, we assume that the
Doppler frequencies are indexed such that fn < fm for n < m. The equality
fn = fN n+1 implies that the underlying AOAs satisfy
n =
N n+1 , where

n (0, ). In turn, the symmetry of S (f ) implies that g () = g ( ), for

(0, ]. Consequently, if S (f ) = S (f ) and the Doppler frequencies fn
satisfy (E.4), then
Z n

g ()d =

g ()d,

n.

(E.5)

Nn+1

Thus, to prove the theorem, it is sufficient to demonstrate that if the AOAs

n
fulfill (E.2), then the previous equality holds if and only if (E.3) is met. The
proof follows immediately by noticing that the requirement in (E.2) entails that
Z n

g ()d =

n1
+
2N

g ()d =

Nn+1

R 1
0

g ()d

Nn+1
Z

2N

Obviously, (E.6) equals (E.7) if and only if

Z 1

(E.6)

g ()d

Z 1

g ()d.

(E.7)

g ()d = 1/(4N ).

Theorem E.2 If the DPSD of the channels diffuse component, S (f ), is symmetrical, and the parameters of the SOC simulator described by the random process
(t) are computed by following the RSAM, then the IQ components of
(t)
are mutually uncorrelated.
Proof To prove this theorem, it is sufficient to demonstrate that the Doppler
frequencies fn and the gains cn of
(t) satisfy the following condition when the
RSAM is applied and S (f ) is symmetrical:
Condition E.1 For each pair of parameters (
cn , fn ), where fn 6= 0, there exists

one and only one pair (

cm , fm ), n 6= m, such that fn = fm and cn = cm .

Appendix E SOC Simulators with Uncorrelated IQ Components

147

Notice that the above is a more general condition for the uncorrelatedness of the
IQ components of
(t) than the Condition 3.1 stated in Section 3.3.3.1.
The proof of the theorem follows immediately by noticing that if S (f ) =
S (f ), and assuming that the even part g () of the AOA distribution has at
most one maximum in [0, ), then the points and u at which g () crosses


a threshold 0, sup{g ()}[0,) from down to up and from up to down,
respectively, meet the relation = u . Consequently, the AOAs
n obtained
by applying the RSAM satisfy the equation [cf. (4.31)]

n =
N n+1 ,

n = 1, . . . , N.

(E.8)

Taking account of the relationship fn = fmax cos(

n ), and given that g () =
g ( ) when S (f ) is symmetrical, we find that [see 4.28]
n )
fn = fmax cos(
= fmax cos(
N n+1 )

cn

N n+1 )
= fmax cos(
= fN n+1 ,
s
g (
n )
= PN
m )
m=1 g (
s
g (
N n+1 )
=
PN
m )
m=1 g (
s
g (
N n+1 )
= PN
m )
m=1 g (
= cN n+1 ,

n = 1, . . . , N

n = 1, . . . , N.

(E.9)

(E.10)

We recall that the Doppler frequencies fn produced by the RSAM fulfill fn 6= fm

for n 6= m [see Sec. 4.4.1.3]. Bearing this in mind, and in view of (E.9) and
(E.10), we can state that the Condition E.1 is fulfilled when the RSAM is applied
and S (f ) = S (f ). This concludes the proof.

148

Channel Simulation Models for Mobile Broadband Communications

Appendix F

Validation of the Expressions

Obtained for the Squared
Envelope ACF of the
Deterministic SOC-Based
Simulation Model
In order to demonstrate the correctness of the analytical expression presented in
Section 3.4.4 for the ACF r2 2 ( ) of the squared envelope 2 (t) of the stochastic
SOC models sample functions, we present in Figs. F.1F.3 a comparison between
the measured ACF of 2 (t) and the solution given in (3.40) for r2 2 ( ). The
graphs of |r2 2 ( )| depicted in those figures were generated by applying the
RSAM with N = 20 to the von Mises PDF of the AOA. The results presented
in Figs. F.1F.3 illustrate the case where the IQ components of the simulation
models random component are mutually uncorrelated as well as the case where
they are cross-correlated. One can observe from Figs. F.1F.3 that the theoretical
curves of r2 2 ( ) match perfectly the empirical one. This can be taken as a proof
of the correctness of the solutions presented in Section 3.4.4 for r2 2 ( ).

149

150

Channel Simulation Models for Mobile Broadband Communications

(a)

ACF of the squared envelope

m = 0
cR = 0

Theory, Eq. (3.42)

Realization
Theory, Eq. (3.41)
Realization
Theory, Eq. (3.41)
Realization

= 20

1.8

1.6

=5
1.4

N
4fmax

=0
1.2

Normalized time difference, fmax

(b)

ACF of the squared envelope

m = 90
1.8

m = 30

m = 0

= 10
cR = 0

Theory, Eq. (3.41)

Realization
Theory, Eq. (3.41)
Realization
Theory, Eq. (3.42)
Realization

1.6

1.4

1.2

N
4fmax
1

0.8
0

Normalized time difference, fmax

Figure F.1: Comparison between the theoretical and the empirical squared envelope ACFs of the SOC-based simulation models sample functions
by considering a Rician factor equal to cR = 0 and applying the
RSAM to the von Mises PDF of the AOA with different pairs of
parameters m and (fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 ,
and N = 20).

Appendix F Validation of the Solutions for the Squared Envelope ACF

151

(a)
Theory, Eqs. (3.40) and (3.42)
Realization
Theory, Eqs. (3.40) and (3.41)
Realization
Theory, Eqs. (3.40) and (3.41)
Realization

ACF of the squared envelope

1.7

m = 0
cR = 2

1.6

= 20

1.5
1.4

=5

1.3
1.2
1.1
1
0.9
0.8

N
4fmax

=0
0.7
0

Normalized time difference, fmax

(b)

ACF of the squared envelope

1.6
1.5

m = 90

1.4

m = 30

1.3

Theory, Eqs. (3.40) and (3.41)

Realization
Theory, Eqs. (3.40) and (3.41)
Realization
Theory, Eqs. (3.40) and (3.42)
Realization

= 10
cR = 2
m = 0

1.2
1.1
1
0.9

N
4fmax

0.8
0.7
0

Normalized time difference, fmax

Figure F.2: Comparison between the theoretical and the empirical squared envelope ACFs of the SOC-based simulation models sample functions
by considering a Rician factor equal to cR = 2 and applying the
RSAM to the von Mises PDF of the AOA with different pairs of
parameters m and (fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 ,
and N = 20).

152

Channel Simulation Models for Mobile Broadband Communications

(a)

ACF of the squared envelope

1.5

Theory, Eqs. (3.40) and (3.42)

Realization
Theory, Eqs. (3.40) and (3.41)
Realization
Theory, Eqs. (3.40) and (3.41)
Realization

m = 0
cR = 4

1.4

= 20

1.3

=5

1.2

=0

1.1

0.9

N
4fmax

0.8

0.7

Normalized time difference, fmax

(b)

ACF of the squared envelope

1.4

m = 90

1.3

Theory, Eqs. (3.40) and (3.41)

Realization
Theory, Eqs. (3.40) and (3.41)
Realization
Theory, Eqs. (3.40) and (3.42)
Realization

= 10
cR = 4
m = 0

1.2

m = 30
1.1

0.9

N
4fmax

0.8
0

Normalized time difference, fmax

Figure F.3: Comparison between the theoretical and the empirical squared envelope ACFs of the SOC-based simulation models sample functions
by considering a Rician factor equal to cR = 4 and applying the
RSAM to the von Mises PDF of the AOA with different pairs of
parameters m and (fmax = 91 Hz, 2 = 1, fp = 65 Hz, = 0 ,
and N = 20).

Bibliography
[3gpSCM]

3GPP - 3GPP2, Spatial Channel Model Ad-hoc Group3GPP TR 25.996.,

Spatial channel model for multiple input multiple output (MIMO) simulations, v7.0.0, 2003-09.

[3gpUMB]

3GPP2, Physical Layer for Ultra Mobile Broadband (UMB) Air Interface
Specification, C.S0084-001-0, 3GPP2, Aug. 2007.

[3Gweb1]

http://www.3GPP.org

[3Gweb2]

http://www.3GPP2.org

[Abd02a]

Abdi A., Barger J. A., and Kaveh M., A parametric model for the distribution of the angle of arrival and the associated correlation function and
power spectrum at the mobile station, IEEE Trans. Veh. Technol., vol. 51,
no. 3, pp. 425-434, May 2002.

[Abd02b]

Abdi A. and Kaveh M., A space-time correlation model for multielement

antenna systems in mobile fading channels, IEEE J. Sel. Areas Commun.,
vol. 20, no. 3, pp. 550-560, Apr. 2002.

[Abic06]

Abichar Z., Peng Y., and Morris Chang J., WiMAX: The emergence of
wireless broadband, IEEE IT Prof., vol. 8, no. 4, pp. 44-48, Jul.-Aug.
2006.

[Alme07]

Almers P., Bonek E., Burr A., Czink N., Debbah M., et al., Survey of channel and radio propagation models for wireless MIMO systems, EURASIP
Journal on Wireless Communications and Networking, vol. 2007, pp. 1-19,
2007, DOI: 10.1155/2007/19070.

[Aspl06]

Asplud H., Alayon Glazunov A., Molisch A. F., Pedersen K. I., Steinbauer
M., The COST 259 directional channel modelPart II: Macrocells, IEEE
Trans. Wirel. Commun., vol. 5, no. 12, pp. 3434-3450, Dec. 2006.

[Auli79]

Aulin T., A modified model for the fading signal at a mobile radio channel,
IEEE Trans. Veh. Technol., vol. VT-28, no. 3, pp. 182-203, Aug. 1979.

[Badd04]

Baddour K. E. and Beaulieu N. C., Accurate simulation of multiple crosscorrelated Rician fading channels, IEEE Trans. Commun., vol. 52, no. 11,
pp. 1980-1987, Nov. 2004.

[Badd05]

Baddour K. E. and Beaulieu N. C., Autoregressive modeling for fading

channel simulation, IEEE Trans. Wirel. Commun., vol. 4, no. 4, pp. 16501662, Jul. 2005.

153

154

Channel Simulation Models for Mobile Broadband Communications

[Beau04]

Beaulieu N. C. and Merani M. L., Generation of multiple Rayleigh fading sequences with specified cross-correlations, European Transactions on
Telecommunications, vol. 15, no. 5, pp. 471-476, Sep. 2004.

[Bell63]

Bello P. A., Characterization of randomly time-variant linear channels,

IEEE Trans. Commun. Syst., vol. CS-11, no. 4, pp. 360-393, 1963.

[Blau06]

Blaunstein N. and Ben-Shimol Y., Spectral properties of signal fading and

Doppler spectra distribution in urban mobile communication links, Wirel.
Commun. Mob. Comput., vol. 6, no. 1, pp. 113-126, Feb. 2006.

[Bolt07]

Bolton W., Yang X., and Guizani M., IEEE 802.20: mobile broadband
wireless access, IEEE Wirel. Commun. Mag., vol. 14, no. 1, pp. 84-95, Feb.
2007.

[Brig88]

Brigham E. O., The Fast Fourier Transform and its Applications, New Jersey: Prentice Hall, 1988.

[Brow96]

Brown C., Do G., and Feher K., Digital ultrafast carrier recovery for interactive transmission systems, IEEE Trans. Consum. Electron., vol. 42,
no. 1, pp. 132-139, Feb. 1996.

[Cai06]

Cai J., Li Z., Hao Y., and J. Cai., Time-variant Doppler frequency estimation and compensation for mobile OFDM systems, IEEE Trans. Consum.
Electron., vol. 52, no. 2, pp. 336-340, May 2006.

[Chen00]

Chen T. A., Fitz M. P., Kuo W. Y., Zoltowski M. D., and Grimm J. H., A
space-time model for frequency nonselective Rayleigh fading channels with
applications to space-time modems, IEEE J. Sel. Areas Commun., vol. 18,
pp. 1175-1190, Jul. 2000.

[Chiz03]

Chizhik D., Ling J., Wolniansky P. W., Valenzuela R. A., Costa N., et al.,
Multiple-input-multiple-output measurements and modeling in Manhattan, IEEE J. Sel. Areas Commun., vol. 21, no. 3, pp. 321-331, Apr. 2003.

[Choi01]

Choi Y. S., Voltz P. J., and Cassara F. A., On channel estimation and
detection for multicarrier signals in fast and selective Rayleigh fading channels, IEEE Trans. Commun., vol. 49, no. 8, pp. 1375-1387, Aug. 2001.

[Chri02]

Chryssomallis M., Simulation of mobile fading channels, IEEE Antennas

Propag. Mag., vol. 44, no. 6, pp. 172-183, 2002.

[Clar68]

Clarke R. H., A statistical theory of mobile radio reception, Bell Syst.

Tech. J., vol. 47, pp. 957-1000, Jul. 1968.

[Cora97]

Corazza G. E., Degli-Esposti V., Frullone M., and Riva G., A characterization of indoor space and frequency diversity ray-tracing modeling, IEEE
J. on Sel. Areas Commun., vol. 14, no. 3, pp. 411-419, Apr. 1996.

[Corr01]

Correia L. M., Wireless Flexible Personalized Communications, COST 259:

European Co-operation in Mobile Radio Research, Chichester: John Wiley
& Sons, 2001.

[Corr06]

Correia L. M., Mobile Broadband Multimedia Networks: Techniques, Models

and Tools for 4G, London: Academic Press, 2006.

Bibliography

155

[Cos207]

COST 207, Digital land mobile radio communications, Office Official Publ.
Eur. Communities, Final Rep., Luxemburg, 1989.

[Edba89]

Edbauer F., Performance of interleaved trellis-coded differential 8-PSK

modulation over fading channels, IEEE J. Sel. Areas Commun., vol. 7,
no. 9, pp. 1340-1346, Dec. 1989.

[Edwa98]

Edwards C. H. and Penney D. E., Calculus with Analytic Geometry, New

Jersey: Prentice Hall, 5th ed., 1998.

[Erce99]

Erceg V., Michelson D. G., Ghassemzadeh S. S., Greenstein L. J., Rustako

A. J. Jr., et al., A model for the multipath delay profile of fixed wireless
channels, IEEE J. Sel. Areas Commun., vol. 17, no. 3, pp. 399-410 1999.

[Erce04]

Erceg V, Schumacher L, Kyritsi P, Molisch A. F., Baum D. S., et al. TGn

channel models, Tech. Rep. IEEE P802.11, Wireless LANs, Garden Grove,
Calif, USA, 2004.

[Erte98]

Ertel R. and Reed J. H., Generation of two equal power correlated Rayleigh
fading envelopes, IEEE Commun. Lett., vol. 2, no. 10, pp. 276-278, Oct.
1998.

[EtsGSM]

ETSI, Radio Transmission and Reception, ETSI. GSM 05.05, ETSI EN 300
910 V8.5.1, 2000.

[Etsi99]

ETSI, Broadband radio access networks (BRAN), HIPERLAN Type2 Technical specification: Physical layer, Ots/bran 0023003, ETSI, Oct. 1999.

[Fleu00]

Fleury B. H., First- and second-order characterization of direction dispersion and space selectivity in the radio channel, IEEE Trans. Inf. Theory,
vol. 46, no. 6, pp. 2027-2044, Sep. 2000.

[Ghos05]

Ghosh A., Wolter D. R., Andrews J. G., and Chen R., Broadband wireless
access with WiMAX/802.16: Current performance benchmarks and future
potentials, IEEE Commun. Mag., vol. 43, no. 2, pp. 129-136, Feb. 2005.

[Gold05]

Goldsmith A., Wireless Communications, New York: Cambridge Univ.

Press, 2005.

[Goza06]

Gozalvez J., Mobile WiMAX rollouts announced, IEEE Veh. Technol.

Mag., vol. 1, no. 3, pp. 53-59, Sep. 2006.

[Gumb53]

Gumbel E. J., Greenwood J. A., and Durand D., The circular normal distribution: Theory and tables, Journal of the American Statistical Association,
vol. 48, no. 261, pp. 131-152, Mar. 1953.

[Gumb54]

Gumbel E. J., Applications of the circular normal distribution, Journal

of the American Statistical Association, vol. 49, no. 266, pp. 267-297, Jun.
1954.

[Guti05]

Gutierrez C. A. and Cabrera M., Issues of the simulation of wireless channels with exponential-decay power-delay profiles, in Proc. 16th IEEE International Symposium on Personal, Indoor, and Mobile Radio Communications, PIMRC05, Berlin, Germany, Sep. 2005; pp. 507-511.

156

Channel Simulation Models for Mobile Broadband Communications

[Guti06]

Gutierrez C. A., Sanchez J, Cabrera M, and Gutierrez L. MC-CDMA based

architecture for the downlink of infrastructure WLANs, in Proc. 17th IEEE
International Symposium on Personal, Indoor, and Mobile Radio Communications, PIMRC06, Helsinki, Finland, Sept. 2006.

[Gut07a]

Gutierrez C. A. and Cabrera M., Deterministic simulation of flat-fading

MIMO wireless channels under non-isotropic scattering conditions, in Proc.
17th IEEE International Symposium on Personal, Indoor, and Mobile Radio
Communications, PIMRC07, Athens, Greece, Sep. 2007.

[Gut07b]

Gutierrez C. A. and Patzold M., Sum-of-sinusoids-based simulation of flatfading wireless propagation channels under non-isotropic scattering conditions, in Proc. 50th IEEE Global Telecommunications Conference, Globecom 2007, Washington, U.S.A., Nov. 2007, pp. 3842-3846.

[Gut09a]

Gutierrez C. A. and Patzold M., Sum-of-cisoids-based simulation of

Rayleigh fading channels in presence of multiple clusters of scatters, Submitted to the 2009 IEEE CERMA, Cuernavaca, Mexico.

[Gut09b]

Gutierrez C. A. and Patzold M., The Riemann sum method for the design
of sum-of-cisoids simulators for Rayleigh fading channels in non-isotropic
scattering environments, Submitted to the 20th IEEE International Symposium on Personal, Indoor, and Mobile Radio Communications, PIMRC09,
Tokio, Japan.

[Han02]

Han J. K., Yook J. G., and Park H. K., A deterministic channel simulation
model for spatially correlated Rayleigh fading, IEEE Commun. Lett., vol. 6,
no. 2, pp. 58-60, Feb. 2002.

[Hanz03]

Hanzo L., M
unster M., Choi B. J., and Keller T., OFDM and MC-CDMA
for Broadband Multi-User Communications, WLANs and Broadcasting,
Chichester: John Wiley & Sons, 2003.

[Hara97]

Hara S. and Prasad R., Overview of multicarrier CDMA, IEEE Commun.

Mag., vol. 35, no. 12, pp. 126-133 1997.

[Hogs05]

Hogstad B. O., Patzold M., Youssef N., and Kim D., A MIMO mobile-tomobile channel model: Part II The simulation model, in Proc. 16th IEEE
International Symposium on Personal, Indoor, and Mobile Radio Communications, PIMRC05, Berlin, Germany, Sep. 2005.

[Hogs08]

Hogstad B. O. and P
atzold M., On the stationarity of sum-of-cisoids-based
mobile fading channel simulators, in Proc. 67th IEEE Vehicular Technology
Conference, VTC2008-spring, Singapore, May 2008.

[Hohe92]

Hoher P., A statistical discrete-time model for the WSSUS multipath channel, IEEE Trans. Veh. Technol., vol. 41, no. 4, pp. 461-468, Nov. 1992.

[Hou04]

Hou Z. and Dubey V. K., Exact analysis for downlink MC-CDMA in

Rayleigh fading channels, IEEE Commun. Lett., vol. 8, no. 2, pp. 90-92,
Feb. 2004.

[IEEE06]

IEEE 802, IEEE Standard for Local and Metropolitan Area Networks, Part
16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems,
Amendment 2: Physical and Medium Access Control Layers for Combined
Fixed and Mobile Operation in Licensed Bands and Corrigendum 2, 2006.

Bibliography

157

[Jake74]

Jakes W., Microwave mobile communications, New York: Wiley, 1974.

[Jeru00]

Jeruchim M. C., Balaban P., and Shanmugan K. S., Simulation of Communication Systems: Modeling, Methodology, and Techniques, Berlin: Kluwer,
2nd ed., ch. 9, 2000.

[Kail59]

Kailath T., Sampling models for linear time-variant filters, Report no. 352,
M.I.T. Research Lab. of Electronics, Cambridge, Mass, May 1959.

[Kott04]

Kotterman W. A. T., Pedersen G. F., and Olsen K., Diversity properties

of multiantenna small handheld terminals, EURASIP Journal on Applied
Signal Processing, vol. 2004, no. 9, pp. 1340-1353, 2004.

[Kotz01]

Kotz S., Kozubowski T. J., and Podgorski K., The Laplace Distribution
and Generalizations: A Revisit with Applications to Communications, Economics, Engineering, and Finance, Berlin: Birkhauser, 2001.

[Laga98]

Lagarias J. C., Reeds J. A., Wright M. H., and Wright P. E., Convergence
properties of the NelderMead simplex method in low dimensions, SIAM
Journal of Optimization, vol. 9, no. 1, pp. 112-147, 1998.

[Lapl74]

Laplace P. S., Memoire sur les probabilite des causes par envenem`ents,

Mem. Acad. Sci. Sav. Etranger,

vol. 6, pp. 621-656, 1774.

[Lee07]

Lee K. T., Create the future with mobile WiMAX, IEEE Commun. Mag.,
vol. 45, no. 5, pp. 10-14, Mar. 2007.

[Leon94]

Leon-Garcia A., Probability and Random Processes for Electrical Engineering. Addison-Wesley, 2nd ed., 1994.

[Li07]

Li B., Qin Y., Low C. P., and Gwee C. L., A survey on mobile WiMAX,
IEEE Commun. Mag., vol. 45, no. 12, pp. 70-75, Dec. 2007.

[Mard99]

Mardia K. V. and Jupp P. E., Directional Statistics, Chichester: John Wiley

& Sons, 1999.

[Matz05]

Matz G., On non-WSSUS wireless fading channels, IEEE Trans. Wireless

Commun., vol. 4, no. 5, pp. 2465-2478, 2005.

[Mayy07]

Ma Y. and Patzold M., Performance comparison of space-time coded

MIMO-OFDM systems using different wideband MIMO channel models,
in Proc. 4th IEEE International Symposium on Wireless Communication
Systems, ISWCS 2007, Trondheim, Norway, Oct. 2007, pp. 762-766.

[Mayy08]

Ma Y. and Patzold M., A wideband one-ring MIMO channel model under

non-isotropic scattering conditions, in Proc. 67th IEEE Vehicular Technology Conference, VTC2008-Spring, Singapore, May 2008, pp. 424429.

[Med98a]

Medbo J. and Berg J. E., Measured radio wave propagation characteristics

at 5 GHz for typical HIPERLAN/2 scenarios, Tech. Rep. 3ERI074a, ETSI,
Sophia-Antipolis, France, 1998.

[Med98b]

Medbo J. and Schramm P., Channel models for HIPERLAN/2, Tech. Rep.
3ERI085B, ETSI, Sophia-Antipolis, France, 1998.

158

Channel Simulation Models for Mobile Broadband Communications

[Mill04]

Miller S. and Childers D., Probability and Random Processes: With Applications to Signal Processing and Communications, London: Elsevier Academic
Press, 2nd ed., 2004.

[Miti04]

Mitilineos S. A., Varlamos P. K., and Capsalis C. N., A simulation method

for bit-error-rate-performance estimation for arbitrary angle of arrival channel models, IEEE Antennas Propag. Mag., vol. 46, no. 2, pp. 158-163, Apr.
2004.

[Moha05]

Mohanty S., VEPSD: A Novel Velocity Estimation Algorithm for NextGeneration Wireless Systems, IEEE Trans. Wirel. Commun., vol. 4, no. 6,
pp. 2655-2660, Nov. 2005.

[Moli05]

Molisch A. F., Wireless Communications, West Sussex: John Wiley & Sons,
2005.

[Moli06]

Molisch A. F., Asplud H., Heddergott R., Steinbauer M., and Zwick T., The
COST 259 directional channel modelPart I: Overview and methodology,
IEEE Trans. Wirel. Commun., vol. 5, no. 12, pp. 3421-3433, Dec. 2006.

[Niss06]

Nissila M. and Pasupathy S., Joint estimation of carrier frequency offset

and statistical parameters of the multipath fading channel, IEEE Trans.
Commun., vol. 54, no. 6, pp. 1038-1048, Jun. 2006.

[Orti07]

Ortiz S., 4G wireless begins to take shape, IEEE Computer, vol. 40, no. 11,
pp. 18-21, Nov. 2007.

[Pae02a]

Patzold M., Mobile fading channels, Chichester: John Wiley & Sons, 2002.

[Pae02b]

Patzold M., Szczepanski A., and Youssef N., Methods for modeling of specified and measured multipath power-delay profiles, IEEE Trans. Veh. Technol., vol. 51, no. 5, Sep. 2002.

[Pae02c]

Patzold M. and Yao Q., A Study of stochastic and deterministic procedures

for the design of simulation models for spatial channels,, in Proc. 13th IEEE
International Symposium on Personal, Indoor, and Mobile Radio Communications, PIMRC02, Lisbon, Portugal, 15-18 Sept. 2002, pp. 1924-1931.

[Pae04a]

Patzold M. and Hogstad B. O., A space-time simulator for MIMO channels

based on the geometrical one-ring scattering model, Wirel. Commun. Mob.
Comput., vol. 4, no. 7, pp. 727-737, Nov. 2004.

[Pae04b]

Patzold M., On the stationarity and ergodicity of fading channel simulators

based on Rices Sum-of-Sinusoids, Int. J. of Wireless Inf. and Networks,
vol. 11, no. 2, pp. 63-69, Apr. 2004.

[Pae08a]

Patzold M., Hogstad B. O., and Youssef N., Modeling, analysis, and simulation of MIMO mobile-to-mobile fading channels, IEEE Trans. Wirel.
Commun., vol. 7, no. 2, pp. 510-520, Feb. 2008.

[Pae08b]

Patzold M. and Gutierrez C. A., Level-crossing rate and average duration

of fades of the envelope of a sum-of-cisoids, in Proc. 67th IEEE Vehicular
Technology Conference, VTC2008-spring, Singapore, Apr. 2008.

Bibliography

159

[Paet96]

Patzold M., Killat U., and Laue F., A deterministic digital simulation
model for Suzuki processes with application to a shadowed Rayleigh land
mobile radio channel, IEEE Trans. Veh. Tech., vol. 45, no. 2, pp. 318-331,
May 1996.

[Paet98]

Patzold M., Killat U., Laue F., and Li Y., On the statistical properties of
deterministic simulation models for mobile fading channels, IEEE Trans.
Veh. Tech., vol. 47, no. 1, pp. 254-269, Feb. 1998.

[Paet01]

Patzold M. and Youssef N., Modelling and simulation of direction-selective

and frequency-selective mobile radio channels, International Journal of

Electronics and Communications, vol. AEU-55,

no. 6, pp. 433-442, Nov.
2001.

[Paet05]

Patzold M. and Hogstad B. O., Design and performance of MIMO channel simulators derived from the two-ring scattering model, in Proc. 14th
IST Mobile & Communications Summit, IST 2005, Dresden, Germany, Jun.
2005, pp. 19-23.

[Paet06]

Patzold M. and Hogstad B. O., A wideband space-time MIMO channel

simulator based on the geometrical one-ring model, in Proc. 64th IEEE
Vehicular Technology Conference, VTC 2006-Fall, Montreal, Canada, Sep.
2006.

[Paet07]

Patzold M. and Talha B., On the statistical properties of sum-of-cisoidsbased mobile radio channel simulators, in Proc. 10th International Symposium on Wireless Personal Multimedia Communications, WPMC 2007,
Jaipur, India, Dec. 2007.

[Papo02]

Papoulis A. and Pillai S. U., Probability, Random Variables and Stochastic

Processes, New York: McGraw-Hill, 4th ed., 2002.

[Pars00]

Parsons J. D., The Mobile Radio Propagation Channel, Chichester: John

Wiley & Sons, 2000.

[Pate05]

Patel C. S., St
uber G. L., and Pratt T. G., Comparative analysis of statistical models for the simulation of Rayleigh faded cellular channels, IEEE
Trans. Commun., vol. 53, no. 6, pp. 1017-1026, Jun. 2005.

[Pede97]

Pedersen K. I., Mongensen P. E., and Fleury B. H., Power azimuth spectrum in outdoor environments, IEE Electron. Lett., vol 33, no. 18, pp. 15831584, Aug. 1997.

[Proa01]

Proakis J. G., Digital Communications, New York: McGraw-Hill, 4th ed.,

ch. 2, 2001.

[Qua08a]

Qualcomm., 3GPP Long-Term Evolution (LTE), White paper, Jan. 2008.

[Rafi08]

Rafiq G. and Patzold M., A Study of the Influence of Shadowing on the

Statistical Properties of the Capacity of Mobile Radio Channels, Wireless
Per. Commun., accepted for publication.

[Rapp02]

Rappaport T. S., Wireless Communications, Principles and Practice, New

York: Prentice-Hall, 2nd ed., 2002.

160

Channel Simulation Models for Mobile Broadband Communications

[Rice44]

Rice S. O., Mathematical analysis of random noise, Bell Syst. Tech. J.,
vol. 23, pp. 282-332, Jul. 1944.

[Rice45]

Rice S. O., Mathematical analysis of random noise, Bell Syst. Tech. J.,
vol. 24, pp. 46-156, Jan. 1945.

[Robe06]

Roberts M. L., Temple M. A., Mills R. F., and Raines R.A., Evolution of
the air interface of cellular communications systems toward 4G realization,
IEEE Commun. Surveys Tuts., vol. 8, no. 1, pp. 2-23, 1st quarter 2006.

[Roys04]

Roy S., Foerster J. R., Somayazulu V. S., and Leeper D. G., Ultrawideband
radio design: the promise of high-speed, short-range wireless connectivity,
Proc. IEEE, vol. 92, no. 2, pp. 295-311, Feb. 2004.

[Saun07]

Saunders S. R. and Aragon-Savala A., Antennas and Propagation for Wireless Communication Systems, Chichester: John Wiley & Sons, 2nd ed., 2007.

[Schu05]

Schulze H. and L
uders C., Theory and Applications of OFDM and CDMA,
West Sussex: John Wiley & Sons, ch. 2, 2005.

[Shiu00]

Shiu D. S., Foschini G. J., Gans M. J., and Kahn J. M., Fading correlation
and its effect on the capacity of multielement antenna systems, IEEE Trans.
Commun., vol. 48, pp. 502-513, Mar. 2000.

[Skla88]

Sklar B., Digital communications: Fundamentals and Applications, Englewood Cliffs, NJ: Prentice-Hall, 1988.

[Soro03]

Sorooshyari S. and Daut D. G., Generation of correlated Rayleigh fading

envelopes for accurate performance analysis of diversity systems, in Proc.
14th IEEE International Symposium on Personal, Indoor, and Mobile Radio
Communications, PIMRC03, Beijing, China, Sep. 2003.

[Spen00]

Spencer Q. H., Jeffs B. D., Jensen M. A., and Swindlehurst A. L., Modeling
the statistical time and angle of arrival characteristics of an indoor multipath
channel, IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp. 347-360, Mar.
2000.

[Stal04]

Stallings W., IEEE 802.11: Wireless LANs from a to n, IEEE IT Prof.,

vol. 6, no. 5, pp. 32-37, Sep.-Oct. 2004.

[Stei87]

Stein S., Fading channel issues in system engineering, IEEE J. Sel. Areas
Commun., vol. SAC-5, no 2, pp. 68-89, 1987.

[Syko00]

Sykora J., Tapped delay line model of linear randomly time-variant WSSUS
channel, Electron. Lett., vol. 36, no. 19, pp. 1656-1657, 2000.

[Teo07]

Teo K. H., Tao Z., and Zhang J., The mobile broadband WiMAX standard, IEEE Signal Process. Mag., vol. 24, no. 5, pp. 144-148, Sep. 2007.

[Upto73]

Upton G. J. G., Single-sample tests for the von Mises distribution,

Biometrika, vol. 60, no. 1, pp. 87-99, Apr. 1973.

[VanN04]

Van Nee R., Prasad R., OFDM for Wireless Multimedia Communications,
Artech House, 2004.

[VonM18]

Von Mises R., Uber

die ganzzahligkeit det atomgewichte und verwandte
fragen, Physikalische Zeitschrift, vol. 19, pp. 490-500, 1918.

Bibliography

161

[Wang07]

Wang C.-X., Patzold M., and Yao Q., Stochastic modelling and simulation of frequency-correlated wideband fading channels, IEEE Trans. Veh.
Technol., vol. 56, no. 3, pp. 1050-1063, May 2007.

[Wats82]

Watson G. S., Distributions on the circle and sphere, Journal of Applied

Probability, vol. 19, pp. 265-280, 1982.

[Yang02]

Yang L L., Hanzo L., Performance of generalized multicarrier DS-CDMA

over Nakagami-m fading channels, IEEE Trans. Commun., vol. 50, no. 6,
pp. 956-966, 2002.

[Yang05]

Yang H., A road to future broadband wireless access: MIMO-OFDM-based

air interface, IEEE Commun. Mag., vol. 33, pp. 53-60, Jan. 2005.

[Yip95]

Yip K.-W. and Ng T.-S., Efficient simulation of digital transmission over

WSSUS channels, IEEE Trans. Commun., vol. 43, no. 12, pp. 2907-2913,
Dec. 1995.

[Youn00]

Young D. J. and Beaulieu N. C., The generation of correlated Rayleigh random variates by inverse discrete Fourier transform, IEEE Trans. Commun.,
vol. 48, pp. 1114-1127, Jul. 2000.

[Zhao03]

Zhao X., Kivinen J., Vainikainen P., and Skog K., Characterization of
Doppler Spectra for Mobile Communications at 5.3 GHz, IEEE Trans.
Veh. Technol., vol. 52, no. 1, pp. 14-23, Jan. 2003.

162

Channel Simulation Models for Mobile Broadband Communications