Modeling and Simulation of Mobile Radio Channels: Guest Editors: Matthias Pätzold, Neji Youssef, and Carlos A. Gutierrez

Modelling and Simulation in Engineering
Guest Editors: Matthias Ptzold, Neji Youssef,

and Carlos A. Gutierrez
Modeling and Simulation of
Mobile Radio Channels
Modeling and Simulation of Mobile
Radio Channels
Modeling and Simulation of Mobile
Radio Channels
Guest Editors: Matthias P atzold, Neji Youssef,
Copyright 2012 Hindawi Publishing Corporation. All rights reserved.
This is a special issue published in Modelling and Simulation in Engineering. All articles are open access articles distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.
Editorial Board
Driss Aboutajdine, Morocco
K. Al-Begain, UK
Adel M. Alimi, Tunisia
Zeki Ayag, Turkey
Sergio Baragetti, Italy
Andrzej Bargiela, UK
Joaquim Barros, Portugal
Salim Belouettar, Luxembourg
Jean-Michel Bergheau, France
Tianshu Bi, China
Philippe Boisse, France
Agostino Bruzzone, Italy
Hing Kai Chan, UK
Jinn Kuen Chen, USA
Min-Chie Chiu, Taiwan
Weizhong Dai, USA
Mingcong Deng, Japan
Lei Ding, China
Tadashi Dohi, Japan
Dimitris Drikakis, UK
Andrzej Dzielinski, Poland
Abdelali El Aroudi, Spain
Enmin Feng, China
Huijun Gao, China
F. Gao, UK
Xiaosheng Gao, USA
Huijun Gao, China
Dariusz J. Gawin, Poland
Ratan K. Guha, USA
Chung-Souk Han, USA
Shinsuke Hara, Japan
Joanna Hartley, UK
Tasawar Hayat, Pakistan
Jing-song Hong, China
Qinglei Hu, China
Bassam A. Izzuddin, UK
MuDer Jeng, Taiwan
Chia-Feng Juang, Taiwan
Iisakki Kosonen, Finland
Stavros Kotsopoulos, Greece
Nikos D. Lagaros, Greece
Hak-Keung Lam, UK
Hongyi Li, China
Dimitrios E. Manolakos, Greece
Shengwei Mei, China
Laurent Mevel, France
Azah Mohamed, Malaysia
Carlos A. Mota Soares, Portugal
Antonio Munjiza, UK
Javier Murillo, Spain
Petr Musilek, Canada
Tomoharu Nakashima, Japan
Gaby Neumann, Germany
Alessandra Orsoni, UK
Javier Otamendi, Spain
Anna Pandol, Italy
Bei Peng, China
Ricardo Perera, Spain
Evtim Peytchev, UK
Zhiping Qiu, China
Luis Carlos Rabelo, USA
Ahmed Rachid, France
Franco Ramrez, Spain
Rubens Sampaio, Brazil
Aiguo Song, China
Rajan Srinivasan, India
S. Taib, Malaysia
Mohamed B. Trabia, USA
Joseph Virgone, France
Bauke Vries, The Netherlands
Guowei Wei, USA
Ligang Wu, China
Feng Xiao, China
Farouk Yalaoui, France
Shigeru Yamada, Japan
Guang-Hong Yang, China
ShengKai Yu, Singapore
Qingling Zhang, China
Contents
Modeling and Simulation of Mobile Radio Channels, Matthias P atzold, Neji Youssef,
Volume 2012, Article ID 160297, 2 pages
Design of Wideband MIMOCar-to-Car Channel Models Based on the Geometrical Street Scattering
Model, Nurilla Avazov and Matthias P atzold
A Three-Dimensional Geometry-Based Statistical Model of 2 2 Dual-Polarized MIMO
Mobile-to-Mobile Wideband Channels, Jun Chen and Thomas G. Pratt
An Accurate Hardware Sum-of-Cisoids Fading Channel Simulator for Isotropic and Non-Isotropic
Mobile Radio Environments, L. Vela-Garcia, J. V azquez Castillo, R. Parra-Michel, and Matthias P atzold
Filter-Based Fading Channel Modeling, Amirhossein Alimohammad, Saeed Fouladi Fard,
and Bruce F. Cockburn
Wideband and Ultrawideband Channel Models in Working Machine Environment,
Attaphongse Taparugssanagorn, Matti H am al ainen, and Jari Iinatti
Hindawi Publishing Corporation
doi:10.1155/2012/160297
Editorial
Modeling and Simulation of Mobile Radio Channels
Matthias P atzold,
1
Neji Youssef,
2
3
1
Department of Information and Communication Technology, University of Agder, 4604 Kristiansand, Norway
2
Departement EPP (Electronique, Physique et Propagation), Ecole Superieure des Communications de Tunis, 2083 Ghazala, Tunisia
3
Faculty of Science, Universidad Autonoma de San Luis Potosi, 78290 San Luis Potosi, SLP, Mexico
Correspondence should be addressed to Matthias P atzold, matthias.paetzold@uia.no
Received 2 December 2012; Accepted 2 December 2012
Copyright 2012 Matthias P atzold et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
The aggregate demand for multimedia services, high mobil-
ity, and global connectivity has resulted in recent years in an
explosion of new technologies for wireless communication
systems. All components of a wireless communication system
ranging from digital modulation schemes over channel
coding techniques up to higher layer protocols are more
or less inuenced by the characteristics of the mobile
radio channel. A thorough understanding of the mobile
radio channel is therefore crucial for the development,
performance optimization, and testing of present as well as
next-generation mobile radio systems. This is one of the
reasons why exploring the mobile radio channel has always
been a key research topic from the very beginning of mobile
communications to the present. Currently, the research on
mobile radio channels involves a variety of challenging topics
such as the modeling of car-to-car channels, multiple-input
multiple-output (MIMO) channels, cooperative channels,
satellite channels, and ultrawideband channels, only to name
a few. The objective of this special issue is to give an overview
of the state-of-the-art research in the fascinating area of
modeling and simulation of mobile radio channels. This
special issue is composed of ve papers.
The paper entitled A Three-Dimensional Geometry-
Based Statistical Model of 2 2 Dual-Polarized MIMO
Mobile-to-Mobile Wideband Channels uses a geometric
scattering approach to derive a three-dimensional (3D)
parametric reference model for wideband dual-polarized
MIMO mobile-to-mobile (M2M) channels. The authors
assume nonisotropic scattering, where the distribution of the
scatterers is characterized in the azimuth direction through
the von Mises distribution and in the elevation direction
by the cosine distribution. The statistical properties of the
reference channel model are controlled by a large number
of physical model parameters, including the velocities of
the transmitter and the receiver, the distance between the
transmitter and the receiver, the 3D antenna pattern gains,
the azimuth and elevation angles of arrival (departure),
the geometrical distribution of the scatterers, the Rician
K-factor, the maximum Doppler frequency, the scattering
loss factor, the cross-polar power discrimination ratio, and
the copolarization power ratio. The spotlight here is on
the derivation and analysis of the time-frequency corre-
lation function of the proposed reference model for 3D
nonisotropic scattering environments. With the knowledge
of the time-frequency correlation function, an essential
prerequisite is provided for the performance analysis and
optimization of mobile-to-mobile communication systems
employing 2 2 dual-polarized architectures.
The paper An Accurate Hardware Sum-of-Cisoids
Fading Channel Simulator for Isotropic and NonIsotropic
Mobile Radio Environments presents a hardware simulator
for mobile fading channels characterized by Doppler power
spectral densities with arbitrary symmetrical and asymmet-
rical shapes. The hardware simulators architecture described
in this paper is based on an ecient implementation of
the sum-of-cisoids principle. According to this principle, a
nite number of complex sinusoids (cisoids) are combined
to generate complex-valued waveforms with statistical prop-
erties resembling those of a given reference channel model.
Each of the cisoids that constitute the sum-of-cisoids model
is synthesized in hardware using a piecewise polynomial
approximation technique, which oers a high-performance
alternative to traditional solutions using look-up tables.
The architecture of the hardware simulator is capable
2 Modelling and Simulation in Engineering
of handling any conguration of the cisoids amplitudes,
frequencies, and phases. The exibility and accuracy of the
authors implementation scheme are demonstrated through
a variety of numerical results of the rst- and second-order
statistics of emulated Rayleigh fading channels assuming
dierent isotropic and nonisotropic scattering conditions.
The authors analyze the computational complexity of the
hardware simulator in an FPGA implementation with 32
cisoids and dierent data path bit-widths. To better assess
the complexity of their proposal, they provide a comparison
with two alternative hardware fading channel simulators,
which are also built on the sum-of-sinusoids principle. Aside
from its eciency and accuracy, the main feature of the
hardware fading channel simulator presented in this paper
is the capability of simulating mobile fading channels under
a wide range of scattering conditions.
The paper Filter-Based Fading Channel Modeling out-
lines a lter-based approach for the design of fading channel
simulators. As distinct from the sum-of-sinusoids concept,
which has been studied extensively in the last two decades for
the modeling and simulation of multipath fading channels,
the lter concept has attracted only little attention in the
past. This is mainly due to the design complexity of digital
lters and its restriction to channel models described by
rational Doppler power spectral densities. This third paper
addresses the design of Rayleigh fading channel simulator
using innite impulse response (IIR) lters for shaping the
Doppler spectrum. Specically, the authors proposed an
iterative technique for the IIR lter design with xed-point
arithmetic by considering both isotropic and nonisotropic
scattering scenarios. The proposed iterative algorithm for
the determination of the lter coecients is described in a
systematic manner. The authors also provide an overview of
alternative methods for the design of digital lters and they
discuss their respective advantages and disadvantages.
The authors of the paper, Wideband and Ultrawideband
Channel Models in Working Machine Environment, focus
on a rather new topic in the area of mobile radio channel
modeling. This paper presents statistical models for wide-
band and ultrawideband (UWB) radio channels in specic
working environments, such as working machine cabins.
From the analysis of measurement data collected in such
kinds of rather small and conned propagation environ-
ments, it turned out that the channel impulse response is
mainly composed of a sum of diuse multipath components
rather than of specular components. In wideband channels,
the path gains associated with the various delay bins are
generally Rayleigh distributed, whereas only a few path gains
exhibit a Rician distribution. This is in contrast to UWB
channels, in which the path gains associated with the delay
bins show a tendency to follow the log-normal distribution.
The results of the measurements were used to estimate the
key channel parameters for the development of stochastic
channel simulators. The developed mobile radio channel
models provide important information to the designers of
wireless communication systems dedicated to operate in
working machine cabins.
In the paper Design of Wideband MIMO Car-to-Car
Channel Models Based on the Geometrical Street Scattering
Model, the authors deal with the modeling and simulation
of channel models for car-to-car (C2C) communication
systems. The starting point is an environment-specic
geometrical street scattering model, which assumes that the
local scatterers are uniformly distributed over two separated
rectangular scattering areas located on opposite sides of the
street. From the geometrical scattering model, the authors
derive a reference model for a wideband MIMOC2Cchannel
model by assuming that the number of scatterers is innite.
The statistical properties of the reference model are analyzed,
where the emphasis lies on the space-time-frequency cross-
correlation function from which many other important
system functions and characteristic quantities can be easily
derived. To validate the proposed channel model, the mean
Doppler shift and the Doppler spread of the reference model
have been matched to measurement data. In a further
step, an ecient sum-of-cisoids channel simulator has been
derived from the reference model. It is demonstrated that the
statistical properties of the sum-of-cisoids channel simulator
match very well those of the reference model. The proposed
geometry-based channel simulator is useful for studying the
eect of local scatterers along the street on the performance
of C2C communication systems.
Matthias P atzold
Neji Youssef
Carlos A. Gutierrez
doi:10.1155/2012/264213
Research Article
Design of Wideband MIMO Car-to-Car Channel Models Based on
the Geometrical Street Scattering Model
Nurilla Avazov and Matthias P atzold
Faculty of Engineering and Science, University of Agder, P.O. Box 509, 4898 Grimstad, Norway
Correspondence should be addressed to Nurilla Avazov, nurilla.k.avazov@uia.no
Received 2 May 2012; Revised 28 August 2012; Accepted 5 September 2012
Academic Editor: Neji Youssef
Copyright 2012 N. Avazov and M. P atzold. This is an open access article distributed under the Creative Commons Attribution
cited.
We propose a wideband multiple-input multiple-output (MIMO) car-to-car (C2C) channel model based on the geometrical street
scattering model. Starting from the geometrical model, a MIMO reference channel model is derived under the assumption of
single-bounce scattering in line-of-sight (LOS) and non-LOS (NLOS) propagation environments. The proposed channel model
assumes an innite number of scatterers, which are uniformly distributed in two rectangular areas located on both sides of the
street. Analytical solutions are presented for the space-time-frequency cross-correlation function (STF-CCF), the two-dimensional
(2D) space CCF, the time-frequency CCF (TF-CCF), the temporal autocorrelation function (ACF), and the frequency correlation
function (FCF). An ecient sum-of-cisoids (SOCs) channel simulator is derived from the reference model. It is shown that the
temporal ACF and the FCF of the SOC channel simulator t very well to the corresponding correlation functions of the reference
model. To validate the proposed channel model, the mean Doppler shift and the Doppler spread of the reference model have
been matched to real-world measurement data. The comparison results demonstrate an excellent agreement between theory and
measurements, which conrms the validity of the derived reference model. The proposed geometry-based channel simulator allows
us to study the eect of nearby street scatterers on the performance of C2C communication systems.
1. Introduction
C2C communications is an emerging technology, which
receives considerable attention due to new trac telematic
applications that improve the eciency of trac ow and
reduce the number of road accidents [1]. The development of
C2C communication technologies is supported in Europe by
respected organizations, such as the European Road Trans-
port Telematics Implementation Coordinating Organization
(ERTICO) [2] and the C2C Communication Consortium
(C2C-CC) [3]. In this context, a large number of research
projects focussing on C2C communications are currently
being carried out throughout the world.
In C2C communication systems, the underlying radio
channel diers from traditional xed-to-mobile and mobile-
to-xed channels in the way that both the transmitter and the
receiver are in motion. In this connection, robust and reliable
trac telematic systems have to be developed and tested,
which calls for new channel models for C2C communication
systems. Furthermore, MIMO communication systems can
also be of great interest for C2C communications due to their
higher throughput [4]. In this regard, several MIMO mobile-
to-mobile (M2M) channel models have been developed and
analyzed under dierent scattering conditions induced by,
for example, the two-ring model [5], the elliptical model
[6], the T-junction model [7], and the geometrical street
model [8, 9]. A 2D reference model for narrowband single-
input single-output (SISO) M2M Rayleigh fading channels
has been proposed by Akki and Haber in [10, 11]. Simulation
models for SISO M2M channels have been reported in [12,
13]. In [5, 14, 15], the 2D reference and simulation models
have been presented for narrowband MIMO M2M channels.
The proposed model in [15] combines the two-ring model
and the elliptical model, where a combination of single- and
double-bounce scattering in LOS propagation environments
is assumed.
All aforementioned channel models are narrowband
M2M channel models. In contrast with narrowband chan-
nels, a channel is called a wideband channel or frequency-
selective channel if the signal bandwidth signicantly exceeds
the coherence bandwidth of the channel. Owing to increasing
demands for high data rate wideband communication
systems employing MIMO technologies, such as MIMO
orthogonal frequency division multiplexing (OFDM) sys-
tems, it is of crucial importance to have accurate and
realistic wideband MIMO M2M channel models. According
to IEEE 802.11p [16], the dedicated frequency bands for
short-range communications [17] will be between 5770 MHz
and 5925 MHz depending on the region. The range 5795
5815 MHz will be devoted to Europe, while 58505925 MHz
and 57705850 MHz will be assigned to North America and
Japan, respectively. Consequently, a large number of C2C
channel measurements have been carried out at dierent
frequency bands, for example, at 2.4 GHz [18], 3.5 GHz
[19], 5 GHz [20, 21], 5.2 GHz [22], and 5.9 GHz [23]. Real-
world measurement campaigns for wideband C2C channels
can be found in [2427]. In the literature, there exist
several papers [2830] with the focus on the modeling
of wideband MIMO M2M channels. A reference model
derived from the geometrical T-junction scattering model
has been proposed in [7] for wideband MIMO vehicle-to-
vehicle (V2V) fading channels. In [29], a three-dimensional
(3D) model for a wideband MIMO M2M channel has been
studied. Its corresponding rst- and second-order statistics
have been investigated and validated on the basis of real-
world measurement data. In the same paper, it has been
shown that 3Dscattering scenarios are more realistic than 2D
scattering scenarios. However, 2Dscattering models are more
complexity ecient, and they provide a good approximation
to 3D scattering models [31]. For those reasons, we propose
in our paper a 2D street scattering model.
In the literature, numerous fundamental channel models
with dierent scatterer distributions, such as the uniform,
Gaussian, Laplacian, and von Mises distribution, have been
proposed to characterize the angle-of-departure (AOD) and
the angle-of-arrival (AOA) statistics. In [32], the author
studied the eect of Gaussian distributed scatterers on the
channel characteristics in a circular scattering region around
a mobile station. The spatial and temporal properties of the
rst arrival path in multipath environments have also been
analyzed in [32]. The authors of [9] assume rectangular
scattering areas on both sides of the street, in which an
innite number of scatterers are uniformly distributed. It has
been observed that the shape of the Doppler power spectral
density (PSD) resembles a Gaussian function if the width of
the scattering area is very large.
In contrast to our previous work in [9], where the focus
was on the derivation of a reference channel model for
narrowband SISO C2C channels, we design in this paper a
wideband MIMO C2C channel model by starting from the
same geometrical street scattering model. We focus on the
statistical characterization of a wideband reference channel
model assuming that an innite number of scatterers are
uniformly distributed within two rectangular areas. The
radio propagation phenomena in street environments are
modelled by a wide-sense stationary uncorrelated scattering
process, where in addition a LOS component is taken into
account. The reference model has been derived from the
geometrical street scattering model assuming that the AOD
and the AOA are dependent due to single-bounce scattering.
To account for the nature of C2C channels, we take the
mobility of both the transmitter and the receiver for granted.
In our model, we consider a 2D street scattering environ-
ment to reduce the computational cost by still guaranteeing
a good match between the reference model and measured
channels. A typical propagation scenario for the proposed
model is illustrated in Figure 1, where the buildings and the
trees are considered as scattering objects. Such a typical dense
urban environment scenario allows us to assume that the
local scatterers are uniformly distributed in a specic area.
An analytical expression will be derived for the STF-CCF
from which the 2D space CCF, the TF-CCF, the temporal
ACF, and the FCF can be obtained directly. To validate the
proposed reference model, the mean Doppler shift and the
Doppler spread of the reference model have been matched
to the corresponding quantities of the measured channel
described in [25] for dierent propagation environments,
such as urban, rural, and highway areas. Furthermore, we
have derived an SOC channel simulator from the reference
model. It is shown that the designed channel simulator
matches the underlying reference model with respect to the
temporal ACF and the FCF.
The rest of this paper is organized as follows. Section 2
describes the geometrical street scattering model. In
Section 3, the reference channel model is derived from the
geometrical street model. Section 4 analyzes the correlation
properties of the reference model, such as the STF-CCF,
the 2D space CCF, the TF-CCF, the temporal ACF, and the
FCF. The computation of the measurement-based model
parameters and the characteristic quantities describing the
Doppler eect are discussed in Section 5. Section 6 describes
briey the simulation model derived from the reference
model. The illustration of some numerical results found for
the correlation functions of the reference model and the
corresponding simulation model is the topic of Section 7.
Finally, Section 8 draws the conclusion of the paper.
2. The Geometrical Street Scattering Model
This section briey describes the geometrical street scattering
model for wideband MIMO C2C channels. The proposed
geometrical model describes the scattering environment in
an urban area, where the scatterers are located in two
rectangular areas on both sides of the street as illustrated in
Figure 2. We consider rectangular grids formed by rows and
columns, where the length and the width of the rectangular
grids are denoted by L
A
= A
1
+ A
2
and B
i
(i = 1, 2),
respectively. The scatterer located in the mth column of
the nth row is denoted by S
(mn)
(m = 1, 2, . . . , M, n =
1, 2, . . . , N). It is assumed that the local scatterers S
(mn)
are
uniformly distributed in the rectangles. The symbols MS
T
and MS
R
in Figure 2 stand for the mobile transmitter and
the mobile receiver, respectively. The symbol D represents
the scalar projection of the distance between the transmitter
and the receiver onto the x-axis. The transmitter (receiver) is
located at a distance y
T1
(y
R1
) from the left-hand side of the
street and at a distance y
T2
(y
R2
) from the right-hand side of
the street. Both the transmitter and the receiver are in motion
Modelling and Simulation in Engineering 3
MS
R
MS
T
Figure 1: A typical propagation scenario along a straight street in urban areas.
and equipped with M
T
transmitter antenna elements and M
R
receiver antenna elements, respectively. The antenna element
spacings at the transmitter and the receiver are denoted by
T
and
R
, respectively. The symbols
(mn)
T
and
(mn)
R
denote the
AOD and the AOA, respectively. The angle
T
(
R
) describes
the tilt angle of the transmitter (receiver) antenna array.
Moreover, it is assumed that the transmitter (receiver) moves
with speed v
T
(v
R
) in the direction determined by the angle
of motion
T
v
(
R
v
).
3. The Reference Model
3.1. Derivation of the Reference Model. In this section, we
derive the reference model for the MIMOC2Cchannel under
the assumption of LOS and NLOS propagation conditions.
From Figure 2, we realize that the (mn)th homogeneous
plane wave emitted from the lth antenna element A
(l)
T
(l =
1, 2, . . . , M
T
) of the transmitter travels over the local scatterer
S
(mn)
before impinging on the kth antenna element A
(k)
R
(k =
1, 2, . . . , M
R
) of the receiver. The reference model is based on
the assumption that the number of local scatterers within
both rectangular areas is innite, that is, M, N . The
temporal, spatial, and frequency characteristics of the refer-
ence model are determined by the M
R
M
T
channel matrix
H( f
, t) = [H
kl
( f
, t)]
MRMT
, where H
kl
( f
, t) denotes the
time-variant transfer function (TVTF) of the channel for the
link between the lth transmitter antenna element A
(l)
T
and the
kth receiver antenna element A
(k)
R
. The TVTF H
kl
( f
, t) can
be expressed as a superposition of the diuse component and
the LOS component as follows:
H
kl
_
f
, t
_
= H
DIF
kl
_
f
, t
_
+ H
LOS
kl
_
f
, t
_
, (1)
where H
DIF
kl
( f
, t) and H
LOS
kl
( f
, t) represent the diuse and

the LOS components of the channel, respectively.
Note that the single-bounce scattering components bear
more energy than the double-bounce scattering components.
Hence, in our analysis, we model the diuse component
H
DIF
kl
( f
, t) by only taking into account the single-bounce

scattering eects, which is in accordance with the assump-
tions made in [28, 33]. From the geometrical street scattering
model shown in Figure 2, we can derive the TVTF of the
diuse component, which results in the following expression:
H
DIF
kl
_
f
, t
_
= lim
M,N
1
_
(c
R
+ 1)MN
M,N
_
m,n=1
a
l,mn
b
k,mn
c
mn
e
j[mn+2( f
(mn)
T
+f
(mn)
R
)t2 f
(mn)
kl
]
,
(2)
where
a
l,mn
= e
j(T/)(MT2l+1) cos(
(mn)
T
T)
, (3)
b
k,mn
= e
j(R/)(MR2k+1) cos(
(mn)
R
R)
, (4)
c
mn
= e
j(2/)(yT
1
/ sin(
(mn)
T
)+yR
1
/ sin(
(mn)
R
))
, (5)
f
(mn)
T
= f
Tmax
cos
_
(mn)
T

T
v
_
, (6)
f
(mn)
R
= f
Rmax
cos
_
(mn)
R

R
v
_
, (7)
kl
(mn)
=
1
c
0
_
D
(l,mn)
T
+ D
(mn,k)
R
_
. (8)
In (6) and (7), the symbols f
Tmax
= v
T
/ and f
Rmax
=
v
R
/ denote the maximum Doppler frequencies associ-
ated with the movement of the transmitter and the
receiver, respectively, and is the wavelength. The symbol
c
R
in (2) represents the Rice factor, which is dened
as the ratio of the power of the LOS component to
the power of the diuse component, that is, c
R
=
E{|H
LOS
kl
( f
, t)|
2
}/E{|H
DIF
kl
( f
, t)|
2
}. The phases
mn
in (2)
denote the phase shift introduced by the scatterer S
(mn)
. It
is assumed that the phases
mn
are independent, identically
distributed (i.i.d.) random variables, which are uniformly
distributed over the interval [0, 2). The symbols
kl
(mn)
and
c
0
represent the propagation delays of the diuse component
and the speed of light, respectively. In (8), the quantity D
(l,mn)
T
A
1 A
2
A
1 A
2
B
1
y
S
(mn)
y
T1
y
R1
y
R2
D
(mn,1)
R
(mn)
T
D
(mn)
T
A
(1)
R
T
A
(1)
T
A
(2)
T
R
D
(1,mn)
T
R
v
V
R
V
T
(mn)
R
D
(mn)
R
T
v
T
0
A
(M
T
)
T
D
B
2
y
T2
A
(M
R
)
R
. . .
.
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
MS
R
MS
T
x
Figure 2: The geometrical street scattering model with local scatterers uniformly distributed in two rectangular areas on both sides of the
street.
stands for the distance from the lth transmitter antenna
element A
(l)
T
to the scatterer S
(mn)
, whereas D
(mn,k)
R
is the
distance between the scatterer S
(mn)
and the kth receiver
antenna element A
(k)
R
. It is assumed that (M
T
1)
T

min{y
T1
, y
T2
} and (M
R
1)
R
min{y
R1
, y
R2
}. These
assumptions, together with the approximation

1 + x
1 + x/2 (x 1), allow us to approximate the two distances
D
(l,mn)
T
and D
(mn,k)
R
as follows:
D
(l,mn)
T
D
(mn)
T
(M
T
2l + 1)
_
T
2
_
cos
_
(mn)
T

T
_
,
(9)
D
(mn,k)
R
D
(mn)
R
(M
R
2k + 1)
_
R
2
_
cos
_
(mn)
R

R
_
,
(10)
where D
(mn)
T
and D
(mn)
R
are given by D
(mn)
T
= y
T1
/ sin(
(mn)
T
)
and D
(mn)
R
= y
R1
/ sin(
(mn)
R
), respectively.
It is noteworthy that one can also nd articles [11, 34],
in which only double-bounce scattering is assumed for M2M
communications. However, by following a similar approach
as in [15], one can easily extend our analysis on the basis
of single-bounce scattering to the case of double-bounce
scattering, and thus also to a combination of single- and
double-bounce scattering.
The TVTF of the LOS component is given by
H
LOS
kl
_
f

, t
_
=
c
R
(c
R
+ 1)
e
j[2( f
(0)
T
+f
(0)
R
)t(2/)Dkl 2 f
kl
(0)
]
,
(11)
where
f
(0)
T
= f
Tmax
cos
_
(0)
T

T
v
_
, (12)
f
(0)
R
= f
Rmax
cos
_
(0)
R

R
v
_
, (13)
D
kl
= D
0
(M
T
2l + 1)
R
2
cos
_
T
_
+ (M
R
2k + 1)
R
2
cos
_
R
_
,
(14)
D
0
=
_
D
2
+
_
y
T1
y
R1
_
2
.
(15)
In (11), f
(0)
T
and f
(0)
R
denote the Doppler shifts of the
LOS component caused by the movement of the transmitter
and the receiver, respectively. The symbols
(0)
T
and
(0)
R
in (12) and (13) represent the AOD and the AOA of
the LOS component, respectively. Finally,
kl
(0)
denotes the
propagation delay of the LOS component. The delay of the
LOS component is dened by
kl
(0)
= D
kl
/c
0
with D
kl
being
the length of the direct path from the lth transmitter antenna
element A
(l)
T
to the kth receiver antenna element A
(k)
R
. The
symbol D
0
in (14) denotes the Euclidean distance between
the transmitter and the receiver. According to [35], the LOS
component H
LOS
kl
( f
, t) is assumed to be a deterministic pro-

cess, while the diuse component H
DIF
kl
( f
, t) is a stochastic
process.
3.2. Derivation of the AOD and the AOA. The position of all
local scatterers S
(mn)
is described by the Cartesian coordinates
(x
m
, y
n
). In the reference model, the coordinates x
m
and y
n
are independent random variables, which are determined
by the distribution of the local scatterers. With reference
to Figure 2, we take into account that due to single-bounce
scattering, the AOD
(mn)
T
and the AOA
(mn)
R
are dependent.
By using the trigonometric identities, we can express the
AOD
(mn)
T
and the AOA
(mn)
R
in terms of the coordinates
(x
m
, y
n
) of the local scatterers S
(mn)
as follows:
(mn)
T
_
x
m
, y
n
_
=
_
g
_
x
m
, y
n
_
, if y
n
J
i
, x
m
[0, A
2
]
(1)
i+1
+ g
_
x
m
, y
n
_
, if y
n
J
i
, x
m
[A
1
, 0]
(16)
(mn)
R
_
x
m
, y
n
_
=
_
f
_
x
m
, y
n
_
, if y
n
J
i
, x
m
[D, A
2
]
(1)
i+1
+ f
_
x
m
, y
n
_
, if y
n
J
i
, x
m
[A
1
, D]
(17)
for i = 1, 2, where J
1
= [y
T1
, y
T1
+B
1
], J
2
= [y
T2
B
2
, y
T2
],
and
g
_
x
m
, y
n
_
= arctan
y
n
x
m
f
_
x
m
, y
n
_
= arctan
y
n
y
T1
+ y
R1
x
m
D
.
(18)
4. Correlation Properties of
the Reference Model
In this section, we derive a general analytical solution for the
STF-CCF, from which other correlation functions, such as
the 2D space CCF, the TF-CCF, the temporal ACF, and the
FCF can easily be derived.
4.1. Derivation of the STF-CCF. According to [10], the STF-
CCF of the links A
(l)
T
A
(k)
R
and A
(l
)
T
A
(k
)
R
is dened as the
correlation between the channel transfer functions H
kl
( f
, t)
and H
k
l
( f
, t), that is,
kl,k
l
(
T
,
R
,
, ) = E
_
H
kl
_
f
, t
_
H
k
_
f
, t +
_
_
=
DIF
kl,k
l
(
T
,
R
,
, )
+
LOS
kl,k
l
(
T
,
R
,
, ),
(19)
where () denotes the complex conjugate operator and E{}
stands for the expectation operator that applies to all random
variables: the phases {
mn
} and the coordinates (x
m
, y
n
)
of the scatterers S
(mn)
. The rst term
DIF
kl,k
l
(
T
,
R
, ,
)
represents the STF-CCF of the diuse component. This
correlation function can be expressed, after substituting (2)
in (19), by
DIF
kl,k
l
(
T
,
R
,
, )
= lim
M,N
1
(c
R
+ 1)MN
M,N
_
m,n=1
E
_
c
(mn)
ll
d
(mn)
kk
e
j2[( f
(mn)
T
+f
(mn)
R
)
kl
(mn)
]
_
,
(20)
where
c
(mn)
ll
= e
j2(T/)(ll
) cos(
(mn)
T
T)
,
d
(mn)
kk
= e
j2(R/)(kk
) cos(
(mn)
R
R)
.
(21)
The quantities f
(mn)
T
, f
(mn)
R
, and
kl
(mn)
are given by (6), (7),
and (8), respectively. We recall that the AOD
(mn)
T
and the
AOA
(mn)
R
can be expressed in terms of the random variables
x
m
and y
n
according to (16) and (17), respectively.
In Section 2, it has been mentioned that all scatterers
are uniformly distributed in the two rectangular areas on
both sides of the street, as illustrated in Figure 2. Hence, the
random variables x
m
and y
n
are also uniformly distributed
over the rectangular areas. If the number of scatterers tends
to innity, that is, M, N , then the discrete random
variables x
m
and y
n
become continuous random variables
denoted by x and y, respectively. Thus, the probability
density functions (PDFs) p
x
(x) and p
y
(y) of x and y,
respectively, are given by
p
x
(x) =
1
L
A
, if x [A
1
, A
2
],
p
y
_
y
_
=
_
_
1
2B
1
, if y
_
y
T1
, B
1
+ y
T1
_
1
2B
2
, if y
_
B
2
y
T2
, y
T2
_
,
(22)
where L
A
= A
1
+ A
2
. Assuming that the random variables
x and y are independent, the joint PDF p
xy
(x, y) of the
random variables x and y can be expressed as a product of
the marginal PDFs p
x
(x) and p
y
(y), that is,
p
xy
_
x, y
_
= p
x
(x) p
y
_
y
_
=
_
_
1
2L
A
B
1
, if x [A
1
, A
2
], y
_
y
T1
, B
1
+ y
T1
_
1
2L
A
B
2
, if x [A
1
, A
2
], y
_
B
2
y
T2
, y
T2
_
.
(23)
The innitesimal power of the diuse component cor-
responding to the dierential axes dx and dy is propor-
tional to p
xy
(x, y)dxdy. As M, N , this innitesimal
contribution must be equal to 1/MN = p
xy
(x, y)dxdy.
Consequently, it follows from (20) that the STF-CCF of the
diuse component can be expressed as
DIF
kl,k
l
(
T
,
R
,
, )
=
1
2L
A
B
1
(c
R
+ 1)
_
yT
1
+B1
yT
1
_
A2
A1
c
DIF
ll
T
, x, y
_
d
DIF
kk
R
, x, y
_
e
j2[( fT(x,y)+fR(x,y))
kl
(x,y)]
dxdy
+
1
2L
A
B
2
(c
R
+ 1)
_
yT
2
B2yT
2
_
A2
A1
c
DIF
ll
T
, x, y
_
d
DIF
kk
R
, x, y
_
e
kl
(x,y)]
dxdy,
(24)
where
c
DIF
ll
T
, x, y
_
= e
j2(T/)(ll
) cos(T(x,y)T)
,
d
DIF
kk
R
, x, y
_
= e
j2(R/)(kk
) cos(R(x,y)R)
,
f
T
_
x, y
_
= f
Tmax
cos
_
T
_
x, y
_
T
v
_
,
f
R
_
x, y
_
= f
Rmax
cos
_
R
_
x, y
_
R
v
_
,
kl
_
x, y
_
=
1
c
0
_
D
(l)
T
_
x, y
_
+ D
(k)
R
_
x, y
_
_
.
(25)
Using the functions in (9) and (10), the distances D
(l)
T
(x, y)
and D
(k)
R
(x, y) can be expressed as
D
(l)
T
_
x, y
_
y
T1
sin
_
T
_
x, y
__
(M
T
2l + 1)
_
T
2
_
cos
_
T
_
x, y
_
T
_
,
D
(k)
R
_
x, y
_
y
R1
sin
_
R
_
x, y
__
(M
R
2k + 1)
_
R
2
_
cos
_
R
_
x, y
_
R
_
.
(26)
In (19), the quantity
LOS
kl,k
l
(
T
,
R
, ,
), which represents the

STF-CCF of the LOS component, can be written as
LOS
kl,k
l
(
T
,
R
,
, ) =
c
R
(c
R
+ 1)
c
(0)
ll
(
T
)
d
(0)
kk
(
R
)e
j2[( f
(0)
T
+f
(0)
R
)
kl
(0)
]
,
(27)
where
c
(0)
ll
(
T
) = e
j2(T/)(ll
) cos(T)
, (28)
d
(0)
kk
(
R
) = e
j2(R/)(kk
) cos(R)
. (29)
The Doppler shifts f
(0)
T
and f
(0)
R
are given by (12) and (13),
respectively.
4.2. Derivation of the 2D Space CCF. The 2D space
CCF
kl,k
l
(
T
,
R
) is dened as
kl,k
l
(
T
,
R
) =
E{H
kl
( f
, t)H
k
l
( f
, t)}, which is equal to the STF-CCF
kl,k
l
(
T
,
R
,
, ) in (19) by setting
and to zero, that is,
kl,k
l
(
T
,
R
) =
kl,k
l
(
T
,
R
, 0, 0)
=
1
2L
A
B
1
(c
R
+ 1)
_
yT
1
+B1
yT
1
_
A2
A1
c
DIF
ll
T
, x, y
_
d
DIF
kk
R
, x, y
_
dxdy
+
1
2L
A
B
2
(c
R
+ 1)
_
yT
2
B2yT
2
_
A2
A1
c
DIF
ll
T
, x, y
_
d
DIF
kk
R
, x, y
_
dxdy
+
c
R
(c
R
+ 1)
c
(0)
ll
(
T
)d
(0)
kk
(
R
).
(30)
4.3. Derivation of the TF-CCF. The TF-CCF of the trans-
mission link from A
(l)
T
(l = 1, 2, . . . , M
T
) to A
(k)
R
(k =
1, 2, . . . , M
R
) is dened by r
kl
(
, ) := E{H
kl
( f
, t)H
kl
( f
, t+)} [36]. The TF-CCF can be obtained directly fromthe

STF-CCF [see (19)] by setting the antenna element spacings
T
and
R
to zero, that is,
r
kl
(
, ) =
DIF
kl,k
l
(0, 0,
, ) +
LOS
kl,k
l
(0, 0,
, )
=
1
2L
A
B
1
(c
R
+ 1)
_
yT
1
+B1
yT
1
_
A2
A1
e
kl
(x,y)]
dxdy
+
1
2L
A
B
2
(c
R
+ 1)
_
yT
2
B2yT
2
_
A2
A1
e
kl
(x,y)]
dxdy
+
c
R
(c
R
+ 1)
e
j2( f
(0)
T
+f
(0)
R
)
e
j2
kl
(0)
.
(31)
4.4. Derivation of the Temporal ACF and the Doppler PSD.
The temporal ACF of the transmission link from A
(l)
T
(l =
1, 2, . . . , M
T
) to A
(k)
R
(k = 1, 2, . . . , M
R
) is dened by r
kl
() :=
E{H
kl
( f
, t)H
kl
( f
, t +)} [36, Page 376]. The temporal ACF

can be obtained directly from the TF-CCF (see (31)) by
setting at
to zero, that is, r

kl
() = r
kl
(, 0), which gives
r
kl
() =
1
2L
A
B
1
(c
R
+ 1)
_
yT
1
+B1
yT
1
_
A2
A1
e
j2[ fT(x,y)+fR(x,y)]
dxdy
+
1
2L
A
B
2
(c
R
+ 1)
_
yT
2
B2yT
2
_
A2
A1
e
j2[ fT(x,y)+fR(x,y)]
dxdy
+
c
R
(c
R
+ 1)
e
j2( f
(0)
T
+f
(0)
R
)
.
(32)
Notice that the expression in (32) reveals that the ACF r
kl
()
is independent of k and l.
Computing the Fourier transform of the temporal ACF
r
kl
() results in the Doppler PSD S
kl
( f ), that is,
S
kl
_
f
_
=
_
r
kl
()e
j2 f
d. (33)
The two most important statistical quantities character-
izing the Doppler PSD S
kl
( f ) are the average Doppler shift
B
(1)
kl
and the Doppler spread B
(2)
kl
[35]. The average Doppler
shift B
(1)
kl
is dened as the rst moment of S
kl
( f ), which can
be expressed as follows:
B
(1)
kl
=
_
f S
kl
_
f
_
df
_
S
kl
_
f
_
df
. (34)
The Doppler spread B
(2)
kl
is dened as the square root of
the second central moment of S
kl
( f ), which can be written
as
B
(2)
kl
=
_
_
_
f B
(1)
kl
_
2
S
kl
_
f
_
df
_
S
kl
_
f
_
df
.
(35)
4.5. Derivation of the FCF. The frequency characteristics of
the reference model are described by the FCF r
kl
(
). The FCF
r
kl
(
) of the transmission link from A

(l)
T
to A
(k)
R
is dened by
r
kl
(
) := E{H
kl
( f
, t)H
kl
( f
, t)} for all l = 1, 2, . . . , M

T
and k = 1, 2, . . . , M
R
. This function can be obtained directly
from the TF-CCF [see (31)] by setting to zero, that is,
r
kl
(
) = r
kl
(0,
), which results in
r
kl
(
) =
1
2L
A
B
1
(c
R
+ 1)
_
yT
1
+B1
yT
1
_
A2
A1
e
j2
kl
(x,y)
dxdy
+
1
2L
A
B
2
(c
R
+ 1)
_
yT
2
B2yT
2
_
A2
A1
e
j2
kl
(x,y)
dxdy
+
c
R
(c
R
+ 1)
e
j2
kl
(0)
.
(36)
In contrast to the temporal ACF r
kl
(), the FCF r
kl
(
)
depends on k and l.
5. Measurement-Based Computation of
the Model Parameters
The objective of this section is to determine the set of model
parameters P = {A
1
, A
2
, B
1
, B
2
, y
T1
, y
T2
, y
R1
, y
R2
, D, f
Tmax
,
f
Rmax
, c
R
} describing the reference model in such a way that
the average Doppler shift B
(1)
kl
and the Doppler spread B
(2)
kl
of the reference model match the corresponding quantities
(B
(1)
kl
and B
(2)
kl
) of the measured channel reported in [25].
To determine the set of model parameters P, we minimize
the following error:
E
min
= W
1
E
B
(1)
kl
+ W
2
E
B
(2)
kl
,
(37)
where W
1
and W
2
denote the weighting factors. The symbols
E
B
(1)
kl
and E
B
(2)
kl
in (37) stand for the absolute errors of
the average Doppler shift and Doppler spread, respectively,
which are dened as
E
B
(1)
kl
= arg min
P
B
(1)
kl
B
(1)
kl
,
(38)
E
B
(2)
kl
= arg min
P
B
(2)
kl
B
(2)
kl
.
(39)
In (38) and (39), the notation arg min
x
f (x) stands for
the argument of the minimum, which is the set of points
of the given argument for which f (x) reaches its minimum
value. At the beginning of the optimization procedure, the
weighting factors W
1
and W
2
are selected arbitrarily, but
such that they satisfy the equality W
1
+ W
2
= 1. If the
error E
B
(i)
kl
(i = 1, 2) in (37) is large, then we reduce
the corresponding weighting factor W
i
and vice versa. We
continue the optimization procedure until the result in (37)
reaches an error oor, meaning that the average Doppler shift
and the Doppler spread of the reference model best match the
measured average Doppler shift and the measured Doppler
spread, respectively.
For the measured channels in [25], the resulting opti-
mized model parameters and the corresponding average
Doppler shift and Doppler spread are listed in Table 1.
The results found for the reference model demonstrate an
excellent tting to real-world measured channels for rural,
urban, and highway propagation areas, which validates the
usefulness of the proposed reference model. It is worth
Table 1: Measurement-based parameters of the geometrical street scattering model and the resulting average Doppler shift and the Doppler
spread.
Model parameters
Propagation environment
Urban LOS Urban NLOS Rural LOS Highway LOS Highway NLOS
A
1
(A
2
) (m) 546.28 (1249) 537.03 (908.3) 546.52 (1236) 547.69 (1207) 546.88 (1193)
B
1
(B
2
) (m) 198.96 (198.77) 76.46 (1.1113) 20.89 (18.25) 199.8 (200) 0.01 (0.01)
f
Tmax
( f
Rmax
) (Hz) 223.55 (219.77) 262.1 (209.97) 463.72 (491.65) 511.68 (442.62) 491.67 (481.97)
y
T1
(y
T2
) (m) 10.42 (7) 2.12 (1.18) 15.28 (4.63) 17.62 (19.78) 1.3 (1.3)
y
R1
(y
R2
) (m) 19.82 (6.6) 20 (7.06) 14.57 (9.4) 19.63 (25) 20 (9.4)
D (m) 238.6 236.7 186.77 896.7 749.6
c
R
0.485 0 0.27 0.4 0
Measured
average Doppler 20 103 201 209 176
shift B
(1)
kl
(Hz) [25]
Theoretical
average Doppler 20 102.67 200.55 208.8 110
shift B
(1)
kl
(Hz)
Measured
Doppler 341 298 782 761 978
spread B
(2)
kl
(Hz) [25]
Theoretical
Doppler 341 298 782.03 760.88 941
spread B
(2)
kl
(Hz)
mentioning that the computed average Doppler shift B
(1)
kl
=
110 Hz and the Doppler spread B
(2)
kl
= 941 Hz do not
closely agree with the measured channel (B
(1)
kl
= 176 Hz
and B
(2)
kl
= 978 Hz) in case of the highway NLOS scenario.
For this scenario, a close agreement can be found for
suciently small values of c
R /
=0.
6. The Simulation Model
The reference model described above is a theoretical model,
which is based on the assumption that the number of
scatterers (M, N) is innite. Owing to an innite realization
complexity, the reference model is non-realizable. However,
the reference model can serve as a ground for the derivation
of stochastic and deterministic simulation models. According
to the generalized principle of deterministic channel mod-
eling [35, Sec. 8.1], a stochastic simulation model can be
derived from the reference model introduced in (1) by using
only a nite number of scatterers. In the literature, several
dierent models exist that allow for a proper simulation
of mobile channels. The SOC model is an appropriate
simulation model for mobile radio channels under non-
isotropic scattering conditions. A detailed description and
the design of SOC models can be found in [37, 38],
respectively. In [38], several parametrization techniques for
SOC models have been discussed and analyzed. Here, we use
the L
p
-norm method (LPNM), which is a high-performance
parameter computation method for the design of SOC
channel simulators.
7. Numerical Results
This section illustrates the analytical results given by (30),
(31), (32), and (36). The correctness of the analytical
results will be veried by simulations. The performance of
the channel simulator has been assessed by comparing its
temporal ACF and the FCF to the corresponding system
functions of the reference model (see (32) and (36)).
As an example for our geometrical street scattering
model, we consider rectangular scattering areas on both sides
of the street with a length of L
A
= A
1
+ A
2
, where A
1
=
50 m and A
2
= 450 m, and a width of B
1
= B
2
= 100 m.
With reference to Figure 2, the position of the transmitter
and the receiver are dened by the distances D = 400 m,
y
T1
= y
R2
= 20 m, and y
T2
= y
R1
= 10 m. For the reference
model, all theoretical results have been obtained by choosing
the following parameters:
T
= 90
,
R
= 90
,
T
v
= 0
R
v
= 180
, and f
Tmax
= f
Rmax
= 91 Hz. The Rice factor c
R
was
chosen from the set {0, 0.5, 1}. The scatterers are uniformly
distributed over the considered rectangular areas. The L
p
-
norm method has been applied to optimize the simulation
model parameters by using a nite number of scatterers
(cisoids). For the simulation model, we use MN = 50 25
scatterers (cisoids) within the rectangle on the left-hand side
as well as on the right-hand side.
In Figure 3, the absolute value of the 2D space CCF
|
11,22
(
T
,
R
)| of the reference model is presented for the
NLOS propagation scenario (c
R
= 0). The results have been
obtained by using (30). From Figure 3, we can observe that
the 2D space CCF decreases as the antenna element spacings
0.8
0.6
0.4
0.2
0
0
0
1
1
1
2
2
3
3
A
1
= 50 m
A
2
= 450 m
B
1
= B
2
= 100 m
D = 400 m
y
T1
= y
R2
= 20 m
y
T2
= y
R1
= 10 m
c
R
= 0
R
/
T
/
2
D

s
p
a
c
e

C
C
F
,
|
1
1
,
2
2
(
T
,
R
)
|
Figure 3: Absolute value of the 2D space CCF |
11,22
(
T
,
R
)| of the
reference model for a NLOS propagation scenario (c
R
= 0).
0.8
0.6
0.4
0.2
0
0
1
1
1
2
2
3
3
A
1
= 50 m
A
2
= 450 m
B
1
= B
2
= 100 m
D = 400 m
y
T1
= y
R2
= 20 m
y
T2
= y
R1
= 10 m
c
R
= 1
R
/
T
/
2
D

s
p
a
c
e

C
C
F
,
|
1
1
,
2
2
(
T
,
R
)
|
Figure 4: Absolute value of the 2D space CCF |
11,22
(
T
,
R
)| of the
reference model for a LOS propagation scenario (c
R
= 1).
increase. For comparison reasons, the absolute value of the
2D space CCF |
11,22
(
T
,
R
)| is depicted in Figure 4 for a
LOS propagation scenario (c
R
= 1). From Figure 4, one
can see that the channel transfer functions H
kl
( f
, t) and
H
k
l
( f
, t) are highly correlated over a large range of antenna

element spacings
T
and
R
. This can be concluded from
the fact that even for large antenna element spacings, for
example,
T
=
R
= 3, the absolute value of the 2D
space CCF |
11,22
(
T
,
R
)| equals approximately one half of
its maximum value. Comparing Figures 3 and 4 shows that
by increasing the Rice factor c
R
, the 2D space CCF also
increases.
Figures 5 and 6 illustrate the TF-CCFs of the refer-
ence model under NLOS and LOS propagation conditions,
respectively. From Figure 5, we can observe that the TF-CCF
decreases as the time and frequency lags increase in NLOS
propagation environments. A comparison of Figures 5 and
6 shows that the absolute value of the TF-CCF under LOS
conditions is in general higher than under NLOS.
Figure 7 depicts the absolute value of the temporal ACF
|r
kl
()| according to (32) if both the transmitter and the
receiver are moving towards each other. A good match
A
1
= 50 m
A
2
= 450 m
B
1
= B
2
= 100 m
D = 400 m
y
T1
= y
R2
= 20 m
y
T2
= y
R1
= 10 m
c
R
= 0
T
im
e
la
g
,
(
s
)
T
F
-
C
C
F
,
|
r
1
1
(
A
)
|
1
0.8
0.6
0.4
0.2
0
0
0.01
0.02
0.03
0.04 10
5
0
F
r
e
q
u
e
n
c
y
la
g
,
A
(
M
H
z
)
Figure 5: Absolute value of the TF-CCF |r
11
(
, )| of the reference
model for a NLOS propagation scenario (c
R
= 0).
A
1
= 50 m
A
2
= 450 m
B
1
= B
2
= 100 m
D = 400 m
y
T1
= y
R2
= 20 m
y
T2
= y
R1
= 10 m
c
R
= 1
T
im
e
la
g
,
(s)
T
F
-
C
C
F
,
|
r
1
1
(
A
)
|
1
0.8
0.6
0.4
0.2
0
0
0.01
0.02
0.03
0.04
10
5
0
F
r
e
q
u
e
n
c
y
la
g
,
A
(
M
H
z
)
Figure 6: Absolute value of the TF-CCF |r
11
(
, )| of the reference
model for a LOS propagation scenario (c
R
= 1).
between the temporal ACF of the reference model and that of
the simulation model can be observed in Figure 7. This gure
demonstrates also that the experimental simulation results
of the temporal ACF match very well with the theoretical
results.
Finally, Figure 8 illustrates the absolute value of the FCF
|r
kl
(
)| for dierent Rice factors c

R
= {0, 0.5, 1} if both the
transmitter and the receiver are moving towards each other.
A close agreement between the reference model and the
simulation model can be seen in Figure 8 for all chosen Rice
factors. One can realize that the experimental simulation
results of the FCF match very well with the theoretical results.
8. Conclusion
In this paper, a reference model for a wideband MIMO
C2C channel has been derived by starting from the geomet-
rical street scattering model. Taking both LOS and NLOS
Reference model
Simulation model
Simulation
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
T
e
m
p
o
r
a
l

A
C
F
,
|
r
k
l
(
)
|
c
R
= 1
c
R
= 0.5
c
R
= 0
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Time lag, (s)
A
1
= 50 m
A
2
= 450 m
B
1
= B
2
= 100 m
D = 400 m
y
T1
= y
R2
= 20 m
y
T2
= y
R1
= 10 m
Figure 7: Absolute values of the ACFs |r
kl
()| (reference model)
and | r
kl
()| (simulation model) for dierent values of the Rice
factor c
R
{0, 0.5, 1}.
Reference model
Simulation model
Simulation
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
A
1
= 50 m
A
2
= 450 m
B
1
= B
2
= 100 m
D = 400 m
y
T1
= y
R2
= 20 m
y
T2
= y
R1
= 10 m
0 1 2 3 4 5 6 7 8 9 10
Frequency lag, A
(MHz)
F
C
F
r
1
1
(
A
)c
R
= 1
c
R
= 0.5
c
R
= 0
Figure 8: Absolute values of the FCFs |r
11
(
)| (reference model)
and | r
11
(
)| (simulation model) for dierent values of the Rice

factor c
R
{0, 0.5, 1}.
propagation conditions into account, we have analyzed the
2D space CCF and the TF-CCF of the reference model. To
nd a proper simulation model, the SOC principle has been
applied. It has been shown that the SOC channel simulator
approximates the reference model with high accuracy with
respect to the temporal ACF and the FCF. An excellent
tting of the average Doppler shift and the Doppler spread
of the reference model to the corresponding quantities
of measured channels has validated the usefulness of the
proposed reference model. Further extensions of the pro-
posed wideband MIMO C2C channel model incorporating
the nonstationarity properties of real-world C2C channels
are planned for future work.
References
[1] F. Qu, F. Y. Wang, and L. Yang, Intelligent transportation
spaces: vehicles, trac, communications, and beyond, IEEE
Communications Magazine, vol. 48, no. 11, pp. 136142, 2010.
[2] http://www.ertico.com/.
[3] http://www.car-to-car.org/.
[4] E. Telatar, Capacity of multi-antenna Gaussian channels,
European Transactions on Telecommunications, vol. 10, no. 6,
pp. 585595, 1999.
[5] M. P atzold, B. O. Hogstad, and N. Youssef, Modeling,
analysis, and simulation of MIMO mobile-to-mobile fading
channels, IEEE Transactions on Wireless Communications, vol.
7, no. 2, pp. 510520, 2008.
[6] M. P atzold and B. O. Hogstad, A wideband MIMO channel
model derived from the geometric elliptical scattering model,
Wireless Communications and Mobile Computing, vol. 8, no. 5,
pp. 597605, 2008.
[7] H. Zhiyi, C. Wei, Z. Wei, M. P atzold, and A. Chelli, Modelling
of MIMO vehicle-to-vehicle fading channels in T-junction
scattering environments, in Proceedings of the 3rd European
Conference on Antennas and Propagation (EuCAP 09), pp.
652656, Berlin, Germany, March 2009.
[8] A. Chelli and M. P atzold, A MIMO mobile-to-mobile
channel model derived from a geometric street scattering
model, in Proceedings of the 4th IEEE International Symposium
on Wireless Communication Systems (ISWCS 07), pp. 792797,
Trondheim, Norway, October 2007.
[9] N. Avazov and M. P atzold, A geometric street scattering
channel model for car-to-car communication systems, in
Proceedings of the International Conference on Advanced Tech-
nologies for Communications (ATC 11), pp. 224230, Da Nang
City, Vietnam, August 2011.
[10] A. S. Akki and F. Haber, A statistical model of mobile-to-
mobile land communication channel, IEEE Transactions on
Vehicular Technology, vol. VT-35, no. 1, pp. 27, 1986.
[11] A. S. Akki, Statistical properties of mobile-to-mobile land
communication channels, IEEE Transactions on Vehicular
Technology, vol. 43, no. 4, pp. 826831, 1994.
[12] R. Wang and D. Cox, Channel modeling for ad hoc mobile
wireless networks, in Proceedings of the 55th IEEE Vehicular
Technology Conference (VTC 02), vol. 1, pp. 2125, Birming-
ham, AL, USA, May 2002.
[13] C. S. Patel, G. L. St uber, and T. G. Pratt, Simulation of
Rayleigh-faded mobile-to-mobile communication channels,
IEEE Transactions on Communications, vol. 53, no. 11, pp.
18761884, 2005.
[14] A. G. Zaji c and G. L. St uber, Space-time correlated mobile-
to-mobile channels: modelling and simulation, IEEE Trans-
actions on Vehicular Technology, vol. 57, no. 2, pp. 715726,
2008.
[15] X. Cheng, C. X. Wang, D. I. Laurenson, H. H. Ghent, and
A. V. Vasilakos, A generic geometrical-based MIMO mobile-
to-mobile channel model, in Proceedings of the International
Wireless Communications and Mobile Computing Conference
(IWCMC 08), pp. 10001005, August 2008.
[16] IEEE 802.11p, Part 11: wireless LAN medium access control
(MAC) and physical layer (PHY) specications amendment
6: wireless access in vehicular environments, IEEE standards
association, June 2010.
[17] Standard specication for telecommunications and informa-
tion exchange between roadside and vehicle systems5 GHz
band dedicated short range communications (DSRC) medium
access control (MAC) and physical layer (PHY) specications,
ASTM E2213-03, September 2003.
[18] G. Acosta, K. Tokuda, and M. A. Ingram, Measured joint
Doppler-delay power proles for vehicle-to-vehicle commu-
nications at 2.4 GHz, in Proceedings of IEEE Global Telecom-
munications Conference (GLOBECOM 04), pp. 38133817,
Dallas, Tex, USA, December 2004.
[19] P. C. F. Eggers, T. W. C. Brown, K. Olesen, and G. F.
Pedersen, Assessment of capacity support and scattering in
experimental high speed vehicle-to-vehicle MIMO links, in
Proceedings of the IEEE 65th Vehicular Technology Conference
(VTC 07), pp. 466470, April 2007.
[20] I. Sen and D. W. Matolak, Vehicle-to-vehicle channel models
for the 5-GHz band, IEEE Transactions on Intelligent Trans-
portation Systems, vol. 9, no. 2, pp. 235245, 2008.
[21] Q. Wu, D. W. Matolak, and I. Sen, 5-GHz-band vehicle-
to-vehicle channels: models for multiple values of channel
bandwidth, IEEE Transactions on Vehicular Technology, vol.
59, no. 5, pp. 26202625, 2010.
[22] J. Maurer, T. F ugen, and W. Wiesbeck, Narrow-band mea-
surement and analysis of the inter-vehicle transmission chan-
nel at 5.2 GHz, in Proceedings of the 55th Vehicular Technology
Conference (VTC 02), vol. 3, pp. 12741278, May 2002.
[23] L. Cheng, B. E. Henty, D. D. Stancil, F. Bai, and P. Mudalige,
Mobile vehicle-to-vehicle narrow-band channel measure-
ment and characterization of the 5.9 GHz dedicated short
range communication (DSRC) frequency band, IEEE Journal
on Selected Areas in Communications, vol. 25, no. 8, pp. 1501
1516, 2007.
[24] A. Paier, J. Karedal, N. Czink, H. Hofstetter, and C. Dumard,
Car-to-car radio channel measurements at 5 GHz: pathloss,
power-delay prole, and delay-Doppler spectrum, in Proceed-
ings of the 4th IEEE International Symposium on Wireless Com-
munication Systems (ISWCS 07), pp. 224228, Trondheim,
Norway, October 2007.
[25] I. Tan, W. Tang, K. Laberteaux, and A. Bahai, Measurement
and analysis of wireless channel impairments in DSRC vehic-
ular communications, in Proceedings of IEEE International
Conference on Communications (ICC 08), pp. 48824888, May
2008.
[26] J. Kunisch and J. Pamp, Wideband car-to-car radio channel
measurements and model at 5.9 GHz, in Proceedings of the
68th IEEE Vehicular Technology (VTC 08), pp. 15, September
2008.
[27] O. Renaudin, V. M. Kolmonen, P. Vainikainen, and C. Oestges,
Wideband measurement-based modeling of inter-vehicle
channels in the 5 GHz band, in Proceedings of the 5th European
Conference on Antennas and Propagation (EUCAP 11), pp.
28812885, April 2011.
[28] A. Chelli and M. P atzold, A wideband multiple-cluster
MIMO mobile-to-mobile channel model based on the geo-
metrical street model, in Proceedings of IEEE 19th Inter-
national Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC 08), pp. 16, Cannes, France,
September 2008.
[29] A. G. Zaji c, G. L. St uber, T. G. Pratt, and S. T. Nguyen,
Wideband MIMO mobile-to-mobile channels: geometry-
based statistical modeling with experimental verication,
IEEE Transactions on Vehicular Technology, vol. 58, no. 2, pp.
517534, 2009.
[30] X. Cheng, C. X. Wang, D. I. Laurenson, S. Salous, and A.
V. Vasilakos, An adaptive geometry-based stochastic model
for non-isotropic MIMO mobile-to-mobile channels, IEEE
Transactions on Wireless Communications, vol. 8, no. 9, pp.
48244835, 2009.
[31] G. D. Durgin, Space-Time Wireless Channels, Prentice Hall,
2002.
[32] S. H. Kong, TOA and AOD statistics for down link Gaussian
scatterer distribution model, IEEE Transactions on Wireless
Communications, vol. 8, no. 5, pp. 26092617, 2009.
[33] Y. Ma and M. P atzold, A wideband one-ring MIMO channel
model under non-isotropic scattering conditions, in Proceed-
ings of IEEE 67th Vehicular Technology Conference (VTC 08),
pp. 424429, Singapore, May 2008.
[34] F. Vatalaro, Doppler spectrum in mobile-to-mobile com-
munications in the precense of three-dimensional multipath
scattering, IEEE Transactions on Vehicular Technology, vol. 46,
no. 1, pp. 213219, 1997.
[35] M. P atzold, Mobile Radio Channels, John Wiley & Sons,
Chichester, UK, 2nd edition, 2011.
[36] A. Papoulis and S. U. Pillai, Probability, Random Variables and
Stochastic Processes, McGraw-Hill, New York, NY, USA, 4th
edition, 2002.
[37] M. P atzold and B. Talha, On the statistical properties of
sum-of-cisoids-based mobile radio channel simulators, in
Proceedings of the 10th International Symposium on Wireless
Personal Multimedia Communications (WPMC 07), pp. 394
400, Jaipur, India, December 2007.
[38] C. A. Guti errez and M. P atzold, The design of sum-of-
cisoids Rayleigh fading channel simulators assuming non-
isotropic scattering conditions, IEEE Transactions on Wireless
doi:10.1155/2012/756508
Research Article
A Three-Dimensional Geometry-Based Statistical Model of 2 2
Dual-Polarized MIMOMobile-to-Mobile Wideband Channels
Jun Chen and Thomas G. Pratt
Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN 46556, USA
Correspondence should be addressed to Thomas G. Pratt, tpratt@nd.edu
Received 27 April 2012; Accepted 27 July 2012
Academic Editor: Carlos A. Gutierrez
Copyright 2012 J. Chen and T. G. Pratt. This is an open access article distributed under the Creative Commons Attribution
cited.
A three-dimensional (3D) model for wide-band dual-polarized (DP) multiple-input-multiple-output (MIMO) mobile-to-mobile
(M2M) channels is proposed. Using geometrical scattering based on concentric spheres at the transmitter (Tx) and at the receiver
(Rx), a 3D parametric reference model for 22 M2M DP multipath fading channels is developed. The channel model assumes the
use of colocated half-wavelength dipole antennas for vertical and horizontal polarizations at both transmit and receive stations.
Model parameters include the velocities of the Tx and Rx nodes, the distance between the nodes, the 3D antenna pattern gains,
the azimuth and elevation angles of arrival and departure, the geometrical distribution of the scatterers, the Rician K-factors
dening the fading envelope distributions, the maximum Doppler frequency, the scattering loss factors, the cross-polar power
discrimination ratio (XPD), and the copolarization power ratio (CPR). Using the proposed model, expressions for joint time-
frequency correlation functions (TFCFs) are derived which are used to investigate system behavior over dierent wide sense
stationary uncorrelated scattering (WSSUS) channel realizations. The numerical results illustrate the sensitivities of the TFCF
to simultaneous time and frequency osets for the 2 2 DP-MIMO architectures.
1. Introduction
For nearly two decades, multiple-input-multiple-output
(MIMO) communications systems [13] have been rig-
orously studied, leading to the adoption of MIMO in
a number of wireless communications standards such as
Long Term Evolution (LTE) and 802.11n. The LTE provides
high-speed wireless communications for mobile phones
and data terminals, while 802.11n supports MIMO, frame
aggregation, and security improvements for wireless local
area networks (WLAN). In these systems, multiple antennas
can be used at both the transmitter (Tx) and receiver (Rx)
to exploit the diversity and channel capacity aorded by
MIMO architectures. In array-based MIMO systems, CP
antennas at mobile stations and access points are ideally
separated by at least one-half wavelength and at cellular base
stations by at least ten wavelengths to achieve signicant
diversity or multiplexing gains [4]. In space-constrained
implementations such as with handheld devices, use of
antenna spacings less than one-half wavelength can lead to
antenna correlations that lend themselves to beamforming
gains, but ohmic coupling eects can deplete any gains that
would otherwise be achieved with these strategies [5]. An
alternative architecture for space-constrained deployments
involves the use of dual-polarized (DP) antennas. This
strategy has the advantage of oering colocated antennas
that provide largely uncorrelated fading responses, although
usually with asymmetrical average powers among the MIMO
channel matrix entries. Hence, the channel behaviors for
DP systems are substantively dierent from conventional
copolarized (CP) channels, the latter which exhibit similar
average powers. The relative performance of DP-MIMO in
comparison to traditional MIMO with CP arrays will be
dictated by the disparities in these responses. While DP
architectures are receiving increasing attention, they have
thus far been studied with much less rigor than CP systems.
To characterize DP systems, especially in wideband
signaling applications, modeling of dispersive input-to-
output polarimetric channel behavior is important. Many
theoretical models [611] provide statistical representations
of polarimetric channels based on cross-polarization
coupling (XPC), but do not characterize polarization-
frequency behavior that is needed to represent wider-
band, frequency-selective channels for the analysis of DP
communications systems. For example, XPC is most often
analyzed in the narrowband sense and yet it has been shown
that the XPC can vary with frequency. It also exhibits
correlation in the frequency dimension in a manner that
depends on the channels temporal dispersion properties
and polarization coupling [12]. Statistical channel models in
literature that address time-selective and frequency-selective
behavior often assume that the time dispersion (due to
time delays) and the frequency dispersion (due to Doppler
spread) are statistically independent, where the time delays
depend on the relative locations of the random scatterers and
the Doppler spreads depend on the motion of the Tx and
Rx antennas [13, 14]. However, this assumption is broken
if both the time delays and the Doppler spreads depend
on the relative location of the random scatterers (e.g., the
angles of departure and the angles of arrivals). Because
these phenomena would not be expected to be statistically
independent, it is of interest to examine joint correlation
functions in time and frequency, which is an application of
the model that we consider in our paper.
Analytical approaches such as Kronecker or eigenbeam
models have also been used for model simulations, and
these have dealt with narrowband channels. An analytical
framework based on a Kronecker model is presented in
[15] to model narrowband DP Rayleigh and Rician fading
channels for arbitrary array sizes. The framework uses a
relatively small number of physical parameters to analyze
the benets of multiple polarization architectures. In other
work, a 3D polarized narrowband spatial channel model
is presented in [7], and the impact of elevation angle on
capacity is reported in [16]. These models assume ideal
dipole antennas and unity cross-polar discrimination ratios.
Another approach to characterize channels for DP
systems involves the use of geometric scattering models.
Geometrical scattering models [1719] have the advantage
that they are able to represent important channel behav-
iors, including mutual eects that arise from polarization
coupling, time-dispersion, angle-dispersion, and frequency-
dispersion. However, they also are more complex. Early two-
dimensional (2D) geometric scattering models were devel-
oped for narrowband CP single-input single-output (SISO)
mobile-to-mobile (M2M) Rayleigh fading channels [2022].
In [23], the polarisation-sensitive geometric modelling is
developed with a direction-of-arrival (DOA) distribution
that depends on the polarisation states of the transmitting
antennas, the receiving antennas and the polarisation prop-
erties of the scatterers. These models were 2D in the sense
that the models treated electromagnetic propagation only in
a xed-elevation plane and did not model elevation antenna
pattern dependencies or distributions of scatterers in the
elevation dimension.
Some three-dimensional (3D) models have been pro-
posed to overcome shortcomings of 2D M2M channel mod-
els, particularly in environments where deployed antenna
heights are lower than surrounding buildings and obstacles.
In [18], a 3D wideband M2M mathematical reference model
was proposed based on a concentric-cylinders geometry
using a superposition of line-of-sight (LOS), single bounce
at the transmit side (SBT), single bounce at the receive
side (SBR), and double-bounce (DB) rays in a variety of
urban environments. The analysis approach proved useful
to characterize M2 M correlations, but to date has only been
applied to the analysis of CP MIMO channels with unity-
gain idealized dipole antennas. Recently, a 3D polarized
channel model has been proposed to treat spatially-separated
orthogonally polarized elements [19]. The model deals
exclusively with the double-bounce ray and narrowband
channels.
In this paper, we develop a 3D geometric scattering
model for 2 2 M2M DP wideband channels based on
concentric spheres to evaluate joint time frequency cor-
relation functions associated with the subchannel fading
envelopes. The use of concentric spheres is motivated by
the ease with which scatterer locations may be identied
for a given radius, azimuth AOD, and elevation AOD,
and by its capability to support the modeling of overhead
reectors. The DP-MIMO channel is constructed assuming a
DP antenna at both the Tx and the Rx, where the polarization
basis at the Rx is matched to the polarization basis used
at the Tx. The received signals can be translated to an
arbitrary orthogonally-polarized basis by applying a unitary
transformation to the received signal vectors [24] without
impacting theoretical performance measures such as capacity
and diversity.
We assume a channel with wide sense stationary uncor-
related scattering (WSSUS) where the channel correlation
function is invariant over time, and the scatterers with
dierent path delays are uncorrelated [25, 26]. Channel
transfer functions for each transmit antenna/receive antenna
pair are derived as a superposition of LOS, SBT, SBR, and
DB component rays. The transfer functions are used to
form time-frequency correlation functions (TFCF) for the
assumed 3D nonisotropic scattering environment, where
scattering distributions are characterized in the azimuth
dimension through the von Mises distribution and in
the elevation dimension by a cosine distribution. The
3D scattering propagation model used to evaluated the
TFCFs simulates the eects of the antenna pattern gains,
the geometrical distribution of scatterers and the asso-
ciated azimuth/elevation angles of arrival and departure,
the K-factor of the fading distributions, the maximum
Doppler frequency, the scattering loss factors, the cross-
polar discrimination (XPD), and the copolarization ratio
(CPR).
The remainder of the paper is organized as follows.
The 3D DP 2 2 MIMO M2M model is presented in
Section 2. In Section 3, we derive transfer functions and
TFCFs for each subchannel of the DP-MIMO channel
and for each ray path type for 3D nonisotropic scattering
environments. Numerical results of the joint correlation
functions associated with the 2 2 DP channel models are
presented in Section 4. We conclude with a summary of our
ndings in Section 5.

(m)
r
Z
S
(m)
t
R
t
Y
(n)
t
(m)
t
v
t
vt
A
(p)
t
O
t
X
A
( p)
t
D
Z
S
(n)
r
Y
vr
(m)
r
(n)
r
R
r
v
r
A
(q)
r
O
r
X
LOS
t
LOS
r
A
( q)
r
Tx
Rx
p,m

vt
LOS
t
(m)
t
(n)
t
m,n

vr
LOS
r

n,q
(n)
r
Figure 1: Concentric-spheres for 3D colocated DP channel model. The signals from the Tx antennas arrive in parallel at the Rx antenna
array.
2. 2 2 DP-MIMOChannel Model Based on
Concentric Spheres
The concentric sphere geometry-based scattering model
assumes a mobile Tx and a mobile Rx, both equipped
with DP antennas using a matched polarization basis. Radio
propagation between the Tx and the Rx is characterized by
3D WSSUS under channels that can include line-of-sight
(LOS) and non-line-of-sight (NLOS) components, where
scattering centers in the latter case reside on concentric
spheres about either the Tx, the Rx, or both. The linearly
time-variant MIMO channel can be represented by a 2 2
impulse response matrix in terms of time t and delay
H(t, ) =
_
h
qp
(t, ) h
q ` p
(t, )
h
` qp
(t, ) h
` q ` p
(t, )
_
. (1)
Figure 1 illustrates the concentric sphere model for a
MIMO M2M channel with DP antennas. One DP antenna
with vertically-polarized and horizontally-polarized compo-
nents denoted by A
(p)
t
and A
( ` p)
t
, respectively, is located at
the center of the Tx sphere (O
t
). A second DP antenna
with corresponding components denoted by A
(q)
r and A
(` q)
r
is located at the center of the Rx sphere (O
r
). At the Tx, M
xed scatterers reside within the volume of a sphere dened
by a radius R
t
. The mth transmit scatterer is denoted by S
(m)
t
,
where 1 m M and resides on the surface of a sphere with
radius R
m
, where R
m
< R
t
. Similarly at the Rx, N scatterers
occupy the spherical volume with a radius R
r
. The nth receive
scatterer is denoted by S
(n)
r , where 1 n N and resides
on the surface of a sphere with radius R
n
, where R
n
< R
r
.
The set of scatterers is comprised of H scatterers (reecting
horizontally polarized waves) and V scatterers (reecting
vertical polarized waves). We assume that the distribution of
the H scatterers and the V scatterers are identical and that
the number of scatterers for each are the same, although this
may not be true in general.
The center of the Tx sphere serves as the global origin of
a rectangular coordinate system. At time t = 0, the Rx is a
distance D from the Tx with XYZ coordinates denoted by
(
x
,
y
,
z
). The height dierence between the Tx and the
Rx antennas is included in the oset
z
. The symbols
p,m
,
m,q
,
p,n
,
n,q
,
m,n
and
p,q
denote distances d(A
(p)
t
, S
(m)
t
),
d(S
(m)
t
, A
(q)
r ), d(A
(p)
t
, S
(n)
r ), d(S
(n)
r , A
(q)
r ), d(S
(m)
t
, S
(n)
r ), and
d(A
(p)
t
, A
(q)
r ), respectively, where d() denotes the distance
between the two coordinates. The symbols
(m)
t
,
(n)
t
are the
elevation angles of departure (EAoD, relative to the XY
plane) to the scatterers S
(m)
t
and S
(n)
r , respectively, whereas
(m)
t
,
(n)
t
are the azimuth angles of departure (AAoD, in
the XY plane relative to the Z-axis) to the scatterers S
(m)
t
and S
(n)
r respectively. Similarly, the symbols
(m)
r ,
(n)
r denote
the elevation angles of arrival (EAoA, relative to the x-y
plane) reected from the scatterers S
(m)
t
and S
(n)
r , respectively,
whereas
(m)
r ,
(n)
r denote the azimuth angles of arrival
(AAoA, in the XY plane relative to the z-axis) reected
from the scatterers S
(m)
t
and S
(n)
r respectively. For the LOS
component between the Tx and the Rx, the symbols
LOS
t
and
LOS
t
are the elevation angle of departure (relative to the XY
plane) and the azimuth angle of departure in the XY plane
relative to the Z-axis, respectively. Similarly, the symbols
LOS
r
and
LOS
r
are the elevation angle of arrival and the azimuth
angle of arrival, respectively. The Tx and Rx are moving with
speeds v
t
and v
r
in directions described by the elevation
angles
vt
and
vr
relative to the XY plane, respectively,
and by the azimuth angles
vt
and
vr
in the XY plane,
respectively.
2.1. Scatterer Distributions. The directions of scatterers are
described by azimuth and elevation angle distributions.
Several dierent distributions, such as uniform, Gaussian,
Laplacian, and von Mises, have been used in the literature to
characterize the azimuth angles of departure and arrival. von
Mises Fisher (VMF) distributions are eective to model both
AAoD and AAoA for spatial fading correlation models [27]
and have the advantage that the probability density function
approximates many of distributions and admits closed-form
solutions [18]. To simplify the modeling, it is assumed that
the azimuth and elevation angles are independent, enabling
the use of a product of distribution functions instead of a
joint distribution. The scatterer distributions are synthesized
using dierent angle distributions for the azimuth and the
elevation dimensions. The Von Mises Fisher probability
density function (pdf) distribution is used for azimuth
dimensions, including for angles of departure (
(m)
t
and
(n)
r )
and angles of arrival (
(n)
t
and
(m)
r ). The von Mises pdf is
dened as [28]
f
_
_
=
exp
_
cos
_

__
2I
0
()
, [, ), (2)
where [, ) is the mean value of the scatterer
directions in the azimuth plane, controls the spread of
scatterers around the mean , and I
0
() is the zero-order
modied Bessel function of the rst kind. When =
0, f ()=1/2 is a uniform distribution, corresponding to
isotropic scattering in azimuth. As increases, the scatterers
become more clustered about the mean angle , and
scattering is nonisotropic.
The random elevation angles of departure and arrival
can be characterized by a uniform, cosine or Gaussian
distribution. In [18, 29], a cosine pdf is used as it may t with
the typical propagation in M2M communications, where the
Tx and Rx are in motion and equipped with low elevation
antennas. A cosine distribution employed for the elevation
angles of departure and arrival is given by [29]:
f () =
_
4|
max
|
cos
_
max
_
, || |
max
|

2
,
0, otherwise.
(3)
where
max
is the maximumelevation angle which we assume
to have a value near 20
. This maximum elevation angle is

typical of M2M wireless communications where both the Tx
and Rx are equipped with low elevation antennas.
The above-mentioned distributions are used to represent
the response from a cluster, which describes a group of scat-
terers located within an isolated solid angle, and the response
from multiple clusters compose the aggregate response from
the channel. In [27], a mixture of Von Mises distributions
is proposed for modeling the 3D direction of scatterers
in the presence of multiple clusters of scatterers over the
propagation channel. Mathematically, each distribution in
the mixture of M VMFs at the Tx or N VMFs at the Rx
can be described as f (
(m)
t
|
(m)
t
,
m
)(1 m M) or
f (
(n)
r |
(n)
r
,
n
)(1 n N). Note that
(m)
t
or
(n)
r
is the
mean azimuth angle of the mth or nth cluster;
m
or
n
is the concentration of the mth or nth cluster. Hence, the
overall density functions of the mixture model consisting
of M VMF distributions for the cluster azimuth angles
(m)
t
at the transmitter and N VMF distributions for the cluster
azimuth angles
(n)
r at the Rx can be described as [27]
f
_
t
_
=
M
_
m=1
m
_
f
(m)
t
|
(m)
t
,
m
_
,
f
_
r
_
=
N
_
n=1
n
_
f
(n)
r
|
(n)
r
,
n
_
,
(4)
where M and N the number of clusters at the Tx and Rx,
respectively;
m
or
n
is dened as the a priori probability
that mth or nth cluster was generated. Similarly, the overall
density functions of the mixture model consisting of the
cosine distributions for the elevation angles
(m)
t
at the Tx
and
(n)
r at the Rx can be expressed as
f (
t
) =
M
_
m=1
m
_
(m)
t
| f
(m)
t
_
,
f (
r
) =
N
_
n=1
n
_
(n)
r
|
(n)
r
_
,
(5)
where
(m)
t
and
(n)
r
are the mean elevation angle of the mth
or nth cluster at the Tx and Rx, respectively.
Signals from the Tx antenna elements that propagate
directly to the Rx antenna elements form the LOS compo-
nent. Signals reected exclusively from the scatterers located
around the Tx before arriving at the Rx antenna elements
are collectively called the SBTcomponent. Similarly, transmit
signals reected only by scatterers located around the Rx
before arriving at the Rx antenna elements form the SBR
component. The DB component is formed from the signals
that are reected from scatterers about both the Tx and
the Rx before arriving at the Rx antenna elements. For
each realization of the WSSUS channel, the channel impulse
responses and TFCFs can be written as superpositions of the
LOS, SBT, SBR, and DB signal components.
The channel impulse responses for the subchannels
A
(p)
t
A
(q)
r , A
( ` p)
t
A
(q)
r , A
(p)
t
A
(` q)
r and A
( ` p)
t
A
(` q)
r
can be written as a superposition of the LOS, SBT, SBR and
DB signals,
h
ab
(t, ) = h
SBR
ab
(t, ) + h
SBR
ab
(t, ) + h
DB
ab
(t, ) + h
LOS
ab
(t, ),
(6)
where the h
l
ab
(t, )(a {p, ` p}, b {q, ` q}, l
{LOS, SBT, SBR, DB}) represent the impulse response func-
tions between antenna elements a and b along the l path
respectively, and h
ab
(t, ) represents the total response
between antenna elements a and b.
2.2. Single Bounce Channel Impulse Response. The time
varying impulse responses of the single-bounce components
are given by
h
SBT
ab
(t, ) =

t, pq
K + 1
1
N
M
_
n=1
a,n,b
g
a,n,b
(t)
_

a,m,b
_
__
G
(v)
a
_
(m)
t
,
(m)
t
_
_
G
(h)
a
_
(m)
t
,
(m)
t
_
_
_
_
_
_
1
_
r
(m)
p` q
_
r
(m)
` pq
1
_
_
_
_
_
_
_
_
_
G
(v)
b
_
(m)
r ,
(m)
r
_
_
G
(h)
b
_
(m)
r ,
(m)
r
_
_
_
_
_
exp
_
j
(m)
ab
_
SBT
m
,
h
SBR
ab
(t, ) =

t, pq
K + 1
1
N
M
_
n=1
a,n,b
g
a,n,b
(t)
_

a,n,b
_
__
G
(v)
a
_
(n)
t
,
(n)
t
_
_
G
(h)
a
_
(n)
t
,
(n)
t
_
_
_
_
_
_
1
_
r
(n)
p` q
_
r
(n)
` pq
1
_
_
_
_
_
_
_
_
_
G
(v)
b
_
(n)
r ,
(n)
r
_
_
G
(h)
b
_
(n)
r ,
(n)
r
_
_
_
_
_
exp
_
j
(n)
ab
_
SBR
n
,
(7)
where the parameter K is the Rician K factor which is a ratio
of the power through the LOS path relative to the power
through the scattered paths. The power ratio coecients
t,ab
and
r,ab
(a {p, ` p}, b {q, ` q}) in (7) specify the
fraction of the power of the single-bounce components along
the subchannels A
(a)
t
A
(b)
r with respect to the average
transmit power P
ab
between the Tx antenna element a and
the Rx antenna element b. The relationships among these,
and related power ratio coecients are shown in (29).
a,m,b
and
a,n,b
denote the signal amplitudes and
a,m,b
and
a,n,b
denote time delays of the multipath components propagating
from antenna element a to antenna element b via transmit
scatterer m and from antenna element a to antenna element
b via receive scatterer n (a {p, ` p}, b {q, ` q}), respectively.
G
(v)
a (, ), G
(h)
a (, ), G
(v)
b
(, ), and G
(h)
b
(, ) (a
{p, ` p}, b {q, ` q}) denote the antenna patterns of the
pth, ` pth, qth, and ` qth antenna element with vertical(v) or
horizontal(h) polarizations, respectively. We assume that a
half-wavelength dipole is used for each element where the
spherical coordinate system dening antenna orientations
are illustrated in Figure 2. The antenna feed is situated at
Z
Y
X
P(r, , )
Dipole
antenna
Figure 2: Half-wavelength dipole antenna and its coordinate
system.
the origin of the coordinate system and the antenna elements
are inclined at the angle from the Z axis in the vertical
ZX plane. A thin dipole is assumed and the element radius
is ignored. The radiation power gain patterns of the half-
wavelength dipole for vertical and horizontal polarizations
are given by [30]
G
(v)
_
,
_
= 1.641
_
cos cos sin sin cos
_
2
cos
2
(/2)
(1
2
)
2
,
G
(h)
_
,
_
=1.641 sin
2
sin
2
cos
2
(/2)
(1
2
)
2
,
(8)
where
= sin cos sin + cos cos . (9)
For illustration, antenna gain patterns of the half-
wavelength dipole with inclination angles of 0 and /2 are
shown in Figures 3 and 4 respectively.
a,m,b
and
a,n,b
denote time delays of the multipath
components propagating from antenna element A
(a)
t
to
antenna element A
(b)
r via the scatterer S
(m)
t
and from antenna
element A
(a)
t
to antenna element A
(b)
r via the scatterer S
(n)
r ,
respectively. g
a,m,b
(t) and g
a,n,b
(t) are the time-varying phase
functions along the path A
(a)
t
S
(m)
t
A
(b)
r and a
S
(n)
r b, respectively. Since the antenna elements are
collocated and share a common phase center, the time delays
a,m,b
or
a,n,b
are equivalent between antennas elements A
(a)
t
and A
(b)
r , that is,
p,m,q
=
p,m, ` q
=
` p,m,q
=
` p,m, ` q
,
p,n,q
=
p,n, ` q
=
` p,n,q
=
` p,n, ` q
.
(10)
Similarly, g
a,m,b
(t) and g
a,n,b
(t) satisfy
g
p,m,q
(t) = g
p,m, ` q
(t) = g
` p,m,q
(t) = g
` p,m, ` q
(t),
g
p,n,q
(t) = g
p,n, ` q
(t) = g
` p,n,q
(t) = g
` p,n, ` q
(t).
(11)
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
2
0
1
2
0
1
2
1
2
1
0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4 X
Z
Y
Figure 3: Gain patterns of half-wavelength dipole antenna with
inclination angle of 0: G
(v)
(, ). The horizontal component
G
(h)
(, ) is zero.
2
1
0
1
2
2
1
0
1
2
0.02
0.04
0.06
0.08
0.1
0.12
2
1
0
1
2
Z
Y
X
(a)
2
1
0
1
2
2
1
0
1
2
0.02
0.04
0.06
0.08
0.1
X Y
Z
2
1
0
1
2
0
(b)
Figure 4: Gain patterns of half-wavelength dipole antenna with
inclination angle of /2: (a) G
(v)
(, ), and (b) G
(h)
(, ).
The random variable
(m)
ab
represents the uniformly-
distributed random phase oset associated with the path to
each scatterer S
(m)
t
between the V or H component of the
Tx antenna elements and the V or H component of the Rx
antenna elements. Similarly, the random variable
(n)
ab
is the
uniformly-distributed phase oset associated with the path
to each scatterer S
(n)
r between the V or H component of the
Tx antenna elements and the V or H component of the Rx
antenna elements.
r
(m)
` pq
, r
(n)
p` q
, r
(n)
` pq
, and r
(n)
` pq
represent the inverse of the channel
cross-polar discrimination ratios along the single bounce
path to scatterer m or scatterer n, and have statistical means
that follow
E
_
r
(m)
p` q
_
= E
_
r
(n)
p` q
_
=
1
XPD
p
,
E
_
r
(m)
` pq
_
= E
_
r
(n)
` pq
_
=
1
XPD
` p
.
(12)
Here, the channel cross-polar discrimination for each
transmit polarization is dened as
XPD
p
=
E
h
qp
2
E
_
h
` qp
2
_
,
XPD
` p
=
E
_
h
` q ` p
2
_
E
_
h
` q ` p
2
_
,
(13)
where E{} denotes the expectation operator.
The XPD depends on channel parameters and the
environment, such as the distance between the Tx and the Rx,
the angles of arrival and departure (both azimuth and
elevation), the delay spread of the multipath components,
and the transmit and receive antenna polarization basis.
We dene a parameter CPR
qp` q ` p
as the copolar power
ratio between the average powers transmitted through the
vertical-vertical subchannel and the average powers trans-
mitted through the horizontal-horizontal subchannel:
CPR
qp, ` q ` p
=
E
_
h
` q ` p
2
_
E
_
h
` q ` p
2
_
. (14)
The CPR depends on the Brewster angle phenomenon [31].
The XPD and CPR, when expressed in decibel (dB),
are often observed as having the normal distribution with
N(, ). In [7], the mean of XPD varies from 0 to 18 dB,
with the standard deviation in order of 3 8 dB. Normally,
the received power in the vertical-to-vertical transmission
is reported to be greater than that in the horizontal-to-
horizontal transmission (CPR > 0 dB). For example, the
mean of the CPR is reported to vary between 0 and
6 dB [15]. Depending on the propagation environment and
transmission conguration, the CPR may be less than 0 dB
when the amplitude of vertically polarized waves is degraded
more than that of horizontally polarized waves or the
transmission power of horizontally polarized waves is greater
than that of vertically polarized waves. When the XPD or
CPR in dB is denoted as X, the expectations in (12) can be
computed as
E
_
_
r
(m)
p` q
_
= E
__
r
(m)
` pq
_
= E
_
10
(X/20)
_
,
E
_
1
CPR
qp, ` q ` p
_
= E
_
10
(X/20)
_
.
(15)
When X has the normal distribution with N(, ), then
the expectation E{10
X/20
} over an interval [x
1
, x
2
] can be
expressed as
E
_
10
(X/20)
_
=
_
x2
x1
10
(x/20)
1
2
2
exp
_
_
x
_
2
2
2
_
dx.
(16)
For simplicity, we adopt the normal XPD model to
characterize cross-polarized components in the 3D modeling
of 2 2 DP-MIMO M2M channels.
When the elevation angles of arrival/departure are
small, the distances d(A
(p)
t
, S
(m)
t
), d(S
(m)
t
, A
(q)
r ), d(A
(p)
t
, S
(n)
r ),
d(S
(n)
r , A
(q)
r ), d(S
(m)
t
, S
(n)
r ) and d(A
(p)
t
, A
(q)
r ) can be approxi-
mated as below
d
_
A
(p)
t
, S
(m)
t
_
=
p,m
R
t
,
d
_
S
(m)
t
, A
(q)
r
_
=
m,q
D,
d
_
A
(p)
t
, S
(n)
r
_
=
p,n
D,
d
_
S
(n)
r
, A
(q)
r
_
=
n,q
R
r
,
d
_
A
(p)
t
, A
(q)
r
_
=
pq
D,
d
_
S
(m)
t
, S
(n)
r
_
=
m,n

_
(R
t
R
r
)
2
+ D
2
.
(17)
Furthermore, using the Taylor series expansion, the
amplitudes of the multipath components,
p,m,q
or
p,m, ` q
is
approximately given by
p,m,q
=
p,m, ` q
=
_
d
_
A
(p)
t
, S
(m)
t
_
d
_
S
(m)
t
, A
(q)
r
_
_
/2
A
m
_
1

2
R
t
D
_
,
(18)
where is the path loss exponent, whose value is normally
in the range of 2 to 4, depending upon the propagation
environment.
` p,m, ` q
or
` p,m,q
is approximated as
` p,m, ` q
=
` p,m,q
=
1
CPR
qp, ` q ` p
_
d
_
A
(p)
t
, S
(m)
t
_
d
_
S
(m)
t
, A
(q)
r
_
_
/2
A
m
1
CPR
qp, ` q ` p
_
1

2
R
t
D
_
,
(19)
where A
m
is dened as
A
m
= D
/2
. (20)
Similarly,
p,n,q
or
p,n, ` q
is approximated as
p,n,q
=
p,n, ` q
=
_
d
_
A
(p)
r , S
(n)
r
_
d
_
S
(n)
r
, A
(q)
r
_
_
/2
A
m
_
1

2
R
r
D
_
,
(21)
` p,n, ` q
or
` p,n,q
is approximated as
` p,n, ` q
=
` p,n,q
=
1
CPR
qp, ` q ` p
_
d
_
A
(p)
r , S
(n)
r
_
d
_
S
(n)
r
, A
(q)
r
_
_
/2
A
m
1
CPR
qp, ` q ` p
_
1

2
R
r
D
_
.
(22)
The time delays
p,m,q
and
p,n,q
are the travel times of the
signals scattered from the Tx scatterer S
(m)
t
, and from the Rx
scatterer S
(n)
r , respectively
p,m,q
=

p,m
+
m,q
c
, (23)
p,n, p
=

p,n
+
n,q
c
, (24)
where c is the speed of the light.
The time-varying phase function g
a,m,b
(t) along the path
A
(p)
t
S
(m)
t
A
(q)
r is given by
g
p,m,q
(t) = exp
_
j
2
c
_
p,m
+
m,q
_
_
exp
_
j2t f
t max
cos
_
(m)
t

vt
_
cos
(m)
t
cos
vt
_
exp
_
j2t f
t max
sin
(m)
t
sin
vt
_
exp
_
j2t f
r max
cos
_
(m)
r

vr
_
cos
(m)
r
cos
vr
_
exp
_
j2t f
r max
sin
(m)
r
sin
vr
_
,
(25)
where f
t max
= vt/
c
and f
r max
= vr/
c
are the maximum.
Doppler frequencies associated with the Tx and Rx
respectively and
c
denotes the carrier wavelength.
The time-varying phase function g
p,m,b
(t) along the path
A
(p)
t
S
(n)
r A
(q)
r is given by
g
p,n,q
(t) = exp
_
j
2
c
_
p,n
+
n,q
_
_
exp
_
j2t f
t max
cos
_
(n)
t

vt
_
cos
(n)
t
cos
vt
_
exp
_
j2t f
t max
sin
(n)
t
sin
vt
_
exp
_
j2t f
r max
cos
_
(n)
r

vr
_
cos
(n)
r
cos
vr
_
exp
_
j2t f
r max
sin
(n)
r
sin
vr
_
.
(26)
The parameters
SBT
m
and
SBR
n
represent scattering loss
factors, and are governed by [32]
= exp
_
8
_
h
sin
i
c
_
2
_
I
0
_
8
_
h
sin
i
c
_
2
_
, (27)
where
h
is the standard deviation of the surface height of
the scatterer,
i
is a given angle of incidence, and I
0
() is the
zero-order modied Bessel function of the rst kind.
2.3. Double-Bounce Channel Impulse Response. The double-
bounce components of the channel impulse responses are
h
DB
pq
(t, )
=
tr, pq
K + 1
1
MN
M
_
m=1
N
_
n=1
a,m,n,b
g
a,m,n,b
(t)
_

a,m,n,b
_
_
_
G
(v)
a
_
(m)
t
,
(m)
t
_
_
G
(h)
a
_
(m)
t
,
(m)
t
_
_
_
_
_
_
1
_
r
(m)
p` q
_
r
(m)
` pq
1
_
_
_
_
_
_
_
_
1
_
r
(n)
p` q
_
r
(n)
` pq
1
_
_
_
_
_
_
_
_
_
G
(v)
b
_
(n)
r ,
(n)
r
_
_
G
(h)
b
_
(n)
r ,
(n)
r
_
_
_
_
_
exp
_
j
(m,n)
ab
_
SBT
m

SBR
n
,
(28)
where the power ratio coecients
tr,ab
(a {p, ` p}, b
{q, ` q}) in (28), specify the ratio of the power of the double-
bounce ray along the subchannels A
(a)
t
A
(b)
r with respect
to the averaged transmit power P
ab
between the Tx antenna
element a and the Rx antenna element b. The power ratio
coecients of the single-bounce and the double-bounce rays
satisfy
t,ab
+
r,ab
+
tr,ab
= 1,
_
a
_
p, ` p
_
, b
_
q, ` q
__
. (29)
a,m,n,b
denotes the amplitude and
a,m,n,b
denotes the time
delay of the multipath components along the path A
(a)
t

S
(m)
t
S
(n)
r A
(b)
r (a {p, ` p}, b {q, ` q}). The amplitudes
of the multipath components
p,m,n,q
and
p,m,n, ` q
are given by
p,m,n,q
=
p,m,n, ` q
d
_
A
(p)
t
, S
(m)
t
_
d
_
S
(m)
t
, S
(n)
r
_
d
_
S
(n)
r
, A
(q)
r
_
_
/2
A
m
_
1

2
R
t
+ R
r
D
_
.
(30)
The amplitudes of the multipath components
` p,m,n, ` q
and
` p,m,n,q
are given by
` p,m,n, ` q
=
` p,m,n,q
=
1
CPR
qp, ` q ` p
d
_
A
(p)
t
, S
(m)
t
_
d
_
S
(m)
t
, S
(n)
r
_
d
_
S
(n)
r
, A
(q)
r
_
_
/2
A
m
1
CPR
qp, ` q ` p
_
1
2
R
t
+ R
r
D
_
.
(31)
As DP antenna pair are colocated at the center of Tx and
the center of Rx without considering the size of antennas, the
time delays
a,m,n,b
are equivalent between antennas A
(a)
t
and
A
(b)
r
p,m,n,q
=
p,m,n, ` q
=
` p,m,n,q
=
` p,m,n, ` q
. (32)
The time delay
p,m,n,q
is the travel time of the signals
scattered from both the Tx scatterer S
(m)
t
and the Rx scatterer
S
(n)
r
p,m,n,q
=

p,m
+
m,n
+
n,q
c
. (33)
Similarly, g
a,m,n,b
(t) satisfy
g
p,m,n,q
(t) = g
p,m,n, ` q
(t) = g
` p,m,n,q
(t) = g
` p,m,n, ` q
(t). (34)
The time-varying function g
p,m,n,q
(t) is given by
g
p,m,n,q
(t)=exp
_
j
2
c
_
p,m
+
m,n
+
n,q
_
_
exp
_
j2t f
t max
cos
_
(m)
t

vt
_
cos
(m)
t
cos
vt
_
exp
_
j2t f
t max
sin
(m)
t
sin
vt
_
exp
_
j2t f
r max
cos
_
(n)
r

vr
_
cos
(n)
r
cos
vr
_
exp
_
j2t f
r max
sin
(n)
r
sin
vr
_
.
(35)
The random variable
(m,n)
ab
is the phase oset of the
double-bounce path scattered by both the Tx scatterer S
(m)
t
and the Rx scatterer S
(n)
r between the V or H component of
the Tx antenna elements and the V or H component of the
Rx antenna elements.
2.4. LOS Channel Impulse Response. The LOS components of
channel impulse responses are
h
LOS
ab
(t, ) =
K
K + 1
LOS
a,b
g
LOS
a,b
(t)
_

LOS
a,b
_
_
G
a
_
LOS
t
,
LOS
t
_
G
b
_
LOS
r
,
LOS
r
_
,
(36)
where
LOS
a,b
(a {p, ` p}, b {q, ` q}) denotes the amplitude
and
LOS
a,b
denotes the time delay of the LOS components
between the antenna element A
(a)
t
at the Tx and the antenna
element A
(b)
r at the Rx. The time-varying function g
LOS
a,b
(t) is
dened as
g
LOS
p,q
(t) =g
LOS
p, ` q
(t) = g
LOS
` p,q
(t) = g
LOS
` p, ` q
(t)
= exp
_
j
2
p,q
_
exp
_
j2t f
t max
cos
_
LOS
t

vt
_
cos
LOS
t
cos
vt
_
exp
_
j2t f
t max
sin
LOS
t
sin
vt
_
exp
_
j2t f
r max
cos
_
LOS
r

vr
_
cos
LOS
r
cos
vr
_
exp
_
j2t f
r max
sin
LOS
r
sin
vr
_
.
(37)
The amplitude of the LOS components
LOS
a,b
is given by
LOS
p,q
=
LOS
p, ` q
D
2
= A
m
,
LOS
` p, ` q
=
LOS
` p,q

1
CPR
qp, ` q ` p
D
2
= A
m
1
CPR
qp, ` q ` p
.
(38)
The time delay
LOS
a,b
is the travel time of the signal from
the Tx antenna element A
(a)
t
and the Rx antenna element
A
(b)
r . Consider the following:
LOS
p,q
=
LOS
p, ` q
=
LOS
` p,q
=
LOS
` p, ` q
=

p,q
c
. (39)
It is assumed that the elevation angles (
LOS
t
,
LOS
r
,
(m)
t
,
(n)
t
,
(m)
r , and
(n)
r ) and the azimuth angles (
LOS
t
,
LOS
r
,
(m)
t
,
(n)
t
,
(m)
r and
(n)
r ) are independent random variables. The
radii R
t
of the Tx sphere and R
r
of the Rx sphere are also
independent. The phase osets
(m)
ab
,
(n)
ab
, and
(m,n)
ab
are
assumed to be uniformly random variables on the interval
[, ) that are independent from the elevation angles,
azimuth angles, and radii of the scattering spheres. Using
the Central Limit Theorem [33, 34], we posit that the
delay-spread functions h
SBT
ab
(t, ), h
SBR
ab
(t, ), h
LOS
ab
(t, ), and
h
DB
ab
(t, ) are zero-mean complex Gaussian random pro-
cesses.
2.5. Time-Variant Transfer Functions. The time-variant
transfer function is the Fourier transform of the channel
impulse response with respect to the delay . Consider the
following:
T
ab
_
t, f
_
= F
{h
ab
(t, )} = T
SBT
ab
_
t, f
_
+ T
SBR
ab
_
t, f
_
+ T
DB
ab
_
t, f
_
+ T
LOS
ab
_
t, f
_
,
(40)
where T
SBT
ab
(t, f ), T
SBR
ab
(t, f ), T
DB
ab
(t, f ), and T
LOS
ab
(t, f ) (a
{p, ` p}, b {q, ` q}) are time-variant transfer functions for the
SBT, SBR, DB and LOS components, respectively.
3. Polarization Matched Time-Frequency
Correlation Functions
Wide-sense stationarity and uncorrelated scatterers are often
assumed to be valid for mobile radio channels [18]. In this
paper, time and frequency dispersion are modeled depen-
dently over a wide sense stationary uncorrelated scattering
channel for a 3D nonisotropic scattering environment. For
such channels, the time-frequency correlation function is an
eective way of characterizing the statistical dependencies
in the temporal and frequency domains associated with the
mobile-to-mobile channels. Since the polarization states of
antennas at both the Tx and Rx are matched, we derive
the matched polarization-basis time-frequency correlation
function to show the relationship between the frequency and
the time of the 3D statistical model for 2 2 DP antennas.
For a WSSUS channel, the TFCF of the time-varying
transfer function T
ab
(a {p, ` p}, b {q, ` q}) is dened in
terms of the time dierence
t
and the frequency separation
ab
_
t
,
f
_
= E
_
T
ab
_
t, f
_
T
ab
_
t +
t
, f +
f
__
, (41)
where a {p, ` p}, b {q, ` q} and p, ` p, q, ` q correspond
to vertical and horizontal antenna polarizations. E[] is the
expectation operator, and denotes the complex conjugate
operation. The TFCFs in (41) can be rewritten as the
superposition of the TFCFs of the SBT, SBR, DB and LOS
components. Consider
ab
_
t
,
f
_
=
SBT
ab
_
t
,
f
_
+
SBR
ab
_
t
,
f
_
+
DB
ab
_
t
,
f
_
+
LOS
ab
_
t
,
f
_
.
(42)
In order to estimate the time frequency correlations, we
make the assumption that the number of scatterers in the 3D
reference model is innite so that the discrete azimuth angles
(
(m)
t
,
(m)
r ,
(n)
t
and
(n)
r ), the discrete elevation angles (
(m)
t
,
(m)
r ,
(n)
t
, and
(n)
r ) and the radii R
t
, R
r
can be represented
as continuous random variables with probability density
functions of f (
(s)
r ), f (
(s)
t
), f (
(s)
r ), f (
(s)
t
), f (R
t
), and
f (R
r
), respectively. The form of these pdf s are as follows.
The Tx azimuth angles are characterized by the von Mises
pdf in (2) as
f
_
(S)
t
_
=
exp
_
(S)
t
cos
_
(S)
t

(S)
t
__
2I
0
_
(S)
t
_
, (43)
where
(S)
t
[, ) is the mean value of angle at which the
Tx scatterers are distributed in the x-y plane and
(S)
t
controls
the spread of Tx scatterers around the mean
(S)
t
. Similarly,
the Rx azimuth angles are characterized by
f
_
(S)
r
_
=
exp
_
(S)
r cos
_
(S)
r
(S)
r
__
2I
0
_
(S)
r
_
, (44)
where
(S)
r [, ) is the mean value of angle at which the
Rx scatterers are distributed in the x-y plane and
(S)
r controls
the spread of Rx scatterers around the mean
(S)
r .
The Tx and Rx elevation angles are characterized by the
cosine pdf in (3) as
f
_
(S)
t
_
=
_
4|
tm
|
cos
_
(S)
t
tm
_
,
(S)
t
|
tm
|

2
,
0, otherwise,
f
_
(S)
r
_
=
_
4|
rm
|
cos
_
(S)
r
rm
_
,
(S)
r
|
rm
|

2
,
0, otherwise,
(45)
where
tm
is the maximum elevation angle for the Tx
scatterers and
rm
is the maximum elevation angle for the
Rx scatterers.
The pdfs used to characterize the radii R
t
and R
r
are given
by
f (R
t
) =
3R
2
t
R
3
t2
R
3
t1
R
t1
R
t
R
t2
,
f (R
r
) =
3R
2
r
R
3
r2
R
3
r1
R
r1
R
r
R
r2
.
(46)
The pdf of phase-oset random variables
(pol)
m ,
(m)
ab
,
(n)
ab
, and
(m,n)
ab
(a {p, ` p} and b {q, ` q}) is given
by a uniform distribution on the interval [, ). Using
these distributions, the TFCFs
SBT
(t, f ),
SBR
(t, f ),
DB
(t, f ), and
LOS
(t, f ) of the time-variant transfer
functions for the corresponding SBT, SBR, DB and LOS
components are derived below.
Using these pdfs it is possible to formulate the TFCFs for
each of the bounce path components. The TFCFs along the
SBT path are given by
SBT
ab
_
t
,
f
_
=

t, pq
A
2
m
K + 1
I
SBT
ab
_
t
,
f
_
,
(47)
where I
SBT
ab
(
t
,
f
) is dened as
I
SBT
ab
_
t
,
f
_
=
1
8|
tm
|I
0
_
(s)
t
_
_
Rt2
Rt1
_
tm
tm
_
_
1

2
R
t
D
_
2
exp
_
j
2
c

f
_
p,m
+
m,q
_
_
3R
2
t
R
3
t2
R
3
t1
cos
_
(s)
t
tm
_
exp
_
(s)
t
cos
_
(s)
t

(s)
t
__
exp
_
j2
t
f
t max
cos
_
(s)
t

vt
_
cos
(s)
t
cos
vt
_
exp
_
j2
t
f
t max
sin
(s)
t
sin
vt
_
exp
_
j2
t
f
r max
cos
_
(s)
r

vr
_
cos
_
(s)
r
_
cos
vr
_
exp
_
j2
t
f
r max
sin
(s)
r
sin
vr
_
exp
_
_
8
_
h
sin
(i)
t
c
_
2
_
_
I
0
_
_
8
_
h
sin
(i)
t
c
_
2
_
_
E
_
1
CPR
qp, ` q ` p
_
__
G
(v)
a
_
(s)
t
,
(s)
t
_
_
G
(h)
a
_
(s)
t
,
(s)
t
_
_
_
_
_
_
1 E
__
r
(m)
p` q
_
E
__
r
(m)
` pq
_
1
_
_
_
_
_
_
_
_
_
G
(v)
b
_
(s)
r ,
(s)
r
_
_
G
(h)
b
_
(s)
r ,
(s)
r
_
_
_
_
_
d
(s)
t
d
(s)
t
dR
t
.
(48)
E[
_
r
(m)
p` q
] and E[
_
r
(m)
` pq
] represent the expectations of the
random variables
_
r
(m)
p` q
and
_
r
(m)
` pq
, respectively, over the
interval [20 dB, 20 dB]. They are calculated using the pdf
of the normal XPD model in (16). E[
_
CPR
qp, ` q ` p
] denotes
the expectation of random variable of copolar ratio over the
interval [0, 6 dB].
At the Rx, the azimuth angle
(s)
r and elevation angle
(s)
r
induced by the scatterer S
(m)
t
are approximated as
(s)
r

3
2
+
R
t
D
sin
(s)
t
,
(s)
r

R
t
D

(s)
t
+

z
D
.
(49)
The angle of incidence
(i)
t
with the surface of Tx scatterer
can be determined by using trigonometric identities in the
triangle O
t
S
(m)
t
O
r
as follows:
(i)
t

1
2
arccos
_
R
2
t
+ d
2
m,r
D
2
2R
t
d
m,r
_
, (50)
where d
m,r
is given by
d
2
m,r
=
_
x
R
t
cos
(s)
t
cos
(s)
t
_
2
+
_
y
R
t
cos
(s)
t
sin
(s)
t
_
2
+
_
z
R
t
sin
(s)
t
_
2
.
(51)
The TFCFs along the SBR path are given by
SBR
ab
_
t
,
f
_
=

r,ab
A
2
m
K + 1
I
SBR
ab
_
t
,
f
_
,
(52)
where I
SBR
ab
(
t
,
f
) is dened as
I
SBR
ab
_
t
,
f
_
=
1
8|
rm
|I
0
_
(s)
r
_
_
Rr2
Rr1
_
rm
rm
_
_
1

2
R
r
D
2
_
2
exp
_
j
2
c

f
_
p,n
+
n,q
_
_
3R
2
r
R
3
r2
R
3
r1
cos
_
(s)
r
rm
_
exp
_
(s)
r
cos
_
(s)
r

(s)
r
__
exp
_
j2
t
f
t max
cos
_
(s)
t

vt
_
cos
(s)
t
cos
vt
_
exp
_
j2
t
f
t max
sin
(s)
t
sin
vt
_
exp
_
j2
t
f
r max
cos
_
(s)
r

vr
_
cos
(s)
r
cos
vr
_
exp
_
j2
t
f
r max
sin
(s)
r
sin
vr
_
exp
_
_
8
_
h
sin
(i)
r
c
_
2
_
_
I
0
_
_
8
_
h
sin
(i)
r
c
_
2
_
_
E
_
1
CPR
qp, ` q ` p
_
__
G
(v)
a
_
(s)
t
,
(s)
t
_
_
G
(h)
a
_
(s)
t
,
(s)
t
_
_
_
_
_
_
1 E
__
r
(n)
p` q
_
E
__
r
(n)
` pq
_
1
_
_
_
_
_
_
_
_
_
G
(v)
b
_
(s)
r ,
(s)
r
_
_
G
(h)
b
_
(s)
r ,
(s)
r
_
_
_
_
_
d
(s)
r
d
(s)
r
dR
r
,
(53)
where E[
_
r
(n)
p` q
] and E[
_
r
(n)
` pq
] represent the expectations of
the random variables
_
r
(n)
p` q
and
_
r
(n)
` pq
, respectively. They are
calculated over the interval [20 dB, 20 dB] using the pdf of
the normal XPD model in (16).
At the Tx, the azimuth angle
(s)
t
and elevation angle
(s)
t
induced by the scatterer S
(n)
r are approximated as
(s)
t

R
r
D
sin
(s)
r
,
(s)
t

R
r
D

(s)
r

z
D
.
(54)
The angle of incidence
(i)
r with the surface of Rx scatterer
can be determined by using trigonometric identities in the
triangle O
t
S
(n)
r O
r
as follows:
(i)
r

1
2
arccos
_
R
2
r
+ d
2
t,n
D
2
2R
r
d
t,n
_
, (55)
where d
t,n
is given by
d
2
t,n
=
_
x
+ R
r
cos
(s)
r
cos
(s)
r
_
2
+
_
y
+ R
r
cos
(s)
r
sin
(s)
r
_
2
+
_
z
+ R
r
sin
(s)
r
_
2
.
(56)
The TFCFs along the DB path are given by
DB
ab
_
t
,
f
_
=

tr,ab
A
2
m
K + 1
I
DB
ab
_
t
,
f
_
,
(57)
where I
DB
ab
(
t
,
f
) is dened as
I
DB
ab
_
t
,
f
_
=
1
64|
tm
||
rm
|I
0
_
(s)
t
_
I
0
_
(s)
r
_
_
Rt2
Rt1
_
Rr2
Rr1
_
tm
tm
_
rm
rm
_
_
1

2
R
r
+ R
t
D
_
2
exp
_
j
2
c

f
_
p,m
+
m,n
+
n,q
_
_
3R
2
r
R
3
r2
R
3
r1
3R
2
t
R
3
t2
R
3
t1
cos
_
(s)
t
tm
_
cos
_
(s)
r
rm
_
exp
_
(s)
t
cos
_
(s)
t

(s)
t
__
exp
_
(s)
r
cos
_
(s)
r

(s)
r
__
exp
_
j2
t
f
t max
cos
_
(s)
t

vt
_
cos
(s)
t
cos
vt
_
exp
_
j2
t
f
t max
sin
(s)
t
sin
vt
_
exp
_
j2
t
f
r max
cos
_
(s)
r

vr
_
cos
(s)
r
cos
vr
_
exp
_
j2
t
f
r max
sin
(s)
r
sin
vr
_
exp
_
_
8
_
h
sin
(i)
t
c
_
2
_
_
I
0
_
_
8
_
h
sin
(i)
t
c
_
2
_
_
exp
_
_
8
_
h
sin
(i)
r
c
_
2
_
_
I
0
_
_
8
_
h
sin
(i)
r
c
_
2
_
_
E
_
1
CPR
qp, ` q ` p
_
__
G
(v)
a
_
(s)
t
,
(s)
t
_
_
G
(h)
a
_
(s)
t
,
(s)
t
_
_
_
_
_
_
1 E
__
r
(m)
p` q
_
E
__
r
(m)
` pq
_
1
_
_
_
_
_
_
_
_
1 E
__
r
(n)
p` q
_
E
__
r
(n)
` pq
_
1
_
_
_
_
_
_
_
_
_
G
(v)
b
_
(s)
r ,
(s)
r
_
_
G
(h)
b
_
(s)
r ,
(s)
r
_
_
_
_
_
d
(s)
r
d
(s)
t
d
(s)
r
d
(s)
t
dR
r
dR
t
.
(58)
The TFCFs along the LOS path are written in forms as
LOS
pq
_
t
,
f
_
=
A
2
m
K + 1
I
LOS
_
t
,
f
_
,
(59)
where I
LOS
(
t
,
f
) is dened as
I
LOS
_
t
,
f
_
=K exp
_
j
2
c

f
pq
_
exp
_
j2
t
f
t max
cos
_

(LOS)
r

vt
_
sin
vt
_
exp
_
j2
t
f
r max
cos
_
(LOS)
r

vr
_
sin
vr
_
.
(60)
Table 1: The list of parameters used in the simulation.
Parameter K f
t max
[Hz] f
r max
[Hz]
h
[mm]
Value 0 3 100 100 0.2
Parameter
x
[m]
y
[m]
z
[m] f
c
[GHz]
t
Value 1500 0 0 1.0 0.2
Parameter
vt

vr

vt

vr

r
Value pi/4 pi/4 /4 /4 0.2
Parameter XPD- XPD- CPR- CPR-
tr
Value 10 dB 6 dB 0 dB 3 dB 0.6
V-V/H-H link
Parameter R
t1
[m] R
t2
[m] R
r1
[m] R
r2
[m]
(s)
t
Value 10 100 10 100 3
Parameter
(s)
r
(s)
t

(s)
r
tm
[deg]
rm
[deg]
Value 3 /4 7/4 20 20
V-H/H-V link
Parameter R
t1
[m] R
t2
[m] R
r1
[m] R
r2
[m]
(s)
t
Value 10 100 10 100 3
Parameter
(s)
r
(s)
t

(s)
r
tm
[deg]
rm
[deg]
Value 3 /4 7/4 20 20
Numerical integration methods are used to evaluate these
TFCFs in the simulation analysis section.
4. Numerical Results and Analysis
In this section, we present numerical results associated with
the analysis of the 2 2 DP-MIMO system. The matched
polarization-basis time-frequency correlation functions are
evaluated using the models described in Section 3. In the
analysis, the propagation paths include LOS, SBT, SBR, and
DB bounce paths, where the DB paths predominate over
the SBT and SBR paths. The temporal dispersion is on the
order of 600 ns, and the Doppler spread is approximately
200 Hz. The mean CP power ratio CPR
qp, ` q ` p
is set to 0 dB
so that the average powers of the V- V link and the H-
H link are approximately identical, a condition that we
recognize will not always hold in operational channels.
Further, we also assume that the von Mises distributions
for the azimuth angles and the cosine distributions for
the elevation angles are equivalent for both H scatterers
(reecting horizontally polarized waves) and V scatterers
(reecting vertical polarized waves), and that the XPDs
probability distribution functions are identical for the H-
V and V- H components. As before, these are assumed for
expediency. The specic simulation parameters are listed in
Table 1. The center frequency is assumed to be 1 GHz.
Figures 5, 6, 7, 8, and 9 provide examples of joint TFCFs
that were computed for specic subchannels and bounce
path components.
SBT
pq
(
t
,
f
),
DB
pq
(
t
,
f
), and
pq
(
t
,
f
)
are shown for the V- V CP components in Figures 5, 6, and
7, respectively. Similarly,
SBT
` pq
(
t
,
f
) and
DB
` pq
(
t
,
f
) are
shown in Figures 8, and 9 for the H-V link. The correlations
along the t axis are driven by the Doppler frequency
spread [35] and correlations along the f axis are driven by
6
8
10
12
14
16
18
0
5
10 0.06
0.04
0.02
0
7
8
9
10
11
12
13
14
15
16
17
t
(s)
T
F
C
F
|
|
(
d
B
)
f
(
M
H
z
)
Time-frequency correlation, V-V, SBT
Figure 5: Normalized time-frequency correlation function of V-V
link for SBT ray.
Time-frequency correlation, V-V, DB2
0
0.02
0.04
0.06
0
5
10
35
30
25
20
15
10
5
0
t
(
s
)
f
(
M
H
z
)
T
F
C
F
|
|
(
d
B
)
30
25
20
15
10
5
Figure 6: Normalized time frequency correlation function of V-V
link for DB ray.
the delay spread associated with each component. The results
indicate that the TFCFs are not separable into a product
of a time correlation function and a frequency correlation
function, a conclusion that we anticipated since the scatter
locations contribute to both the time dispersion and the
Doppler spread. The peak correlation value at t = f = 0
in each plot is determined by the relative power of the paths
being considered. The physical model parameters that play
a role in generating temporal and Doppler spread include
the radii R
r1
, R
r2
, R
t1
, and R
t2
associated with the Rx and Tx
spheres of scatterers, the maximum elevation angles
rm
and
tm
, the mean value of the azimuth angles
(s)
r and
(s)
t
, the
spread factor of scatterers
(s)
r and
(s)
t
, the angles dening
the direction of the velocity,
vr
,
vr
,
vt
, and
vt
, and the
maximum Doppler frequencies f
r max
and f
t max
.
0
0.02
0.04
0.06
0
5
10
20
15
10
5
0
t (s)
f (MHz)
T
F
C
F
|
|
(
d
B
)
16
14
12
10
8
6
4
2
0
Time-frequency correlation, V-V, ALL
Figure 7: Normalized time frequency correlation function of V-V
link for the superposition of SBT, SBR and DB rays.
0
0.02
0.04
0.06
0
5
10
58
56
54
52
50
48
46
56
55
54
53
52
51
50
49
48
47
t
(
s
)
Time-frequency correlation, VH, SBT
f
(
M
H
z
)
T
F
C
F
|
|
(
d
B
)
Figure 8: Normalized time-frequency correlation function of V-H
link for SBT ray.
Frequency correlation functions (for a xed t) are
driven primarily by the delay prole of the channel.
Small delay proles yield frequency correlation functions
that decorrelate at higher frequencies. The DB component
will exhibit the lowest decorrelation frequency since this
component has the most signicant delay spread. In the
concentric-sphere based scattering model shown in Fig-
ure 1, the maximum delay spread is determined by the
maximum distance of signal traveling along the double-
bounce path induced by the Tx scatterer or the Rx scatterer.
It can be roughly calculated as (2R
t
+ 2R
r
)/c, where c is
the speed of light. The coherence bandwidth W
c
in Hz
is given approximately by the inverse proportion of the
time delay spread
d
. The minimum coherence bandwidths
versus the radii of the Tx and Rx spheres used in the
channel model are estimated and presented in Figures
0
0.02
0.04
0.06
0
5
10
30
25
20
15
10
5
0
t
(
s
)
f (MHz)
T
F
C
F
|
|
(
d
B
)
25
20
15
10
5
Time-frequency correlation, V-H, DB2
Figure 9: Normalized time-frequency correlation function of V-H
link for DB ray.
11 and 12. The coherence bandwidth decreases when the
radii of the Tx or Rx spheres increase. In Figure 7,
if we approximate the coherence bandwidth as the fre-
quency separation over which the magnitude of correlation
coecients is larger than a half of the its maximum at
the zero frequency separation, the coherence bandwidth
for the test prole used in our analysis is approximately
2.5 MHz.
The time correlation function dependencies on the
maximum Doppler frequency are illustrated for the double-
bounce ray with f = 0, in Figure 10. Large Doppler
frequencies lead to small decorrelation times. We anticipate
that the double-bounce components will decorrelate at least
as quickly as the single-bounce components due to the fact
that the Doppler spread of this component will be greater
than or equal to the Doppler spread of the SBT component,
a trend that is conrmed in Figures 8 and 9.
The use of non-zero XPD values leads to cross-polarized
components (V-H and H-V) with power contributions
lower than the CP components (V-V or H-H). Due to
the similarity of parameters used to dene the co- and
crosspolarized channels, the cross-polarized response is
approximately related to the CP ones by a scale factor.
This result can be observed from cross-polarized parts of
TFCFs in Section 3. The normalized correlation values for
the SBT and DB components at f = t = 0 are
plotted as a function of the mean XPD value for both
matched polarization and cross polarization components in
Figure 13. The normalized correlation values for the matched
polarizations are constant since the correlations and powers
are independent of . For the crosspolarization components,
the normalized correlation is seen to decrease as increases,
largely due to changes in relative power. At = 0, when
the scatterers yield the same average power at both matched
and cross-polarized components, the normalized correlation
is identical for both the matched and cross-polarization
links.
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
0.05
0
0.05
0.1
0.15
0.2
Time delay
t
(seconds),
f
= 0
M
P
B

t
i
m
e

c
o
r
r
e
l
a
t
i
o
n

f
u
n
c
t
i
o
n
s
Max doppler freq. = 25 Hz
Figure 10: Normalized time-correlation functions varying with the
time delay and the maximum Doppler Tx/Rx frequencies for DB
ray. The conguration is dened in Table 1 with
t
=
r
=
tr
=1/3.
20
40
60
80
100
20
40
60
80
100
0
5
10
15
20
25
30
35
40
R
a
d
iu
s in
T
x
(m
) R
a
d
iu
s
in
R
x
(
m
)
C
o
h
e
r
e
n
c
e

b
a
n
d
w
i
d
t
h

(
M
H
z
)
5
10
15
20
25
30
35
Figure 11: The minimum coherence bandwidth of 2 2 DP
system varies with upper radii of spheres of Tx and Rx scatterers
respectively.
5. Conclusion
In this paper, a 3D geometrical propagation model has
been proposed for DP-MIMO mobile-to-mobile Rayleigh
and Rician fading channels. The parametric channel model
incorporates a number of physical parameters to characterize
the channel, including antenna patterns, azimuth/elevation
angles of arrival and departure, geometrical distribution
of scatterers, K-factor, the maximum Doppler frequency,
scattering loss factor, crosspolar discrimination, and copolar-
ization power ratio. Using the key parameters of the model,
10 20 30 40 50 60 70 80 90 100
0
5
10
15
20
25
30
35
40
Radius (m)
C
o
h
e
r
e
n
c
e

b
a
n
d
w
i
d
t
h

(
M
H
z
)
Figure 12: The minimum coherence bandwidth of 22 DP system
varies with upper radii of spheres of scatterers in both Tx and Rx.
0 2 4 6 8 10 12 14 16 18 20
30
25
20
15
10
5
0
T
F
C
F
|
|
(
d
B
)
of XPD distribution (dB)
SBT-V-V
DB2-V-V
SBT-V-H
DB2-V-H
Figure 13: Normalized time-correlation functions of SBT and DB
rays vary with the mean XPD and the conguration is set up in
Table 1.
matched polarization-basis time-frequency correlation func-
tions were formulated and numerically computed for
WSSUS 3D non-isotropic scattering environments. The
numerical results show that the joint TFCFs are not sep-
arable into independent time-correlation and frequency-
correlation functions. The normalized TFCFs can be parsed
into bounce path components for matched polarization and
copolarization links, leading to power-normalized marginal
TFCFs for the channel realization. The exibility of the
model enables control of channel parameters to achieve a
variety of multipath fading environments to investigate 2 2
DP architectures.
Our intention in future research is to validate the
proposed channel model using measurements. A series of
experiments are planned to verify and justify our model.
Disclaimer
The views and conclusions contained in this paper are those
of the authors and should not be interpreted as representing
the ocial policies, either expressed or implied,of the Oce
of Naval Research or the U. S. Government.
Acknowledgment
This work was sponsored by the oce of Naval Research
under grant number N00014-11-1-0607.
References
[1] A. J. Paulraj and T. Kailath, Increasing capacity in wireless
broadcast systems using distributed transmission/directional
re-ception (dtdr), US Patent 5, 345, 599, 1994.
[2] G. J. Foschini and M. J. Gans, On limits of wireless com-
munications in a fading environment when using multiple
antennas, Wireless Personal Communications, vol. 6, no. 3, pp.
311335, 1998.
[3] D. Tse and P. Viswanath, Fundamentals of Wireless Communi-
cation, Cambridge University Press, 2005.
[4] C. Oestges, V. Erceg, and A. J. Paulraj, Propagation modeling
of MIMO multipolarized xed wireless channels, IEEE Trans-
actions on Vehicular Technology, vol. 53, no. 3, pp. 644654,
2004.
[5] N. W. Bikhazi, MIMO communication capacity: antenna
coupling and precoding for incoherent detection [Ph.D. thesis],
CiteSeer, 2006.
[6] P. Soma, D. S. Baum, V. Erceg, R. Krishnamoorthy, and
A. J. Paulraj, Analysis and modeling of Multiple-Input
Multiple-Output (MIMO) radio channel based on outdoor
measurements conducted at 2.5GHz for xed BWA applica-
tions, in Proceedings of the IEEE International Conference on
Communications (ICC 02), vol. 1, pp. 272276, May 2002.
[7] M. Sha, M. Zhang, A. L. Moustakas et al., Polarized MIMO
channels in 3-D: models, measurements and mutual informa-
tion, IEEE Journal on Selected Areas in Communications, vol.
24, no. 3, pp. 514526, 2006.
[8] R. G. Vaughan, Polarization diversity in mobile communica-
tions, IEEE Transactions on Vehicular Technology, vol. 39, no.
3, pp. 177186, 1990.
[9] P. Kyritsi, D. C. Cox, R. A. Valenzuela, and P. W. Wolniansky,
Eect of antenna polarization on the capacity of a multiple
element system in an indoor environment, IEEE Journal on
Selected Areas in Communications, vol. 20, no. 6, pp. 1227
1239, 2002.
[10] T. B. Sorensen, A. O. Nielsen, P. E. Mogensen, M. Tolstrup,
and K. Steensen, Performance of two-branch polarization
antenna diversity in an operational GSMnetwork, in Proceed-
ings of the 48th IEEE Vehicular Technology Conference (VTC
98), vol. 3, pp. 741746, May 1998.
[11] T. Neubauer and P. C. F. Eggers, Simultaneous character-
ization of polarization matrix components in pico cells,
in Proceedings of the IEEE VTS 50th Vehicular Technology
Conference (VTC 99), vol. 3, pp. 13611365, September 1999.
[12] T. G. Pratt, H. Tapse, B. Walkenhorst, and G. Acosta-Marum,
A modied XPC characterization for polarimetric channels,
IEEE Transactions on Vehicular Technology, vol. 60, no. 7,
Article ID 5871726, pp. 29042913, 2011.
[13] P. Hoeher, Statistical discrete-time model for the WSSUS
multipath channel, IEEE Transactions on Vehicular Technol-
ogy, vol. 41, no. 4, pp. 461468, 1992.
[14] E. Chiavaccini and G. M. Vitetta, GQR models for multipath
rayleigh fading channels, IEEE Journal on Selected Areas in
[15] C. Oestges, B. Clerckx, M. Guillaud, and M. Debbah, Dual-
polarized wireless communications: from propagation models
to system performance evaluation, IEEE Transactions on
Wireless Communications, vol. 7, no. 10, pp. 40194031, 2008.
[16] M. Sha, M. Zhang, P. J. Smith, A. L. Moustakas, and
A. F. Molisch, The impact of elevation angle on MIMO
capacity, in Proceedings of the IEEE International Conference
on Communications (ICC 06), vol. 9, pp. 41554160, July
2006.
[17] T. R. ETSI, 125 996 v. 6.1. 0,universal mobile telecommu-
nications system (umts); spatial channel model for multiple
input multiple output (mimo) simulations (3gpp tr 25.996
version 6.1. 0 release 6), Further study is required to design
an adaptive or hybrid MCN structure to improve both the
handover eciency and the throughput of cell-edge and cell-
boundary users and to resolve the network issues, 2003.
[18] A. G. Zaji c and G. L. St uber, Three-dimensional modeling
and simulation of wideband MIMO mobile-to-mobile chan-
nels, IEEE Transactions on Wireless Communications, vol. 8,
no. 3, pp. 12601275, 2009.
[19] M. T. Dao, V. A. Nguyen, Y. T. Im, S. O. Park, and G. Yoon, 3D
polarized channel modeling and performance comparison of
MIMO antenna congurations with dierent polarizations,
IEEE Transactions on Antennas and Propagation, vol. 59, no. 7,
pp. 26722682, 2011.
[20] R. Wang and D. Cox, Channel modeling for Ad Hoc
mobile wireless networks, in Proceedings of the 55th Vehicular
Technology Conference (VTC 02), vol. 1, pp. 2125, May 2002.
[21] C. S. Patel, G. L. St uber, and T. G. Pratt, Simulation of
Rayleigh-faded mobile-to-mobile communication channels,
IEEE Transactions on Communications, vol. 53, no. 11, pp.
18761884, 2005.
[22] J. Maurer, T. F ugen, and W. Wiesbeck, Narrow-band mea-
surement and analysis of the inter-vehicle transmission chan-
nel at 5.2 GHz, in Proceedings of the 55th Vehicular Technology
Conference (VTC 02), vol. 3, pp. 12741278, May 2002.
[23] Y. I. Wu and K. T. Wong, Polarisation-sensitive geometric
modelling of the distribution of direction-of-arrival for uplink
multipaths, IET Microwaves, Antennas and Propagation, vol.
5, no. 1, pp. 95101, 2011.
[24] J. D. Kraus and R. J. Marhefka, Antennas for All Applications,
vol. 1, McGraw-Hill, New York, NY, USA, 2002.
[25] H. L. Van Trees, Detection, Estimation, and Modulation Theory:
Detection, Estimation, and Linear Modulation Theory, Wiley,
1968.
[26] J. D. Parsons and P. J. D. Parsons, The Mobile Radio Propaga-
tion Channel, vol. 81, Wiley, 2000.
[27] K. Mammasis, R. W. Stewart, and J. S. Thompson, Spatial
Fading Correlation model using mixtures of von Mises Fisher
distributions, IEEE Transactions on Wireless Communications,
vol. 8, no. 4, pp. 20462055, 2009.
[28] E. J. Gumbel, Applications of the circular normal distribu-
tion, JAMA, vol. 49, no. 266, pp. 267297, 1954.
[29] J. D. Parsons and A. M. D. Turkmani, Characterisation of
mobile radio signals. Model description, IEE Proceedings, Part
I, vol. 138, no. 6, pp. 549556, 1991.
[30] T. Taga, Analysis for mean eective gain of mobile antennas
in land mobile radio environments, IEEE Transactions on
Vehicular Technology, vol. 39, no. 2, pp. 117131, 1990.
[31] P. Kyritsi and D. C. Cox, Propagation characteristics of
horizontally and vertically polarized electric elds in an
indoor environment: simple model and results, in Proceedings
of the IEEE 54th Vehicular Technology Conference (VTC 01),
vol. 3, pp. 14221426, October 2001.
[32] L. Boithias and L. J. Libols, Radio Wave Propagation, North
Oxfordshire Academy, London, UK, 1987.
[33] J. G. Proakis, Digital Communications, McGraw-Hill, New
York, NY, USA, 2001.
[34] J. M. Wozencraft and I. M. Jacobs, Principles of Communication
Engineering, vol. 32, Wiley, New York, NY, USA, 1965.
[35] T. S. Rappaport et al., Wireless Communications: Principles and
Practice, Prentice Hall, Upper Saddle River, NJ, USA, 1996.
doi:10.1155/2012/542198
Research Article
An Accurate Hardware Sum-of-Cisoids Fading Channel Simulator
for Isotropic and Non-Isotropic Mobile Radio Environments
L. Vela-Garcia,
1
J. V azquez Castillo,
1
R. Parra-Michel,
1
and Matthias P atzold
2
1
Department of Telecommunications, CINVESTAV I.P.N., 45019 Zapopan, JAL, Mexico
2
Faculty of Engineering and Science, University of Agder, P.O. Box 509, 4898 Grimstad, Norway
Correspondence should be addressed to J. V azquez Castillo, jvazquezc@gdl.cinvestav.mx
Received 27 April 2012; Revised 13 July 2012; Accepted 27 July 2012
Academic Editor: Carlos A. Gutierrez
Copyright 2012 L. Vela-Garcia et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The rapid technological development in the eld of wireless communications calls for devices capable of reproducing and
simulating the behavior of the channel under realistic propagation conditions. This paper presents a hardware fading channel
simulator that is able to generate stochastic processes characterized by symmetrical and asymmetrical Doppler power spectral
densities (PSDs) depending on the assumption of isotropic or non-isotropic scattering. The concept of the proposed hardware
simulator is based on an implementation of the sum-of-cisoids (SOC) method. The hardware simulator is capable of handling
any conguration of the cisoids amplitudes, frequencies, and phases. Each of the cisoids that constitutes the SOC model is
implemented using a piecewise polynomial approximation technique. The investigation of the higher-order statistics of the
generated fading processes, like the level-crossing rate (LCR) and the average duration of fades (ADF), shows that our design
is able to reproduce accurately the key features of realistic channel models that are considered as candidates for the latest wireless
communication standards.
1. Introduction
The increasing demand for high data rate wireless com-
munication systems makes it mandatory to develop devices
enabling the test and the performance evaluation of such sys-
tems with a high level of accuracy. A channel simulator oers
the possibility of reproducing the most important channel
characteristics without the necessity of conducting eld tests
in specic environments, which reduces signicantly the
system development time and the costs of the system design
and the verication process.
Flat-fading wireless channels are modeled in the complex
baseband as complex stochastic processes characterized by
a specic autocorrelation function (ACF) or, alternatively,
by the corresponding power spectral density (PSD). Con-
sidering a base station to mobile link, these functions are
related to the angle-of-arrival (AOA) distribution of the
incoming plane waves (in the mobile to base station link,
the angle-of-departure (AOD) distribution determines the
ACF and PSD) in a multipath propagation environment
[1]. The uniform distribution of the AOA (i.e., the uniform
AOA probability density function (PDF) over [, )) is
related to environments characterized by isotropic scattering
conditions, where the ACF is determined by the zeroth-
order Bessel function corresponding to the Jakes PSD [2].
This model was originally proposed by Clarke [3]. How-
ever, realistic propagation environments do not fulll the
isotropic scattering conditions, which result in a nonuniform
distribution of the AOA, leading to an asymmetrical PSD
that can exhibit a variety of shapes (see [48] for more
details).
For the reproduction of the statistical properties of
real mobile radio channels on hardware platforms, several
channel emulators have been proposed in the literature (see
[911]), where the fading samples have been generated by
one of the following two methods: the lter method or
the sum-of-sinusoids (SOS) method. In most cases, the
generation of the fading waveforms has been accomplished
under the assumption of isotropic scattering. In [11], a
hardware channel simulator based on the lter method
assuming non-isotropic scattering conditions is presented.
However, the level of computational complexity in lter-
based emulators is greater than in SOS-based emulators,
as SOS techniques do not require the realization of white
Gaussian random number generators (see [12]). There-
fore, SOS-based simulators provide an attractive alternative
because of their computational eciency and comfortable
reconguration capabilities.
In this paper, a novel architecture for generating stochas-
tic processes using the SOS principle is presented. The
challenge of this study is to develop a simulator, which is
able to handle both symmetrical and asymmetrical PSDs in
a parametric architecture, allowing the emulation of more
realistic channels, which are required in modern wireless
communication standards. In order to achieve this, the sum-
of-cisoids (SOC) method rather than the SOS approach
will be used as proposed in [1]. The resulting hardware
architecture allows the implementation of any SOC or SOS
simulation model, permitting the conguration of the gains,
frequencies, and phases of each complex sinusoid for any
chosen deterministic or stochastic parameter computation
method. In addition, this provides the methodological
exibility for adding specular and non-specular components.
The sinusoid evaluation is performed using a piecewise poly-
nomial approximation, to enable the generation of sinusoid
samples at high rates with a high frequency resolution of
less than 1 Hz, which is desired for the development of
modern emulators based on the SOS/SOC principle. From
this perspective, the polynomial approximation technique
is employed here as a high-performance alternative to the
direct sinusoid computation method, which is essential for
traditional architectures using look-up tables (LUT) that are
described in [9, 10].
The rest of this article is organized as follows. In Section
2, a brief description of the SOC method is given. The
hardware architecture of a at-fading simulator is described
in Section 3. In Section 4, the study of two methods for
computing the SOC parameters is provided. In Section 5,
the statistical properties of the fading samples generated by
the proposed hardware architecture are discussed. This is
followed by the performance analysis in Section 6, which
elaborates on the hardware resource costs of the pro-
posed architecture in comparison to other existing SOS-
based simulators. Finally, the conclusions are drawn in
Section 7.
2. Review of the SOC Model
In this section, a brief description of the SOC model for the
generation of stochastic processes is presented. Additionally,
mathematical expressions for the rst and second order
statistics of the SOC model are reviewed, which provides a
reference for comparing several given PSD functions with
the corresponding approximation results obtained for the
proposed at-fading channel simulator.
2.1. Basic Description of SOC Models. In the complex
equivalent baseband, a multipath at-fading channel can be
modeled by an SOC process (t), which is a superposition of
N cisoids of the form:
(t) =
N
_
n=1
c
n
e
j(2 fnt+n)
, (1)
where c
n
, f
n
, and
n
are the gains, the Doppler frequencies,
and the phases of the nth propagation path, respectively.
Considering that c
n
and f
n
are constants, and only
n
are
independent, identically distributed (i.i.d.) random variables
with a uniform distribution over [0, 2), then (t) is a zero-
mean complex stochastic process with variance
2

=
N
n=1
c
2
n
[1]. Under certain conditions, it can be shown [13] that the
SOC process (t) tends to a complex Gaussian process (t)
with variance 2
2
0
if N .
Let the line-of-sight (LOS) component be given by [1]
s = e
j
, (2)
where the amplitude 0 and the phase
are constant
quantities, then the sum of the scattered components (t)
and the LOS component s results in the following nonzero-
mean complex random process:

(t) = (t) + s. (3)

The envelope of
(t), which is denoted by
(t) =
(t)
, (4)
provides an accurate and ecient simulation model for Rice
processes.
2.2. Statistical Properties of SOC Models. In [13], it has been
shown that the PDF p
(r) of

(t) is given by
p
(r) = r(2)
2
_
0
_
_
N
n=1
J
0
(2|c
n
|x)
_
_
J
0
(2rx)J
0
_
2x
_
xdx,
r 0,
(5)
where J
0
() denotes the 0th order Bessel function of the rst
kind. If the gains c
n
are given by c
n
=
0
2/N, then the PDF

in (5) tends to the Rice density
p
(r) =
r
2
0
e
(r
2
+
2
)/2
2
0
I
0
_
r
2
0
_
, r 0 (6)
as N , where I
0
() is the 0th order-modied Bessel
function of the rst kind.
Likewise, the cumulative distribution function (CDF)
P
(r) of

(t) has been derived in [14], where the following
result can be found:
P
(r) = 2r
_
0
_
_
N
n=1
J
0
(2|c
n
|x)
_
_
J
1
(2rx)J
0
_
2x
_
dx,
r 0.
(7)
In a similar fashion, the CDF P
(r) of

(t) converges to
the Rice CDF as N [1, 14]. The Rice CDF P
(r) can be
expressed as
P
(r) = 1 M
Q
_

0
,
r
0
_
, r 0, (8)
where M
Q
(, ) represents the Marcum Q-function [1].
Owing to the fact that the PDF and the CDF do not reveal
any information about how fast the channel changes with
time, it is necessary to consider the second-order statistics
of the envelope

(t). In this respect, the level-crossing rate
(LCR) and average duration of fades (ADF) are of special
interest. The LCR N
(r) describes how often the envelope
(t) crosses on average a specied signal level r within one

second, and the ADF T
(r) describes how long on average

the envelope

(t) remains below a specied signal level
r. These statistical quantities depend on the propagation
environment and the Doppler shifts caused by the movement
of the mobile station. The LCR N
(r) of

(t) can be obtained
by using Rices formula [15, 16]:
N
(r) =
_
0
zp
(r, z) d z, r 0. (9)
Equation (9) requires the knowledge of the joint PDF
p
(z, z) of the envelope

(t) and its time derivative

(t). It
should be mentioned that the relation in (9) holds for any
stationary random process characterized by the joint PDF
p
(z, z).
In [17], an exact solution has been presented for the LCR
N
(r) of the envelope

(t). This exact-solution factors in that
the inphase and quadrature components of the SOC process
(t) as well as their time derivatives are mutually correlated.
The LCR N
(r) of

(t) in (4) is given as
N
(r) = 4r
_
zmax
0
z
_
0
_
0
_
2
0
J
0
_
2v
1
_
_
_
N
n=1
J
0
_
2c
n
_
v
2
1
+
_
2 f
n
v
2
_
2
4 f
n
v
1
v
2
sin
_
_
_
e
j2rv1 cos()
e
j2 zv2
v
1
ddv
1
dv
2
d z, r 0,
(10)
where z
max
denotes the maximum of

(t), which is given by

z
max
= max{
(t)} = 2
N
n=1
| f
n
c
n
|. The LCR N
(r) of SOC
processes

(t) approaches the LCR N
(r) of Rice processes

(t) as N , that is,
lim
N
N
(r) = N
(r) =

2
p
(r), r 0,
(11)
where p
(r) is the Rice PDF presented in (6). The symbol

represents the negative curvature of the ACF r
i i
() of the
underlying real-valued Gaussian process
i
(t) (i = 1, 2) at
the origin = 0. In case of isotropic scattering, the quantity
is given by = 2( f
max
0
)
2
[1].
The corresponding ADF T
(r) of

(t) can be obtained
after substituting (7) and (10) in
T
(r) =
P
(r)
N
(r)
. (12)
From the discussion above, it is obvious that the ADF
T
(r) of

(t) converges to the ADF T
(r) of Rice processes

(t) as N .
In addition to the symmetrical Jakes PSD:
S
Jakes

_
f
_
=
_
2
0
f
max
_
1
_
f / f
max
_
2
,

f
max
0,

> f
max
(13)
which characterizes isotropic scattering environments, it is
desirable to have a fading channel simulator that is able to
generate fading process characterized by both symmetrical
and asymmetrical PSDs.
In this sense, the proposed fading channel simulator
shall also incorporate the characteristics of non-isotropic
scattering environments. The utility of the von Mises PDF
as a parametric model for the distribution of the AOA has
been supported by dierent studies focussing non-isotropic
propagation scenarios [18]. The resulting PSD includes the
symmetrical Jakes PSD as a special case. In our study,
the Doppler PSD obtained by using the von Mises AOA
distribution will be used as a reference model for the PSD of
non-isotropic scattering environments.
When using the von Mises AOA distribution, then the
Doppler PSD can be written as follows [19]
S
VM
_
f
_
=
exp
_
cos(
0
)
_
f / f
max
__
cosh
_
sin(
0
)
_
1
_
f / f
max
_
2
_
f
max
I
0
()
_
1
_
f / f
max
_
2
,
f
max
,
(14)
where 0 is a concentration parameter, which controls the
angular spread of the AOA, and
0
denotes the mean AOA
of the incoming plane waves. Equation (14) assumes that the
PDF of the AOA follows the von Mises distribution dened
by
p
VM
() =
exp[ cos(
0
)]
2I
0
()
, [, ). (15)
According to [19], the LCR of Rayleigh processes (t)
assuming the von Mises AOA distribution can be written as
N
VM
(r) =
_
I
2
0
() I
2
1
() + cos(2
0
)
_
I
0
()I
2
() I
2
1
()
_
I
0
()
2 f
max
r exp
_
r
2
_
,
(16)
Complex sinusoid
evaluation module
Sinusoid
parameter
generator
Frequencies
Phases
Gains
Global control
unit
P
o
l
y

C
n
t
r
l

Instruction
F
r
e
q

C
n
t
r
l

P
h
a
s
e
s

C
n
t
r
l

G
a
i
n
s

C
n
t
r
l

Synchronous
multiplier
Output
register
Output
register
inphase
quadrature Cisoid
quadrature
component
Cisoid
inphase
component
Accumulator
register
Accumulator
register
Datapath
Poly-evaluator
module
Poly-evaluator
module
Partial sum
of the cisoids
nth inphase
component sample
n
t
h

q
u
a
d
r
a
t
u
r
e
c
o
m
p
o
n
e
n
t

s
a
m
p
l
e
+
+
x
1 x
2
f
n
n
c
n
c
n
cos(t + )
sin(t + )
Q(32, 30)
Q(32, 30)
Q(32, 30)
Q
(
3
2
,
3
0
)
Q
(
3
2
,
3
0
)
Q(32, 29) Q(32, 28)
Q(32, 28)
Q(32, 28)
Q(32, 29)
Q(32, 28)
Q
(
3
2
,
3
0
)
S
S
Figure 1: General architecture of the SOC fading channel simulator.
and the corresponding ADF is given by
T
VM
(r) =
I
0
()
_
I
2
0
() I
2
1
() + cos(2
0
)
_
I
0
()I
2
() I
2
1
()
_
exp
_
r
2
_
1
2 f
max
r
.
(17)
For the case of isotropic scattering ( = 0), the expression in
(16) reduces to the LCRN
(r) =
2 f
max
r exp(r
2
), and the
ADF in (17) simplies to T
(r) = (exp(r
2
)1)/(
2 f
max
r).
3. Hardware Architecture
In this section, the hardware channel simulator based on
the SOC method is presented for both isotropic and non-
isotropic scattering environments. The general architecture
of the proposed SOC fading channel simulator is illustrated
in Figure 1. The datapath of the architecture is drawn by
using the notation Q(WL,WF), where WL represents the
datapath wordlength, and WF denotes the fractional part of
the wordlength. The SOC fading channel simulator design
is composed of four main modules: the complex sinusoid
evaluation module, the sinusoid parameter generator, the
global control unit, and the datapath.
In order to compute the SOC according to (1), the
proposed fading channel simulator requires the initialization
of the SOC parameters (c
n
, f
n
,
n
), as well as the number
of cisoids N, which are stored in the sinusoid parameter
generator. Once the simulator has been congured, the
complex sinusoid evaluation module computes the nth cisoid
by exp( jx
1
) if a positive Doppler frequency is provided.
Otherwise, exp( jx
2
) is computed for the case of a negative
Doppler frequency, where x
1
= 2 f
n
t +
n
and x
2
=
(2 f
n
t) +
n
. This fact is obvious due to the following
relation:
exp
_
jt
_
= cos(t) j sin(t). (18)
Once the corresponding nth cisoid has been determined,
it is multiplied by the gain c
n
before accumulating the
resultant value in the accumulator register blocks, as shown
in Figure 1.
The summation of the cisoids as well as the initialization
of the SOC parameters of the simulator are handled by the
global control unit. It is important to point out that the
proposed architecture is parametric in the sense that it is
independent of the number of cisoids N and the choice of
the model parameters ( f
n
, c
n
, and
n
).
3.1. Piecewise Polynomial Approximation Technique. In case
of the SOS method, the most common hardware implemen-
tation technique described in the literature for the evaluation
of cosine and sine functions is the one that utilizes LUT. The
main characteristic of this technique is that it uses memory
blocks for storing the sinusoids. However, the use of LUT
limits the evaluation of the SOS process when covering a
wide range of Doppler frequencies at high data rates (i.e.,
Doppler frequencies in the range from mHz up to kHz at
megasamples per second (MSPS)). This limitation is due
to the limited number of samples that can be stored in
block memories for generating the sinusoids with the lower
Doppler frequencies. For example, for the generation of
0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
2
2.5
3
3.5
4
A
b
s
o
l
u
t
e

e
r
r
o
r
Time (s)
10
3
Degree 2 polynomial approximation
Degree 1 polynomial approximation
0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
A
m
p
l
i
t
u
d
e
1
0.5
Time (s)
Figure 2: Approximation of a cosine function with degree 1 and degree 2 polynomials.
a sinusoid with a Doppler frequency of 1 kHz at 1 MSPS,
it is necessary to store 1000 samples in a memory table.
On the other hand, in order to generate a sinusoid with a
Doppler frequency of 0.1 Hz at 1 MSPS, it is required to
store 10 megasamples in the LUT. Techniques for saving
memory resources are described in the literature (see [20] for
more details). It is important to remark that the waveform
with the lowest Doppler frequency generated by the sinusoid
generator determines the frequency resolution that can be
achieved by using LUT. This implies that huge memories
are required in order to obtain a high frequency resolution.
As a consequence, the ACF of the generated SOC process
depends directly on the accuracy of the Doppler frequencies
of the generated complex sinusoids. Likewise, higher-order
statistics, such as the LCR and ADF, can be aected as a result
of the level of accuracy in the generated waveforms [21].
An alternative solution that avoids the limitations
imposed by LUT is provided by using a piecewise polynomial
approximation technique for the sinusoid generation. This
technique rests upon the approximation of a function f (x),
which is in our case a sine or cosine function, by means of the
evaluation of polynomials p
(x), which are dened by the

polynomial degree d and the type of segmentation. The type
of segmentation denes the number of intervals j and the
bounds of such intervals [a
j
, b
j
). Each polynomial is dened
as follows:
p
(x) = p
d
x
d
+ p
d1
x
d1
+ + p
1
x + p
0
. (19)
An ecient procedure to evaluate the polynomials p
(x)
is by employing Horners rule as follows:
p
(x) =
__
p
d
x + p
d1
_
x +
_
x + p
0
, (20)
where x is the polynomial input, and p
k
(k = 0, 1, . . . , d) are
the polynomial coecients.
In order to obtain the coecients p
k
(k = 0, 1, . . . , d), two
kinds of approximation techniques can be employed, namely,
the approximation technique that minimizes the least square
error between f (x) and p
(x), called the least square approx-

imation, and the approximation technique that minimizes
the worst-case error between f (x) and p
(x), called the

least maximum approximation or minimax approximation.
In [22], several methods are described for approximating a
continuous function f (x) with a prespecied precision.
The piecewise polynomial approximation technique has
been utilized for generating random variables in channel
simulators designed by using the Monte Carlo method,
where its implementation in hardware platforms makes it
possible to approximate dierent functions eciently [23].
The scalability provided by a polynomial approximation
architecture allows the ecient use of multiple modules
in a channel simulator [23], which represents a good
design strategy for the development of real-world channel
simulators. In this work, the polynomial coecients are
computed by using the least square approximation as the
error metric.
For the evaluation of the sinusoidal functions, the use
of a uniform segmentation is feasible. Figure 2(a) shows two
approximations of the function cos(2x), where x is within
the range [0, 1). In the rst approximation, 16 piecewise
polynomials of degree 1 are considered, whereas the second
approximation uses 16 piecewise polynomials of degree 2.
It can be seen in Figure 2(b) that the approximation using
polynomials of degree 2 results in a much lower absolute
error compared with the one using polynomials of degree 1.
Hence, a sinusoid generator that uses a polynomial approxi-
mation of degree 2 is preferred for the generation of cisoids.
It should be noted that for the hardware implementation of
a polynomial of degree d, d multipliers and d adders are
required if a pipeline structure is applied, otherwise only one
multiplier and one adder are required when using an iterative
scheme.
3.2. Direct Digital Synthesizer Based on a Piecewise Polyno-
mial Approximation. After analyzing the piecewise polyno-
mial approximation, it is indispensable to visualize this tech-
nique for the generation of cisoids with dierent Doppler
Accumulator
Adder
Phase generator Tuning
word
Polynomial
evaluator
Sinusoid
output
Clk
converter
x
[0, 1)
ACC
Phase-to-wave
Figure 3: Architecture for the complex sinusoid generation module.
frequencies. Therefore, the structure of a direct digital
synthesizer (DDS) employing the polynomial approximation
technique is shown in Figure 3.
Besides the dierence in the sinusoid generator concept
(sinusoid storing versus sinusoid approximation), there is
also a dierence between the phase generator of the DDSs.
While the phase generator of the LUT-based oscillator
directly generates the memory addresses, the phase generator
of the polynomial approximation method computes values
in the interval [0, 1) for the case of the approximated
function cos(2x). This feature makes it possible to generate
sinusoids for a wide range of frequencies. For the application
of the SOS method, it is important that a high Doppler
frequency resolution can be achieved for lower as well as for
higher Doppler frequencies.
The frequency resolution of the DDS based on the
polynomial approximation is determined by the sampling
frequency F
c
and the quantization factor Qbits, which
determines the precision of the input x after its quantization
with Qbits. The minimum frequency achievable by the DDS
as well as the output frequency F
0
are determined by the
following expressions:
F
0min
=
F
c
2
Qbits
, (21a)
F
0
= F
0
_
Qbits
_
=
F
c
2
Qbits
Qbits
, (21b)
where
Qbits
is the tuning word that species the output
frequency F
0
. With regard to the conguration of the
waveforms phase, the polynomial-based DDS has the same
advantages as the LUT-based DDS. However, the phase for
the polynomial approximation approach is dened in the
interval [0, 1). For the case of a datapath with a bit-width
of 32 bits, the parameter Qbits is equal to 31 bits, and the
remaining bit is used as a sign bit. Hence, the minimum
frequency achievable by the DDS at an output data rate of
1 MSPS is equal to F
0min
= 0.4 mHz. In the case of an output
rate of 4096 samples per second, the frequency resolution will
be F
0min
= 1.9 Hz. In comparison with a LUT-based DDS,
the same frequency resolutions can be achieved by using a
LUT memory of 2
31
samples if a complete sinusoid is stored,
or 2
31
/4 samples if a quarter of a sinusoid is stored. It can be
inferred from the discussion above that the usage of a LUT-
based DDS for achieving a high frequency resolution F
0min
of
less than 1 Hz is not a practical option.
The SOC method requires the evaluation of both cosine
and sine waveforms (see (18)). This results in the use of either
one or two polynomial-based DDS modules. Figure 4 illus-
trates the architecture of the poly-evaluator module, which
consists of three blocks: the address generator unit (AGU),
the datapath of the evaluation, which applies Horners rule,
and the piecewise polynomial coecients used for storing the
polynomial coecients. When the value x is generated, this
value is passed through the AGU module to compute the
address of the corresponding polynomial coecients. Once
the address is obtained, the evaluation is performed while the
argument x and the three coecients are being passed across
the datapath, as it can be seen at the bottom of Figure 4. The
cisoid evaluation requires two clock cycles when employing a
single poly-evaluator module. However, when two modules
are used, the cisoid evaluation is achieved within a single
clock cycle. In addition to the cisoid evaluation, the poly-
evaluator module has been applied for the generation
of Gaussian random variables [24], which demonstrates
the usefulness of the piecewise polynomial approximation
technique.
4. Parameter Computation Methods
In order to demonstrate the performance of the proposed
hardware fading channel simulator under dierent propa-
gation scenarios, two methods were chosen to compute the
parameters of the SOC model: the extended method of exact
Doppler spread (EMEDS) for the case of isotropic scattering
and the modied method of equal areas (MMEA) for non-
isotropic scattering.
4.1. The EMEDS. The EMEDS was proposed in [25] as an
extension of the method of exact Doppler spread (MEDS)
[26]. The MEDS was originally proposed for computing
the model parameters of SOS Rayleigh fading channel
simulators, while the EMEDS has been developed for the
parametrization of SOC Rayleigh fading channel simulators.
Currently, the EMEDS is one of the principal methods for
computing the model parameters of SOC channel models.
This method has been highly recognized in the case of
isotropic scattering.
AGU
ROM
Address
AGU
address
5
D Q
E
D Q
E
FF
FF
D Q
E
D Q
E
FF
FF
FSM
C
o
u
n
t
e
r
Start
Sequence control
0 1 1 1 0
Local control
ROM ROM
Piecewise polynomial coefcients
Sinusoid
output
Polynomial
evaluator
Sinusoid
output
Q(32, 30)
x
argument
x
argument
p
2
coecient
x
argument
p
1
coecient
p
0
coecient
x argument
x argument x argument
+
+
Horners rule
P2 P0 P1
x
argument
p
2
p
1
p
0
[0, 1)
[0, 1)
(p
2
x + p
1
)x
+ p
0
Q(32, 25) Q(32, 25) Q(32, 25)
Q(32, 30)
Q(32, 30)
Q(32, 30)
Q(32, 25) Q(32, 26)
Q(32, 26)
Q(32, 25)
Q(32, 25)
Q(32, 26) Q(32, 26)
Q(32, 30)
Q(32, 30)
Figure 4: Architecture of the poly-evaluator module.
For the Jakes PSD, the EMEDS proposes to compute the
path gains c
n
and Doppler frequencies f
n
as follows [1, 25]:
c
n
=
0
2
N
,
(22)
f
n
= f
max
cos
_
2
N
_
n
1
4
__
(23)
for n = 1, 2, . . . , N.
4.2. The MMEA. Currently, there are only some fewmethods
available for computing the model parameters of SOC
models under non-isotropic scattering conditions [1]. The
MMEA provides a reasonably good solution for this kind
of problems. The MMEA [27, 28] is a modication of the
method of equal areas (MEA) reported in [29]. The MMEA
methodology requires the computation of an inverse func-
tion for obtaining a set of Doppler frequencies. When using
the MMEA, it is necessary to compute rst the cumulative
power function (CPF) of a given asymmetrical PSD, and
then one has to nd the inverse CPF (ICPF). An algorithm
was developed using MATLAB in order to obtain the
ICPF function. This algorithm uses a piecewise polynomial
approximation technique. It is completely parameterizable,
allowing the conguration of the polynomial degree, the
number of segments at which the ICPF is approximated,
the segmentation technique (uniform or non-uniform seg-
mentation), and the method for obtaining the polynomial
coecients (least square or minimax approximation).
Once the ICPF is obtained, the Doppler frequencies are
computed according to the following expression:
f
n
= F
1
f
_
2
N
_
n
1
2
_
_
, (24)
where F
1
f
() represents the ICPF. The path gains c
n
are
computed by using (22).
5. Test Results
Using the parameter computation methods described in Sec-
tion 4, the accuracy of the proposed architecture will be eval-
uated in the present section. For this reason, the hardware
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.1
0.2
0.3
0.4
0.5
0.6
Signal level, r
P
D
F
,
p
^
(
r
)
(a)
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.2
0.4
0.6
0.8
1
Signal level, r
C
D
F
,
P
^
(
r
)
(b)
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.2
0.4
0.6
0.8
1
Signal level, r
N
o
r
m
a
l
i
z
e
d

L
C
R
,
N
^
(
r
)
/
f
m
a
x
(c)
0 5 10
10
0
10
1
20 15 10 5
Signal level, r (dB)
N
o
r
m
a
l
i
z
e
d

L
C
R
,
N
^
(
r
)
/
f
m
a
x
(d)
10
1
10
0
10
1
10
2
0 5 10 20 15 10 5 25
Reference model (theory)
Simulation model (theory)
Simulation model (experiment)
Hardware implementation
N
o
r
m
a
l
i
z
e
d

A
D
F
,
T
^
(
r
)
f
m
a
x
(e)
Figure 5: PDF, CDF, LCR, and ADF comparison results of the generated envelope process under isotropic scattering (Jakes PSD) using
2
0
= 1, N = 32, and f
max
= 91 Hz.
channel simulator has been congured and tested according
to the criteria described in the following subsection.
5.1. Test Conguration. In addition to the path gains c
n
and
Doppler frequencies f
n
, the phases
n
have been computed
using a uniform random number generator, as reported
in [30]. For the generation of Rayleigh processes, the LOS
component in (2) has been set to zero, that is, m = 0. The
number of cisoids N was set to 32, the segment number
for the sine and cosine function generation was 32, and the
datapath in the general architecture was equal to WL = 32
bits. This conguration has been chosen for comparison
purposes with other available architectures (see Section 6).
The maximum Doppler frequency f
max
was xed to 91 Hz.
The variance
2
0
was equal to unity in case of isotropic
scattering (Jakes PSD). For non-isotropic scattering (von
Mises PSD), the parameter
2
0
was set to 0.5. Finally, the
statistical properties were obtained by averaging the results
over 10 trials, and updating only the phases for each trial.
0 0.5 1 1.5 2 2.5 3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Signal level, r
P
D
F
,
p
^
(
r
)
(a)
0 0.5 1 1.5 2 2.5 3
0
0.2
0.4
0.6
0.8
1
Signal level, r
C
D
F
,
P
^
(
r
)
(b)
0 0.5 1 1.5 2 2.5 3
0
0.2
0.4
0.6
0.8
1
Signal level, r
Reference model (theory-Jakes)
Reference model (theory-vM)
= 0,
0
= 0
= 1,
0
=

3
= 5,
0
=

4
N
o
r
m
a
l
i
z
e
d

L
C
R
,
N
^
(
r
)
/
f
m
a
x
(c)
0 5 10
10
1
10
2
10
3
10
0
25 20 15 10 5
= 0,
0
= 0
= 5,
0
=

4
= 1, 0 =

3
N
o
r
m
a
l
i
z
e
d

L
C
R
,
N
^
(
r
)
/
f
m
a
x
(d)
10
1
10
1
10
0
0 5 25 20 15 10 5
= 5,
0
=

4
= 1,
0
=

3
= 0,
0
= 0
N
o
r
m
a
l
i
z
e
d

A
D
F
,
T
^
(
r
)
f
m
a
x
(e)
Figure 6: PDF, CDF, LCR, and ADF comparison results of the process generated envelope under non-isotropic scattering (von Mises PSD)
using
2
0
= 0.5, N = 32, f
max
= 91 Hz, and a parameter conguration pair of ( = 0,
0
= 0), ( = 1,
0
= /3), and ( = 5,
0
= /4).
5.2. Hardware Test Results. The comparison results of the
proposed hardware fading channel simulator and the refer-
ence model for isotropic scattering (Jakes PSD) are presented
in Figure 5. The PDF, CDF, and LCR of the reference model
are described by (6), (8), and (11), respectively. The results
obtained for the reference model are shown in form of
continuous lines. The ADF of the reference model has been
found by computing the quotient of the expressions in (8)
and (11).
Equations (5), (7), (10), and (12) are plotted in Figure 5
as dashed lines, which represent the theoretical results
of the simulation model. The experimental results of
the implemented SOC simulation model in oating-point
representation have been computed using MATLAB. This
model has been implemented for reasons of comparison
with the hardware implementation. The envelope generated
by the SOC model can be identied in the plots by dot
markers. Finally, the square markers are used for illustrating
the behavior of the proposed hardware fading channel
simulator.
Figure 5(a) shows the PDF plots of the generated enve-
lope process for N = 32 cisoids. This PDF converges to the
Rayleigh density as N increases. However, the resulting PDF
obtained with the hardware output samples has a perfect
match with the corresponding PDFs of the theoretical and
experimental simulation models. The CDF comparison is
presented in Figure 5(b). The normalized LCR is plotted
in Figure 5(c) and its corresponding plot in decibels (dB)
is shown in Figure 5(d). Finally, Figure 5(e) shows the
comparison of the ADFs.
Figure 6 illustrates the results when the proposed archi-
tecture is congured for producing samples under non-
isotropic scattering conditions using the von Mises AOA
distribution. The parameters and
0
were set according to
the following congurations: ( = 0,
0
= 0), ( = 1,
0
=
/3), and ( = 5,
0
= /4). The PDF and CDF plots of
the generated envelope process are shown in Figures 6(a) and
6(b), respectively. The resulting LCR for the dierent cong-
urations of and
0
is presented in Figure 6(c), and its cor-
responding plot in decibels (dB) is illustrated in Figure 6(d).
Figure 6(e) shows the plot corresponding to the ADF. It can
be observed that for = 0 and
0
= 0, the graphs of the
reference model dened for the cases of isotropic and non-
isotropic scattering perfectly overlap with the experimental
results.
6. Performance Analysis
The performance of the proposed hardware channel simula-
tor was analyzed using 32, 24, 18, and 16 bits for WL. The
implementation was synthesized in a Xilinx Virtex5 FPGA
(XC5VLX110T-3FF1136). The used hardware resources of
the hardware architecture are summarized in Table 1. The
four random-access memories (RAMs) are needed for stor-
ing the gains c
n
(one RAMblock), the Doppler frequencies f
n
(one RAM block), and the cosine and sine phases (one RAM
block each). It is important to mention that the RAM sizes
are parameterizable in accordance to the number of cisoids
Table 1: Hardware resource analysis of the SOC fading channel
simulator for an implementation in a virtex5 XC5VLX110T-
3FF1136 using N = 32 cisoids.
Used elements
Datapath bit-width
16 18 24 32
Number of slice registers 609 675 873 1086
Number of slice LUT 708 768 975 978
Number of RAMs (N WL) 4 4 4 4
Number of ROMs (32 WL bits) 3 2 3 2 3 2 3 2
Number of multipliers 6 6 6 6
Number of adders 16 16 16 16
N. On the other hand, the read-only memories (ROMs)
allocate the coecients of the polynomial approximation. In
this case, the number of segments has been congured to 32.
Consequently, only three ROMs of 32 memory locations (for
each quadrature component) are needed for a single poly-
evaluator module. Likewise, the multipliers are distributed
as follows: two multipliers for each poly-evaluator module
and two multipliers for the multiplication between the gain
c
n
and the cisoid.
Table 2 shows the performance comparison of the
proposed architecture with the hardware fading channel
simulators. These simulators, which are also using the SOS
principle, are the most representative and novel works
described in [9, 10]. In [9], a MIMO channel simulator has
been implemented using 32 stochastic processes. The inphase
and quadrature phases of the channel model are updated for
each realization using a random walk process. Moreover, in
[10], a triply selective SOS fading channel simulator has been
implemented, where the sinusoids are generated in parallel
fashion. It should be noted that a detailed comparison of the
hardware resources is dicult to achieve. This is due to the
dierences in hardware implementation, as well as due to the
FPGA technologies utilized in the references [9, 10].
In order to ensure a fair comparison, the hardware
resources of the fading channel simulators described in
[9, 10] have been inferred from the architecture schematics
described by the authors. Therefore, Table 2 only con-
siders the hardware resources of a single SOS Rayleigh
fading channel simulator module. From Table 2, it can be
observed that the principal resource saved by using the
proposed architecture lies in the ROMs, which are needed
for implementing the sine/cosine functions of the SOS
models for all compared methods. This saving is due to
the fact that our architecture only stores the polynomial
coecients for the cosine/sine function generation, while the
other architectures have to store a rather large number of
sinusoid samples. Furthermore, the use of the polynomial
approximation technique makes it possible to achieve a high
frequency resolution with regard to the sinusoid genera-
tion, which greatly improves the statistics of the generated
process. In this work, the frequency resolution achieved is
7.63 Hz. This value has been computed using (21a), where
the fading generation rate is F
c
= 16, 384 samples per
second.
Table 2: Performance comparison of fading channel simulators (Rayleigh fading).
Used elements
Implementation
[9] [10] [Proposed]
Number of sinusoids 32 16 32
Clock. freq. (MHz) 224 50 90
Number of multipliers 1 2 16 + 2 6
Number of adders 9 4 16 + 2 16
Number of RAMs 4 (32 16 bits) Unknown 4 (N WL bits)
Number of ROMs 2 (1024 and 4096 16 bits) D 2 16 16 bits
1
6 (32 WL bits)
Frequency resolution Unknown Unknown 7.63 Hz
SOS-Based SOS SOS SOC
Isotropic scattering Yes Yes Yes
Non-isotropic and asymmetrical scattering No No Yes
Note: The hardware resources have been deduced from [9, 10] by considering a single Rayleigh process.
1
The symbol D is the LUT size, which is not provided in [10].
It should be highlighted that [9] requires only a single
multiplier in its architecture. However, this restricts the
architecture to SOS models in which all sinusoids have
the same amplitude. This is in contrast to the proposed
architecture that can be used for any SOS/SOC model, even
if the sinusoids do not have the same amplitude. Besides
the hardware dierences in previous works and the ones
proposed here, the most important dierence lies in the
functionality: only the proposed simulator has the capability
of generating waveforms characterized by both symmetrical
and asymmetrical PSDs.
7. Conclusions
In this paper, an ecient hardware SOC fading simulator
was presented. This exible and parametric architecture is
capable of generating stochastic processes with symmetrical
and asymmetrical PSDs. The use of a piecewise polynomial
approximation technique allows generating cisoids with a
high frequency resolution, which results in highly accurate
rst- and second-order statistics of the generated envelope
processes. This allows the reproduction of realistic channel
scenarios, as required in the latest wireless communications
standards. In addition, the architecture described in this
paper is simple and easy to implement, considering that
only small hardware resources are needed in comparison
with other hardware SOS simulators. Therefore, the pro-
posed architecture can be considered as a key module for
development of more elaborated channel simulators, such as
wideband and MIMO channel simulators.
References
[1] M. P atzold, Mobile Radio Channels, John Wiley & Sons,
Chichester, UK, 2nd edition, 2011.
[2] W. C. Jakes, Ed., Microwave Mobile Communications, IEEE
Press, Piscataway, NJ, USA, 1994.
[3] R. H. Clarke, A statistical theory of mobile-radio reception,
Bell System Technical Journal, vol. 47, pp. 9571000, 1968.
[4] M. P atzold, Y. Li, and F. Laue, A study of a land mobile
satellite channel model with asymmetrical Doppler power
spectrum and lognormally distributed line-of-sight compo-
nent, IEEE Transactions on Vehicular Technology, vol. 47, no.
1, pp. 297310, 1998.
[5] W. R. Braun and U. Dersch, A physical mobile radio channel
model, IEEE Transactions on Vehicular Technology, vol. 40, no.
2, pp. 472482, 1991.
[6] J. P. Rossi, J. P. Barbot, and A. J. Levy, Theory and
measurement of the angle of arrival and time delay of UHF
radiowaves using a ring array, IEEE Transactions on Antennas
and Propagation, vol. 45, no. 5, pp. 876884, 1997.
[7] J. G. Wang, A. S. Mohan, and T. A. Aubrey, Angles-of-arrival
of multipath signals in indoor environments, in Proceedings
of the IEEE 46th Vehicular Technology Conference, pp. 155159,
May 1996.
[8] W. C. Y. Lee, Finding the approximate angular probability
density function of wave arrival by using a directional
antenna, IEEE Transactions on Antennas and Propagation, vol.
AP-21, no. 3, pp. 328334, 1973.
[9] S. F. Fard, A. Alimohammad, and B. F. Cockburn, An FPGA-
based simulator for high path count Rayleigh and Rician
fading, IEEE Transactions on Vehicular Technology, vol. 59, no.
6, pp. 27252734, 2010.
[10] F. Ren and Y. R. Zheng, A novel emulator for discrete-time
MIMO triply selective fading channels, IEEE Transactions on
Circuits and Systems, vol. 57, no. 9, pp. 25422551, 2010.
[11] S. F. Fard, A. Alimohammad, B. Cockburn, and C. Schlegel,
A single FPGA lter-based multipath fading emulator, in
Proceedings of the IEEE Global Telecommunications Conference
(GLOBECOM 09), Honolulu, Hawaii, USA, December 2009.
[12] R. Parra-Michel, V. Kontorovitch, A. Orozco-Lugo, and
M. Lara, Computational complexity of narrowband and
wideband channel simulators, in Proceedings of the IEEE
58th Vehicular Technology Conference (VTC03), pp. 143148,
Orlando, Fla. USA, 2003.
[13] M. P atzold and B. Talha, On the statistical properties of
sum-of-cisoids-based mobile radio channel simulators, in
Proceedings of the 10th International Symposium on Wireless
Personal Multimedia Communications (WPMC 07), pp. 394
400, Jaipur, India, 2007.
[14] M. P atzold and C. A. Guti errez, Level-crossing rate and
average duration of fades of the envelope of a sum-of-
cisoids, in Proceedings of the IEEE 67th Vehicular Technology
Conference-Spring (VTC 08), pp. 488494, Singapore, May
2008.
[15] S. O. Rice, Mathematical analysis of random noise, Bell
System Technical Journal, vol. 23, pp. 282332, 1944.
[16] S. O. Rice, Mathematical analysis of random noise, Bell
System Technical Journal, vol. 24, pp. 46156, 1945.
[17] M. P atzold, J. V azquez, C. Guti errez, and R. Parra-Michel,
An exact solution for the level-crossing rate and the average
duration of fades of the Envelope of sum-of-cisoids, in
Proceedings of the Conference on Electronics Engineering and
Computer Science (CIIECC 12), pp. 3040, Guadalajara,
Mexico, 2012.
[18] A. Abdi, J. A. Barger, and M. Kaveh, A parametric model
for the distribution of the angle of arrival and the associated
correlation function and power spectrum at the mobile
station, IEEE Transactions on Vehicular Technology, vol. 51,
no. 3, pp. 425434, 2002.
[19] A. Abdi, K. Wills, H. A. Barger, M. S. Alouini, and M.
Kaveh, Comparison of the level crossing rate and average
fade duration of Rayleigh, Rice, and Nakagami fading models
with mobile channel data, in Proceedings of the 52nd Vehicular
Technology Conference (VTC 00), pp. 18501857, September
2000.
[20] L. Cordesses, Direct digital synthesis: a tool for periodic wave
generation, IEEE Signal Processing Magazine, vol. 21, no. 4,
pp. 5054, 2004.
[21] M. P atzold and D. Kim, Test procedures and performance
assessment of mobile fading channel simulators, in Pro-
ceedings of the 59th IEEE Semiannual Vehicular Technology
Conference (VTC 04), pp. 254260, Milan, Italy, 2004.
[22] J. M. Muller, Elementary Functions, Algorithms and Implemen-
tation, Birkh auser, Boston, Mass, USA, 2th edition, 2002.
[23] R. Zarate-Martinez, F. Pena-Campos, J. V azquez Castillo,
and R. Parra-Michel, Arbitrary distribution random variable
generator for channel emulators, in Proceedings of the IEEE
International Conference on Recongurable Computing and
FPGAs (ReConFig 11), pp. 339344, Cancun, Mexico, 2011.
[24] J. V azquez Castillo, A. Castillo Atoche, O. Longoria-Gandara,
and R. Parra-Michel, An ecient Gaussian random number
architecture for MIMO channel emulators, in Proceedings of
the IEEE Workshop on Signal Processing System (SIPS 11), pp.
316321, Beirut, Lebanon, October 2011.
[25] B. O. Hogstad, M. P atzold, N. Youssef, and D. Kim, A MIMO
mobile-to-mobile channel model: part IIthe simulation
model, in Proceedings of the 16th International Symposium on
Personal, Indoor and Mobile Radio Communications (PIMRC
05), pp. 562567, Berlin, Germany, September 2005.
[26] M. P atzold, U. Killat, F. Laue, and Y. Li, On the statistical
properties of deterministic simulation models for mobile
fading channels, IEEE Transactions on Vehicular Technology,
vol. 47, no. 1, pp. 254269, 1998.
[27] C. A. Guti errez-Diaz-De-Leon and M. P atzold, Sum-of-
sinusoids-based simulation of at fading wireless propagation
channels under non-isotropic scattering conditions, in Pro-
ceedings of the 50th Annual IEEE Global Telecommunications
Conference (GLOBECOM 07), pp. 38423846, Washington,
DC, USA, November 2007.
[28] C. A. Guti errez-Diaz-De-Leon and M. P atzold, Ecient sum-
of-sinusoids-based simulation of mobile fading channels with
asymmetrical Doppler power spectra, in Proceedings of the
4th IEEE International Symposium on Wireless Communication
Systems (ISWCS 07), pp. 246251, Trondheim, Norway,
October 2007.
[29] M. P atzold, U. Killat, and F. Laue, A deterministic digital
simulation model for suzuki processes with application to
a shadowed Rayleigh land mobile radio channel, IEEE
Transactions on Vehicular Technology, vol. 45, no. 2, pp. 318
331, 1996.
[30] J. V azquez Castillo, L. Vela-Garcia, A. Castillo Atoche, J. Lopez
Estrada, and R. Parra-Michel, High-speed low-power parallel
random number generator for wireless channel emulators,
in Proceedings of the Symposium on Circuits and Systems
(LASCAS 12), pp. 14, Playa del Carmen, Mexico, 2012.
doi:10.1155/2012/705078
Research Article
Filter-Based Fading Channel Modeling
Amirhossein Alimohammad,
1
Saeed Fouladi Fard,
2
and Bruce F. Cockburn
3
1
Department of Electrical and Computer Engineering, San Diego State University, San Diego, CA 92182, USA
2
Ukalta Engineering, 4344 Enterprise Square, 10230 Jasper Avenue, Edmonton, AB, Canada T5J 4P6
3
Department of Electrical and Computer Engineering, University of Alberta, Edmonton, AB, Canada T6G 2V4
Correspondence should be addressed to Amirhossein Alimohammad, aalimohammad@mail.sdsu.edu
Received 28 April 2012; Accepted 6 August 2012
Academic Editor: Neji Youssef
Copyright 2012 Amirhossein Alimohammad et al. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
A channel simulator is an essential component in the development and accurate performance evaluation of wireless systems. A key
technique for producing statistically accurate fading variates is to shape the at spectrum of Gaussian variates using digital lters.
This paper addresses various challenges when designing real and complex spectrum shaping lters with quantized coecients for
ecient realization of both isotropic and nonisotropic fading channels. An iterative algorithmfor designing stable complex innite
impulse response (IIR) lters with xed-point coecients is presented. The performance of the proposed lter design algorithm
is veried with 16-bit xed-point simulations of two example fading lters.
1. Introduction
Wireless communication systems must be designed to oper-
ate over radio channels in a wide variety of expected envi-
ronments. Testing wireless transceivers is challenging due
to unrepeatable and uncontrollable channel conditions. The
initial performance verication of communication systems
at the early stages of the design cycle is performed based on
the channel characteristics dened by the underlying wireless
communication standard. Therefore, accurate emulation of
fading channels is a key step in the design and verication of
wireless communication systems.
In the multipath propagation scenario, the received
signal contains dierent faded copies of the transmitted
signal. The eect of the multipath fading channel on the
baseband signal can be modeled with a time-variant linear
system with the following impulse response [1]:
(; t) =
Np1
_
n=0
n
|c
n
(t)|e
jcn(t)
(
n
(t)), (1)
where N
p
is the number of independent paths,
n
is the
average attenuation of the nth path, and c
n
(t) and
n
(t)
denote the complex gain and delay of the nth path. Each
path gain c
n
(t) is commonly modeled as a complex Gaussian
wide-sense stationary (WSS) process [2].
To simulate a fading channel, we need to generate a
suitable sequence of complex path gains {c
n
(t)} and then
superimpose delayed replicas of the transmitted samples
with the given delays {
n
(t)} and path attenuations {
n
}.
Two major techniques have been widely used for simulating
fading channels. In the rst approach, the so-called sum-
of-sinusoids (SoS) model, the fading process is modeled as
the superposition of a suciently large number of sinusoidal
waves. This approach was originally proposed by Clarke [3]
and later simplied by Jakes [4]. Over the past four decades
several modied SoS-based models have been proposed (e.g.,
[5, 6]).
The second approach, which is used in this paper, is
called the lter-based scheme. In this approach, to generate
the in-phase c
i
(t) and quadrature components c
q
(t) of
complex fading process c
n
(t) = c
i,n
(t) + jc
q,n
(t) with a
particular correlation between variates, a complex zero-
mean and unit-variance Gaussian random process n(t) =
n
i
(t) + jn
q
(t) with independent samples is passed through
a spectrum shaping lter (SSF) [7] with an appropriate
frequency response H( f ). A linear ltering operation on the
complex Gaussian samples with at power spectral density
(PSD) yields samples that also have a Gaussian distribution,
with spectrum S
y
( f ) = S
n
( f ) |H( f )|
2
, where S
n
( f ) is the
PSD of the input samples and S
y
( f ) is the PSD of the output
samples.
Compared to the SoS-based method, the lter-based
fading simulation method is much trickier to design and
implement. A lter-based simulator needs to be designed
carefully because of possible instability and possible nite
word-length eects when implemented with xed-point
arithmetic. On the other hand, lter-based fading simulation
has several advantages over the SoS-based method. First,
with the lter-based method, it is possible to simulate a
wide range of power spectral densities. Second, the lters can
be designed to provide a high level of statistical accuracy.
Finally, the generated samples have very accurate Gaussian
distribution.
In this paper, we present various considerations for
designing stable xed-point spectrum shaping lters for
modeling both isotropic and nonisotropic fading channels.
We propose an iterative lter design technique. The least-
squares cost function is suggested in polar coordinates and
augmented with a barrier function that keeps the poles (and
potentially also the zeros) within the unit circle to enforce
lter stability.
The rest of this paper is organized as follows. Section 2
discusses various considerations when designing isotropic
fading channels. Section 3 presents lter design techniques
for isotropic fading channels. Section 4 discusses the model-
ing of nonisotropic fading channels. In Section 5, we present
our stable real and complex lter design techniques for stable
xed-point implementations. Finally, Section 6 makes some
concluding remarks.
2. Isotropic Fading Channels
Isotropic scattering refers to the case in which the distri-
bution of the incident directions of the received multipath
signals, or angle of arrival (AoA), are equally distributed.
Assuming an isotropic scattering Rayleigh fading channel
with an omnidirectional antenna at the receiver, the path
gains are modeled using a unit-variance zero-mean complex
Gaussian process c(t) = c
i
(t) + jc
q
(t) [2]. Both c
i
(t) and c
q
(t)
are Gaussian-distributed independent stochastic processes
that have the same autocorrelation function (ACF) R
ci ,ci
() =
R
cq,cq
() = J
0
(2 f
D
)/2, where f
D
is the maximum Doppler
frequency and J
o
() is the zeroth-order Bessel function of
the rst kind [4]. The PSD associated with either the in-
phase or quadrature component of a complex fading signal
has the well known Jakes U-shaped band-limited form with
independent in-phase and quadrature samples [8]:
S
c
_
f
_
=
_
_
1
_
f
2
D
f
2
if

< f
D
,
0 otherwise.
(2)
Fading samples can be generated by passing a stream of inde-
pendent Gaussian samples through a SSF with the magnitude
response equal to the square root of the magnitude of the
PSD of the desired fading process (i.e., |H( f )| = |S
c
( f )|
1/2
).
The ltering process can be carried out in the time or
frequency domains. In the frequency domain, the Gaussian
samples are multiplied by |S
c
( f )|
1/2
. Then, an inverse fast
Fourier transform (IFFT) can be applied to the resulting
discrete spectrum to obtain time series fading samples [9].
The resulting series is still Gaussian by virtue of the linearity
of the IFFT, and it has the desired Jakes spectrum. One
major disadvantage of the IFFT method is its block-oriented
nature, which requires all fading coecients to be generated
and stored before the data is sent through the channel.
This implies signicant memory requirements and precludes
unbounded continuous transmission, which is usually pre-
ferred in long running characterization applications such as
hardware fading simulation.
In the time domain, the SSF can be implemented with
either nite-impulse response (FIR) lters [10, 11] or innite
impulse response (IIR) lters [12, 13]. When implementing
fading channel simulators using FIR and IIR lters, it is
important to note that the degree of the FIR lter is
related to the time span of the truncated signal held in the
lter and inversely proportional to the Doppler frequency.
Specically, implementation of an extremely narrowband
digital lter with a sharp cuto and a very large attenuation
in the stop-band requires a high-order FIR lter. Meeting
the same specications with an IIR lter typically requires
fewer hardware resources than an FIR lter. In fact, utilizing
both feedforward and feedback polynomials in an IIR lter
permits steeper frequency roll-os to be implemented for a
given lter order than an FIR lter [14]. However, an FIR
lter has no feedback and is thus inherently stable. As the
coecients are quantized in any xed-point implementation,
the resulting numerical error is fed back in the IIR lter,
possibly causing instability. Moreover, such eects can cause
signicant deviations from the expected response. Thus we
must ensure that the designed SSF is stable.
The autoregressive (AR) modeling approach has also
been proposed for generating fading processes by passing
the noise through an all-pole IIR lter [15]. To produce
samples with accurate statistics, the AR model needs a large
lter order, which greatly increases the number of required
multiplications per output sample. Also, implementation of
the AR fading simulator demands highly accurate oating-
point variables, which makes it unappealing for compact
xed-point implementations.
3. SpectrumShaping Filter Design for
Isotropic Fading Channels
In the design of the SSF, IIR lters are widely preferred to
FIR lters due to their reduced computational complexity
and their higher stopband attenuation. Moreover, since no
constraints are imposed on the phase response, we choose
to use IIR lters. We approximate the desired magnitude
response of the SSF with an IIR lter of order . The
magnitude response of the SSF is expressed as
H
_
e
j
_
=
/2
q=1
_
g
q
1 + b
1,q
e
j
+ b
2,q
e
j2
1 + a
1,q
e
j
+ a
2,q
e
j2
_
, (3)
which is equivalent to the magnitude response of /2
cascaded canonic second-order sections (SOSs), also known
as biquads. In (3), b
1,q
, b
2,q
, a
1,q
, and a
2,q
denote the
coecients and g
q
denotes the real-valued scaling factor of
the qth biquad.
For a typical wireless communication scenario, the
Doppler frequency f
D
is signicantly lower than the signal
sample rate F
s
. Thus, SSF would have an extremely narrow
bandwidth and a very sharp cut-o. We can reduce the
complexity of SSF by designing it at a much lower sampling
frequency, F
1
F
s
thereby reducing the required computa-
tions and also improving the accuracy of the designed lter.
The resulting low-rate signal can then be interpolated to
reach the target sampling frequency F
s
= F
1
Tg
j=1
I
j
= F
s
,
where T
g
denotes the number of interpolation stages and
I
j
is the interpolation factor at the jth interpolation stage.
Given the desired frequency response S
1/2
c
( f ) = (
2
( f
2
D

f
2
))
1/4
, | f | < f
D
, we can nd the IIR lter coecients
using the MATLAB function iirlpnorm. This function uses
double-precision oating-point variables and calculates the
optimal values by minimizing the p-norm [16].
Due to the higher hardware cost and complexity of
oating-point hardware, xed-point arithmetic is often
preferred in very-large-scale integration (VLSI) and eld-
programmable gate array (FPGA) implementations. After
the initial lter design, the lter coecients and scaling
factors are then quantized for xed-point implementation.
For a compact hardware implementation, variables should
be implemented with the minimum possible xed-point
word-length. However, reducing the word-length impacts
the response, and potentially the stability, of the designed
IIR lter. When the lter coecients are quantized from
oating-point to xed-point, the poles and zeros of the sys-
tem function typically shift to new positions in the z-plane.
Unfortunately, this step can perturb the implemented fre-
quency response from its intended response. If the designed
IIR lter is extremely sensitive to coecient changes, the
lter response might not meet the target specications or
the lter might even become unstable [17]. At this step,
the lter has to be tested with xed-point variables, and
scaling factors g
q
are determined that suciently limit the
magnitude of the generated samples to keep them within the
representable range. To make sure that the lters are stable
under quantization eects, we have designed the lters in
xed-point format using Filter Design Toolbox [18], which
oers bit-true implementations of SOSs with section scaling
and reordering to obtain maximum accuracy.
An important observation is that if the stopband atten-
uation of the shaping lter is not suciently high, then
the out-of-band noise that passes through the lter will
degrade the accuracy of the statistics of the generated fading
variates. Specically, since designing a narrowband lter
with a sharp cuto and large attenuation invariably leads to
a high-order lter, to obtain the closest approximation to
the desired frequency response with a relatively small lter
order, we only minimized the approximation error in the
passband of the SSF. The low-pass lters utilized downstream
in the interpolator stages can then be designed with extra
attenuation over the transition region to ensure a sharp
cut-o.
To generate the Rayleigh fading process, independent
samples of a zero-mean complex Gaussian process, generated
by the Gaussian noise generator (GNG) block in Figure 1,
pass through an SSF with a magnitude response equal to
the square root of the magnitude of (2). Samples generated
by the shaping lter at a low sampling frequency need
to be oversampled and passed through lowpass lters in
order to obtain the target sample rate F
s
. Let P denote
the integer upsampling factor. Interpolation of the signal
{x[n]} is performed by inserting P 1 zeros between each
pair of successive samples of {x[n]} and then passing the
resulting stream through a low-pass lter with cut-o fre-
quency /P radian/sec. In order to reduce the computational
complexity, we perform the interpolation in multiple stages.
Since only the amplitude response aects the correlation
properties and no restrictions are imposed on the phase
response, we use an elliptic IIR low-pass lter (EILPF) in
the rst stage. The lowpass lter has a symmetric frequency
response and hence its poles and zeros appear in complex
conjugate pairs, therefore, this lter can be realized using
cascaded biquads. The SSF output is upsampled I
1
times and
passed to the EILPF generating F
2
= I
1
F
1
samples per
second. The samples are further upsampled and interpolated
using fading-specic interpolation low-pass lters (SILPFs)
to obtain the desired output sample rate F
s
. Since the
maximum Doppler frequency is typically much smaller than
the sampling frequency, we use a multiplication-free SILPF
with frequency response:
D
lz
P
_
e
j
_
=
_
1 e
jP
P Pe
j
_
lz
, l
z
1,
(4)
where l
z
is the number of cascaded stages.
We propose to design the SSF at a sampling frequency F
1
,
where 4 f
D
< F
1
8 f
D
. Choosing F
1
in this range satises
the minimum Nyquist rate while keeping the computational
complexity low. In addition, we have the opportunity to
exploit power-of-2 interpolation factors to further reduce
the hardware complexity and simplify the lter design. The
generated samples from the SSF are upsampled I
1
= 16
times and passed through the EILPF. Note that since the
SILPF stages are designed to operate on narrowband signals,
the rst interpolation stage is positioned prior to the SILPF
stages. Then, the samples are passed through successive
SILPFs. The ith SILPF interpolates the signal 2
ki
times.
Based on the processing architecture, the relation between
F
1
and the target output sampling rate is F
s
= 16 F
1

Tg
i=1
2
ki
. From here we have F
1
= 2
(4+Sg )
F
s
, where S
g
=
Tg
i=1
k
i
is an integer value in the range log
2
(F
s
/ f
D
) 7
S
g
< log
2
(F
s
/ f
D
) 6. Based on the maximum interpolation
factor 2
Kmax
, where K
max
= 4 + max{S
g
}, each SILPF
is assigned a specic interpolation factor. The minimum
Doppler frequency that can be simulated by this system is
f
min
D
= 2
(Kmax+3)
F
s
. (5)
F
1
F
1
F
2
F
3
F
s
GNG SSF
SILPF SILPF EILPF I
1
I
2
I
Tg

Figure 1: Block diagram of the fading simulator.
0 0.5 1 1.5 2 2.5 3 3.5 4
1
0.5
0
0.5
1
Lag (seconds)
A
C
F
Theoretical autocorrelation
Measured autocorrelation
Figure 2: ACF of the generated fading samples with f
D
= 0.6 Hz,
F
s
= 2.5 MHz over 60 seconds.
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.2
0.4
0.6
0.8
c
p
C
(
c
)
Reference Rayleigh PDF
Measured PDF
Figure 3: PDF of the generated fading samples with f
D
= 200 Hz,
F
s
= 10 MHz.
To ensure accuracy, we measured the ACF and the
probability density function (PDF) of the generated xed-
point fading samples against the ideal functions. Figure 2
shows the autocorrelation of the real part of the generated
fading process. The Doppler frequency is set to f
D
= 0.6 Hz
and the sample rate is F
s
= 2.5 million samples per second.
This gure conrms a close match between the desired
response and the generated results over up to 60 seconds. In
another example, we measured the PDF for the amplitude of
the generated samples with f
D
= 200 Hz and F
s
= 10 MHz.
The plots in Figure 3 show that the measured PDF accurately
matches the ideal Rayleigh PDF.
4. Modeling Nonisotropic Fading Channels
Isotropic scattering assumption has been challenged [19]
due to the blockage of some propagation directions and
antenna directivity, resulting in a nonuniform PDF for
AoA at the receiver. Several nonuniform PDFs have been
proposed in the literature to represent the AoA including
the geometrically based PDFs [20], Gaussian PDF [21],
quadratic PDF [22], Laplace PDF [23], cosine PDF [24], and
von Mises PDF [25]. The von Mises PDF, which includes the
uniform AoA distribution as a special case, is supported with
empirical measurements of narrowband fading channels in
[25]. Also it is argued that the von Mises PDF is attractive
because it can approximate other nonuniform PDFs and can
provide mathematical convenience for analysis [25].
When the scattering encountered in the propagation
environment is nonisotropic [26, 27], the complex envelope
of the fading process is
c(t) = c
i
(t) + jc
q
(t)
= lim
N
1
N
N
_
n=1
n
e
j(2 fD cos(n)+n)
,
(6)
where
n
, n = 1, . . . , N are independent and identically
distributed (i.i.d.) angles of arrival of the incoming wave
at the receiver antenna with distribution p
(),
n
, n =
1, . . . , N are i.i.d. phases with uniform distribution over
[, ), and
n
, n = 1, . . . , N are deterministic normalized
complex constants that satisfy

N
n=1
|
n
|
2
= N. The PSD
function associated with c(t) is given by [25]
S
c
_
f
_
=
e
cos( ` )( f / fD)
cosh
_
sin
_
`
_
_
1
_
f / f
D
_
2
_
I
0
()
_
1
_
f / f
D
_
2
,
(7)
where controls the beam width, ` denotes the average
AoA of the scattered component, and I
m
() is the mth order
modied Bessel function of the rst kind. To obtain (7),
it is assumed that the AoA of the scattered component is
distributed with the von Mises/Tikhonov distribution [28]
as follows:
p
_
=
exp
_
cos
_
`
__
2I
0
()
, [, ). (8)
Note that when the beam-width parameter is zero, the AoA
has a uniform distribution over [, ), and (7) reduces to
Jakes U-shaped spectrum S
c
( f ) = (
_
1 ( f / f
D
)
2
)
1
.
One of the important steps for generating the fading
samples is designing the SSF. While most of the previously
proposed lter design techniques might be appropriate
for simulations of isotropic fading, they are not the best
candidates for nonisotropic scattering scenarios, where AoA
is not uniformly distributed or when the antennas are
not omnidirectional. In contrast to isotropic fading, the
PSD of fading samples in nonisotropic scattering (7) is
not symmetric in general, and hence the lter coecients
are potentially complex [29] (e.g., the TGn fading channel
models for simulating IEEE 802.11 radio propagation [30]).
We will refer to lters with real coecients as real lters
and lters with complex coecients complex lters.
Various methods such as nonlinear optimization [31],
linear programming [32], and semidenite programming
[33] have been suggested for designing IIR digital lters
in the complex Chebyshev sense. Least-squares methods
have also been applied extensively to design IIR lters [34].
For lter stabilization, several approaches have also been
proposed. One method, proposed in [35], is to start the
optimization from a stable point and then control the step
size so that the solution never leaves the stable region.
However, this method is computationally expensive and not
easy to implement with traditional optimization procedures.
In a second method, explicit constraints are imposed on
the coecients of the denominator of the transfer function
[36]. This technique, however, has some limitations that
aect the lter quality [37]. Finally, in a third method, the
least squares cost function is modied so that the minimum
always falls in the stable region [37]. This is ensured by
adding a barrier function to the original cost function to
avoid lter instability. To design the barrier function, an all-
pole proxy transfer function is formed consisting of all of the
lter poles. The barrier function is basically the sum of the
squared amplitude of a section of the impulse response of the
proxy transfer function. If the lter is unstable, the trailing
tail of the response will have (large) nonzero values.
For ecient lter implementation, the lter design
method needs to produce lters that can be implemented
with the minimumword-length while providing the required
accuracy. The bit-precision selected for implementing the
SSF plays an important role in the accuracy, stability, and
eciency of the lter implementation. Various methods have
been proposed for designing IIR lters with xed-point
coecients [3841]. In [38], the optimal xed-point real-
valued lter is found by reformulating a cost function to
include the hardware complexity. The cost function is then
minimized using the simulated annealing method [42]. In
this approach, no constraints are imposed on the precision
of the intermediate signals, which can result in inecient
implementations. The lter design method proposed in
[39] is based on FIR-to-IIR lter conversion. However, the
resulting IIR lter has signicantly higher complexity than
its FIR prototype. In [40], IIR lter design is framed as a
combinatorial optimization problem that is solved using a
genetic algorithm. The design approach in [41] proposes bit-
ipping and the downhill-simplex method for xed-point
IIR lter design.
5. SpectrumShaping Filter Design for
Nonisotropic Fading Channels
We represent the IIR lter as a product of rst-order sections
(FOSs) [43]:
H
_
e
j
_
= A
k=1
1 r
k
e
j(k)
1 s
k
e
j(k)
, (9)
where A is a positive scaling factor, r
k
e
jk
and s
k
e
jk
are the
kth complex zero and pole, respectively, and is the number
of FOSs, that is, the lter order. Here, we focus on designing
IIR lters with a prescribed amplitude response. When the
amplitude of the frequency response is symmetric, the poles
and zeros of (9) appear as complex conjugate pairs and the
IIR lter can be implemented using /2 biquads.
We assume that the desired amplitude response is
represented with 2 M samples y
d
i
, where
y
d
i
=
_
G
_
e
j2ui
_
if 2u
i
is in the passband
, otherwise,
(10)
and where G(e
j2ui
) is the desired response, u
i
[0.5, 0.5]
is the normalized sampling frequency, and > 0 is the
attenuation in the stopband. Similar to the work in [44], we
express H(e
j
) = AF(x; e
j
), where F(x; e
j
) represents the
product of FOSs in (9) and the column vector x of length 4
contains r
k
, s
k
,
k
, and
k
for k = 1, . . . , . Next, to nd the
lter parameters we dene the cost function as follows:
q(A, x) =
2M1
_
i=0
v
i
_
log
_
A
F
_
x; e
j
_
_
log
_
y
d
i
__
2
+ B
_
; ; x
_
,
(11)
where the weight vector v = [v
1
, v
2
, . . . , v
2M
]
T
allows us
to emphasize the error minimization for certain frequency
bands. Note that the sum of squared errors on a logarith-
mic scale is augmented by a parametric barrier function
B(; ; x). Function B(; ; x) is included to keep the poles
(and zeros, if necessary) within the unit circle to enforce lter
stability and is dened as
B
_
; ; x
_
=
2
_
k=k0
b
_
; ; x
k
_
, (12)
where
b
_
; ;
_
=
_
_
0 if || ,
_
||
1
_
2
if < || 1,
2
1
||
_
1 +
1
_
if || > 1.
(13)
In (12), determines how fast the barrier function grows
outside of the unit circle and the parameter 1 determines
an outer boundary for the poles and zeros. When k
0
= + 1,
the barrier function tries to keep the poles within a circle of
radius . Setting k
0
= 1, on the other hand, forces both the
poles and zeros into the same boundary. The barrier function
(12) is especially useful when designing lters for xed-point
implementation since it can be parameterized to keep the
poles and zeros at any desired safe distance from the unit
circle. Moreover, using this technique, the quantization noise
can be reduced to acceptable levels. It can be shown that the
variance of the quantization noise that originates in the kth
FOS when implemented in direct-form-I (DF-I) is
2
f
(k) =
7 2
2(k1)
6
_
1 s
2
k
_
,
(14)
where
k
is the number of bits used to quantize the
coecients (and the intermediate variables) at the kth stage.
To derive (14), it is assumed that the quantization noise
after each multiplier is uniformly distributed, wide-sense
stationary white noise that is uncorrelated with both the
input signal and the quantization noise in other stages. We
also assume that the samples are truncated and represented
in 2s-complement. Controlling the maximumabsolute value
of the lter poles limits the quantization eects to an
acceptable level.
The coecients of the IIR lter are calculated using
an iterative optimization algorithm. At each iteration, the
optimum scaling factor A
o
is calculated as
A
o
=
2M1
i=0
_
y
d
i
F(x; e
j2ui
)
_
(vi /
2M1
i=0
vi )
.
(15)
This expression for A
o
is found by dierentiating (11) with
respect to A and setting the resulting expression to zero. To
calculate the gradient vector
g(A
o
, x) =
_
q(A
o
, x)
x
1
, . . . ,
q(A
o
, x)
x
4
_
T
,
(16)
the partial derivative of q(A
o
, x) with respect to the kth
component x
k
of x can be calculated as
q(A
o
, x)
x
k
= 2
2M1
_
i=0
_
_
_
_
_
_
v
i
log
_
A
o
F
_
x; e
ji
_
/ y
d
i
_
F(x; e
ji
)
F
_
x; e
ji
_
x
k
_
_
_
_
_
_
+
B
_
; ; x
_
x
k
,
(17)
where
i
= 2u
i
,
F
_
x; e
j
_
r
k
=
F
_
x; e
j
_
r
k
cos(
k
)
1 r
k
e
j(k)
2
,
F
_
x; e
j
_
k
=
F
_
x; e
j
_
r
k
sin(
k
)
1 r
k
e
j(k)
2
,
F
_
x; e
j
_
s
k
=
F
_
x; e
j
_
cos
_
_
s
k
1 s
k
e
j(k)
2
,
F
_
x; e
j
_
k
=
F
_
x; e
j
_
s
k
sin
_
1 s
k
e
j(k)
2
,
(18)
Require: ; ; ; v = [v
i
], and y
d
= [y
d
i
]
for i = 0, . . . , 2M1
Initialize k = 0, x
0
, E
0
= 20 I
44
while|x
k+1
x
k
| do
nd A
o
k
from (15)
nd g
k
= g(A
o
k
; x
k
)
k
=
_
g
T
k
E
k
g
k
` g
k
= g
k
/
k
x
k+1
= Q[, x
k
(1/(4 + 1))E
k
` g
k
]
E
k+1
= ((4)
2
/((4)
2
1))(E
k
(2/(4 + 1))E
k
` g
k
` g
T
k
E
k
)
k = k + 1
end while
Algorithm 1: Iterative calculation of the lter coecients [43].
and each partial derivative of the barrier function is given by
B
_
; ; x
_
x
k
=
_
_
0 if |x
k
| or k<k
0
or k>2,
2
_
|x
k
|
_
_
1
_
2
sign(x
k
) if < |x
k
| 1,
2
1
sign(x
k
) if |x
k
| 1.
(19)
Algorithm 1 summarizes the steps for our proposed
iterative lter design. We utilized the ellipsoid algorithm
[45] here for its simplicity; however, other optimization
techniques could be used. This lter design algorithm can
be parameterized to provide a close approximation of the
desired response. The desired response y
d
= [y
d
i
], the xed-
point format used for the lter coecients, a weight
vector v, and the outer boundary for zeros and poles are
passed to Algorithm 1. The lter design procedure can start
with a reasonable order for the initial approximation. The
lter order can be increased gradually if the desired lter
characteristics are not met. The algorithm starts from an
arbitrary point x
0
contained within the unit sphere and a
relatively large (20 times in this algorithm) initial ellipsoid
matrix E
0
= 20 I
44
, where I denotes the identity matrix.
The algorithm then searches for the optimal solution within
the present ellipsoid of feasible points. This algorithm then
converges on the optimal solution by successively reducing
the size of the ellipsoid by
k
until it is small enough (i.e.,
the algorithm has converged) or when |x
k+1
x
k
| is within
a chosen accuracy . The function Q[, x] represents the
quantization eects that aect each element of x in the
Cartesian coordinate system (coecients are transferred to
Cartesian coordinates, quantized, and then transferred back
to the polar coordinates). Note that stable real IIR lters
can be designed with the above algorithm as well. To design
such lters, the sample update is only performed for half of
10
0
10
20
30
40
50
60
0.5 0.4 0.3 0.2 0.1 0 0.1 0.2 0.3 0.4 0.5
Desired
Normalized frequency (rad/Hz)
(a)
(b)
Fixed-point design
F
r
e
q
u
e
n
c
y

r
e
s
p
o
n
s
e

(
d
B
)
Figure 4: Frequency response of the designed xed-point lters and
the desired responses.
the poles and zeros, and the others are simply the complex
conjugates of the updated samples.
An important point to note is that IIR lters are naturally
susceptible to arithmetic overow and instability due to the
inherent feedback. Both the design and implementation of
digital IIR lters must be carried out carefully to avoid such
pitfalls. While scaling techniques are commonly used to keep
the lter variables in range [46], a poor choice of scaling
factor can result in a loss of signal precision and an increase
in quantization noise. Another technique is to use more
bits to represent intermediate signals. This method, however,
is inecient on DSPs with a xed word-length. Moreover,
simply adding extra bits increases resource utilization in
FPGAand ASICimplementations. We suggest to use (i) pole-
zero ordering and/or (ii) augmenting auxiliary poles and
zeros techniques for minimizing the range of intermediate
signals that can be eectively used for reducing the signal
range, overow probability, and resource utilization in
hardware implementations [47].
Considering a small section of an IIR lter, overows are
mainly caused by oscillations around resonant frequencies.
Assuming a limited input signal range, we can reduce
the signal range by reducing the maximum oscillation
magnitude. The oscillation frequency is mainly determined
by lter poles. Consider a single FOS of an IIR lter with
only one pole at frequency e
j2 fpole
. The output range of this
section can be reduced signicantly if the input signal to this
section is attenuated around frequency e
j2 fpole
. In pole-zero
ordering technique, we need to implement the IIR lter with
the DF-I structure so that the input signal experiences a zero
before it is aected by the pole. We thus sort the lter sections
according to their pole magnitude in an ascending order and
match the lter poles with the closest zeros. Further, signals
are scaled in the dierent stages to assure lter stability.
The second method for reducing the signal range is to
attenuate the input signal around the resonant frequencies
of an IIR lter e
j2 fNat
, that could potentially result in
oscillation or overow. This imposed distortion can be later
compensated for with additional poles and/or zeros. This
technique is applicable only if the input signal does not have a
major frequency component around e
j2 fNat
. One example is
the implementation of narrowband low-pass lters with an
approximate resonance frequency around DC. If the input
signal does not have a DC component, it can be rst passed
through a high-pass lter D(e
j
) = 1e
j
(dierence) prior
to being passed through the lter. The lter output can then
be later compensated by passing the output signal through
the integrator I(e
j
) = (1 e
j
)
1
. The coecient
[1, 1) is intentionally added here since quantization noise
and computational errors can render a perfect integrator
(i.e., = 1) unstable.
When employed along with pole-zero ordering, the
augmented poles and zeros are not included in the ordering
process, and instead they keep their position in the DF-I
structure, that is, the augmented zero appears rst and the
pole appears last. This technique, when used in conjunction
with pole-zero ordering and scaling, can provide ecient,
accurate, and compact implementation of real and complex
IIR lters.
To demonstrate the performance of our lter design
procedure, we design two IIR lters with = 12 FOSs. Here,
the SSF is approximated with
H
_
e
j
_
=
k=1
g
k
1 b
1,k
e
j
1 a
1,k
e
j
, (20)
where g
k
is the positive scaling factor, b
1,k
and a
1,k
are the
kth complex zero and pole, respectively, and is the lter
order. This lter can be realized as a cascade of FOSs. For
isotropic scattering, however, poles and zeros of (20) appear
in complex conjugate pairs and the shaping lter can be
implemented using /2 canonic second-order sections. We
set the parameters = 5 and = 0.99, that is, the poles and
zeros are bounded within a circle of radius 0.99. For all of
the FOSs, the number of bits for representing each coecient
is set to = 12. Figure 4 shows the frequency responses of
the designed lters (with out-of-band attenuation = 0.001)
as well as the desired responses. In this gure, lter (a) is a
lowpass lter with normalized pass frequency f
p
= 0.3, and
lter (b) is a complex lter with frequency response H( f ) =
+ f
3
for | f | 0.2 and H( f ) = for | f | > 0.2. As Figure 4
shows, the frequency responses of the designed xed-point
lters closely match the desired responses. Figure 5 shows the
position of the poles and zeros for the designed lters. Note
that all of the poles and zeros lie within a circle of radius
= 0.99.
To illustrate the eectiveness of these range reduction
techniques, we simulated an order = 6 Elliptic low-
pass lter with sample rate F
s
= 4800 Hz, pass frequency
F
pass
= 1200 Hz, stop frequency F
stop
= 1500 Hz, passband
ripple A
pass
=1 dB, and stopband attenuation A
stop
= 50 dB.
We measured the maximum absolute range of variables
by passing 10
8
uncorrelated zero-mean Gaussian samples
through the designed lter. It is assumed that the input signal
(white Gaussian noise) does not have a DC component.
Table 1 shows the maximum absolute signal ranges for
dierent lter implementations. The output of each FOS
or SOS is scaled to lie within [1, +1]. As this table shows,
the direct-form-II (DF-II) implementation requires the
most number of bits (at least nine bits for the integer part).
I
m
a
g
i
n
a
r
y

p
a
r
t
Real part
1 0.5 0 0.5 1
1
0.5
0
0.5
1
(a)
I
m
a
g
i
n
a
r
y

p
a
r
t
Real part
1 0.5 0 0.5 1
1
0.5
0
0.5
1
(b)
Figure 5: Positions of the poles and zeros in the unit circle for the designed lters.
Table 1: Maximum absolute signal range in dierent sections of a
low-pass lter implemented with dierent techniques.
1st section 2nd section 3rd section
DF-II 182.22 422.82 421.44
DF-I + ord. 35.43 20.93 9.79
DF-I + ord. + aug. 20.87 14.81 11.04
The DF-I implementation of this lter with the pole-zero
ordering (DF-I + ord.) reduces the signal range signicantly.
Moreover, augmenting a zero at DC (DF-I + ord. +
aug.) can further reduce the signal range such that the
minimum number of integer bits is reduced to ve. In this
example, the range reduction technique saves four bits in
word-length, which can signicantly reduce the hardware
complexity.
6. Conclusions
In this paper we discussed various considerations when
designing accurate isotropic and nonisotropic fading chan-
nels. We investigated the lter-based approach, which can
be customized to accurately provide the statistical prop-
erties required for simulating dierent fading scenarios.
We proposed an iterative technique that can be used to
design both complex and real IIR lters with xed-point
coecients. We augmented the least-squares cost function
with a specic barrier function to control the location of the
poles (and potentially also the zeros) within the unit circle to
ensure numerical stability. We also suggested two techniques
for reducing the signal range. Signal range reduction is
particularly important in hardware implementation since
it can directly reduce hardware complexity. Simulation
results showed that the proposed lter design technique can
signicantly reduce the required word-length for xed-point
IIR lter implementations.
References
[1] G. L. Stuber, Principles of Mobile Communication, Kluwer
Academic Publishers, New York, NY, USA, 2001.
[2] P. Bello, Characterization of randomly time-variant linear
channels, IEEE Transactions on Communications, vol. 11, no.
4, pp. 360393, 1963.
[3] R. H. Clarke, A statistical theory of mobile-radio reception,
Bell System Technical Journal, vol. 47, pp. 9571000, 1968.
[4] W. C. Jakes, Microwave Mobile Communications, Wiley-IEEE
Press, Piscataway, NJ, USA, 1994.
[5] Y. R. Zheng and C. Xiao, Simulation models with correct
statistical properties for Rayleigh fading channels, IEEE
Transactions on Communications, vol. 51, no. 6, pp. 920928,
2003.
[6] A. Alimohammad, S. Fouladi Fard, B. F. Cockburn, and C.
Schlegel, An improved SOS-based fading channel emulator,
in Proceedings of the IEEE 66th Vehicular Technology Conference
(VTC07), pp. 931935, Baltimore, Md, USA, October 2007.
[7] J. I. Smith, Computer generated multipath fading simulation
for mobile radio, IEEE Transactions on Vehicular Technology,
vol. 24, no. 3, pp. 3940, 1975.
[8] R. H. Clarke and W. L. Khoo, 3-D mobile radio channel
statistics, IEEE Transactions on Vehicular Technology, vol. 46,
no. 3, pp. 798799, 1997.
[9] D. J. Young and N. C. Beaulieu, The generation of correlated
Rayleigh random variates by inverse discrete fourier trans-
form, IEEE Transactions on Communications, vol. 48, no. 7,
pp. 11141127, 2000.
[10] K. W. Yip and T. S. Ng, Ecient simulation of digital
transmission over WSSUS channels, IEEE Transactions on
[11] R. A. Goubran, H. M. Hafez, and A. U. H. Sheikh, Implemen-
tation of a real-time mobile channel simulator using a DSP
chip, IEEE Transactions on Instrumentation and Measurement,
vol. 40, no. 4, pp. 709714, 1991.
[12] C. Komninakis, A Fast and Accurate Rayleigh Fading Sim-
ulator, in Proceedings of the IEEE Global Telecommunica-
tions Conference (GLOBECOM03), pp. 33063310, December
2003.
[13] A. Alimohammad, S. Fouladi Fard, B. F. Cockburn, and C.
Schlegel, A compact single-FPGA fading-channel simulator,
IEEE Transactions on Circuits and Systems II, vol. 55, no. 1, pp.
8488, 2008.
[14] L. B. Jackson, Digital Filters and Signal Processing, Kluwer
Academic Publishers, 1989.
[15] K. E. Baddour and N. C. Beaulieu, Autoregressive modeling
for fading channel simulation, IEEE Transactions on Wireless
[16] A. Antoniou, Digital Filters, Analysis, Design, and Applications,
McGraw-Hill, 1993.
[17] J. Carletta, R. Veillette, F. Krach, and Z. Fang, Determining
appropriate precisions for signals in xed-point IIR lters, in
Proceedings of the 40th Design Automation Conference, pp. 656
661, June 2003.
[18] Filter design toolbox for use with Matlab, Users Guide, The
Mathworks, 2005.
[19] M. P atzold, Y. Li, and F. Laue, A study of a land mobile
satellite channel model with asymmetrical Doppler power
spectrum and lognormally distributed line-of-sight compo-
nent, IEEE Transactions on Vehicular Technology, vol. 47, no.
1, pp. 297310, 1998.
[20] P. Petrus, J. H. Reed, and T. S. Rappaport, Eects of
directional antennas at the base station on the Doppler
spectrum, IEEE Communications Letters, vol. 1, no. 2, pp. 40
42, 1997.
[21] J. Salz and J. H. Winters, Eect of fading correlation on
adaptive arrays in digital mobile radio, IEEE Transactions on
[22] K. Anim-Appiah, Complex envelope correlations for non-
isotropic scattering, Electronics Letters, vol. 34, no. 9, pp. 918
919, 1998.
[23] Q. Spencer, M. Rice, B. Jes, and M. Jensen, Statistical
model for angle of arrival in indoor multipath propagation,
in Proceedings of the 1997 47th IEEE Vehicular Technology
Conference, pp. 14151419, Phoenix, Ariz, USA, May 1997.
[24] M. D. Austin and G. L. Stuber, Velocity adaptive hando
algorithms for microcellular systems, IEEE Transactions on
[25] A. Abdi, J. A. Barger, and M. Kaveh, A parametric model
for the distribution of the angle of arrival and the associated
correlation function and power spectrum at the mobile
station, IEEE Transactions on Vehicular Technology, vol. 51,
no. 3, pp. 425434, 2002.
[26] C. A. Guti errez and M. P atzold, The design of sum-of-
cisoids Rayleigh fading channel simulators assuming non-
isotropic scattering conditions, IEEE Transactions on Wireless
[27] C. A. Guti errez, M. Ptzold, A. Sandoval, and C. Delgado-Mata,
An ergodic sum-of-cisoids simulator for multiple uncorre-
lated Rayleigh fading channels under generalized scattering
conditions, IEEE Transactionson Wireless Communications,
vol. 61, no. 5, pp. 23752382, 2012.
[28] K. V. Mardia, Statistics of Directional Data, Academic Press,
New York, NY, USA, 1972.
[29] A. V. Oppenheim, R. W. Schafer, and J. R. Buck, Discrete Time
Signal Processing, Prentice Hall, 1999.
[30] V. Erceg, L. Schumacher, P. Kyritsi et al., TGn channel
models, Tech. Rep. IEEE 802.11-03/940r4, Garden Grove,
Calif, USA, 2004.
[31] H. G. Martinez, Design of recursive digital lters with
optimum magnitude and attenuation poles on the unit circle,
IEEE Transactions on Acoustics, Speech, and Signal Processing,
vol. 26, no. 2, pp. 150156, 1978.
[32] X. Chen and T. W. Parks, Design of IIR lters in the complex
domain, IEEE Transactions on Acoustics, Speech, and Signal
Processing, vol. 38, no. 6, pp. 910920, 1990.
[33] W. S. Lu, Design of stable minimax IIR digital lters
using semidenite programming, in Proceedings of the IEEE
Internaitonal Symposium on Circuits and Systems, pp. 355358,
Geneva, Switzerland, May 2000.
[34] L. B. Jackson, Frequency-domain Steiglitz-McBride method
for least-squares IIR lter design, ARMA modeling, and
periodogram smoothing, IEEE Signal Processing Letters, vol.
15, pp. 4952, 2008.
[35] A. G. Deczky, Synthesis of recursive digital lters using the
minimum p-error criterion, IEEE Transactions on Audio and
Electroacoustics, vol. 20, no. 4, pp. 257263, 1972.
[36] A. Tarczy nski and G. D. Cain, New algorithm for designing
near-optimal Chebyshev IIR and FIR lters, in Proceedings of
the 38th IEEE Midwest Symposium on Circuits and Systems, pp.
584587, Rio de Janeiro, Brazil, August 1995.
[37] A. Tarczy nski, G. D. Cain, E. Hermanowicz, and M. Rojewski,
A WISE method for designing IIR lters, IEEE Transactions
on Signal Processing, vol. 49, no. 7, pp. 14211432, 2001.
[38] T. W. Fox, Fixed-point IIR digital lters with simultane-
ous hardware usage, group delay and magnitude frequency
response specications, in Proceedings of the Canadian Con-
ference on Electrical and Computer Engineering, pp. 17041707,
Saskatoon, Canada, May 2005.
[39] L. B. Jackson, A narrowband IIR digital lter with low
sensitivity and roundo noise, IEEE Signal Processing Letters,
vol. 10, no. 6, pp. 164167, 2003.
[40] G. Vanuytsel, P. Boets, L. Van Biesen, and S. Temmerman,
Ecient hybrid optimization of xed-point cascaded IIR
lter coecients, in Proceedings of the 19th IEEE Instrumen-
tation and Measurement Technology Conference, pp. 793797,
May 2002.
[41] A. Krukowski and I. Kale, Two approaches for xed-point
lter design: bit-ipping algorithmand constrained downhill
simplex method, in Proceedings of the Signal Processing and Its
Applications Symposium, vol. 2, pp. 965968, 1999.
[42] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, Optimization
by simulated annealing, Science, vol. 220, no. 4598, pp. 671
680, 1983.
[43] A. Alimohammad, S. Fouladi Fard, and B. F. Cockburn,
Accurate simulation of nonisotropic fading channels with
arbitrary temporal correlation, IET Communications, vol. 6,
no. 5, pp. 557564, 2012.
[44] K. Steiglitz, Computer-aided design of recursive digital
lters, IEEE Trans Audio Electroacoust, vol. AU-18, no. 2, pp.
123129, 1970.
[45] R. G. Bland, D. Goldfarb, and M. J. Todd, The ellipsoid
method: a survey, Operations Research, vol. 29, no. 6, pp.
10391091, 1981.
[46] A. K. Aboagye, Overow avoidance techniques in cascaded
IIR lter implementations on the TMS320 DSPs, Texas
Instruments Application Report, 1999.
[47] L. B. Jackson, Digital Filters and Signal Processing: with
MATLAB Exercises, Kluwer Academic Publishers, 1996.
doi:10.1155/2012/702917
Research Article
Wideband and Ultrawideband Channel Models in
Working Machine Environment
Attaphongse Taparugssanagorn, Matti H am al ainen, and Jari Iinatti
Centre for Wireless Communications, University of Oulu, 90570 Oulu, Finland
Correspondence should be addressed to Attaphongse Taparugssanagorn, attaphongset@gmail.com
Received 23 April 2012; Revised 17 July 2012; Accepted 30 July 2012
Academic Editor: Matthias P atzold
Copyright 2012 Attaphongse Taparugssanagorn et al. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
We present statistical models for wideband and ultrawideband (UWB) radio channels in a working machine cabin environment.
Based on a set of measurements, it was found that such a small and conned space causes mostly diuse multipath scattering
rather than specular paths. The amplitude of the channel impulse responses in the wideband case is mostly Rayleigh distributed
small-scale fading signal, with only a few paths exhibiting Ricean distributions, whereas the ones in the UWB case tend to be
log-normally distributed. For the path amplitude, we suggest an exponential decay prole, which has a constant slope in dB
scale, with the corresponding parameters for the UWB case. For the wideband case, a twofold exponential decay prole provides
excellent ts to the measured data. It was also noted that the root-mean-square (RMS) delay spread is independent of the line-of-
sight/obstructed line-of-sight situations of the channel. The multipath components contributing signicant energy play a major
role in such a small environment if compared to the direct path. In addition, the radio channel gains are attenuated with the
presence of a driver inside the cabin.
1. Introduction
A wireless sensor network (WSN), which is a wireless net-
work consisting of spatially distributed autonomous devices
using sensors to cooperatively monitor physical or environ-
mental conditions [1], today covers very widely in many
industrial and civilian application areas, including industrial
process monitoring and control, health monitoring, environ-
ment and habitat monitoring, home automation, and trac
control. Wireless intra-vehicle communications system has
been proposed as a new type of communications system
that can provide onboard passengers with bidirectional high-
speed data exchange services for both communications and
entertainment [2, 3]. For the purpose of safety, comfort and
convenience, WSNs are being deployed in the new models
of automotives to collect such information as temperature,
speed, pressure, and so on. This idea of exploiting WSNs is
not limited to applications in a working machine only. The
information and communication system used in a working
machine makes it easier for driver to control the operation,
for example, in a wood processing guiding the trunk
cutting according to orders input from the system. This can
guarantee quick processing during logging, and high produc-
tivity. This information system is expected to be wirelessly
controlled by the driver. Using wireless communication links
instead of wired link can save a big amount of installing costs.
In addition, measuring vibrations acting on the human body,
so-called human vibration, is the prevention of health risks
and the evaluation of comfort, for example, in vehicles [4].
Wireless vibration sensing application embedded to wireless
body area networks (WBANs) can also be included in the
future system.
The design of information and communication systems
requires a good understanding of the corresponding radio
wave propagation channel. The characteristics of an indoor
radio propagation is found to be its site-specic behavior.
Many existing radio channel measurement campaigns and
the corresponding channel models have been restricted to
typical residential and oce environments, which are highly
limited by the attenuation of walls [5, 6]. Only few radio
channel measurements were done inside a vehicle [7, 8]. To
our knowledge, there is no work done in a working machine
environment, which can have the dierent radio channel
characteristics due to its small and conned space.
This paper discusses on experimental radio channel
measurements that are carried out in a working machine
environment within two dierent frequency bands, that
is, the industrial, scientic, and medical (ISM) radio
frequency band (2.42.4835 GHz) and the ultrawideband
(UWB) frequency band (3.1 and 10 GHz). The statistical
radio channel models based on the measurement results are
then developed. The novelty of this paper comes from the
new environment whose radio channel characteristics are
modeled.
Section 2 of the paper is devoted to the description
of the measurement setup including the scenarios. Sec-
tion 3 addresses the characterization of the measured radio
channels. The average channel impulse responses for each
scenario are analyzed and discussed. The parameters of the
statistical channel model are also presented in this section.
Section 4 deals with the implementation of the channel
model and its evaluation. The conclusions of this work are
given in Section 5.
2. Measurement Campaigns
2.1. Measurement Setup. The measurements were carried out
in a frequency domain using Agilent 8720ES S-parameter
vector network analyzer (VNA) [9]. The network analyzer
is operated in a transfer function measurement mode, where
port 1 and port 2 are the transmitting and the receiving ports,
respectively. Thus, the measurements result as a frequency
response of the channel.
The measurements were carried out in the frequency
band between 2.42.4835 GHz, which is in the ISM band,
and is used, that is, by the standards IEEE 802.11 and
802.15.4 dedicated for wireless local area network (WLAN)
communication. The measurements in an UWB frequency
band between 3.110 GHz were also conducted. Therefore,
the measured bandwidths B were 83.5 MHz and 6.9 GHz,
respectively. Consequently, the corresponding delay resolu-
tions are 11.97 ns and 0.14 ns. The maximum number of
frequency points per sweep M is 1601, which can then be
used to calculate the maximum detectable delay
max
of the
channel as
max
=
(M 1)
B
. (1)
Using (1), the maximum detectable delay,
max
of the chan-
nels for two dierent bandwidths are 19.16 s and 231.88 ns,
which correspond to 5.75 km and 69.56 m, respectively, in
a free-space distance. Maximum detectable delay is just a
measurement system-based parameter. It is introduced to
show that the measurement system is able to detect all
propagation paths having signicant contribution to the
received power. However, the usable signal energy is in the
shorter delays than the maximum detectable delay as can be
seen later on.
The antennas for the measurements in WLAN band 2.4
2.4835 GHz were HGA7S 7 dBi high gain antennas [10]. The
SkyCross SMT-3TO10 M-A antennas [11] were used for the
measurements in UWB band. Both types of the antennas are
linearly polarized. The SkyCross SMT-3TO10 M-A antennas
Table 1: Measurement system parameters.
Parameters Value
Frequency band
ISM (2.42.4835 GHz),
UWB (3.110 GHz)
Bandwidth
83.5 MHz,
6.9 GHz
Number of sample points within the
band
1601
Maximum detectable delay
19.16 s (5.75 km),
231.88 ns (69.56 m)
Sweep time 800 ms
Average noise oor 108 dBm
Transmit power 5 dBm
Tx and Rx cables loss 7.96 dB
1.49 m
0.5 m
0.54 m
0.20 m
0.20 m
1.7 m
1.1 m
1.32 m
0.12 m 0.6 m 0.48 m
1.08 m
Rx2
Rx4
Rx3
Rx8
Rx5
Rx7
Rx6
Rx9
Rx16
Rx11
Rx13
Rx12
Rx15
Rx17
Rx14
Rx10
Tx
Mirror
Rx1
Figure 1: Arrangement of the antennas positions for measure-
ments inside a working machine cabin.
are azimuthally omnidirectional, whereas the HGA7S 7 dBi
high gain antennas are designed for 360 degrees of coverage
[10]. The sweep time of the VNA depends on the frequency
points within the sweep band, being automatically adjusted
by the network analyzer, that is, radio channels were
measured in a frequency domain over the frequency band of
interest. This corresponds to a conventional S
21
parameter
(S
21
parameter refers to the signal exiting at Port 2 for
the signal incident at Port 1 of VNA) measurement setup,
where the radio channel is the device under test (DUT). The
transmit power used in this study was 5 dBm (equivalent to
3.16 mW) as the VNA output port. Table 1 summarizes the
measurement system parameters.
2.2. Channel Measurement Description. The measurements
were conducted inside a cabin of a working machine with
the dimensions as illustrated in Figure 1. The transmit
(Tx) antenna and the receive (Rx) antenna are, in general,
set to face each other, and the line-of-sight (LOS) path
always exists except when the links are blocked by the seat.
The arrangement of the antennas positions is shown in
Figure 1. The position of the Tx antenna is xed at the
seat on 1.08 m above the cabin oor. The positions of the
Rx antenna were moved to 17 dierent spots, namely, from
the position Rx1 till the position Rx17. At each position,
100 consecutive measurements of the channel were taken to
improve statistical reliability. This number of typically used
in static cases. In order to study radio channels inside a cabin
in a working machine for real situations, the following three
scenarios are investigated:
(i) scenario 1: without a driver, when the engine is o.
(ii) scenario 2: with a driver sitting, when the engine is
o.
(iii) scenario 3: with a driver sitting, when the engine is
on.
3. Channel Measurement Results and
Data Analysis
Measured S
21
-parameters, that is, channel transfer functions,
are converted to the time domain, that is, to channel impulse
responses h(), using an inverse fast Fourier transform
(IFFT). A Hamming window is used to reduce sidelobes. The
channel impulse response is generally given by a tapped delay
line model [12] as
h() =
L1
_
l=0
a
l
(
l
) exp
_
j
l
_
, (2)
where a
l
is the amplitude of the lth path,
l
is the arrival time
of the lth path,
l
is the phase for the lth path, and L is the
number of paths. (t) stands for the Dirac function, which is
the impulse symbol and is sometimes called as Diracs delta
function. The phase
l
is modeled by a uniform distribution
U{0, 2}. The statistical parameters dening the channel
model are presented in the following paragraphs.
3.1. Average Channel Impulse Response. One hundred indi-
vidual realizations of the channel impulse response were
averaged for each position. The reason for using one hundred
realizations is since this is typically used to guarantee
statistical reliability. This number of sweeps is recorded for
all the individual links measured, so the total number of
sweeps carried out during the measurement campaign is
much higher. When using frequency domain measurement
technique, the environment needs to be static, which means
that one hundred consecutive measurements per link is
enough.
To proceed further analysis, all measured channel
impulse responses are rst truncated above the noise thresh-
old, set to the noise level adding 10 dB (about 98 dBm). The
only paths above the noise threshold are taken into account.
Some examples are presented here in order to explain the
analysis. The large-scale features (i.e., path loss) and also
small-scale features (due to the change in pulse shape) of
the channels are taken into account. The large-scale models
are necessary for network planning and link budget design
while small-scale models are necessary for ecient receiver
design and performance analysis. This kind of models is not
0 20 40 60 80 100 120
Delay (ns)
A
v
e
r
a
g
e

c
h
a
n
n
e
l

i
m
p
u
l
s
e

r
e
s
p
o
n
s
e

(
d
B
m
)
Rx1
Rx8
Reection from the building
110
100
90
80
70
60
50
Figure 2: Average channel impulse responses for the UWB band in
scenario 1: without a driver, when the engine is o. The positions of
Rx antenna are Rx1 and Rx8.
publicly available at open literature for working machines
cabin environment so far.
As preliminarily shown in [13], the measured channel
impulse responses for the ISM band have two constant slopes
in dB scale, which correspond to a typical exponential decay
prole in a linear scale. This shows that there are many
reections rather than a signicant specular path. This is
excellently suited to characterize diuse scattering [14].
The average channel impulse responses for the UWB
band (6.9 GHz bandwidth) in scenario 1, without a driver,
when the engine is o shown in Figure 2 were also
represented in [13]. The UWB measurements allow us to
check the presence of many paths arriving at the receiver
within the delay resolution and the maximum delay. Firstly,
the channel impulse responses have a constant slope in dB,
which corresponds to a typical exponential decay prole in a
linear scale. The other advantage of the UWB measurements
is the high temporal resolution, which makes it possible to
see the rst path more precisely. The rst path arrival time
can approximately translate into the distance between Tx
and Rx antennas. Therefore, the distances between Tx and
Rx1, Rx3, Rx8 as well as Rx17 are 1.17 m, 1.35 m, 1.05 m,
and 0.81 m, respectively (3.9 ns, 4.5 ns, 3.5 ns, and 2.7 ns,
resp.). The Rx1 and Rx3 are in the obstructed line-of-sight
(OLOS) case; thus the rst path is not the strongest path.
Depending on the position, the constructive multipaths can
create a stronger path to the receiver. The Rx8 and Rx17 are
LOS cases; therefore, the rst path is also the strongest path.
Figure 3 illustrates the average measured channel impulse
responses for the ISM band, Rx5 (OLOS), in three dierent
scenarios. When the driver is inside the cabin, the radio waves
in all multipaths from the Tx close to the body of the driver
(3 cm) are attenuated. In addition, the rst path arrival
time is longer than the one in the case of no driver inside
the cabin.
0 20 40 60 80 100
Delay (ns)
Without a driver, engine off
With a driver, engine off
With a driver, engine on
A
v
e
r
a
g
e

c
h
a
n
n
e
l

i
m
p
u
l
s
e

r
e
s
p
o
n
s
e

(
d
B
m
)
110
100
90
80
70
60
50
Figure 3: Average measured channel impulse responses for the ISM
band, Rx5 (OLOS), in three dierent scenarios.
Note that in the following, the analysis has been carried
out against several distributions, and the best ts have
been selected. Moreover, all the parameters used in the
models were extracted from the measured data during the
postprocessing. From the extracted parameters, we develop
measurements-based deterministic channel models for both
ISM band and UWB band cases.
3.1.1. Exponential Decaying Factor (a
l
). As explained above,
the path amplitude a
l
in (2) is approximately modeled by
an exponential decay prole with a Ricean factor
0
and an
exponential decaying factor for UWB case as
10 log
10
|a
l
|=
_
_
0, l = 0
0
+10log
10
_
exp
_
t
l
__
, 1l L1.
(3)
For ISM case, the channel impulse responses have two
constant slopes in dB, and thus the path amplitude a
l
in (2) is
approximately modeled by twofold exponential decay prole
with two Ricean factors
01
and
02
as well as two exponential
decaying factors
1
and
2
as
10 log
10
|a
l
|=
_
_
0, l =0
01
+10 log
10
_
exp
_
t
l
1
__
, 1l l
1
02
+10 log
10
_
exp
_
t
l
2
__
, l
1
+1l L1,
(4)
where l
1
is the last path of the rst part.
An example from each frequency band is presented in
Figure 4 showing the average channel impulse response for
the ISM band and for Rx17 (LOS) in the rst scenario
0 5 10 15 20 25 30 35
Delay (ns)
Measured
LS tting
110
90
80
70
60
50
40
A
v
e
r
a
g
e

t
r
u
n
c
a
t
e
d

c
h
a
n
n
e
l

i
m
p
u
l
s
e

r
e
s
p
o
n
s
e

(
d
B
m
)
Figure 4: Average channel impulse responses for the ISM band for
Rx17 (LOS) and the least square linear tting curve in dB scale.
0 10 20 30 40 50 60
Measured
LS tting
Delay (ns)
A
v
e
r
a
g
e

t
r
u
n
c
a
t
e
d

c
h
a
n
n
e
l

i
m
p
u
l
s
e

r
e
s
p
o
n
s
e

(
d
B
m
)
100
90
95
85
80
75
70
65
60
55
Figure 5: Average channel impulse responses for the UWB band for
Rx17 (LOS) and the least square linear tting curve in dB scale.
(without a driver, the engine is o) and the twofold least
square (LS) linear tting curve in dB scale. The correspond-
ing factors
01
and
02
are 44 dB and 81 dB, respectively,
as well as the exponential decaying factor
1
and
2
are 2.33
and 24.39, respectively. Figure 5 shows the average channel
impulse response for the UWB band and for Rx17 (LOS) and
the LS linear tting in dB. The corresponding factor
0
and
the exponential decaying factor are 66 dB and 21.74.
3.1.2. Amplitude Variation (). This part represents the
amplitude variation over the mean value of the path
amplitude. The cumulative distribution functions (CDFs) in
Figures 6 and 7 depict that the amplitude variations of the
0
0.2
0.4
0.6
0.8
1
Empirical CDF for amplitude variation
0
0.2
0.4
0.6
0.8
1
Part 1
Part 2
P
r
o
b
a
b
i
l
i
t
y
>
a
b
s
c
i
s
s
a
P
r
o
b
a
b
i
l
i
t
y
>
a
b
s
c
i
s
s
a
0 5 10 10 5
0 5 10 10 5
(dB)
(dB)
Measured
Log-normal tting
Figure 6: CDFs of the amplitude variation over the mean value of
the path amplitude and the log-normal tting curve for the ISM
band and for Rx17 (LOS).
0 5 10 15
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Empirical CDF for amplitude variation
Measured
Log-normal tting
(dB)
20 15 10 5
P
r
o
b
a
b
i
l
i
t
y
>
a
b
s
c
i
s
s
a
Figure 7: CDFs of the amplitude variation over the mean value of
the path amplitude and the log-normal tting curve for the UWB
band and for Rx17 (LOS).
average channel impulse responses are well tted by the log-
normal distribution with zero mean and standard deviation
of 6.69 dB and 4.50 dB, respectively.

3.1.3. Number of Paths (L). The CDFs of the number of paths
L (above the noise threshold) are excellently tted by Poisson
distribution with a certain mean
L
. An example for Rx14
(LOS) in the UWB band shown in Figure 8, respectively. The
number of paths decreases with the presence of the driver.
In the case of the ISM band, the CDFs of the number of
200 250 300 350 400 450 500
Empirical CDF
Measured
Poisson tting
0.1
0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Without a driver
engine o
= 391
With a driver
engine o
= 334
With a driver
engine on
= 305
Number of resolvable components, L
P
r
o
b
a
b
i
l
i
t
y
L
>
a
b
s
c
i
s
s
a
Figure 8: CDFs of the number of paths L for Rx14 (LOS) in the
UWB band measurements.
5 10 15 20 25 30 35 40 45 50 55
Empirical CDF
Gaussian tting
Number of resolvable components, L
With a driver
engine o
= 13
With a driver
engine on
= 27
0.1
0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Without a driver
engine o
= 43
= 1.93
= 1.6
= 2
Measured
P
r
o
b
a
b
i
l
i
t
y
L
<
a
b
s
c
i
s
s
a
Figure 9: CDFs of the number of paths L for Rx14 (LOS) in the ISM
band measurements.
paths L are tted by Gaussian distribution with certain mean
L
and standard deviation
L
as an example in Figure 9 for
Rx14 (LOS). The number of paths presented in the paper is
for channel modeling purpose. Nonetheless, they are much
more than the typical number of paths in system design, for
example, the number of paths within 10 dB of the strongest
path and the number of paths containing 85 percent of the
energy, which give approximately 3550 paths in the UWB
band case. As can be seen, the impact of the vibration (of the
engine) on the number of the paths is clear, especially, in the
case of OLOS unlike the other channel characteristics.
Rx9
Rx12 Rx14 Rx8
Rx11 Rx13
Rx16 Rx15 Rx17
1 2 3 4 5 6
RMS delay spread (ns)
P
r
o
b
a
b
i
l
i
t
y

R
M
S

d
e
l
a
y

s
p
r
e
a
d
>
a
b
s
c
i
s
s
a
0.1
0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Empirical CDF
Figure 10: CDFs of the measured RMS delay spreads for LOS in the
UWB band measurements.
3.2. Small-Scale Channel Characterization
3.2.1. Delay Dispersion. In order to compare dierent mul-
tipath channels and to develop general design guidelines,
parameters which grossly quantify the multipath channel
are used. The mean excess delay
m
and root-mean-square
(RMS) delay spread
RMS
are commonly used to imply the
time delay dispersive properties of a wideband multipath
channel parameters that can be determined from a power
delay prole. The received signal will suer spreading in
time if compared to the transmitted signal. This dispersion
can also imply frequency selective fading and intersymbol
interference (ISI). No serious ISI is likely to occur if the
symbol duration is longer than, say, ten times the RMS delay
spread [14]. The RMS delay spread is the standard deviation
(or root-mean-square) value of the delay of reections,
weighted proportional to the energy in the reected waves
dened as [14]
RMS
=
L
i=1
(
i

m
)
2
|h(
i
)|
2
L
i=1
|h(
i
)|
2
,
(5)
where
m
is expressed as
m
=
L
i=1

i
.|h(
i
)|
2
L
i=1
|h(
i
)|
2
.
(6)
Figures 10 and 11 depict the CDFs of the measured RMS
delay spreads for LOS and OLOS situations, respectively, for
the UWB band measurements. All values are in the range
of 1.75.8 ns, which are very small if compared to typical
indoor RMS delay spread varying between 14 and 18 ns [14].
We can see that the RMS delay spread is independent of the
LOS/OLOS situations of the channel. Neither of the distance,
this is because the cabin space is very small. The multipath
components contributing signicant energy play a major
role in such a small environment if compared to the direct
path. This is very important on the selection of the antenna
position and also to the receiver structure.
2 2.5 3 3.5 4 4.5 5 5.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Empirical CDF
Rx6 Rx10 Rx1 Rx3 Rx4 Rx2 Rx7 Rx5
P
r
o
b
a
b
i
l
i
t
y

R
M
S

d
e
l
a
y

s
p
r
e
a
d
>
a
b
s
c
i
s
s
a
Figure 11: CDF of the measured RMS delay spreads for OLOS in
the UWB band measurements.
0 0.5 1 1.5 2 2.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Empirical CDF
P
r
o
b
a
b
i
l
i
t
y

R
M
S

d
e
l
a
y

s
p
r
e
a
d
>
a
b
s
c
i
s
s
a
= 0.97 ns, = 0.36 ns
= 0.55 ns, = 0.08 ns
= 1.72 ns, = 0.01 ns
Figure 12: CDFs of the measured RMS delay spreads for Rx12
(LOS) in the UWB band measurements in all three scenarios.
An example for the case of the cabin with a driver, when
the Rx antenna is at Rx12 (LOS), is shown in Figure 12.
The RMS delay spread decreases by approximately 1.2 ns,
since the presence of people in the cabin attenuates the
path gains. Corresponding CDFs for OLOS case (Rx4) are
shown in Figure 13. The standard deviations of the RMS
delay spreads, when the engine is on (cabin vibration), are
larger than the ones, when the engine is o. This shows that
the radio channels dier considerably for the case with and
without driver.
3.2.2. Amplitude Distribution. In the ISM band, the power
received for each delay tap is the vectorial sum of many
multipath components arriving in the corresponding delay
0.5 1 1.5 2 2.5 3 3.5 4 4.5
Empirical CDF
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P
r
o
b
a
b
i
l
i
t
y

R
M
S

d
e
l
a
y

s
p
r
e
a
d
>
a
b
s
c
i
s
s
a
= 2.65 ns, = 0.6 ns
= 4.08 ns, = 0.01 ns
= 2.62 ns, = 0.43 ns
Figure 13: CDFs of the measured RMS delay spreads for Rx4
(OLOS) in the UWB band measurements in all three scenarios.
bin. The central limit theorem is therefore valid, and the
amplitudes of the delay bins exhibit a Rayleigh (in nonLOS
or OLOS case) or a Ricean (in LOS case) distribution [15].
On the contrary, only fewmultipath components (MPCs) fall
into each resolvable delay in the UWB band. Therefore, the
central limit theorem is not applicable and the amplitudes of
the delay bins consequently do not exhibit a Rayleigh or Rice
distributions.
In the ISM band case, in all scenarios, the rst delay bins
tend to be Ricean distribution due to their high K factors
[16], whereas the other delay bins can be assumed to be
Rayleigh distributed due to their very small K factors.
For LOS case in the UWB band, as explained above, only
few MPCs fall into each resolvable delay and the central
limit theorem is therefore not applicable. The measured
amplitudes are rather log-normally distributed [17].
3.3. Large-Scale Channel Characterization. Path loss (PL) is
an important large-scale characteristic for link budget calcu-
lation and system design. In the conventional narrowband
channel, Friis transmission formula is used to model the
received signal power, P
r
in free space given by [18]
P
r
(d) =
P
t
G
t
G
r
2
(4d)
2
, (7)
where P
t
is the transmitted power, and G
t
and G
r
are
the Tx and the Rx antenna gains, respectively. is the
wavelength, and d is the Tx-Rx separation distance. The
Friis transmission formula shows that the received signal
power falls o as the square of the Tx-Rx separation
distance. From (7), the term PL is given by (/4d)
2
which
predicts that signal power will decrease with the square of
increasing frequency and distance. These show the presence
of both distance and frequency dependency in path loss.
Table 2: Average path loss (PL) within the measured band.
Measurement scenarios and cases Mean path loss (dB)
UWB band
Scenario 1, LOS 73.31 dB
Scenario 1, OLOS 77.32 dB
ISM band
Nevertheless, the frequency is constant within a bandwidth
of interest and is thus negligible. However, for UWB systems,
the frequency dependency in PL can be signicant due to
the large bandwidth (>500 MHz). The channel would distort
the signal spectrum of the signal which is directly related
to the signal distortions, that is, pulse shape distortion. PL
can be obtained directly from the measured channel transfer
function H( f ). The distance dependency of the path loss is
calculated by time snapshot and frequency averaging over the
total bandwidth at each Rx position as
PL(d) =
1
MN
N
_
i=1
M
_
j=1
H
_
f
i
, t
j
; d
_
2
, (8)
where H( f
i
, t
j
; d) denotes the jth channel transfer function
time snapshot at frequency f
i
in a distance d. N and M
denote the number of sample points within the band and the
number of snapshots, respectively.
The mean PLs for both LOS and OLOS cases are
calculated from (8) and averaged over the results from all
Rx positions in each case. The results for each case and each
scenario are summarized in Table 2. As expected, the PL
increases in the OLOS case and with the presence of a driver.
Moreover, it is noted that the PL in the case when the engine
is on is not very dierent than in the case when the engine is
o. This follows the initial assumption.
4. Channel Modeling and Evaluation
In this section, we describe an experimental based deter-
ministic channel model which are developed for both ISM
band and UWB band cases. As shown in Figures 14 and
15 in Section 3.2.2, the amplitude of the rst ve delay bin
in the ISM band case for the LOS case tends to be Ricean
distributed. The remaining delay bins can be assumed to
be Rayleigh distributed due to their small K-factors. For
the OLOS case, only the rst delay bin tends to be Ricean
distributed, whereas the rest can be assumed to be Rayleigh
distributed. In the UWB case, the amplitudes of all delay
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Signal amplitude (dBmV)
Empirical CDF
Measured
Ricean tting
Delay bin 1, Delay bin 5, Delay
bin 15,
Delay bin 40,
Delay
bin 45,
P
r
o
b
a
b
i
l
i
t
y

s
i
g
n
a
l

a
m
p
l
i
t
u
d
e
>
a
b
s
c
i
s
s
a
K = 3.1
K = 3.9
K = 9.8
K = 25.1
K = 11.7
110 100 90 80 70 60 50 40 30
Figure 14: CDFs of the amplitudes of ve delay bins, 1, 5, 15, 40
and 45 with corresponding K factors for LOS case at Rx12.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Signal amplitude (dBmV)
P
r
o
b
a
b
i
l
i
t
y

s
i
g
n
a
l

a
m
p
l
i
t
u
d
e

>

a
b
s
c
i
s
s
a
Empirical CDF
Measured
Log-normal tting
Delay bin 1
Delay bin 7
Delay bin 50
Delay bin 160
Delay bin 200
Delay bin 300
110 100 80 90 70 60 50
Figure 15: CDFs of the amplitudes of four delay bins, 1, 5, 15, and
40 with log-normal tting for OLOS case at Rx1, without a driver
and the engine o.
bins tend to be log normally distributed. To achieve that
a random vector x, where each component is the value of
each delay bin with the corresponding distribution, either
Ricean or Rayleigh or log normal up to the cases, with zero
(dB) mean is generated. Subsequently, the amplitude vector
of the channel |h| is obtained by scaling the random vector
with an exponential decay prole with the corresponding
Ricean factor
0
and exponential decaying factor described
in Section 3.1.1. Finally, the phase vector of the channel is
assumed to be uniformly distributed U{0, 2}.
As an example, the channels for the case of Rx17 (LOS)
in the ISM band and the case of Rx17 (LOS) in the UWB
Table 3: Channel parameters for simulation for the case of Rx 17
(LOS).
Parameters ISM band UWB band
L
(paths) 53 441
01
(dB) 44
02
(dB) 81
0
(dB) 66
1
2.33
2
24.39
21.74
Amplitude
distribution
Ricean with K-factor =
21.35, 16.70, 16.28, 14.40,
and 13.89 dB for the delay
bin 15, Rayleigh for the
remaining delay bins
log normal
for all delay bins
(dB) 2.97 and 3.04 4.50

band (Figures 4 and 5) are simulated. For the former case, it
is done by using (3) and the simulation parameters presented
in Table 3. For the latter case, (4) is used with the simulation
parameters in Table 3.
The average channel impulse response and the RMS delay
spread
RMS
are used for validating the simulated model as
depicted in Figures 16 and 17. As we can see, the simulated
channels have a good agreement with the measured channels.
5. Conclusions
In order to deploy wireless information and communication
systems in such a conned environment like in a cabin
of a working machine, a good understanding of the radio
propagation channels, and a simple corresponding channel
model are required. To our knowledge, there is no work done
in a working machine environment, which have dierent
radio channel characteristics due to its small and conned
space. The measurement campaigns inside the working
machine cabin were taken for two dierent frequency ranges,
that is, in 2.42.4835 GHz ISM band used, for example,
by the standards IEEE 802.11 and 802.15.4 dedicated for
wireless local area network communication, as well as, in an
ultra wideband frequency band 3.110 GHz. Three dierent
scenarios for each frequency band, that is, (1) without a
driver, the engine is o, (2) with a driver, the engine is o,
and (3) with a driver, the engine is on, were carried out.
The characteristics of the propagation channels were
quite dierent from a traditional indoor environment;
namely, there are many reections rather than a signicant
specular path. This is excellently suited to characterize diuse
scattering. The UWB measurements have the high temporal
resolution advantage, which consequently makes us able
to see the rst path accurately. This is translated into the
distance between Tx and Rx antennas. With the presence
of the driver inside the cabin, the radio channel gains are
attenuated. In addition, the rst path arrival time comes later
than the one in the case of no driver inside the cabin. There
was no signicant dierence found between the cases when
0 10 20 30 40 50
Delay (ns)
110
100
90
80
70
60
50
40
A
v
e
r
a
g
e

t
r
u
n
c
a
t
e
d

c
h
a
n
n
e
l

i
m
p
u
l
s
e

r
e
s
p
o
n
s
e

(
d
B
m
)
Measured
Simulated
Figure 16: Average measured and simulated channel impulse
responses for Rx17 (LOS) in the ISM band.
Empirical CDF
Measured
Simulated
4.5 5 6 6.5 5.5 7
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P
r
o
b
a
b
i
l
i
t
y

R
M
S

d
e
l
a
y

s
p
r
e
a
d
>
a
b
s
c
i
s
s
a
Figure 17: CDF of the RMS delay spreads of the measured and
simulated channels for Rx17 (LOS) in the UWB band.
the engine is on and o, only the number of the paths are
dierent, especially, in the case of OLOS; that is, the vibration
(of the engine) has only the impact on the number of paths,
whereas the impact on other channel properties is trivial.
In both frequency ranges, the exponential decay prole
was modeled for the path with the corresponding parame-
ters. For the amplitude distribution, many MPCs arrive in
the corresponding delay bin in the ISM band, the central
limit theorem is therefore valid, and the amplitudes of the
delay bins exhibit a Rayleigh (in non-LOS or OLOS case)
or Ricean (in LOS case) distributions. On the contrary, only
few MPCs fall into each resolvable delay in the UWB band.
Therefore, the central limit theorem is not applicable and
the amplitudes of the delay bins consequently do not exhibit
a Rayleigh or Ricean distribution, but rather log-normal
distribution. For the delay dispersion, the RMS delay spread
was independent of the LOS/OLOS situations of the channel.
Neither of the distance, this is because the cabin space is
very small. The MPCs contributing signicant energy play
a major role in such a small environment if compared to the
direct path. This is very important on the selection of the
antenna position and the receiver design. For the path loss,
the frequency dependency of the path loss in the UWB band
was rstly observed. Secondly, the mean path loss increases in
the OLOS case and with the presence of the driver. Moreover,
it was noted that the mean path loss in the case, when the
engine is on, is not very dierent than in the case, when
the engine is o. The measurements obtained were used to
estimate the corresponding channel parameters in order to
build statistical simulated channel models.
The developed radio channel models give a great addi-
tional information to wireless communications designs in a
working machine cabin environment.
Acknowledgment
The authors would like to thank Mr. Ari Isola for his
contribution during the measurements. This work was
funded by the Finnish Funding Agency for Technology and
Innovation through the European Regional Development
Fund.
References
[1] K. R omer and F. Mattern, The design space of wireless sensor
networks, IEEE Wireless Communications, vol. 11, no. 6, pp.
5461, 2004.
[2] T. Elbatt, C. Saraydar, M. Ames, and T. Talty, Potential for
intra-vehicle wireless automotive sensor networks, in 2006
IEEE Sarno Symposium, usa, March 2006.
[3] G. Leen and D. Heernan, Vehicles without wires, Comput-
ing and Control Engineering Journal, vol. 12, no. 5, pp. 205
211, 2001.
[4] AN21E Application Note Human Vibration Measurement EC
Directive 2002/44/EC.
[5] D. Cassioli, M. Z. Win, and A. F. Molisch, The ultra-wide
bandwidth indoor channel: from statistical model to simula-
tions, IEEE Journal on Selected Areas in Communications, vol.
20, no. 6, pp. 12471257, 2002.
[6] S. S. Ghassemzadeh, R. Jana, C. W. Rice, W. Turin, and V.
Tarokh, Measurement and modeling of an ultra-wide band-
width indoor channel, IEEE Transactions on Communications,
vol. 52, no. 10, pp. 17861796, 2004.
[7] W. Niu, J. Li, and T. Talty, Intra-Vehicle UWB channel
measurements and statistical analysis, in 2008 IEEE Global
Telecommunications Conference, GLOBECOM 2008, pp. 3402
3406, usa, December 2008.
[8] W. Xiang, A vehicular ultra-wideband channel model for
future wireless intra-vehicle communications (IVC) Systems,
in 2007 IEEE 66th Vehicular Technology Conference, VTC 2007-
Fall, pp. 21592163, usa, October 2007.
[9] http://www.alliancetesteq.com/.
[10] http://tinyurl.com/cyszob/.
[11] http://skycross.com/Products/PDFs/SMT-3TO10M-A.pdf.
[12] I. Oppermann, M. H am al ainen, and J. Iinatti, UWB Theory
and Applications, John Wiley & Sons, New York, NY, USA,
2004.
[13] A. Taparugssanagorn, M. H am al ainen, and J. Iinatti, UWB
and wideband channel models for working machine environ-
ment, in Proceedings of the 73rd IEEE Vehicular Technology
Conference Spring (VTC12-Spring), Budapest, Hungary, 2011.
[14] A. Richter, Estimation of radio channel parameters: models
and algorithms [Ph.D. thesis], Technische Universitt Ilmenau,
Ilmenau, Germany, 2005.
[15] A. Muqaibel, A. Safaai-Jazi, A. Attiya, B. Woerner, and S. Riad,
Path-loss and time dispersion parameters for indoor UWB
propagation, IEEE Transactions on Wireless Communications,
vol. 5, no. 2, pp. 550558, 2006.
[16] M. Abramowitz and I. A. Stegun, Handbook of Mathematical
Functions, National Bureau of Standards, 1964, Dover Publica-
tions, New York, NY, USA, 1965.
[17] C. Tepedelenlio glu, A. Abdi, and G. B. Giannakis, The
Ricean K factor: estimation and performance analysis, IEEE
Transactions on Wireless Communications, vol. 2, no. 4, pp.
799810, 2003.
[18] E. Limpert, W. A. Stahel, and M. Abbt, Log-normal distribu-
tions across the sciences: keys and clues, BioScience, vol. 51,
no. 5, pp. 341352, 2001.

Modeling and Simulation of Mobile Radio Channels: Guest Editors: Matthias Pätzold, Neji Youssef, and Carlos A. Gutierrez

Uploaded by

Copyright:

Available Formats

Modeling and Simulation of Mobile Radio Channels: Guest Editors: Matthias Pätzold, Neji Youssef, and Carlos A. Gutierrez

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Modeling and Simulation of Mobile Radio Channels: Guest Editors: Matthias Pätzold, Neji Youssef, and Carlos A. Gutierrez

Uploaded by

Copyright:

Available Formats

Modelling and Simulation in Engineering

Guest Editors: Matthias Ptzold, Neji Youssef,

, t) represent the diuse and

, t) by only taking into account the single-bounce

, t) is assumed to be a deterministic pro-

, t), that is,

), which represents the

, t)}, which is equal to the STF-CCF

and to zero, that is,

, t+)} [36]. The TF-CCF can be obtained directly fromthe

, t +)} [36, Page 376]. The temporal ACF

to zero, that is, r

) of the transmission link from A

, t)} for all l = 1, 2, . . . , M

, t) are highly correlated over a large range of antenna

)| for dierent Rice factors c

)| (simulation model) for dierent values of the Rice

. This maximum elevation angle is

(t) = (t) + s. (3)

(t), which is denoted by

2/N, then the PDF

(r) describes how often the envelope

(t) crosses on average a specied signal level r within one

(r) describes how long on average

(z, z) of the envelope

(r) of the envelope

(t), which is given by

(r) of Rice processes

(r) is the Rice PDF presented in (6). The symbol

(r) of Rice processes

(x), which are dened by the

(x), called the least square approx-

(x), called the

of 6.69 dB and 4.50 dB, respectively.

(dB) 2.97 and 3.04 4.50

You might also like