A Three Dimensional Parametric Model for

Wideband MIMO Mobile-to-Mobile Channels

Alenka G. Zaji c and Gordon L. St uber
School of Electrical and Computer Engineering
Georgia Institute of Technology, Atlanta, GA 30332 USA
AbstractA three-dimensional (3-D) geometrical propagation
model for wideband multiple-input multiple-output (MIMO)
mobile-to-mobile (M-to-M) communications is proposed. Based
on the geometrical model, a 3-D parametric reference model
for wideband MIMO M-to-M multipath fading channels is
developed. From the reference model, the space-time-frequency
correlation function and the space-Doppler power spectral den-
sity are derived for a 3-D non-isotropic scattering environment.
Finally, some simulation results are presented and compared with
measured data. The close agreement between the theoretical and
empirical curves conrms the utility of the proposed wideband
model.
I. INTRODUCTION
Mobile-to-mobile (M-to-M) radio propagation channels
arise in inter-vehicular communications, mobile ad-hoc wire-
less networks, and relay-based cellular radio networks. The
statistical properties of M-to-M channels are quite differ-
ent from conventional xed-to-mobile (F-to-M) cellular land
mobile radio channels [1]. M-to-M communication systems
are equipped with low elevation antennas and have both the
transmitter (T
x
) and receiver (R
x
) in motion. Akki and Haber
[1] proposed a reference model for single-input single-output
(SISO) M-to-M Rayleigh fading channels. The reference mod-
els for narrowband multiple-input multiple-output (MIMO)
M-to-M channels have been proposed in [2], [3]. All these
models assume that the eld incident on the T
x
or R
x
antenna
array is composed of a number of waves travelling only in
the horizontal plane. This assumption is acceptable only for
certain environments, e.g., rural areas. However, it does not
seem appropriate for an urban environment where the T
x
and
R
x
antenna arrays are often located in close proximity to
and lower than surrounding buildings. Recently, we proposed
a three-dimensional (3-D) reference model for narrowband
MIMO M-to-M multipath fading channels [4].
This paper presents a 3-D parametric reference model for
wideband MIMO M-to-M channels. To describe our 3-D
reference model, we rst introduce a 3-D geometrical model
for wideband MIMO M-to-M channels, referred to as the
concentric-cylinders model. This model is extension of the
two-cylinder model for M-to-M channels proposed in [4].
0
Prepared through collaborative participation in the Collaborative Technology
Alliance for Communications & Networks sponsored by the U.S. Army
Research Laboratory under Cooperative Agreement DAAD19-01-2-0011. The
U.S. Government is authorized to reproduce and distribute reprints for
Government purposes notwithstanding any copyright notation thereon.
Previously reported models for narrowband MIMO M-to-M
channels [2] - [4] consider radio propagation in outdoor macro-
cells, assuming that all rays are only double-bounced. This
paper proposes a parametric reference model that employs the
concentric-cylinders model and constructs the input delay-
spread function as a superposition of the line-of-sight (LoS),
single-bounced, and double-bounced rays. The parametric na-
ture of the model makes it adaptable to a variety of propagation
environments, i.e., outdoor micro- and macro-cells. From the
new reference model, we derive a space-time-frequency cor-
relation function (stf-cf) and a space-Doppler power spectral
density (sD-psd) for a 3-D non-isotropic scattering environ-
ment. Finally, we present some simulation results for the
sD-psd and compare them with measured data in [5]. The
close agreement between the theoretical and empirical curves
conrms the utility of the proposed wideband model.
The remainder of the paper is organized as follows.
Section II introduces the geometrical concentric-cylinders
model. Section III presents a 3-D parametric reference model
for wideband MIMO M-to-M channels. Section IV derives
the stf-cf and the sD-psd for 3-D non-isotropic scattering.
Section V presents some simulation results to verify theoretical
derivations. Finally, Section VI provides some concluding
remarks.
II. A GEOMETRICAL CONCENTRIC-CYLINDERS MODEL
In this section, we introduce a 3-D geometrical model for
wideband MIMO M-to-M channels, called the concentric-
cylinders model. The concentric-cylinders model is an
extension of the two-cylinder model for narrowband M-
to-M channels proposed in [4]. We consider a wideband
MIMO communication system with L
t
transmit and L
r
receive
omnidirectional antenna elements. It is assumed that both the
T
x
and R
x
are in motion and equipped with low elevation
antennas. The radio propagation in outdoor micro- and macro-
cells is characterized by 3-D wide sense stationary uncorre-
lated scattering (WSSUS) with either line-of-sight (LoS) or
non-line-of-sight (NLoS) conditions between the T
x
and R
x
.
Fig. 1 shows the concentric-cylinders model with LoS,
single- and double-bounced rays for a MIMO M-to-M channel
with L
t
= L
r
= 2 antenna elements. This 2 2 antenna con-
guration will be used later to construct uniform linear antenna
arrays with an arbitrary number of antennas. The concentric-
cylinders model denes four cylinders, two around the T
x
and another two around the R
x
, as shown in Fig. 1. Around
Appears in the Proceedings of IEEE GLOBECOM'07, November 2007.
the T
x
, M xed omnidirectional scatterers occupy a volume
between cylinders of radii R
t1
and R
t2
. It is assumed that the
M scatterers lie on L cylindric surfaces of radii R
t1
R
(l)
t

R
t2
, where 1 l L. The l
th
cylindric surface contains
M
(l)
xed omnidirectional scatterers, and the (m, l)
th
transmit
scatterer is denoted by S
(m,l)
T
. Similarly, around the R
x
, N
xed omnidirectional scatterers occupy a volume between
cylinders of radii R
r1
and R
r2
. It is assumed that N scatterers
lie on K cylindric surfaces of radii R
r1
R
(k)
r
R
r2
, where
1 k K. The k
th
cylindric surface contains N
(k)
xed
omnidirectional scatterers, and the (n, k)
th
receive scatterer is
denoted by S
(n,k)
R
. The distance between the centers of the T
x
and R
x
cylinders is D. It is assumed that the radii R
t2
and R
r2
are much smaller than the distance D, i.e., max{R
t2
, R
r2
}
D (local scattering condition for outdoor micro- and macro-
cells). Furthermore, it is assumed that the distance D is
smaller than 4R
t1
R
r1
L
r
/((L
t
1)(L
r
1)) (channel does
not experience keyhole behavior [6]), where denotes the
carrier wavelength. The spacing between antenna elements at
the T
x
and R
x
is denoted by d
T
and d
R
, respectively. It is
assumed that d
T
and d
R
are much smaller than the radii R
t1
and R
r1
, i.e., max{d
T
, d
R
} min{R
t1
, R
r1
}. Angles
T
and
R
describe the orientation of the T
x
and R
x
antenna
array in the x-y plane, respectively, relative to the x-axis.
Similarly, angles
T
and
R
describe the elevation of the
T
x
s antenna array and the R
x
s antenna array relative to
the x-y plane, respectively. The T
x
and R
x
are moving with
speeds v
T
and v
R
in directions described by angles
T
and
R
,
respectively. The symbols
(m,l)
T
and
(n,k)
T
are the azimuth
y
x
D
T
v
T

) , ( l m
T

T

) , ( l m
T

) , ( l m
T
S
z
) (l
t
R
) , ( k n
R

'
R
O
) , ( k n
R
S
R
v
) , ( k n
R

R

p
H
p
H~
q
H
l m p , ,

k n l m , , ,

q k n , ,

) ( p
T
A
)
~
( p
T
A
)
~
(q
R
A
) (q
R
A
1 t
R
2 t
R
z
1 r
R
2 r
R
T
O
R
O
) (k
r
R
q l m , ,

k n p , ,

) , ( k n
T

) , ( l m
R

q p,

LoS
Rq

Fig. 1. The concentric-cylinders model with LoS, single- and double-

bounced rays for a MIMO M-to-M channel with Lt = Lr = 2 antenna
elements.
angles of departure (AAoD) of the waves that impinge on the
scatterers S
(m,l)
T
and S
(n,k)
R
, whereas
(m,l)
R
and
(n,k)
R
are the
azimuth angles of arrival (AAoA) of the waves scattered from
S
(m,l)
T
and S
(n,k)
R
, respectively. Similarly, the symbols
(m,l)
T
and
(n,k)
R
denote the elevation angle of departure (EAoD)
and the elevation angle of arrival (EAoA), respectively. The
symbols
p,m,l
,
m,l,q
,
p,n,k
,
n,k,q
,
m,l,n,k
, and
p,q
denote
distances A
(p)
T
- S
(m,l)
T
, S
(m,l)
T
- A
(q)
R
, A
(p)
T
- S
(n,k)
R
, S
(n,k)
R
-
A
(q)
R
, S
(m,l)
T
- S
(n,k)
R
, and A
(p)
T
- A
(q)
R
, respectively, as shown
in Fig. 1. Finally,
LoS
Rq
denotes the AAoAs of the LoS paths.
III. A 3-D PARAMETRIC MODEL FOR WIDEBAND MIMO
MOBILE-TO-MOBILE CHANNELS
In this section, we derive a parametric reference model
for wideband MIMO M-to-M multipath fading channels. The
MIMO channel is described by an L
r
L
t
matrix H(t, ) =
[h
ij
(t, )]
LrLt
of the input delay-spread functions. From the
geometrical concentric-cylinders model, the input delay-spread
function of the link A
(p)
T
A
(q)
R
is a superposition of the LoS,
single-bounced, and double-bounced rays, and can be written
as follows
h
pq
(t, ) = (1)
h
SBT
pq
(t, ) + h
SBR
pq
(t, ) + h
DB
pq
(t, ) + h
LoS
pq
(t, ).
To simplify further derivations, we will use the time-variant
transfer function instead of the input delay-spread function.
The time-variant transfer function is dened as the Fourier
transformation of the input delay-spread function and can be
written as
T
pq
(t, f) = F

{h
pq
(t, )} = T
SBT
pq
(t, f) + T
SBR
pq
(t, f)
+ T
DB
pq
(t, f) + T
LoS
pq
(t, f). (2)
The single-bounced components of the time-variant transfer
function are, respectively,
T
SBT
pq
(t, f) =

T
lim
M
L

l=1
M
(l)

m=1

m,l
g
m,l
(t)e
j2f
m,l
, (3)
T
SBR
pq
(t, f) =

R
lim
N
K

k=1
N
(k)

n=1

n,k
g
n,k
(t)e
j2f
n,k
, (4)
where parameters
m,l
,
n,k
,
m,l
, and
n,k
denote amplitudes
of the multipath components and time delays, respectively.
Functions g
m,l
(t) and g
n,k
(t) are dened as follows
g
m,l
(t) = e
j
2

(
p,m,l
+
m,l,q
)
(5)
e
j2t
_
fTmax cos
_

(m,l)
T
T
_
+fRmax cos
_

(m,l)
R
R
__
+j
m,l
,
g
n,k
(t) = e
j
2

(
p,n,k
+
n,k,q
)
(6)
e
j2t
_
fTmax cos
_

(n,k)
T
T
_
+fRmax cos
_

(n,k)
R
R
__
+j
n,k
,
where f
Tmax
= v
T
/ and f
Rmax
= v
R
/ are the maximum
Doppler frequencies associated with the T
x
and R
x
, respec-
tively, and is the carrier wavelength. The amplitudes
m,l
and
n,k
are approximated as

m,l

pq
_
1

2
R
(l)
t
D
_
_
M(K
pq
+ 1)
,
n,k

pq
_
1

2
R
(k)
r
D
_
_
N(K
pq
+ 1)
, (7)
where
pq
= D
/2
_
P
pq
/4, P
pq
is the power transmitted
through the subchannel A
(p)
T
A
(q)
R
, K
pq
is the Rice factor of
the subchannel A
(p)
T
-A
(q)
R
, and is the path loss exponent.
Finally, the time delays
m,l
and
n,k
are dened as the travel
times of the waves scattered from the scatterers S
(m,l)
T
and
S
(n,k)
R
, i.e.
m,l
= D/c
0
+ R
(l)
t
(1 cos
(m,l)
T
)/c
0
cos
(m,l)
T
and
n,k
= D/c
0
+R
(k)
r
(1+cos
(n,k)
R
)/c
0
cos
(n,k)
R
, respec-
tively, where c
0
is the speed of light. The double-bounced
component of the time-variant transfer function is
T
DB
pq
(t, f) = (8)

TR
lim
M,N
L,M
(l)

l,m=1
K,N
(k)

k,n=1

m,l,n,k
g
m,l,n,k
(t)e
j2f
m,l,n,k
,
where
m,l,n,k
and
m,l,n,k
are the amplitude of the multipath
component and the time delay, respectively. The function
g
m,l,n,k
(t) is dened as
g
m,l,n,k
(t) = e
j
2

(
p,m,l
+
m,l,n,k
+
n,k,q
)+j
m,l,n,k
(9)
e
j2t
_
fTmax cos
_

(m,l)
T
T
_
+fRmax cos
_

(n,k)
R
R
__
.
The amplitude
m,l,n,k
is approximated as

m,l,n,k


pq
_
MN(K
pq
+ 1)
_
1

2
R
(l)
t
+ R
(k)
r
2D
_
. (10)
The time delay
m,l,n,k
is dened as the travel time of
the wave impinged on the scatterer S
(m,l)
T
and scattered
from the scatterer S
(n,k)
R
, i.e.
m,l,n,k
= D/c
0
+ R
(l)
t
(1
cos
(m,l)
T
)/c
0
cos
(m,l)
T
+R
(k)
r
(1 +cos
(n,k)
R
)/c
0
cos
(n,k)
R
.
Parameters
T
,
R
, and
TR
specify how much the single- and
double-bounced rays contribute in the total power P
pq
, i.e.,
these parameters satisfy
T
+
R
+
TR
= 1. It is assumed
that the angles of departure (
(m,l)
T
and
(m,l)
T
), the angles
of arrival (
(n,k)
R
and
(n,k)
R
), and the radii R
(l)
t
and R
(k)
r
are random variables, and that the angles of departure and
radii R
(l)
t
are independent from the angles of arrival and radii
R
(k)
r
. Note that the AAoA,
(m,l)
R
, is a function of the AAoD,

(m,l)
T
, and the AAoD,
(n,k)
T
, is a function of the AAoA,

(n,k)
R
and, hence, they are not independent angular variables.
Additionally, it is assumed that the phases
m,l
,
n,k
, and

m,l,n,k
are random variables uniformly distributed on the
interval [, ) and independent from the angles of departure,
the angles of arrival, and the radii of the cylinders. The LoS
component of the time-variant transfer function is
T
LoS
pq
(t, f) =

K
pq
K
pq
+ 1

LoS
g
LoS
(t)e
j2fLoS
, (11)
where
LoS

pq
,
LoS
=

D
2
+ H
2
/c
0
, g
LoS
(t) =
e
j2t[fRmax cos(
LoS
Rq
R)+fTmax cos(
LoS
Rq
T )]j
2

p,q
, and
H = (H
p
H
p
H
q
+ H
q
)/2. Finally, the distances

m,l,q
,
p,n,k
,
p,m,l
,
n,k,q
,
m,l,n,k
, and
p,q
can be expressed
as functions of the random variables
(m,l)
T
,
(m,l)
T
,
(n,k)
R
,

(n,k)
R
,
LoS
Rq
, R
(l)
t
, and R
(k)
r
as follows:

m,l,q
D (0.5L
r
+ 0.5 q)(d
Ry

(l)
T
sin
(m,l)
T
d
Rx
),

p,n,k
D (0.5L
t
+ 0.5 p)(d
Ty

(k)
R
sin
(n,k)
R
+ d
Tx
),

p,m,l
R
(l)
t
(0.5L
t
+ 0.5 p)
_
d
Tz
sin
(m,l)
T
+
d
Tx
cos
(m,l)
T
cos
(m,l)
T
+ d
Ty
sin
(m,l)
T
cos
(m,l)
T

n,k,q
R
(k)
r
(0.5L
r
+ 0.5 q)
_
d
Rz
sin
(n,k)
R
+
d
Rx
cos
(n,k)
R
cos
(n,k)
R
+ d
Ry
sin
(n,k)
R
cos
(n,k)
R

m,l,n,k
D,

p,q
D (0.5L
t
+ 0.5 p)d
Tx

(0.5L
r
+ 0.5 q)d
R
cos
R
cos(
LoS
Rq

R
), (12)
where parameters p and q take values from the sets p
{1, . . . , L
t
} and q {1, . . . , L
r
}, d
Tx
= d
T
cos
T
cos
T
,
d
Ty
= d
T
cos
T
sin
T
, d
Rx
= d
R
cos
R
cos
R
, d
Ry
=
d
R
cos
R
sin
R
, d
Tz
= d
T
sin
T
, d
Rz
= d
R
sin
R
,
(l)
T
=
R
(l)
t
/D, and
(k)
R
= R
(k)
r
/D. Derivations of the expressions
in (12) are omitted for brevity.
IV. SPACE-TIME-FREQUENCY CORRELATION FUNCTION
AND SPACE-DOPPLER POWER SPECTRUM OF THE 3-D
REFERENCE MODEL
The normalized space-time-frequency correlation function
between two time-variant transfer functions T
pq
(t, f) and
T
p q
(t, f) is dened as
R
pq, p q
(t, f) =
E[T
pq
(t, f)

T
p q
(t + t, f + f)]
_
E[|T
pq
(t, f)|
2
]E[|T
p q
(t, f)|
2
]
, (13)
where ( )

denotes complex conjugate operation, E[ ] is the

statistical expectation operator, p, p {1, . . . , L
t
}, and q, q
{1, . . . , L
r
}. Since the number of local scatterers in the ref-
erence model is innite, the discrete AAoDs,
(m,l)
T
, EAoDs,

(m,l)
T
, AAoAs,
(n,k)
R
, EAoAs,
(n,k)
R
, and radii R
(l)
t
and R
(k)
r
can be replaced with continuous random variables
T
,
T
,
R
,

R
, R
t
, and R
r
. To characterize the random azimuth angles

T
and
R
, we use the von Mises probability density function
(pdf) dened as [7] f() = exp [k cos( )]/2I
0
(k),
where [, ), I
0
( ) is the zeroth-order modied
Bessel function of the rst kind, [, ) is the mean
angle at which the scatterers are distributed in the xy plane,
and k controls the spread of scatterers around the mean. To
characterize the random elevation angles
T
and
R
, we use
the pdf [8]
f() =
_

4|m|
cos
_

m
_
, | ||
m
|

2
0 , otherwise
, (14)
where
m
is the maximum elevation angle and lies in the
range 0

|
m
| 20

[9]. Such elevation angles are

typical for street-canyon propagation, which is prevalent
in M-to-M communications (e.g., two vehicles driving along
streets)
1
. To characterize the radii R
t
and R
r
, we use the
pdfs f(R
t
) = 2R
t
/(R
2
t2
R
2
t1
) and f(R
r
) = 2R
r
/(R
2
r2

R
2
r1
), respectively. Using trigonometric transformations, the
equality
_

exp {a sin(c) + b cos(c)}dc = 2I

0
_
a
2
+ b
2
_
1
Note that elevation angles higher than 20

have been found for over

the roof propagation [9], which is characteristic for xed-to-mobile cellular
communications where the base-station is elevated above the roofs of build-
ings.
[10, eq. 3.338-4], and the results in [4], the stf-cfs of the single-
and double-bounced components can be closely approximated
as
R
SBT
pq, p q
(d
T
, d
R
, t, f) (15)

T
I
0
(k
T
)
cos
_
2

Tm
d
Tz
_
1
_
4TmdTz

_
2
e
j
2

(q q)dRx
e
j2tfRmax cos R
_
Rt2
Rt1
I
0
_
_
x
2
SBT
+ y
2
SBT
_
2R
t
_
1
Rt
D
_
e
j
2
c
0
f(D+Rt)
R
2
t2
R
2
t1
dR
t
,
R
SBR
pq, p q
(d
T
, d
R
, t, f) (16)

R
I
0
(k
R
)
cos
_
2

Rm
d
Rz
_
1
_
4RmdRz

_
2
e
j
2

(p p)dTx
e
j2tfTmax cos T
_
Rr2
Rr1
I
0
_
_
x
2
SBR
+ y
2
SBR
_
2R
r
_
1
Rr
D
_
e
j
2
c
0
f(D+Rr)
R
2
r2
R
2
r1
dR
r
,
R
DB
pq, p q
(d
T
, d
R
, t, f) (17)
A
DB
_
Rt2
Rt1
e
j
2
c
0
fRt
R
t
I
0
_
_
x
2
DB
+ y
2
DB
_
dR
t

_
Rr2
Rr1
e
j
2
c
0
fRr
R
r
I
0
_
_
w
2
DB
+ z
2
DB
__
1
R
r
D
_
dR
r
+
A
DB
_
Rr2
Rr1
e
j
2
c
0
fRr
R
r
I
0
_
_
w
2
DB
+ z
2
DB
_
dR
r

_
Rt2
Rt1
e
j
2
c
0
fRt
R
t
I
0
_
_
x
2
DB
+ y
2
DB
__
1
R
t
D
_
dR
t
,
where x
SBT
j2[(p p)d
Tx
/ + tf
Tmax
cos
T
+
fR
t
/c
0
] + k
T
cos
T
, y
SBT
j2[(p p)d
Ty
/+(q
q)d
Ry

T
/+tf
Tmax
sin
T
+tf
Rmax

T
sin
R
]+k
T
sin
T
,
x
SBR
j2[(q q)d
Rx
/+tf
Rmax
cos
R
fR
r
/c
0
] +
k
R
cos
R
, y
SBR
j2[(q q)d
Ry
/+(p p)d
Ty

R
/+
tf
Rmax
sin
R
+tf
Tmax

R
sin
T
] +k
R
sin
R
, x
DB

j2[(p p)d
Tx
/+tf
Tmax
cos
T
+fR
t
/c
0
]+k
T
cos
T
, y
DB

j2[(p p)d
Ty
/+tf
Tmax
sin
T
]+k
T
sin
T
, z
DB
j2[(q
q)d
Rx
/+tf
Rmax
cos
R
fR
r
/c
0
]+k
R
cos
R
, w
DB

j2[(q q)d
Ry
/+tf
Rmax
sin
R
]+k
R
sin
R
, and A
DB
=

TR
cos(2
Tm
d
Tz
/) cos(2
Rm
d
Rz
/)2e
j2fD/c0
/
I
0
(k
T
)I
0
(k
R
)(1(4
Tm
d
Tz
/)
2
)(1(4
Rm
d
Rz
/)
2
)(R
2
t2

R
2
t1
)(R
2
r2
R
2
r1
). The integrals in (15) - (17) must be evaluated
numerically, because they do not have closed-form solutions.
Using (11) and approximation
LoS
Rq
=
LoS
R q
, the stf-cf
of the LoS component can be approximated as
R
LoS
pq, p q
(d
T
, d
R
, t, f)
_
K
pq
K
p q
e
j
2

[(p p)dTx(q q)dRx]

e
j2t[fTmax cos T fRmax cos R]
2
c
0
f

D
2
+H
2
. (18)
Finally, the normalized stf-cf between two time-variant transfer
functions T
pq
(t, f) and T
p q
(t, f) becomes a summation of the
stf-cfs in (15) - (18).
The space-Doppler power spectral density (sD-psd) of the
time-variant transfer function is the Fourier transformation of
the space-time correlation function R
pq, p q
(d
T
, d
R
, t, f =
0). Using the equality [10, eq. 6.677-3] and after extensive
calculations, the sD-psds of the single-bounced components
become
F
t
{R
SBT
pq, p q
(d
T
, d
R
, t, 0)} =

T
I
0
(k
T
)
cos
_
2

Tm
d
Tz
_
1
_
4Tm
dTz

_
2
exp {j2( + f
Rmax
cos
R
)A
SBT
j2p
xSBR
}
f
Tmax
_
1 [( + f
Rmax
cos
R
)/f
Tmax
]
2
I
SBT
cosh
_
B
SBT
_
1 [( + f
Rmax
cos
R
)/f
Tmax
]
2
_
, (19)
F
t
{R
SBR
pq, p q
(d
T
, d
R
, t, 0)} =

R
I
0
(k
R
)
cos
_
2

Rm
d
Rz
_
1
_
4Rm
dRz

_
2
exp {j2( f
Tmax
cos
T
)A
SBR
+ j2p
xSBT
}
f
Rmax
_
1 [( f
Tmax
cos
T
)/f
Rmax
]
2
I
SBR
cosh
_
B
SBR
_
1 [( f
Tmax
cos
T
)/f
Rmax
]
2
_
, (20)
where F{ } denotes the Fourier transformation, cosh( )
is the hyperbolic cosine, | + f
Rmax
cos
R
| f
Tmax
and
| f
Tmax
cos
T
| f
Rmax
. Parameters in equations (19)
and (20) are dened as follows: A
SBT
= (2p
xSBT
q
xSBT
+
2p
ySBT
q
ySBT
w
xSBT
)/2f
Tmax
, B
SBT
= k
T
sin(
T

T
)+
j2p
xSBT
q
ySBT
j2p
ySBT
q
xSBT
), p
xSBT
= (p p)d
Tx
/,
q
xSBT
= cos
T
, w
xSBT
= jk
T
cos(
T

T
), p
ySBT
=
(p p)d
Ty
/+(q q)d
Ry
(R
t1
+0.5(R
t2
R
t1
))/D, q
ySBT
=
sin
T
, w
ySBT
= jk
T
sin(
T

T
), A
SBR
= (2p
xSBR
q
xSBR
+
2p
ySBR
q
ySBR
+w
xSBR
)/2f
Rmax
, B
SBR
= k
R
sin(
R

R
)+
j2p
xSBR
q
ySBR
j2p
ySBR
q
xSBR
, p
xSBR
= (q q)d
Rx
/,
q
xSBR
= cos
R
, w
xSBR
= jk
R
cos(
R

R
), p
ySBR
=
(q q)d
Ry
/+(p p)d
Ty
(R
r1
+0.5(R
r2
R
r1
))/D, q
ySBR
=
sin
R
, w
ySBR
= jk
R
sin(
R

R
), and I
SBT/SBR
= [(3
2R
t2/r2
/D)R
2
t2/r2
(3 2R
t1/r1
/D)R
2
t1/r1
]/3(R
2
t2/r2

R
2
t1/r1
). Using similar reasoning as above, the sD-psd of the
double-bounced component becomes
F
t
{R
DB
pq, p q
(d
T
, d
R
, t, 0)} =

TR
I
DB
I
0
(k
T
)I
0
(k
R
)
(21)
e
j(2px
DB
qx
DB
+2py
DB
qy
DB
jkT cos(T T ))/fTmax

cos
_
2

Tm
d
Tz
_
1
_
4Tm
dTz

_
2
cosh
_
B
DB
_
1 (/f
Tmax
)
2
_
f
Tmax
_
1 (/f
Tmax
)
2
e
j(2pz
DB
qz
DB
+2pw
DB
qw
DB
jkR cos(RR))/fRmax

cos
_
2

Rm
d
Rz
_
1
_
4Rm
dRz

_
2
cosh
_
C
DB
_
1 (/f
Rmax
)
2
_
f
Rmax
_
1 (/f
Rmax
)
2
,
where denotes convolution, p
xDB
= (p p)d
Tx
/, q
xSBT
=
cos
T
, p
yDB
= (p p)d
Ty
/, q
ySBT
= sin
T
, p
zDB
=
(q q)d
Rx
/, q
zDB
= cos
R
, p
wDB
= (q q)d
Ry
/,
q
wDB
= sin
R
, || f
Tmax
+ f
Rmax
, I
DB
= (R
2
t2

R
2
t1
)(0.5R
2
r2
R
3
r2
/(3D) 0.5R
2
r1
+R
3
r1
/(3D)) +(R
2
r2

R
2
r1
)(0.5R
2
t2
R
3
t2
/(3D) 0.5R
2
t1
+ R
3
t1
/(3D)), B
DB
=
k
T
sin(
T

T
)+j2p
xDB
q
yDB
j2p
yDB
q
xDB
, and C
DB
=
k
R
sin(
R

R
) +j2p
zDB
q
wDB
j2p
wDB
q
zDB
. By calcu-
lating the Fourier transformation of the space-time correlation
function in (18), we obtain the sD-psd of the LoS component:
F
t
{R
LoS
pq, p q
(d
T
, d
R
, t, 0)} =
_
K
pq
K
p q
(22)
e
j
2

[(p p)dTx(q q)dRx]

( + f
Tmax
cos
T
f
Rmax
cos
R
),
where ( ) denotes the Dirac delta function. Finally, the sD-
psd between two time-variant transfer functions T
pq
(t, f = 0)
and T
p q
(t, f = 0) becomes a summation of the sD-psds in
(19) - (22).
V. SIMULATION RESULTS
In this section, we present some simulation results to verify
theoretical derivations. Figs. 2 and 3 show the sD-psds for
several MIMO systems. Here, we analyze the radio propaga-
tion in outdoor M-to-M micro- and macro-cells, assuming 3-D
non-isotropic scattering (k
T
= k
R
= 2 for curves in Fig. 2
and k
T
= k
R
= 9 for curves in Fig. 3) and line-of-sight
(K = 2) conditions between the T
x
and R
x
. In all simulations
we use the following parameters:
Tm
=
Rm
= 15

,
-400 -300 -200 -100 0 100 200 300 400
-50
-40
-30
-20
-10
0
sD-psd d
T
=d
R
=0
sD-psd d
T
=d
R
=0.5
sD-psd d
T
=d
R
=1
sD-psd d
T
=d
R
=2
Measured sD-psd d
T
=d
R
=0
N
o
r
m
a
l
i
z
e
d

P
o
w
e
r

S
p
e
c
t
r
a
l

D
e
n
s
i
t
y

[
d
B
]
Doppler Frequency [Hz]

Fig. 2. The theoretical and measured sD-psd characteristic for the outdoor
M-to-M micro-cells.

-400 -300 -200 -100 0 100 200 300 400
-70
-60
-50
-40
-30
-20
-10
0
sD-psd d
T
=d
R
=0
sD-psd d
T
=d
R
=0.5
sD-psd d
T
=d
R
=1
sD-psd d
T
=d
R
=2
Measured sD-psd d
T
=d
R
=0
N
o
r
m
a
l
i
z
e
d

P
o
w
e
r

S
p
e
c
t
r
a
l

D
e
n
s
i
t
y

[
d
B
]
Doppler Frequency [Hz]

Fig. 3. The theoretical and measured sD-psd characteristic for the outdoor
M-to-M macro-cells.

T
=
R
= /4,
T
=
R
= /3,
T
=
R
= 0,
T
= 0,

R
= , = 0.3 m, R
t1
= R
r1
= 30 m, R
t2
= R
r2
=
300 m, D = 10 km, = 4, L
t
= L
r
= 2, d
T
= d
R

{0, 0.5, 1, 2}, and f
Tmax
= f
Rmax
= 400 Hz. Fig. 2
assumes that the single-bounced rays bear more energy than
the double-bounced rays, i.e.,
T
=
R
= 0.3 and
TR
= 0.4,
which is characteristic for the outdoor M-to-M micro-cell
propagation. We can observe that this spectrum is similar to
the U-shaped spectrum of F-to-M cellular channels. Fig. 3
considers the macro-cell propagation (i.e.
T
=
R
= 0.0001
and
TR
= 0.9998). In this case, the sD-psd differs from
the U-shaped spectrum of cellular channels. Finally, Figs. 2
and 3 compare our theoretical Doppler spectra obtained for
d
T
= d
R
= 0 with measured Doppler spectra for SISO system.
The measured results are taken from Figs. 4(b) and 4(d) of [5].
The close agreement between the theoretical and empirical
curves conrms the utility of the proposed wideband model.
VI. CONCLUSIONS
This paper introduced a concentric-cylinders geometrical
propagation model. Based on this geometrical model, a 3-
D parametric reference model for wideband MIMO M-to-M
fading channels was proposed. From the reference model, the
space-time-frequency correlation function and space-Doppler
power spectral density were derived for a 3-D non-isotropic
scattering environment. Finally, simulation results for the sD-
psd compare very well with measured data in [5].
DISCLAIMER
The views and conclusions contained in this document
are those of the authors and should not be interpreted as
representing the ofcial policies, either expressed or implied,
of the Army Research Laboratory or the U. S. Government.
REFERENCES
[1] A.S. Akki and F. Haber, A statistical model for mobile-to-mobile land
communication channel, IEEE Trans. on Veh. Tech., vol. 35, pp. 210,
Feb. 1986.
[2] M. P atzold, B.O. Hogstad, N. Youssef, and D. Kim, A MIMO mobile-
to-mobile channel model: Part I-the reference model, Proc. IEEE
PIMRC05, vol. 1, pp. 573578, Germany, Sept. 2005.
[3] A.G. Zaji c and G.L. St uber, Space-Time Correlated MIMO Mobile-to-
Mobile Channels, Proc. IEEE PIMRC06, Finland, Sept. 2006.
[4] A.G. Zaji c and G.L. St uber, A Three-Dimensional MIMO Mobile-to-
Mobile Channel Model, IEEE WCNC07, Hong Kong, Mar. 2007.
[5] G. Acosta, K. Tokuda, and M. A. Ingram, Measured joint Doppler-delay
power proles for vehicle-to-vehicle communications at 2.4 GHz, Proc.
IEEE GLOBECOM04, vol. 6, pp. 38133817, Dallas, TX, Nov. 2004.
[6] D. Gesbert, H. B olcskei, D.A. Gore, and A.J. Paulraj, Outdoor MIMO
wireless channels: models and performance prediction, IEEE Trans. on
Commun., vol. 50, pp. 19261934, Dec. 2002.
[7] 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 Trans. on Veh. Tech.,
vol. 51, pp. 425434, May 2002.
[8] J.D. Parsons and A.M.D. Turkmani, Characterisation of mobile radio
signals: model description, IEE Proc. I, Commun., Speach, and Vision,
vol. 138, pp. 549556, Dec. 1991.
[9] A. Kuchar, J.-P. Rossi, and E. Bonek, Directional macro-cell channel
characterization from urban measurements, IEEE Trans. on Antennas
and Propagation, vol. 48, pp. 137146, Feb. 2000.
[10] I.S. Gradshteyn and I.M. Ryzhik, Table of Integrals, Series, and Products
5
th
ed.. A. Jeffrey, Ed. San Diego CA: Academic, 1994.