IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO.
2, MARCH 2008
Space-Time Correlated Mobile-to-Mobile
Channels: Modelling and Simulation
Alenka G. Zajić, Student Member, IEEE, and Gordon L. Stüber, Fellow, IEEE
Abstract—A single- and double-bounced two-ring parametric
reference model is proposed for multiple-input multiple-output
(MIMO) mobile-to-mobile (M-to-M) Ricean fading channels.
From this model, a closed-form joint space-time correlation func-
tion and a space-Doppler power spectrum are derived for a
two-dimensional (2-D) non-isotropic scattering environment. Also,
space-time correlation functions for the in-phase and quadrature
components of the complex faded envelope are derived, assuming
a 2-D isotropic scattering environment. Finally, two new sum-of-
sinusoids based simulation models for MIMO M-to-M Ricean
fading channels are proposed. The statistics of the simulation
models are verified by simulation. The results show that the
simulation models are a good approximation of the reference
model and that they outperform existing simulation models.
Index Terms—Deterministic and statistical methods, Doppler
spectrum, fading channel simulator, mobile-to-mobile channels,
Ricean and Rayleigh fading, sum-of-sinusoids.
I. I NTRODUCTION
M OBILE ad-hoc wireless networks, intelligent trans-
portation systems, and relay-based cellular networks
all use mobile-to-mobile (M-to-M) communication channels,
where both the transmitter (Tx ) and the receiver (Rx ) are in
motion and equipped with low elevation antennas. M-to-M
channels differ from conventional fixed-to-mobile (F-to-M)
cellular radio channels, where the base-station is stationary,
elevated, and relatively free of local scattering. Akki and Haber
[1], [2] showed that the received envelope on M-to-M channels
is Rayleigh faded under non line-of-sight conditions (NLoS),
but the statistical properties differ from F-to-M channels.
They were the first to propose a reference model for single-
input single-output (SISO) M-to-M Rayleigh fading channels.
Methods for simulating SISO M-to-M channels have been
proposed in [3]–[5]. Channel measurements for narrowband
and wideband mobile-to-mobile communications have been
reported in [6] and [7], respectively. Recently, using the “two-
ring” geometrical model in [8], Pätzold et al. proposed a refer-
Manuscript received May 30, 2006; revised December 15, 2006, April 8,
2006, and June 4, 2007. The review of this paper was coordinated by
Prof. Y. Ma. This work was prepared through collaborative participation in the
Collaborative Technology Alliance for Communications and Networks spon-
sored 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. The views and conclusions contained in this document are
those of the authors and should not be interpreted as representing the official
policies, either expressed or implied, of the Army Research Laboratory or the
U.S. Government.
The authors are with the School of Electrical and Computer Engineering,
Georgia Institute of Technology, Atlanta, GA 30332 USA (e-mail: alenka@ece.
gatech.edu; stuber@ece.gatech.edu).
Digital Object Identifier 10.1109/TVT.2007.905591
0018-9545/$25.00 © 2008 IEEE
715
ence model for narrowband MIMO M-to-M channel [9]. Based
on this reference model, they also proposed a deterministic
sum-of-sinusoids (SoS) model called the Modified Method of
Exact Doppler Spreads (MMEDS) [10].
Previously reported models for MIMO M-to-M channels
[9], [10] consider radio propagation in outdoor macro-cells,
assuming that all rays are only double-bounced. This paper first
proposes a new parametric reference model that employs a two-
ring model and constructs the received complex faded envelope
as a superposition of the line-of-sight (LoS), single-bounced,
and double-bounced rays. The parametric nature of the model
makes it adaptable to a variety of propagation environments,
i.e., outdoor micro—and macro-cells. For example, our refer-
ence model describes the radio propagation in outdoor macro-
cells as a combination of single- and double-bounced rays,
taking into account that the double-bounced rays bear more en-
ergy than the single-bounced rays. In particular, if there are no
single-bounced and LoS rays, our model simplifies to the model
proposed in [9] and [10]. From the new reference model, we
derive a closed-form joint space-time correlation function for
a two-dimensional (2-D) non-isotropic scattering environment.
For 2-D isotropic scattering and no single-bounced and LoS
rays, our space-time correlation function is shown to reduce
to the one derived in [9]. Furthermore, we derive the space-
time correlation functions of the in-phase (I) and quadrature
(Q) components of the complex faded envelope for the 2-D
isotropic scattering environment. Finally, we derive the space-
Doppler power spectrum of the complex faded envelope assum-
ing a 2-D non-isotropic scattering environment.
The reference models assume an infinite number of scat-
terers, which prevents practical implementation. Hence, we
propose an ergodic statistical (deterministic) SoS simulation
model for a 2-D isotropic scattering environment. We employ a
“two-ring” model that combines the LoS, single-bounced, and
double-bounced rays and has orthogonal functions as the I and
Q components of the complex faded envelope. The statistical
properties of our model are verified by simulations. Further-
more, by allowing amplitudes, phases, and Doppler frequencies
to be random variables, the deterministic model is modified to
better match statistical properties of the reference model. This
model is called the statistical simulation model. The statistical
properties of the statistical model vary for each simulation trial,
but will converge to desired ensemble averaged properties when
averaged over a sufficient number of simulation trials. The sta-
tistical properties of this model are also verified by simulations.
Finally, we compare the performance and complexity of our
new models with the MMEDS model. For fair comparison,
we remove the LoS and single-bounced components of the
complex faded envelope from our models. Our deterministic
model performs similar to the MMEDS model [10], but requires
716 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008

a smaller number of scatterers and has a shorter simulation

time. Compared to the deterministic and the MMEDS model,
the statistical properties of the statistical model match those
of the reference model over a wider range of normalized time
delays while using smaller number of scatterers.
The remainder of the paper is organized as follows. Section II
describes the communication system and presents the new
parametric reference model for MIMO M-to-M channels.
Section III presents the derivation of the closed-form joint
space-time correlation function for 2-D non-isotropic scatter-
ing, the derivation of the space-time correlation functions of
the I and Q components of the complex faded envelope for
2-D isotropic scattering, and the derivation of the space- Fig. 1. Single- and double-bounced two-ring model for MIMO M-to-M
Doppler power spectrum of the complex faded envelope for channel with Lt = Lr = 2 antenna elements.
2-D non-isotropic scattering. Section IV reviews the MMEDS
model proposed in [10] and details the deterministic and statis-
tical simulation models. Section V presents simulation results
for our new models and compares them to the MMEDS model
in [10]. Section VI provides some concluding remarks.

II. A N EW P ARAMETRIC R EFERENCE M ODEL FOR

MIMO M OBILE - TO -M OBILE C HANNELS
This paper considers a narrowband single-user MIMO com-
munication system with Lt transmit and Lr receive omnidirec-
tional antenna elements. It is assumed that both the Tx and
Rx are in motion and equipped with low elevation antennas.
This paper considers the radio propagation in outdoor micro- Fig. 2. LoS paths in the 2 × 2 channel of Fig. 1.
and macro-cells, which is characterized by 2-D scattering with
either line-of-sight (LoS) or non-line-of-sight (NLoS) condi-
tions between the transmitter and receiver. The MIMO channel the angles of departure (AoD) of the waves that impinge on
(m) (n) (m) (n)
can be described by an Lr × Lt matrix H(t) = [hij (t)]Lr ×Lt the scatterers ST and SR , whereas αR and αR are the
of complex faded envelopes. (m)
angles of arrival (AoA) of the waves scattered from ST and
The geometry of the proposed model is shown in Figs. 1 (n)
SR , respectively. The symbols pm , mq , pn , nq , and mn de-
and 2. Fig. 1 shows the two-ring model with single- and double- (p) (m) (m) (q) (p) (n) (n)
bounced rays for a MIMO M-to-M channel with Lt = Lr = 2 note distances AT − ST , ST − AR , AT − SR , SR −
(q) (m) (n)
antenna elements. Fig. 2 shows the LoS paths for the channel in AR , and ST − SR , respectively, as shown in Fig. 1. In
Fig. 1. This elementary 2 × 2 antenna configuration will be used LoS
Fig. 2, the symbols αRq and pq denote the AoAs of the LoS
later to construct uniform linear antenna arrays with arbitrary (p) (q)
paths and the distance AT − AR , respectively.
number of antennas. The two-ring model defines two rings of From the geometrical model described above, the received
fixed scatterers, one around the Tx and another around the Rx , (p) (q)
complex faded envelope of the link AT − AR is a superpo-
as shown in Fig. 1. Around the transmitter, M omnidirectional
sition of the LoS, the single-bounced, and the double-bounced
scatterers lie on a ring of radius Rt , and the mth transmit
(m) rays, and can be written as follows:
scatterer is denoted by ST . Similarly, around the receiver,
N omnidirectional scatterers lie on a ring of radius Rr and
(n) hpq (t) = hSBT SBR DB LoS
pq (t) + hpq (t) + hpq (t) + hpq (t) (1)
the nth receive scatterer is denoted by SR . The distance
between the Tx and Rx is D. It is assumed that the radii Rt
and Rr are much smaller than the distance D between the Tx where the single-bounced components of the received complex
and Rx , i.e., max{Rt , Rr }
D (local scattering condition). faded envelope are, respectively
Furthermore, it is assumed that the distance D is smaller than
4Rt Rr Lr /(λ(Lt − 1)(Lr − 1)) (channel does not experience

ηT Ωpq 
M
1
√ ejφm −j λ (pm +mq )
2π
keyhole behavior [11]), where λ denotes the carrier wavelength. hSBT
pq (t) = lim
The spacing between two adjacent antenna elements at the Tx Kpq + 1 M →∞ m=1 M
and Rx is denoted by δT and δR , respectively. It is assumed   (m)   (m) 
j2πt fT max cos αT −γT +fR max cos αR −γR
that δT and δR are much smaller than the radii Rt and Rr , ×e (2)
i.e., max{δT , δR }
min{Rt , Rr }. Angles θT and θR describe

ηR Ωpq 
N
1
the orientation of the Tx and Rx antenna array, respectively, hSBR (t) = lim √ ejφn −j λ (pn +nq )
2π
pq
relative to the x-axis. Similarly, the Tx and Rx are moving Kpq + 1 N →∞ n=1 N
with speeds vT and vR in directions described by angles γT   (n)   (n) 
(m) (n) j2πt fT max cos αT −γT +fR max cos αR −γR
and γR , respectively. In Fig. 1, the symbols αT and αT are ×e (3)
ZAJIĆ AND STÜBER: SPACE-TIME CORRELATED MOBILE-TO-MOBILE CHANNELS: MODELLING AND SIMULATION 717

the double-bounced component of the received complex faded respectively

envelope is  

(m)

ηT Ωpq 
M j π (L +1−2p)δT cos θT −αT
e λ t

M,N hSBT
pq (t) = lim √
ηT R Ωpq 1 Kpq + 1 M →∞m=1
e−j λ (pm +mn +nq ) M
2π
hDB
pq (t) = lim √  
Kpq+1 M,N→∞m,n=1 M N jφm −j 2π π (m)
(D+Rt )+j λ (Lr +1−2q)δR ∆T sin θR sin αT −cos θR
  (m)   (n)  ×e 
λ
   
j2πt fT max cos αT −γT +fRmax cos αR −γR +jφmn
×e (4) ×e
j2πt fT max cos αT
(m)
−γT +fR max ∆T sin γR sin αT
(m)
−cos γR

(12)
and the LoS component of the received complex faded

envelope is ηR Ωpq
hSBR
pq (t) = lim

Kpq + 1 N →∞
 
Kpq Ωpq j2πtfT max cos(π−αLoS N ej λ (L + 1 − 2q)δ cos α(n) − θ
π
hLoS
pq (t) = e Rq −γT )  r R R R
Kpq + 1 × √
N
× ej2πtfR max cos(αRq −γR )−j λ pq . (5)  
LoS 2π n=1
(n)
jφ −j 2π π
(D+Rr )+j λ (Lt +1−2p)δT ∆R sin θT sin αR +cos θT
×e n λ    
(n) (n)
In (2)–(5), Ωpq and Kpq denote the transmitted power and j2πt fR max cos αR −γR +fT max ∆R sin γT sin αR +cos γT
×e
(p) (q)
the Ricean factor of the subchannel AT − AR , respectively. (13)
Parameters ηT , ηR , and ηT R specify how much the single-

ηT R Ωpq e
M,N −j 2π
λ (D+Rt +Rr )+jφmn
and double-bounced rays contribute in the total power Ωpq , hDB √
pq (t) = lim
i.e., these parameters satisfy ηT + ηR + ηT R = 1. Frequencies Kpq + 1 M,N →∞ m,n=1 MN
fT max = vT /λ and fR max = vR /λ are the maximum Doppler  (m)
 π  (n)

λ (Lt +1−2p)δT cos θT −αT
jπ +j λ (Lr +1−2q)δR cos αR −θR
frequencies associated with the Tx and Rx , respectively, and λ ×e     
(m) (m) (n)
j2πt fT max cos αT −γT +fR max cos αR −γR
is the carrier wavelength. It is assumed that the AoDs (αT and ×e (14)
(n) (m) (n)

αT ) and the AoAs (αR and αR ) are random variables. Note Kpq Ωpq −j 2π D
(m)
that double-bounced rays have the AoD, αT , independent hLoS
pq (t) = e λ
1 + Kpq
(n)
from the AoA, αR , [11], whereas single-bounced rays have
× ej λ [(Lt +1−2p)δT cos θT +(Lr +1−2q)δR cos(αRq −θR )]
π LoS
(m) (m)
the AoA, αR , dependent on the AoD, αT , and the AoD,
× ej2πtfT max cos(π−αRq −γT )+j2πtfR max cos(αRq −γR ) . (15)
LoS LoS
(n) (n)
αT , dependent on the AoA αR (see Appendix). Finally, it is
assumed that the phases φm , φn , and φmn are random variables
uniformly distributed on the interval [−π, π) and independent The complex faded envelope in (1) also can be written as
(I) (Q)
from the angles of departure and the angles of arrival. hpq (t) = hpq (t) + jhpq (t), where
As shown in the Appendix, the distances pm , mq , pn , nq ,  SBT   SBR 
pq (t) =  hpq (t) +  hpq (t)
h(I)
(m)
mn , and pq can be expressed as functions of the angles αT ,
   LoS 
(n)
αR , and αRq LoS
as follows: pq (t) +  hpq (t)
+  hDB (16)
 SBT   SBR 
  pq (t) =  hpq (t) +  hpq (t)
h(Q)
(m)    LoS 
pm ≈ Rt − (0.5Lt + 0.5 − p)δT cos θT − αT (6) +  hDB
pq (t) +  hpq (t) (17)
mq ≈ D − (0.5Lr + 0.5 − q)

(m) are the in-phase (I) and quadrature (Q) components of the
× δR ∆T sin θR sin αT − cos θR (7)
complex faded envelope, and {·} and {·} denote the real
pn ≈ D − (0.5Lt + 0.5 − p) and imaginary operation, respectively.

(n)
× δT ∆R sin θT sin αR + cos θT (8)
 
(n)
nq ≈ Rr − (0.5Lr + 0.5 − q)δR cos αR − θR (9) III. S PACE -T IME C ORRELATION F UNCTIONS AND
  S PACE -D OPPLER P OWER S PECTRUM
(m) (n)
sin αR − αR
mn = m    ≈ D (10) Assuming a 2-D non-isotropic scattering environment, we
(m) (n) now derive the closed-form space-time correlation function of
sin χ + αR − αR
the complex faded envelope in (1). We also show that this
pq ≈ D − (0.5Lt + 0.5 − p)δT cos θT
 LoS  space-time correlation function reduces to the one derived in [9]
− (0.5Lr + 0.5 − q)δR cos αRq − θR (11) if there are no single-bounced and LoS rays and 2-D isotropic
scattering is assumed. Furthermore, we derive the space-time
where parameters p and q take values from the sets p ∈ correlation functions of the I and Q components of the complex
{1, . . . , Lt } and q ∈ {1, . . . , Lr }, respectively, ∆T = Rt /D, faded envelope for a 2-D isotropic scattering environment.
and ∆R = Rr /D. Finally, we derive the space-Doppler power spectral density
Using (6)–(11), the single-bounced, double-bounced, and of the complex faded envelope, assuming a 2-D non-isotropic
LoS components of the complex faded envelope become, scattering environment.
718 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008

(n)
The normalized space-time correlation function between two AoAs, αR , can be replaced with continuous random variables
complex faded envelopes hpq (t) and hp̃q̃ (t) is defined as αT and αR with probability density functions (pdf) f (αT )
and f (αR ), respectively. Hence, the space-time correlation
E [hpq (t)hp̃q̃ (t + τ )∗ ]
Rpq,p̃q̃ (δT , δR , τ ) =

functions of the single- and double-bounced components can
E |hpq (t)|2 E |hp̃q̃ (t)|2 be rewritten as
 ∗
  ∗
 π
E hSBT SBT
pq (t)hp̃q̃ (t+τ ) E hDB DB
pq (t)hp̃q̃ (t+τ ) ej λ (p̃−p)δT cos(θT −αT )
2π
SBT
=
 +
Rpq,p̃q̃ (δT , δR , τ ) = ηT
 2
 2  
2
 2
 DB(t)
E hpq (t) E hp̃q̃ (t) E hDB  −π
SBT SBT (t) E  h 
× e−j2πτ fT max cos(αT −γT ) ej λ (q̃−q)δR (∆T sin θR sin αT −cos θR )
pq p̃q̃ 2π

   
E hSBR SBR
pq (t)hp̃q̃ (t+τ )
∗
E hLoS LoS
pq (t)hp̃q̃ (t+τ )
∗ × e−j2πτ fR max (∆T sin γR sin αT −cos γR ) f (αT )dαT (22)
+
 +
π
 2
 2  
2
 2
 LoS(t) ej λ (q̃−q)δR cos(αR −θR )
2π
SBR
 SBR 
E hpq (t) E hp̃q̃ (t)SBR E hLoSpq (t)  E h  Rpq, p̃q̃ (δT , δR , τ ) = ηR
p̃q̃
−π
× ej λ (p̃−p)δT (∆R sin θT sin αR +cos θT ) e−j2πτ fR max cos(αR −γR )
2π
(18)
where (·)∗ denotes complex conjugate operation, E[·] is the × e−j2πτ fT max (∆R sin γT sin αR +cos γT ) f (αR )dαR (23)
π π
statistical expectation operator, p, p̃ ∈ {1, . . . , Lt }, and q, q̃ ∈ DB
{1, . . . , Lr }. Rpq, p̃q̃ (δT , δR , τ ) = ηT R f (αT )f (αR )
Using (12), (13), and (18), the space-time correlation func- −π −π
−j2πτ [fT max cos(αT −γT )+fR max cos(αR −γR )]
tions of the single-bounced components can be written as ×e
× ej λ [(p̃−p)δT cos(αT −θT )+(q̃−q)δR cos(αR −θR )] dαT dαR .
2π

SBT
Rpq, p̃q̃ (δT , δR , τ ) (24)
 
+ τ )∗ ηT 
M
E hSBT SBT
pq (t)hp̃q̃ (t
=
 = lim Prior work uses several different scatterer distributions, such
 2
  M →∞ M
   SBT 2 m=1 as uniform [12], von Mises [13], Gaussian, and Laplacian
E hpq (t) E hp̃q̃ (t)
SBT
[14]. In this section we use the von Mises pdf because it
      approximates many of the afore mentioned distributions (e.g.,
(m) (m)
j 2π (p̃−p)δT cos θT−αT +(q̃−q)δR ∆T sinθR sinαT −cos θR
×E e λ uniform and Gaussian), matches well measured results in [15],
  (m)   (m)
 and in contrast to afore mentioned distributions, leads to closed-
−j2πτ fT max cos αT −γT +fRmax ∆T sinγR sinαT −cosγR form solutions for many useful situations. The von Mises pdf is
×e
defined as [13]
(19) 1
∆
SBR
Rpq, p̃q̃ (δ T , δ R , τ ) f (θ) = exp [k cos(θ − µ)] (25)
 ∗
2πI0 (k)
ηR 
N
E hSBR SBR
pq (t)hp̃q̃ (t + τ )
=
 = lim where θ ∈ [−π, π), I0 (·) is the zeroth-order modified Bessel
 2
  N →∞ N
 SBR 2 function of the first kind, µ ∈ [−π, π) is the mean angle
E hSBR  n=1
pq (t) E h p̃q̃ (t)
at which the scatterers are distributed on the ring, and k
   (n)    controls the spread of scatterers around the mean. When k = 0,
(n)
j 2π (q̃−q)δR cos αR −θR +(p̃−p)δT ∆R sin θT sin αR +cos θT
×E e λ f (θ) = 1/(2π) is a uniform distribution yielding 2-D isotropic
  (n)
  (n)  scattering. As k increases, the scatterers become more clustered
−j2πτ fT max ∆R sin γT sin αR +cos γT +fRmax cos αR −γR around angle µ and the scattering becomes increasingly non-
×e .
isotropic. By denoting the von Mises pdf for the Tx and Rx
(20) scatterers as f (αT ) = exp [kT cos(αT − µT )]/(2πI0 (kT ))
and f (αR ) = exp [kR cos(αR − µR )]/(2πI0 (kR )), respec-
Using (14) and (18), the space-time correlation function of the
tively, the space-time correlation functions of the single- and
double-bounced component becomes
double-bounced components become, respectively
DB
Rpq, p̃q̃ (δT , δR , τ ) π
 ∗ ηT 2π
cos(θT −αT )
ηT R  
SBT
E hDB DB
pq (t)hp̃q̃ (t + τ )
M N
Rpq, p̃q̃ (δT , δR , τ ) = ej λ (p̃−p)δT

=
 = lim 2πI0 (kT )

 2
  M,N →∞ M N −π
 DB 2
E hDB  m=1n=1
pq (t) E  h p̃q̃ (t) ×e λ (q̃−q)δR (∆T sin θR sin αT −cos θR ) kT cos(αT −µT )
j 2π
e
   (m)   (n)  × e−j2πτ [fT max cos(αT −γT )+fR max (∆T sin γR sin αT −cos γR )] dαT
−j2πτ fT max cos αT −γT +fR max cos αR −γR
×E e (26)
  (n)     π
ηR
ej λ (q̃−q)δR cos(αR −θR )
2π
(m) SBR
λ (q̃−q)δR cos αR −θR +(p̃−p)δT cos θT −αT
j 2π Rpq, p̃q̃ (δT , δR , τ ) =
×e . 2πI0 (kR )
−π
(21)
× ej λ (p̃−p)δT (∆R sin θT sin αR +cos θT ) ekR cos(αR −µR )
2π

Since the number of local scatterers in the reference models × e−j2πτ [fT max (∆R sin γT sin αR +cos γT )+fRmax cos(αR −γR )] dαR
(m)
described in Section II is infinite, the discrete AoDs, αT , and (27)
ZAJIĆ AND STÜBER: SPACE-TIME CORRELATED MOBILE-TO-MOBILE CHANNELS: MODELLING AND SIMULATION 719

π π and wDB = j2π(q̃ − q)δR sin θR /λ − j2πτ fR max sin γR +

ηT R
DB
Rpq, p̃q̃ (δT , δR , τ ) = 2
ekT cos(αT −µT ) kR sin µR , respectively.
4π I0 (kT )I0 (kR ) Using (6) and (18), the space-time correlation function of the
−π −π
kR cos(αR −µR ) j 2π
λ [(p̃−p)δT cos(αT −θT )+(q̃−q)δR cos(αR −θR )]
LoS component can be written as
×e e
× e−j2πτ [fT max cos(αT −γT )+fR max cos(αR −γR )] dαT dαR . 
LoS
Rpq, p̃q̃ (δT , δR , τ, t) = Kpq Kp̃q̃ e−j2π(pq −p̃q̃ )/λ
(28)
× e−j2πτ [fT max cos(π−αRq̃ −γT )+fR max cos(αRq̃ −γR )]
LoS LoS

Using trigonometric transformations, (26) and (27) become,

× ej2πtfT max [cos(π−αRq −γT )−cos(π−αRq̃ −γT )]
LoS LoS

respectively,
−γR )−cos(αLoS
Rq̃ −γR )]
× ej2πtfR max [cos(αRq
LoS
. (34)
ηT e−j
2π
λ (q̃−q)δR cos θR +j2πτ fR max cos γR
SBT
Rpq, p̃q̃ (δT , δR , τ ) = Based on the assumption max{Rt , Rr }
D, from Fig. 2 we
2πI0 (kT )
LoS
obtain αRq LoS
= αRq̃ ≈ π which in turn simplifies (34) to
π
× exSBT cos αT +ySBT sin αT dαT (29)  2π
cos θT −j 2π
p̃q̃ (δT , δR , τ ) ≈
LoS
Rpq, Kpq Kp̃q̃ ej λ δT λ δR cos θR
−π
× ej2πτ fT max cos γT −j2πτ fR max cos γR . (35)
2π
cos θT −j2πτ fT max cos γT
SBR ηR ej λ (p̃−p)δT
Rpq, p̃q̃ (δT , δR , τ ) =
2πI0 (kR ) Finally, the normalized space-time correlation function
π Rpq,p̃q̃ (δT , δR , τ ) between two complex faded envelopes hpq (t)
× exSBR cos αR +ySBR sin αR dαR (30) and hp̃q̃ (t) becomes a summation of the normalized space-
SBT SBR
time correlation functions Rpq, p̃q̃ (δT , δR , τ ), Rpq,p̃q̃ (δT , δR , τ ),
−π DB LoS
Rpq, p̃q̃ (δT , δR , τ ), and Rpq,p̃q̃ (δT , δR , τ ), defined in (31)–(33)
where parameters xSBT , ySBT , xSBR , and ySBR are xSBT = and (35), respectively.
j(2π/λ)(p̃ − p)δT cos θT − j2πτ fT max cos γ T + kT cos µ T , Many existing correlation functions are special cases of the
ySBT = j(2π/λ)((p̃ − p) δT sin θT + (q̃ − q)δR ∆T sin θR ) − MIMO M-to-M space-time correlation function in (18). The
j2πτ (fT max sin γT +fR max ∆T sin γR )+kT sin µT , xSBR = simplest special case of (18) is Clarke’s temporal correla-
j (2π/λ)(q̃ − q) δR cos θR − j2πτ fR max cos γR + kR cos µR , tion function J0 (2πfR max τ ) [17], obtained for K = 0 (NLoS
and ySBR = j(2π/λ)((p̃−p)δT ∆R sin θT +(q̃−q)δR sin θR )− condition), kR = 0 (2-D isotropic scattering around Rx ),
j2πτ (fT max ∆R sin γT + fR max sin γR ) + kR sin µ√
R . Finally,
fT max = 0, kT = 0, ηT = 0, ηT R = 0 (fixed Tx , no scattering
π around Tx ), and δT = δR = 0 (single-antenna Tx and Rx ),
using the equality −π ea sin(c)+b cos(c) dc = 2πI0 ( a2 + b2 )
[16, eq. 3.338–4], the space-time correlation functions of the where J0 (·) is the first kind zeroth-order Bessel function.
single-bounced components become Expressions for other space-time correlation functions based
on the “one-ring” model [18], [19] can be similarly obtained.
SBT −j λ (q̃−q)δR cos θR +j2πτ fR max cos γR
2π
The temporal correlation function for M-to-M channels, assum-
Rpq, p̃q̃ (δT , δR , τ ) = ηT e
  ing 2-D isotropic scattering, J0 (2πfT max τ )J0 (2πfR max τ )
I0 x2SBT + ySBT 2 [1] is obtained for K = 0, ηT = ηR = 0, kT = kR = 0,
× (31) and δT = δR = 0. Similarly, the spatial correlation function
I0 (kT ) for M-to-M channels J0 (2π(p̃ − p)δT /λ)J0 (2π(q̃ − q)δR /λ)
SBR j 2π
λ (p̃−p)δT cos θT −j2πτ fT max cos γT [20] is obtained for K = 0, ηT = ηR = 0, kT = kR =
Rpq, p̃q̃ (δT , δR , τ ) = ηR e
  0, and τ = 0. Finally, the space-time correlation function
I0 x2SBR + ySBR
2 for M-to-M channels,assuming 2-D isotropic scattering,
× . (32) J0 ( x2DB + yDB 2 )J ( z 2
0
2
DB + wDB ) [9] is obtained for K =
I0 (kR )
0, ηT = ηR = 0, and kT = kR = 0.
Since αT and αR are independent random variables, the The space-time correlation functions of the I and Q com-
double integral in (28) reduces to the product of two single ponents of the complex faded envelope can be obtained by
integrals. By grouping the terms in (28) into those contain- substituting (16) and (17) into (18). Using similar derivations
ing αT and those containing αR , and by using trigonometric as above, it can be shown that for a large number of scatterers
transformations and the equality [16, eq. 3.338–4], the space- (M , N → ∞) and 2-D isotropic scattering the space-time
time correlation function of the double-bounced component correlation functions of the I and Q components of the complex
becomes faded envelope are
   
(I/Q,I/Q)
I0 x2DB + yDB
2 I0 zDB 2 + w2
DB
Rpq,p̃q̃ (δT , δR , τ ) =  {Rpq,p̃q̃ (δT , δR , τ )}
DB
Rpq, p̃q̃ (δT , δR , τ ) = ηT R = ηT cos (2πτ fR max cos γR −2π(q̃−q)(δR /λ) cos θR )
I0 (kT )I0 (kR )  
(33)
×J0 2 2
xiSBT +yiSBT + ηR
where parameters xDB , yDB , zDB , and wDB are xDB =
j2π(p̃ − p)δT cos θT /λ − j2 π τ fT max cos γ T + k T cos µ T , ×cos (2π(p̃−p)(δT /λ) cos θT −2πτ fT max cos γT )
   
yDB = j2π(p̃−p)δT sin θT /λ−j2πτ fT max sin γT +kT sinµT ,
×J0 2 2
xiSBR +yiSBR + ηT R J0 2 2
xiDB +yiDB
zDB = j2π(q̃−q)δR cosθR /λ−j2πτ fR max cos γR +kR cosµR,
720 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008

  
 α2 − β 2 [16, eq. 6.677–3], the sD-psds of the single-bounced
×J0 2
ziDB 2
+wiDB + Kpq Kp̃q̃
components become
×cos [2π(δT /λ) cos θT −2π(δR /λ) cos θR  SBT  ηT
+ 2πτ fT max cos γT −2πτ fR max cos γR ] (36) F Rpq, p̃q̃ (δT , δR , τ ) =
I0 (kT )
(I,Q)
Rpq,p̃q̃ (δT , δR , τ ) =  {Rpq,p̃q̃ (δT , δR , τ )} exp{−j2πpxSBR −j2π(f −fR max cos γR )ASBT }
× 
= ηT sin (2πτ fR max cos γR −2π(q̃−q)(δR /λ) cos θR )
  πfT max 1− [(f −fR max cos γR )/fT max ]2

×J0 x2iSBT +yiSBT
2 +ηR
×cosh (kT sin(µT −γT)+j2πpxSBT qySBT −j2πpySBT qxSBT )
× sin (2π(p̃−p)(δT /λ) cos θT −2πτ fT max cos γT )   
  2
 × 1− (f −fR max cos γR )/fT max (40)
×J0 x2iSBR +yiSBR
2 + Kpq Kp̃q̃
 SBR  ηR
×sin [2π(δT /λ) cos θT −2π(δR /λ) cos θR F Rpq,p̃q̃ (δT , δR , τ ) =
I0 (kR )
+2πτ fT max cos γT −2πτ fR max cos γR ] (37) exp {j2πpxSBT −j2π(f +fT max cos γT )ASBR }
×   2
where parameters yiSBT , yiSBR , xiSBT , xiSBR , xiDB , yiDB , πfR max 1− (f +fT max cos γT )/fR max

ziDB , and wiDB are equal to yiSBT = 2π[(p̃ − p)(δT /λ) ×
sin θT + (q̃ − q) (δ R / λ) ∆ T sin θ R − τ ( f T max sin γ T + ×cosh (kR sin(µR −γR )+j2πpxSBR qySBR−j2πpySBR qxSBR)
yiSBR = 2π[((q̃ − q)(δR /λ) sin θR +   
fR max ∆T sin γR )], 2
(p̃ −p)(δT /λ)∆R sin θT )−τ (fT max ∆R sinγT +fR max sinγR )], × 1− (f +fT max cos γT )/fR max (41)
xiSBT = 2π[(p̃ −p)(δT /λ)cos θT −τ fT max cos γT ], xiSBR =
2π[(q̃ − q)(δR /λ) cos θR − τ fR max cos γR ], xiDB = 2π[(p̃ − where cosh(·) is the hyperbolic cosine, |f − fR max cos γR | ≤
p)(δT /λ) cos θT − τ fT max cos γT ], yiDB = 2π[(p̃ − p)(δT / fT max and |f + fT max cos γT | ≤ fR max .
λ) sin θT −τ fT max sin γT ], ziDB = 2π[(q̃−q)(δR /λ) cos θR − Note that existing power spectral densities derived as-
τ fR max cos γR ], wiDB = 2π[(q̃ − q)(δR /λ) sin θR − suming “one-ring” model, are special cases of (41). The
τ fR max sin γR ]. The derivations are omitted for brevity. simplest
The space-Doppler power spectral density (sD-psd) of the  special case of (41) is Clarke’s power spectrum
1/(π fR2 max − f 2 ), |f | ≤ fR max [17], obtained for kR =
complex faded envelope is the Fourier transformation of the 0 (2-D isotropic scattering around Rx ), fT max = 0, kT = 0
space-time correlation function. We start the derivation for (fixed Tx , no scattering around Tx ), and δT = δR = 0 (single-
the sD-psds of thesingle-bounced components  by noting that antenna Tx and Rx ). Similarly, the 2-D non-isotropic power
the functions I0 ( x2SBT + ySBT 2 ) and I0 ( x2SBR + ySBR2 ) spectral density in (3) of [13] is obtained for fT max = 0 and
can be written as δT = δR = 0.
 We start the derivation for the sD-psd of the double-bounced
component by noting that
I0 j2πfT max
    

 F I0 x2DB +yDB2 2 +w 2
 2 I0 zDB DB
pxSBT qySBT pySBT qxSBT wySBT       
× (τ ASBT)2+ − +
fT max fT max 2πfT max = F I0 2
xDB +yDB 2 F I0 2
zDB +wDB 2 (42)
 (38)
where  denotes convolution. Using (42) and similar reason-
I0 j2πfR max ing as above, the sD-psd of the double-bounced component
becomes


 2  DB  ηT R
p q p q w  F Rpq, 
p̃q̃ (δT , δR , τ ) =
xSBR ySBR y x ySBR
× (τ−ASBR)2+ − SBR SBR + I0 (kT )πfT max 1−(f /fT max )2
fRmax fRmax 2πfRmax
×e−j(2πp x
 DB DB
qx +2πpyDB qyDB +jkT cos(γT −µT ))f /fT max
(39)
×cosh (kT sin (µT −γT )+j2πpxDB qyDB −j2πpyDB qxDB )
where pxSBT = (p̃−p)(δT /λ) cos θT , qxSBT = cos γT , ASBT = 

(2πp x SBT q x SBT + 2πp y SBT q y SBT + w x,SBT )/2πf T max , × 1−(f /fT rmmax ) 2
pySBT = [(p̃−p)δT sin θT +(q̃−q)δR ∆T sin θR ]/λ, qySBT ≈
sin γT , ASBR = (2πpxSBR qxSBR +2πpySBR qySBR +wxSBR )/ e−j(2πpzDB qzDB +2πpwDB qwDB +jkR cos(γR −µR ))f /fR max
2πfR max , wxSBT = −jkT cos(γT − µT ), wySBT =  
I0 (kR )πfR max 1−(f /fR max )2
−jkT sin(γT − µT ), pxSBR = (q̃−q)(δR /λ) cos θR , qxSBR = 
cos γR , wxSBR = −jkR cos(γR − µR ), pySBR = [(p̃ − p) × ×cosh (kR sin(µR −γR )+j2πpzDB qwDB −j2πpwDB qzDB )
δT ∆R sin θT + (q̃ − q)δR sin θR ]/λ, qySBR ≈ sin γR , and 
wySBR = −jkR sin(γR − µR ). Using (38), (39), and 
∞ √ × 1−(f /fR max ) 2 (43)
equality 0 I0 (α t2 + u2 ) cos(βt)dt = cosh(u α2 − β 2 )/
ZAJIĆ AND STÜBER: SPACE-TIME CORRELATED MOBILE-TO-MOBILE CHANNELS: MODELLING AND SIMULATION 721

Fig. 3. Comparison of the normalized Doppler power spectra in (43) and (44) Fig. 4. Normalized space-Doppler power spectra characteristic for the out-
for δT = δR = 0 and different rations s = fR max /fT max . door M-to-M micro-cells. The curves are obtained with parameters s =
fR max /fT max = 1, θT = π/3, θR = π/4, γT = γR = π/2, ηT = ηR =
0.4, ηT R = 0.2, kT = kR = 0, K = 10, and Lt = Lr = 2.
where pxDB = (p̃ − p)(δT /λ) cos θT , qxSBT = cos γT ,
pyDB = (p̃ − p)(δT /λ) sin θT , qySBT = sin γT , pzDB =
(q̃ − q)(δR /λ) cos θR , qzDB = cos γR , pwDB = (q̃ −
q)(δR /λ) sin θR , qwDB = sin γR , and |f | ≤ fT max + fR max .
For 2-D isotropic scattering and δT = δR = 0 (SISO sys-
tem), the sD-psd of the double-bounced component has the
closed-form expression [1]
 DB  1
F Rpq, p̃q̃ (0, 0, τ ) = 2 √
 π sf 

T max
 2
1 + s f
×K  √ 1−  (44)
2 s (1 + s)fT max

where K[·] is the complete elliptic integral of the first kind

and s is the ratio of the maximum Doppler frequencies fR max
and fT max , i.e., s = fR max /fT max . To validate the derived
sD-psd in (43), we compare this expression for 2-D isotropic
scattering and δT = δR = γT = γR = 0 with the sD-psd in
(44). For different ratios s = fR max /fT max , Fig. 3 shows good
agreement between the sD-psds in (43) and (44).
By calculating the Fourier transformation of the space-time
correlation function in (35), we obtain the space-Doppler power
spectral density of the LoS component
Fig. 5. Normalized space-Doppler power spectra characteristic for the out-
door M-to-M macro-cells. The curves are obtained with parameters s =
fR max /fT max = 1, θT = π/3, θR = π/4, γT = γR = π/2, ηT = ηR =
0.1, ηT R = 0.8, kT = kR = 0, K = 10, and Lt = Lr = 2.
receiver. In Fig. 4, we assume that the single-bounced rays bear
more energy than the double-bounced rays, i.e., ηT = ηR = 0.4
 LoS 
F Rpq, p̃q̃ (δT , δR , τ )

= Kpq Kp̃q̃ ej λ [(p̃−p)δT cos θT −(q̃−q)δR cos θR ]
2π
× δ(f − fT max cos γT + fR max cos γR ) (45)
where δ(·) denotes the Dirac delta function.
Finally, the sD-psd Spq,p̃q̃ (δT , δR , f ) between two com-
plex faded envelopes hpq (t) and hp̃q̃ (t) becomes a sum-
SBT SBR
mation of the sD-psds Spq, p̃q̃ (δT , δR , f ), Spq,p̃q̃ (δT , δR , f ),
DB LoS
Spq,p̃q̃ (δT , δR , f ), and Spq,p̃q̃ (δT , δR , f ), defined as in (40),
(41), (43), and (45), respectively.
In Figs. 4 and 5, we plot the sD-psd, Spq,p̃q̃ (δT , δR , f ), for
several MIMO systems with parameters s = fR max /fT max =
1, δT = δR ∈ {0.5λ, 1λ, 5λ, 10λ}, θT = π/3, θR = π/4,
γT = γR = π/2, and Lt = Lr = 2. In these figures, we an-
alyze the radio propagation in outdoor micro- and macro-
cells, assuming 2-D isotropic scattering (kT = kR = 0) and
line-of-sight (K = 10) conditions between the transmitter and
and ηT R = 0.2, which is characteristic of outdoor micro-cell
propagation. We can observe that this spectrum is similar to the
U-shaped spectrum of F-to-M cellular channels. In Fig. 5, we
consider the macro-cell propagation, i.e. the double-bounced
rays bear more energy than the single-bounced rays (ηT =
ηR = 0.1 and ηT R = 0.8). In this case, the sD-psd differs from
the U-shaped spectrum of cellular channels. Finally, our the-
oretical Doppler spectra closely match the measured Doppler
spectra in [6] and [7].
IV. MIMO M OBILE - TO -M OBILE S IMULATION M ODELS
The reference models for MIMO M-to-M channels described
in [9] and Section II assume an infinite number of scatterers,
which prevents practical implementation. It is desirable to
design simulation models with a finite number of scatterers,
while still matching the statistical properties of the reference
models.
722 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008

 
A. MMEDS Model (m)
×cos αT −γT +Kq ∆T sin θR sin αT
(m)

Hogstad et al. [10] were the first to propose a determinis-

(m)
tic MIMO M-to-M sum-of-sinusoids (SoS) model called the + 2πtfR max ∆T sin γR sin αT +φm
Modified Method of Exact Doppler Spreads (MMEDS). The N
2PR
MMEDS model describes radio propagation in outdoor macro- + √ cos[βn +Kp cos θT +2πtfT max cos γT ]
cells by assuming that all rays are double-bounced. By choosing N
n=1
 
only the phases to be random variables, the statistical properties (n)
×cos Kq cos αR −θR +2πtfR max
of this model converge to those of the reference model in a  
(n) (n)
single simulation trial. Using the reference model in [9] and ×cos αR −γR +Kp ∆R sin θT sin αR
assuming 2-D scattering, they defined the received complex

(n)
faded envelope as + 2πtfT max ∆R sin γT sin αR +φn

M,N
2P  

M,N
ap,m cm,n bn,q + √ T R cos Kp cos θT −αT(m)
hpq (t) = √ m,n=1
MN
m,n=1
MN  

 (m) (n)
 (m)
+ 2πtfT max cos αT −γT
j 2πfR max t cos(αT −γT )+2πfR max t cos(αR −γR )+φmn
×e (46)  
(n)
×cos Kq cos αR −θR +2πtfR max
(m)
where ap,m = exp {jπ(Lt + 1 − 2p)δT cos(αT − θT )/λ},  

(n)
(m) (n) ×cos αR −γR +φmn
cm,n = exp {j2π(Rt cos αT − Rr cos αR )/λ}, bn,q =   LoS 
(n)
exp {jπ(Lr + 1 − 2q)δR cos(αR − θR )/λ}, and parameters + PLoS cos Kp cos θT + Kq cos αRq θR
 LoS LoS

p and q take values from the sets p ∈ {1, . . . , Lt } and + 2πt fT + fR (49)
q ∈ {1, . . . , Lr }, respectively. The parameters θT , θR , γT ,  2PT
M
γR , fT max , and fR max are defined as in Section II. With the h(Q)
pq (t) = √ sin[βm −Kq cos θR −2πtfR max cos γR ]
MMEDS model, the phases φmn are independent random M
m=1
 
variables uniformly distributed on the interval [0, 2π), and the ×cos Kp sin θT −αT
(m)
+2πtfT max
(m) (n)
angles of departure αT and the angles of arrivals αR are  
(m) (m)
chosen as follows: ×sin αT −γT +Kq ∆T sin θR cos αT

(m)
(m) 2π + 2πtfR max ∆T sin γR cos αT +φm
αT = (m − 0.5) + γT (47)
M N
2π 2PR
(n)
αR = (n − 0.5) + γR (48) + √ sin[βn +Kp cos θT +2πtfT max cos γT ]
N N
n=1
 
(n)
for m = 1, . . . , M , n = 1, . . . , N , respectively. ×cos Kq sin αR −θR +2πtfR max
As we will show, the MMEDS model has the disadvantage  
(n) (n)
that the statistical properties match those of the reference ×sin αR −γR +Kp ∆R sin θT cos αR

model only for a small range of normalized time delays (0 ≤ (n)

+ 2πtfT max ∆R sin γT cos αR + φn
fT max Ts ≤ 5). Furthermore, the model requires large number
of scatterers (i.e., M , N = 40) to match statistical properties of 
M,N
2P  
the reference model. Finally, the MMEDS model assumes that + √ T R sin Kp cos θT −αT(m)
MN
all rays in outdoor macro-cells are double-bounced, which is m,n=1
 

not always the case. (m)

+ 2πtfT max cos αT −γT
 
(n)
×sin Kq sin αR −θR +2πtfR max
B. New Simulation Models  

(n)
×sin αR −γR +φmn
In this section, we propose more general simulation mod-   LoS 
els that describe propagation in outdoor micro- and macro- + PLoS sin Kp cos θT +Kq cos αRq −θR
cells. In our models, the complex faded envelope is designed  
+ 2πt fTLoS +fRLoS (50)
as a superposition of the LoS, the single-bounced, and the
double-bounced rays. Assuming 2-D isotropic scattering and are the in-phase (I) and quadrature (Q) components of the
using the reference model described in Section II, we propose
 faded envelope, βm and β
complex n are the random path gains,
the following function as a received complex faded envelope: PT = ηT Ωpq /(Kpq + 1), PR = ηR Ωpq /(Kpq + 1), PT R =
(I) (Q) 
hpq (t) = hpq (t) + jhpq (t), where η Ω /(Kpq +1), Kq = 2π(0.5Lr +0.5−q)δR /λ, PLoS =
 T R pq
Kpq Ωpq/(1+Kpq ), Kp = 2π(0.5Lt +0.5−p)δT /λ, fTLoS =
M
2P fT max cos(π−αRqLoS
−γT ) and fRLoS = fR max cos(αRq
LoS
− γR ).
h(I)
pq (t) = √ T cos[βm −Kq cos θR −2πtfR max cos γR ]
M The I and Q components in (49) and (50) are constructed
m=1
  as follows: the single-bounced components of the complex
(m)
×cos Kp cos θT −αT +2πtfT max faded envelope are constructed using a similar method to one
ZAJIĆ AND STÜBER: SPACE-TIME CORRELATED MOBILE-TO-MOBILE CHANNELS: MODELLING AND SIMULATION 723

proposed in [21] for the SISO F-to-M model, whereas the

double-bounced components of the complex faded envelope are
constructed using a similar method to one proposed in [5] for
the SISO M-to-M model.
We first propose an ergodic statistical (deterministic) model,
which needs only one simulation trial to obtain the desired
statistical properties. The complex faded envelope is hpq (t) =
(I) (Q) (I) (Q)
hpq (t) + jhpq (t), where functions hpq (t) and hpq (t) are
defined in (49) and (50). For this model, the phases φmn are
generated as independent random variables uniformly distrib-
uted on the interval [−π, π). The AoDs, AoAs and random path
gains are chosen as follows:
(m) π
αT = (m − 0.5) + γT (51)
M
(n) 2π
αR = (n − 0.5) + γR (52)
N Fig. 6. Theoretical and simulated normalized space-time correlation functions
π(m − 0.5) of the I and Q components of the deterministic model for the radio propagation
βm = (53) in outdoor micro-cells.
M
2π(n − 0.5)
βn = (54)
N
for m = 1, . . . , M , n = 1, . . . , N , respectively. Note that the
AoDs, AoAs, and random path gains differ from those in
[21]. For M , N → ∞, our deterministic model can be shown
to exhibit properties (36) and (37) of the reference model.
Derivations are omitted for brevity.
By allowing amplitudes, phases, and Doppler frequencies to
be random variables, our deterministic model can be modified
to match statistical properties of the reference model over a
wider range of normalized time delays, while at the same time
requiring a smaller number of scatterers around the Tx and Rx .
The statistical properties of this model vary for each simulation
trial, but will converge to desired ensemble averaged properties
when averaged over a sufficient number of simulation trials.
(I) (Q)
The complex faded envelope is hpq (t) = hpq (t) + jhpq (t), Fig. 7. Theoretical and simulated normalized space-time correlation functions
(I) (Q)
where functions hpq (t) and hpq (t) are defined as in (49) and of the I and Q components of the deterministic model for the radio propagation
(50), respectively. The angles of departures and the angles of in outdoor macro-cells.
arrivals are chosen as follows:
  Figs. 6 and 7 present the space-time correlation functions
(m) 2πm ψ − π (δT = δR = 0.5λ) of the I and Q components of the complex
αT = 0.5 + (55)
M M faded envelope, respectively, for system with Lt = Lr = 4 an-
(n) 2πn θ − π tennas, using M = 22 and N = 40 scatterers in the deterministic
αR = + (56) model. In Fig. 6, we assume that the single-bounced rays
N N
bear more energy than the double-bounced rays, i.e., ηT = 0.7,
for m = 1, . . . , M , n = 1, . . . , N , respectively. The parameters ηR = 0.2, and ηT R = 0.1. In Fig. 7, we assume that the double-
φmn , βm , βn , θ, and ψ are independent random variables bounced rays bear more energy than the single-bounced rays,
uniformly distributed on the interval [−π, π). Our statistical i.e., ηT = ηR = 0.2 and ηT R = 0.6. Results show that the
model can be shown to exhibit properties (36) and (37) of the space-time correlation functions of the deterministic model
reference model. Derivations are omitted for brevity. approach the theoretical ones in the range of normalized time
delays, 0 ≤ fT max Ts ≤ 6.
Figs. 8 and 9 present the space-time correlation functions
V. S IMULATION R ESULTS
(δT = δR = 0.5λ) of the I and Q components of the complex
This section presents simulation results of our new mod- faded envelope, respectively, for a system with Lt = Lr = 4
els. In all simulations, we use a normalized sampling period antennas, using N = M = 16 scatterers and Nstat = 50 sim-
fT max Ts = 0.01 (fT max = fR max are the maximum Doppler ulation trials in the statistical model. In Fig. 8, we assume
frequencies and Ts is the sampling period). The orientations of that ηT = 0.5, ηR = 0.4, and ηT R = 0.1, whereas in Fig. 9,
the Tx and Rx antenna arrays are chosen to be θT = θR = π/2. we assume that ηT = ηR = 0.05 and ηT R = 0.9. Results show
The angles of motion for the Tx and Rx are chosen to be that the space-time correlation functions of the statistical model
γT = π/4 and γR = 0, respectively. The Rice factor is K = 0.2 approach the theoretical ones in the range of normalized time
and the AoA of the specular component is αRqLoS
= π. delays, i.e. 0 ≤ fT max Ts ≤ 10.
724
Fig. 8. Theoretical and simulated normalized space-time correlation functions
of the I and Q components of the statistical model for the radio propagation in
outdoor micro-cells.
Fig. 9. Theoretical and simulated normalized space-time correlation functions
of the I and Q components of the statistical model for the radio propagation in
outdoor macro-cells.
Here we compare the performance and complexity of our
new models with the MMEDS model. For fair comparison,
we remove the LoS and single-bounced components of the
complex faded envelope from our models. Fig. 10 compares
the temporal auto-correlation functions of the complex faded
envelopes obtained by the deterministic, the statistical, and the
MMEDS model. For the MMEDS model and the deterministic
model, the temporal correlation functions are obtained using
M = 40, N = 38 and M = 22, N = 40 scatterers, respec-
tively. For the statistical model, the temporal correlation func-
tion is obtained using N = M = 14 scatterers and Nstat = 50
trials. Fig. 10 shows that the temporal correlation function of
the MMEDS model approaches the theoretical one for normal-
ized time delays in the range 0 ≤ fT max Ts ≤ 6. The temporal
correlation function of the deterministic model approaches
the theoretical one in the same range of the normalized time
delays as the MMEDS model, but requires a smaller number
of scatterers. Finally, the temporal correlation function of the
statistical model approaches the theoretical one in a wider range
of normalized time delays (i.e., 0 ≤ fT max Ts ≤ 10) than the
deterministic and the MMEDS model.
Fig. 11 compares the space-time correlation functions of
the complex faded envelopes for a system with Lt = Lr = 2
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008
Fig. 10. Normalized temporal auto-correlation function (δT = δR = 0) of
the complex faded envelope of the statistical, deterministic, MMEDS, and
reference model.
Fig. 11. Normalized space-time correlation function (δT = δR = 1λ) of
the complex faded envelope of the statistical, deterministic, MMEDS, and
reference model.
antennas, obtained by the deterministic, the statistical, and the
MMEDS model. For the MMEDS model and the deterministic
model, the space-time correlation functions are obtained us-
ing the same number of scatterers as in Fig. 10 and antenna
distances δT = δR = 1 λ. For the statistical model, the space-
time correlation function is obtained using N = M = 22 scat-
terers, antenna distances δT = δR = 1 λ, and Nstat = 50 trials.
Fig. 11 shows that the space-time correlation function of the
MMEDS model approaches the theoretical one for normalized
time delays in the range 0 ≤ fT max Ts ≤ 5. The space-time
correlation function of the deterministic model approaches
the theoretical one in a somewhat wider range of normalized
time delays (i.e., 0 ≤ fT max Ts ≤ 6), while requiring a smaller
number of scatterers than the MMEDS model. The space-time
correlation function of the statistical model approaches the
theoretical one in the widest range of normalized time delays
(i.e., 0 ≤ fT max Ts ≤ 10).
Simulation results presented in Fig. 11 indicate that larger
distances between antenna elements require a larger number of
scatterers to match theoretical statistics for the same normalized
time delays as in Fig. 10. For the statistical model, instead of
using many scatterers (e.g., 40) in all simulations, we propose
to adaptively select the number of scatterers depending on
ZAJIĆ AND STÜBER: SPACE-TIME CORRELATED MOBILE-TO-MOBILE CHANNELS: MODELLING AND SIMULATION 725

TABLE I statistical SoS simulation models for MIMO M-to-M fading

C OMPLEXITY OF D IFFERENT M ODELS
channels are proposed. The statistical properties of the sim-
ulation models are verified by simulations. The results show
that the simulation models are a good approximation of the
reference model and that they outperform existing simulation
models.

A PPENDIX
D ERIVATIONS OF E QUATIONS (6)–(11)
From Fig. 1, using the cosine law, the distances pm , mq ,
pn , and nq can be written as

2pm = [(0.5Lt + 0.5 − p)δT ]2 + RT2

 
(m)
− 2(0.5Lt + 0.5 − p)δT Rt cos θT − αT (57)
2
the distances between antenna elements. If we assume that 2mq = [(0.5Lr + 0.5 − q)δR ] + 2m
 
a mean square error (MSE) of ≤ 10−3 is required for the (m)
− 2(0.5Lr + 0.5 − q)δR m cos αR − θR (58)
simulated space-time correlation function of the complex faded
2
envelope (relative to the theoretical one), we need at least M = 2pn = [(0.5Lt + 0.5 − p)δT ] + 2n
14 + 2δT /0.3λ and N = 14 + 2δR /0.3λ scatterers.1 This  
(n)
− 2(0.5Lt + 0.5 − p)δT n cos θT − αT (59)
methodology can also be applied to the deterministic and
2
MMEDS models. However, to obtain similar statistics as with 2nq = [(0.5Lr + 0.5 − q)δR ] + 2
RR
the statistical model, the number of scatterers needs to be larger,  
(n)
i.e., 70 + 10δT /R /0.3λ . − 2(0.5Lr + 0.5 − q)δR Rr cos αR − θR . (60)
To compare complexity of the various models, Table I
(m)
summarizes the number of simulation trials and the number By applying the sine law to the triangles ∆OT ST OR and
of scatterers required to obtain similar statistical properties, (n)
∆OT SR OR , respectively, we obtain following identities:
the number of operations in one simulation trial needed to
generate the complex faded envelope and the relative simulation m Rt
=  
times (including averaging over Nstat simulation trials) needed (m)
sin αT sin π − αR
(m)
to generate the complex faded envelopes, in Matlab on a
Pentium III laptop. Here, we count only the frequently executed D
=    (61)
operations and the number of random variables. Table I shows (m) (m)
sin π − αT − π − αR
that the deterministic model has a lower complexity then the
Rr n
MMEDS model, whereas the statistical model has the highest
(n)
=  
(n)
complexity. Both of our new models require a smaller number sin αT sin π − αR
of scatterers than the MMEDS model, which leads to shorter D
simulation times. =    . (62)
(n) (n)
From Table I and Figs. 9–11 we can conclude that, our de- sin π − αT − π − αR
terministic model performs similar to the MMEDS model, but
has lower complexity and shorter simulation time. On the other (m)
From Fig. 1, we observe that π − αR ≤ arcsin(Rt /D)
hand, our statistical model, with slight increase in complexity, (n)
and αT ≤ arcsin(Rr /D). Based on the assumption
outperforms the deterministic and the MMEDS models. Finally,
min{Rt , Rr }
D, we can conclude that arcsin(Rt /D) ≈
our deterministic and statistical model are more general and
Rt /D = ∆T and arcsin(Rr /D) ≈ Rr /D = ∆R , and
can be used to model various outdoor micro- and macro-cell (m) (n)
propagation environments. consequently π − αR and αT are small angles. From (61)
√ (62), using approximations sin x ≈ x, cos x ≈ 1, and
and
1 + x ≈ 1 + x/2, for small x, the distances pm , mq , pn ,
VI. C ONCLUSION and nq become
This paper proposed the parametric reference model for (Lt + 1 − 2p)δt  
(m)
MIMO mobile-to-mobile fading channels. From this model, pm ≈ Rt − cos θT − αT (63)
2
the closed-form joint space-time correlation function and (Lr + 1 − 2q)δR

(m)
the space-Doppler power spectrum for a 2-D non-isotropic mq ≈D − ∆T sin θR sin αT − cos θR
2
scattering environment are derived. Furthermore, the space- (64)
time correlation functions of the in-phase and quadrature (Lt + 1 − 2p)δT (n)

components of the complex faded envelope for a 2-D isotropic pn ≈D − ∆R sin θT sin αR + cos θT
2
scattering environment are derived. Finally, deterministic and (65)
(Lr + 1 − 2q)δR  
(n)
nq ≈ Rr − cos αR − θR . (66)
1 Operation
· denotes rounding up to the next integer. 2
726 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008

From Fig. 1, by applying the sine law to the triangle [11] D. Gesbert, H. Bölcskei, D. A. Gore, and A. J. Paulraj, “Outdoor MIMO
(m) (n) wireless channels: Models and performance prediction,” IEEE Trans.
∆ST SR OR , we obtain the following identity: Commun., vol. 50, no. 12, pp. 1926–1934, Dec. 2002.
[12] J. Salz and J. H. Winters, “Effect of fading correlation on adaptive
Rr  arrays in digital mobile radio,” IEEE Trans. Veh. Technol., vol. 43, no. 4,
=  mn  pp. 1049–1057, Nov. 1994.
sin χ sin α (m) (n)
R − αR [13] A. Abdi, J. A. Barger, and M. Kaveh, “A parametric model for the distri-
 bution of the angle of arrival and the associated correlation function and
=  m  . (67) power spectrum at the mobile station,” IEEE Trans. Veh. Technol., vol. 51,
(m) (n)
sin π − χ − αR − αR no. 3, pp. 425–434, May 2002.
[14] K. I. Pedersen, P. E. Mogensen, and B. H. Fleury, “Power azimuth spec-
trum in outdoor environments,” Electron. Lett., vol. 33, no. 18, pp. 1583–
From (67), using approximation sin x ≈ x for small x, the 1584, Aug. 1997.
distance mn can be written as [15] K. I. Pedersen, P. E. Mogensen, and B. H. Fleury, “A stochastic model of
the temporal and azimuthal dispersion seen at the base station in outdoor
  propagation environments,” IEEE Trans. Veh. Technol., vol. 49, no. 2,
(m) (n)
sin αR − αR pp. 437–447, Mar. 2000.
mn = m    ≈ D. (68) [16] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
(m) (n)
sin χ + αR − αR Products, 5th ed. A. Jeffrey, Ed. San Diego, CA: Academic, 1994.
[17] R. H. Clarke, “A statistical theory of mobile-radio reception,” Bell Syst.
Tech. J., vol. 47, no. 6, pp. 957–1000, Jul./Aug. 1968.
From Fig. 2, using the cosine low, the distance pq can be [18] D. Shiu, G. J. Foschini, M. J. Gans, and J. M. Khan, “Fading correlation
written as and its effect on the capacity of multielement antenna systems,” IEEE
Trans. Commun., vol. 48, no. 3, pp. 502–513, Mar. 2000.
[19] A. Abdi and M. Kaveh, “A space-time correlation model for multielement
2pq = [(0.5Lr + 0.5 − q)δR ]2 + 2 antenna systems in mobile fading channels,” IEEE J. Sel. Areas Commun.,
 LoS 
− 2(0.5Lr + 0.5 − q)δR  cos αRq − θR (69) vol. 20, no. 3, pp. 550–560, Apr. 2002.
[20] H. Kang, G. L. Stüber, T. G. Pratt, and M. A. Ingram, “Studies on the
capacity of MIMO systems in mobile-to-mobile environment,” in Proc.
where 2 = [(0.5Lt + 0.5 − p)δT ]2 +D2√ − 2(0.5Lt + 0.5 − IEEE WCNC, Atlanta, GA, Mar. 2004, vol. 1, pp. 363–368.
p)δT D cos θT . Using the approximation 1 + x ≈ 1 + x/2, [21] A. G. Zajić and G. L. Stüber, “Efficient simulation of Rayleigh fading
with enhanced de-correlation properties,” IEEE Trans. Wireless Commun.,
the distance pq becomes vol. 5, no. 7, pp. 1866–1875, Jul. 2006.
[22] G. L. Stüber, Principles of Mobile Communication 2E. Norwell, MA:
pq ≈D − (0.5Lt + 0.5 − p) δT cos θT Kluwer, 2001.
 LoS 
− (0.5Lt + 0.5 − q) δR cos αRq − θR . (70)
Alenka G. Zajić (S’99) received the B.Sc. and
M.Sc. degrees from the School of Electrical Engi-
ACKNOWLEDGMENT neering, University of Belgrade, Belgrade, Serbia,
in 2001 and 2003, respectively. From 2001 to 2003,
The authors would like to thank the anonymous reviewers she was a Design Engineer for Skyworks Solutions
Inc., Fremont, CA. She is currently pursuing the
whose feedback helped improve the quality of this paper. Ph.D degree with the School of Electrical and Com-
puter Engineering, Georgia Institute of Technology
(Georgia Tech), Atlanta.
R EFERENCES Since 2004, she has been a Graduate Research
[1] A. S. Akki and F. Haber, “A statistical model of mobile-to-mobile land Assistant with the Wireless Systems Laboratory,
communication channel,” IEEE Trans. Veh. Technol., vol. VT-35, no. 1, Georgia Tech. Her research interests are in wireless communications and
pp. 2–7, Feb. 1986. applied electromagnetics.
[2] A. S. Akki, “Statistical properties of mobile-to-mobile land communica- Ms. Zajić was recipient of the Dan Noble Fellowship in 2004, awarded by
tion channels,” IEEE Trans. Veh. Technol., vol. 43, no. 4, pp. 826–831, Motorola Inc. and the IEEE Vehicular Technology Society, for quality impact
Nov. 1994. in the area of vehicular technology.
[3] R. Wang and D. Cox, “Channel modeling for ad hoc mobile wireless
networks,” in Proc. IEEE VTC, Birmingham, AL, May 2002, vol. 1,
pp. 21–25. Gordon L. Stüber (S’81–M’82–SM’96–F’99)
[4] C. S. Patel, G. L. Stüber, and T. G. Pratt, “Simulation of received the B.A.Sc. and Ph.D. degrees in
Rayleigh-faded mobile-to-mobile communication channels,” IEEE Trans. electrical Engineering from the University of
Commun., vol. 53, no. 11, pp. 1876–1884, Nov. 2005. Waterloo, Waterloo, ON, Canada, in 1982 and 1986,
[5] A. G. Zajić and G. L. Stüber, “A new simulation model for mobile-to- respectively.
mobile Rayleigh fading channels,” in Proc. IEEE WCNC, Las Vegas, NV, In 1986, he joined the School of Electrical and
Apr. 2006, vol. 3, pp. 1266–1270. Computer Engineering, Georgia Institute of Technol-
[6] J. Maurer, T. Fügen, and W. Wiesbeck, “Narrow-band measurement and ogy, Atlanta, where he is the Joseph M. Pettit Chair
analysis of the inter-vehicle transmission channel at 5.2 GHz,” in Proc. Professor.
IEEE VTC, Birmingham, AL, May 2002, vol. 3, pp. 1274–1278. Dr. Stüber was co-recipient of the IEEE Vehicular
[7] G. Acosta, K. Tokuda, and M. A. Ingram, “Measured joint Doppler-delay Technology Society Jack Neubauer Memorial Award
power profiles for vehicle-to-vehicle communications at 2.4 GHz,” in in 1997 for the best systems paper. He became an IEEE Fellow in 1999 for con-
Proc. IEEE GLOBECOM, Dallas, TX, Nov. 2004, vol. 6, pp. 3813–3817. tributions to mobile radio and spread spectrum communications. He received
[8] G. J. Byers and F. Takawira, “Spatially and temporally correlated MIMO the IEEE Vehicular Technology Society James R. Evans Avant Garde Award in
channels: Modeling and capacity analysis,” IEEE Trans. Veh. Technol., 2003 for contributions to theoretical research in wireless communications. He
vol. 53, no. 3, pp. 634–643, May 2004. served as General Chair and Program Chair for several conferences, including
[9] M. Pätzold, B. O. Hogstad, N. Youssef, and D. Kim, “A MIMO mobile- VTC’96, ICC’98, MMT”00, CTW’02, and WPMC’02. He is a past Editor for
to-mobile channel model: Part I—The reference model,” in Proc. IEEE IEEE TRANSACTIONS ON COMMUNICATIONS (1993–1998) and served on the
PIMRC, Berlin, Germany, Sep. 2005, vol. 1, pp. 573–578. IEEE Communications Society Awards Committee (1999–2002). He was an
[10] B. O. Hogstad, M. Pätzold, N. Youssef, and D. Kim, “A MIMO mobile- elected member of the IEEE Vehicular Technology Society Board of Governors
to-mobile channel model: Part II—The simulation model,” in Proc. IEEE (2001–2003, 2004–2006) and received the Outstanding Service Award from the
PIMRC, Berlin, Germany, Sep. 2005, vol. 1, pp. 562–567. IEEE Vehicular Technology Society.