Implementation of the COST 273 Directional Channel Model
in Microcell Scenarios
Ivo Sousa, Maria Paula Queluz and Ant´onio Rodrigues
Instituto de Telecomunicac¸˜oes/Instituto Superior T´ecnico, Technical University of Lisbon, Lisbon, Portugal
Keywords:
COST 273 Directional Channel Model, Microcells, Wireless Communications.
Abstract:
This paper presents a tutorial on how to implement the COST 273 Directional Channel Model (DCM) for
microcell scenarios. Special care has been taken to present all the parameters models and values required
by the DCM, being some of them proposed in this work because they were missing in the related literature
and are essential. The results and comparison with experimental data of an implementation example are also
presented, which prove that this DCM is suitable for wireless systems development, especially those that
exploit spatial aspects of radio channels, like for example Multiple-Input Multiple-Output (MIMO) systems.
1 INTRODUCTION
Wireless communication systems can be imple-
mented with multiple antennas at the transmitter and
at the receiver, in order to exploit spatial differences
of a radio channel. These systems, commonly known
as Multiple-Input Multiple-Output (MIMO) systems,
allow to improve the spectral efficiency by offering a
spatial multiplexing gain, or allow to improve the link
reliability by offering a diversity gain (Foschini and
Gans, 1998; Telatar, 1999). These gains are achieved
without increased transmit power or additional band-
width, which is MIMO systems major advantage.
For a correct MIMO system development it is
important to understand and characterize the propa-
gation phenomena between Base Stations (BSs) and
Mobile Stations (MSs). This phenomena depends
not only on the wavelength and distance between
BSs and MSs, but also on all the Interacting Objects
(IOs) present in the surrounding environment where
waves bounce. These IOs are designated as scatter-
ers since they induce scattering, where one or more
non-uniformities in the medium force radio waves to
deviate from a straight trajectory. To properly sim-
ulate the radio wave propagation and incorporate re-
alistic features, like time-variant characteristics, fre-
quency selective responses and correct spatial repre-
sentation of the scenarios, Directional Channel Mod-
els (DCMs) have been developed, which aim to be
easy-to-use methods in order to avoid extensive mea-
surement campaigns. A survey about DCMs classifi-
cation can be found in (Almers et al., 2007).
The aim of this paper is to provide a tutorial on the
COST 273 DCM implementation for microcells. All
parameters models and values required by the DCM
are presented, including some that had to be empiri-
cally defined because they were missing in the liter-
ature and are essential. Following this introduction,
Section 2 describes the COST 273 DCM applied to
microcells. An implementation example is presented
in section 3, followed by final remarks in section 4.
2 COST 273 DCM – MICROCELLS
COST is an intergovernmental framework for Euro-
pean Cooperation in Science and Technology, allow-
ing a European coordination of nationally-funded re-
search. DCMs were developed within the Information
and Communication Technologies COST domain.
The COST 273 DCM (Correia, 2006) is a general
model, since it is defined for several radio environ-
ments and uses an identical generic model for all cell
types (macro-, micro- and picocells). This DCM is
the successor of the COST 259 DCM (Correia, 2001),
being the main update a more realistic modeling of
the multiple-bounce scattering effect (especially im-
portant for micro-and picocellscenariossimulations).
The latest COST DCM is the COST 2100 DCM (Ver-
done and Zanella, 2012), but since this DCM is very
recent, it is not yet fully parametrized, fully imple-
mented and widely used (for example, parameters for
outdoor scenarios remain important missing parts).
349
Sousa I., Paula Queluz M. and Rodrigues A..
Implementation of the COST 273 Directional Channel Model in Microcell Scenarios.
DOI: 10.5220/0004058303490356
In Proceedings of the International Conference on Signal Processing and Multimedia Applications and Wireless Information Networks and Systems
(WINSYS-2012), pages 349-356
ISBN: 978-989-8565-25-9
Copyright
c
 2012 SCITEPRESS (Science and Technology Publications, Lda.)

Table 1: External parameters.
Parameter Symbol Validity range
Carrier freq. [GHz] f
c
1–5
BS/MS height [m] h
BS
/h
MS
3–10/1.5
BS/MS position [m] ~r
BS
/~r
MS
(0,0,h
BS
)/Any
Cell radius [m] R
cell
Any
2.1 Model General Structure
The COST 273 DCM follows an approach where
physical positions of the transmitter, receiver, scat-
terers and their effect on the electromagnetic waves
are defined so that the Double-Directional Channel
Impulse Response (DDCIR) is obtained as a super-
position of MultiPath Components (MPCs). In other
words, the DDCIR is given by (Molisch et al., 2006)
h
(t,τ,Ω,Ψ) =
L(~r)
∑
l=1
a
l
δ(τ−τ
l
)δ(Ω−Ω
l
)δ(Ψ−Ψ
l
)
(1)
where t is the absolute time, τ is the delay variable, Ω
and Ψ are spatial angles characterizing the MPCs’ Di-
rection of Arrival (DoA) and Direction of Departure
(DoD), respectively,~r denotes the location of the re-
ceiver antenna with respect to the transmitter antenna,
a
l
represents the l
th
MPC complex amplitude (polari-
metric 2 ×2 matrix) and δ(·) is the Dirac delta func-
tion.
The description of the DDCIR in (1) is antenna
independent. The non-directional time-variant Chan-
nel Impulse Response (CIR) can be obtained by in-
tegrating over the DoDs and DoAs weighted by the
transmitter and receivercomplex polarimetricantenna
gains
~
G
T
(Ψ) and
~
G
R
(Ω), respectively, i.e.,
h(t,τ) =
Z
Ψ
Z
Ω
h(t,τ,Ω, Ψ)
~
G
T
(Ψ)
~
G
R
(Ω)dΩdΨ
(2)
2.2 Parameters Models and Settings
2.2.1 External Parameters
Table 1 presents the external parameters required by
the COST 273 DCM for microcells (Correia, 2006),
which are user-supplied but have validity ranges.
The COST 273 DCM uses as path-loss model for
microcells the same one as the COST 259 DCM,
which is described in (Feuerstein et al., 1994) as
a one- or two-slope log-distance law, depending on
whether there is Line-of-Sight (LoS) or not (NLoS):
L
p,LoS
(d) =





10n
1
log
10
(d)
+L
0
(1m) [dB] if 1 < d < d
f
10n
2
log
10
(d/d
f
) + 10n
1
log
10
(d
f
)
+L
0
(1m) [dB] if d > d
f
Table 2: Path-loss coefficients (for 2 GHz).
Coefficient Value
n
1
/n
2
/n
3
2.2/3.3/2.6
BS
MS
Single cluster
Twin cluster
Local cluster
•: MPC
Figure 1: General description of the COST 273 DCM.
L
p,NLoS
(d) = 10n
3
log
10
(d) + L
0
(1m) [dB]
L
0
(1m) = 20log
10
(4π(1m)/λ) [dB] (3)
where d is the distance between the BS and the MS, λ
denotes the wavelength (λ = c
0
/ f
c
, being c
0
the speed
of light) and the breakpoint value d
f
is given by
d
f
=
1
λ
s
(Γ
2
−Λ
2
)
2
−2(Γ
2
+ Λ
2
)

λ
2

2
+

λ
2

4
(4)
with Γ = h
BS
+h
MS
and Λ = h
BS
−h
MS
. The values of
the empirical regression coefficients n
1
, n
2
and n
3
for
2 GHz are given in Table 2 (Correia, 2001).
2.2.2 Clusters General Considerations
In the COST 273 DCM first the scatterers are stochas-
tically distributed in the physical environment and
then the corresponding MPCs are computed using a
simple ray tracing technique. Each MPC is character-
ized by its delay and angles (azimuth and elevation).
Since there are MPCs with similar delay and angles,
they can be grouped into a cluster, which enables the
use of less parameters to describe the channel.
As Figure 1 shows, three kinds of clusters are used
in this DCM: local clusters, single-interaction clus-
ters and twin clusters. Local and single-interaction
clusters account for single-bounce scatterers, while
twin clusters allow to simulate the multiple-bounce
behavior. For microcells, there is an equal number of
single and twin clusters (Correia, 2006).
2.2.3 Number of Clusters and Visibility Region
Many clusters can exist in the radio environment but
only some are active, meaning that only those con-
tribute to theCIR. TheVisibility Region (VR) concept
WINSYS2012-InternationalConferenceonWirelessInformationNetworksandSystems
350

Table 3: Visibility region parameters.
Parameter Value
N
C
/R
C
[m]/L
C
[m] 4/50/20
a
C
b
C
h
C
Figure 2: Clusters geometry.
is used to determine whether or not a cluster is active,
where each VR is associated with only one cluster: if
the MS is in a VR, then the corresponding cluster is
active; otherwise the cluster is non-active. VRs are
circular regions with identical size on the xy plane.
The relative power of the cluster m associated to a
certain VR is scaled by a factor A
2
m
, with A
m
given by
A
m
(~r
MS
) =
1
2
−
1
π
arctan
2
√
2(L
C
+ d
MS,VR
−R
C
)
√
λL
C
!
(5)
where R
C
is the VR radius, L
C
is the size of the transi-
tion region and d
MS,VR
is the distance between the MS
and the VR center. For an expected number of active
clusters equal to N
C
, the VRs area density needs to be
ρ
C
= (N
C
−1)/

π(R
C
−L
C
)
2

m
−2

(6)
where the N
C
−1 term comes from the fact that there
is always an active cluster around a MS, i.e., A
1
= 1.
The VRs are uniformly distributed in the simulation
environment. Table 3 presents the values of N
C
, R
C
and L
C
for microcells (Correia, 2006).
2.2.4 Local Clusters Generation
One cluster is always present around a MS and it is
always active. It is named local cluster since its cen-
ter is co-located with the MS. This cluster contains
single-bounce MPCs that introduce a large azimuth
spread with low delay at the MS.
All kinds of clusters can be seen as ellipsoids in
space, each one characterized by the extents in space
a
C
, b
C
and h
C
, Figure 2. In the model implementa-
tion, a local cluster is assumed to be circular in the xy
plane, i.e., the ground plane (a
C
= b
C
), being its spa-
tial spread only determined by its delay spread (a
C
)
and elevation spread (h
C
). The computation of these
spreads, along with the MPCs positioning within the
cluster, will be seen later on.
2.2.5 Single-interaction Clusters Generation
The positioning of a single-interaction cluster is ob-
tained using a simple geometric approach. First an
Table 4: Single-interaction clusters position parameters.
Parameter Value
r
min
[m]/σ
r,C
[m]/σ
φ,C
[deg] 5/48.4/22.5
Table 5: Twin clusters position parameters.
Parameter Value [deg]
ζ
BS
/ζ
MS
22.5/180
imaginary line is drawn from the BS to the center of
the VR associated with the cluster, then the azimuth
angle between this line and the cluster is obtained
from a Gaussian distribution with a standard devia-
tion σ
φ,C
. The distance in the xy plane from the BS to
the cluster is given by the exponential distribution
f
r
C
(r
C
) =
(
0 if r
C
< r
min
1
σ
r,C
e
−(r
C
−r
min
)/σ
r,C
if r
C
≥ r
min
(7)
The values for r
min
, σ
r,C
and σ
φ,C
are presented in Ta-
ble 4. They were empirically chosen since values for
microcell scenarios were not found in the literature.
The proposed σ
r,C
value guarantees that the distance
r
C
is less than or equal to 150 m in 95% of the times.
The above procedure only fixes the position of the
cluster in the xy plane. The COST 273 DCM does
not suggest any approach for the cluster height, so for
the model implementation an uniformdistribution be-
tween the ground and the BS height is proposed.
The spatial spread (Figure 2) of single-interaction
clusters is determined by its delay, azimuth and eleva-
tion spread (a
C
, b
C
and h
C
, respectively).
2.2.6 Twin Clusters Generation
Twin clusters are represented by two clusters with the
same IOs distribution and number, corresponding one
cluster to the BS side and the other to the MS side.
This representation allows independent modeling of
the angular dispersion at the BS and at the MS. First,
the azimuth of a cluster seen from the BS (MS) side
is computed from an uniform distribution within the
interval [−ζ
BS
,+ζ
BS
] ([−ζ
MS
,+ζ
MS
]), where the an-
gle 0 represents the imaginary line from the BS to
the center of the VR associated with the twin cluster.
These parametersvalues are givenin Table 5 (Correia,
2006). The distance in the xy plane from the cluster
to the BS (VR center where the MS is) is given by
d
BS/MS
= (∆τc
0
)/

4tanϕ
BS/MS

(8)
being ∆τ and ϕ
BS
(ϕ
MS
) the delay and azimuth spread
seen from the BS (MS), respectively.
Again, this procedure only fixes the xy plane posi-
tion of the cluster. Like the single-interaction clusters
case, we propose an uniform distribution between the
ImplementationoftheCOST273DirectionalChannelModelinMicrocellScenarios
351

Table 6: Cluster spread parameters.
Parameter Value
µ
τ
[ns]/σ
τ
[dB]/σ
S
[dB] 13/14/2.9
µ
ϕ
BS
[deg]/µ
ϕ
MS
[deg]
2.3/2.3
σ
ϕ
BS
[dB]/σ
ϕ
MS
[dB]
3.4/3.4
µ
θ
BS
[deg]/µ
θ
MS
[deg] 1.3/1.3
σ
θ
BS
[dB]/σ
θ
MS
[dB] 3.3/3.3
groundand the BS height (independentfor each of the
two representations) for model implementation.
The two representations of a twin cluster have the
same delay spread, being it circular in the xy plane
(a
C
= b
C
), while each representation (BS and MS)
sees a different elevation spread (h
C,BS
and h
C,MS
).
2.2.7 Clusters Dispersion
For any type of clusters, the delay spread, angular
spreads and shadow fading of a cluster m are corre-
lated random variables, given respectively by
∆τ
m
= µ
τ
(d/1000)
0.5
10
σ
τ
Z
m
/10
(9)
β
m
= µ
β
10
σ
β
Y
m
/10
(10)
S
m
= 10
σ
S
X
m
/10
(11)
where Z
m
, Y
m
and X
m
are correlated Gaussian random
variables, with zero mean and unit variance, and β
represents one of the angular spreads, i.e., azimuth or
elevation spread for the BS side (ϕ
BS
or θ
BS
) or for the
MS side (ϕ
MS
or θ
MS
). The parameters values needed
to compute a cluster spread are presented in Table 6
(Correia, 2006). The correlated random processes can
be computed using the Cholesky factorization, char-
acterized by the cross-correlationcoefficients givenin
Table 7 (Correia, 2006). After computing the spreads
of a cluster, its ellipsoid (Figure 2) can then be char-
acterized. The delay spread dimension is given by
d
τ
= ∆τc
0
/2 (12)
while the azimuth (β = ϕ
BS/MS
) and elevation
(β = θ
BS/MS
) spreads dimensions are given by
d
β
= d
C,BS/MS
tanβ (13)
where d
C,BS
(d
C,MS
) is the distance between the cluster
and the BS (VR center where the MS is). Table 8
summarizes the clusters spatial characterization.
2.2.8 IOs Positioning
The COST 273 DCM uses for microcells 7 MPCs per
cluster. The positioning of the IOs (each IO corre-
sponds to one MPC) differs if they represent single-
bounce scatterers (local and single-interaction clus-
ters) or multiple-bounce scatterers (twin clusters).
Table 7: Cross-correlation coefficients.
Coefficient Value
ρ
τ−ϕ
BS
/ρ
τ−ϕ
MS
/ρ
τ−S
0.1/0.1/0.04
ρ
ϕ
BS
−S
/ρ
ϕ
MS
−S
/ all other −0.2/−0.2/0
Table 8: Cluster spatial spread.
Type a
C
b
C
h
C
Local ∆τc
0
/2 = a
C
d
C,BS
tanθ
BS
Single ∆τc
0
/2 d
C,BS
tanϕ
BS
d
C,BS
tanθ
BS
T. (BS) ∆τc
0
/2 = a
C
d
C,BS
tanθ
BS
T. (MS) =Twin (BS) d
C,MS
tanθ
MS
For single-bounce scatterers, the distance between
the cluster center and an IO is based on the truncated
one-sided Gaussian distribution (Laurila et al., 1998)
f
r
IO
(r
IO
) =
(
1
√
2π
e
−r
2
/2
if 0 ≤ r
IO
≤ r
T
0 otherwise
(14)
where r
T
= 3 (Correia, 2006). Uniform distributions
are assumed for the azimuth (ϕ
IO
) and elevation (θ
IO
)
directions. The relative position of an IO with respect
to the cluster center is obtained by computing r
IO
, ϕ
IO
and θ
IO
, followed by a conversion to a Cartesian co-
ordinate system (xyz); after that, these values (x
IO
, y
IO
and z
IO
) are multiplied by the cluster ellipsoid spatial
spread (a
C
, b
C
and h
C
from Table 8).
For twin clusters, the COST 273 DCM uses an IO
distribution in all dimensions based on the truncated
Gaussian distribution (shown here for one dimension)
f
x
IO
(x
IO
) =
(
1
√
2π
e
−x
2
IO
/2
if |x
IO
| ≤ x
T
0 if |x
IO
| > x
T
(15)
where x
T
= 3 (Correia, 2006). The relative position of
an IO is obtained as follows: first x
IO
, y
IO
and z
IO
are
computed using (15) (these random generated values
are used for both representations of the twin cluster –
BS side and MS side – to guarantee consistent delay
and angular spreads); then they are multiplied by the
cluster spatial spread (a
C
, b
C
and h
C
from Table 8).
To match the spatial spread with the delay and an-
gular spreads viewed from the BS (VR center where
the MS is, for the MS side representationof twin clus-
ters), clusters are rotated in order to become oriented
towards the BS (away from the MS, for the MS side
representation of twin clusters). This rotation is at-
tained by multiplying the previously computed IOs
relative positions by the following rotation matrix


cosΦcosΘ −sinΦ cosΦsinΘ
sinΦcosΘ cosΦ sinΦsinΘ
−sinΘ 0 cosΘ


(16)
WINSYS2012-InternationalConferenceonWirelessInformationNetworksandSystems
352

Table 9: MPCs power parameters.
Parameter Value
k
τ
[dB/µs]/τ
B
[µs] 40/0.5
K
MPC
[dB]/µ
K
/σ
K
2/7/2.3
where (Φ, Θ) = (Φ
BS
,Θ
BS
), being Φ
BS
and Θ
BS
the
azimuth and elevation angles of the cluster seen from
the BS (for the MS side representation of twin clusters
(Φ,Θ) = (Φ
MS
+π,Θ
MS
+π), being Φ
MS
and Θ
MS
the
azimuth and elevation angles of the cluster seen from
the VR center where a MS is).
2.2.9 MPCs Power
The COST 273 DCM assumes that the power attenu-
ation of a cluster m is a function of its delay (τ
m
) with
respect to the LoS delay (τ
0
), i.e., the longer the de-
lay, the smaller is the power that the cluster carries.
Hence, the power attenuation of a cluster is given by
P
m
= max
n
e
−k
τ
(τ
m
−τ
0
)
,e
−k
τ
(τ
B
−τ
0
)
o
(17)
where k
τ
is the decaying parameter and τ
B
is the cut-
off delay. The delay of a cluster is given by
τ
m
= (d
C
m
,BS
+ d
C
m
,MS
)/c
0
+ τ
C
m
,link
(18)
where τ
C
m
,link
is either the cluster-link delay between
the two representation of twin clusters (computed
using a similar geometric relationship) or assumes
τ
C,link
= 0 for local and single-interaction clusters.
The power P
MPC
of each MPCs within a cluster is
characterized through a Ricean distribution
f
Rice
(w) =
w
σ
2
K
I
0

wA
K
σ
2
K

e
−(w
2
+A
2
K
)/(2σ
2
K
)
(19)
where I
0
(·) denotes the modified Bessel function of
the first kind with order zero, and the Rice factor
K
MPC
is related to the parameters A
K
and σ
K
by
K
MPC
= A
2
K
/(2σ
2
K
) (20)
The complexamplitudeof the MPC l in the cluster
m is given by (for one polarization component)
a
m,l
=
q
L
p
A
2
m
S
m
P
m
P
MPC,m,l
e
−j2πf
c
τ
m,l
(21)
where τ
m,l
is the delay of the MPC given by
τ
m,l
= (d
MPC
m,l
,BS
+ d
MPC
m,l
,MS
)/c
0
+ τ
C
m
,link
(22)
For LoS situations the DDCIR has an extra MPC
a
LoS
=
p
L
p
P
LoS
e
−j2πf
c
τ
0
(23)
where P
LoS
is the LoS power factor drawn from a log-
normal distribution with mean µ
K
and standard devi-
ation σ
K
. The values of the parameters introduced in
this subsection are given in Table 9 (Correia, 2006).
Table 10: Polarization coefficients relations.
Parameter Value [dB]
µ
XPD
/µ
VV/HH
/µ
VH/HV
8.5/0.3/−0.5
σ
XPD
/σ
VV/HH
/σ
VH/HV
1.8/3.2/1.8
Table 11: Autocorrelation distances.
Parameter Value [m]
L
S
/L
τ
/L
ϕ
BS
/L
θ
BS
/L
ϕ
MS
/L
θ
MS
5/5/50/50/25/25
2.2.10 Polarization
The polarization is characterized by the matrix

P
VV
P
VH
P
HV
P
HH

(24)
where the entries characterize the powers of each po-
larization component. The ratio
XPD = (P
VV
+ P
HH
)/(P
VH
+ P
HV
) (25)
is log-normally distributed, with mean µ
XPD
and stan-
dard deviation σ
XPD
. Other relations between the po-
larization coefficients are also modeled log-normally
with the values presented in Table 10 (Correia, 2006).
2.2.11 Other Considerations
The COST 273 DCM described previously is envi-
sioned for singe-link scenarios, i.e., where only one
BS and one MS are present. In order to study the
correlation between different links, one can extend
the model to multi-MS scenarios by dropping multi-
ple MSs into the simulation environment and use the
same VRs and corresponding clusters for each MS.
The MSs positional data can be given by any mo-
bility model. When introducing movement, the shad-
owing as well as the delay and angular spreads are
characterized by the spatial autocorrelation function
ACF(x,x
′
) = e
−|x−x
′
|/L
x
(26)
with the respective autocorrelation distances L
x
shown in Table 11 (Correia, 2006).
Another important aspect of the model’s imple-
mentation is the simulation area, i.e., the area where
the MSs can be and where the VRs are (the cor-
responding clusters can be anywhere since they are
stochastically determined). Since the clusters attenu-
ation power given by (17) should not be greater than
one, the MSs should not be separated from the BS by
more than τ
B
×c
0
= 150 m (Table 9). Recalling that
the VRs are uniformly distributed in the simulation
area, this simulation area should be at least extended
by the radius of the VRs (from Table 3, R
C
= 50 m)
ImplementationoftheCOST273DirectionalChannelModelinMicrocellScenarios
353

from the farthest MS away from the BS. This avoids
effects introduced by the simulation area limits.
Since one spatial dimension of any cluster is al-
ways characterized by its delay spread, the stochastic
generation of this parameter can yield a large unreal-
istic cluster size. To avoid this situation, a maximum
value for the delay spread of 120 ns (Correia, 2001)
should be applied when using expression (9).
3 IMPLEMENTATION EXAMPLE
Validation of a channel model simulator is very im-
portant to ensure that the outputs are realistic and can
be used for MIMO systems development, for exam-
ple. To this end, a large number of channel realiza-
tions were generated using a multi-MS scenario sim-
ulator developed in MATLAB
R

based on the COST
273 DCM described previously. Those results were
then compared with measurements available in the lit-
erature. This procedure allows not only to validate if
the different sub-models for cluster and MPC behav-
ior are well combined, but it also allows to infer if the
values used for cluster positioning that were proposed
in this work are adequate.
3.1 Simulation Parameters
The test environment is characterized by a BS (10 m
height) located in the center of a 100 m × 100 m mi-
crocell. MSs (1.5 m height) are evenly positioned in
the microcell, being separated between them by 1 m,
which makes a total of 10201 MSs. An operating fre-
quency of 2 GHz is used and the VRs can be in a
200 m × 200 m area (center co-located with the BS),
in order to avoid simulation area limits effects.
Each simulation run has a different distribution of
VRs and different stochastic parameters, hence cor-
responds to a different radio channel situation. Also,
no mobility model is used, so each run represents a
snapshot of the radio environment.
3.2 Simulation Results
The following simulation results are based on 200
simulations runs. Table 12 presents the environment
characterization related with measurements values
available in the literature. The indicated references
correspond to: [1]-present simulator; [2]-(Kozono
and Taguchi, 1993); [3]-(Devasirvatham, 1988); [4]-
(Zhao et al., 2002); [5]-(3GPP, 2011); [6]-(Pajusco,
1998); [7]-(Correia, 2006).
Table 12: Reference scenarios.
Reference R
cell
[m] f
c
[GHz] h
BS
/h
MS
[m]
[1] 71 2 10/1.5
[2] 1000 1.5 5/1.5
[3] 150 0.85 9.1/1.8
[4] 200 5.3 12/2
[5] 500 1.9 12.5/1.5
[6] 300 2.2 7/∗
[7] † 5.3 ‡/∗
† – only info available is that it is a microcell environment
‡ – only info available is that BS antenna is on rooftop
∗ – only info available is that MSs are on the street
Table 13: Delay spread comparison.
Statistical measure
Delay spread [ns]
[1] [2] [3] [4]
median – LoS 6 — 30 31
median – NLoS 14 — 50 —
maximum – LoS 68 250 62 64
maximum – NLoS 297 300 330 —
3.2.1 Delay Spread
The delay spread can be used as a measure of the mul-
tipath propagation phenomenon. Given measured de-
lay profiles, the delay spread is calculated using the
formula presented in (Kozono and Taguchi, 1993).
Table 13 presents some delay spread values ob-
tained by the simulator along with measurements val-
ues available in the literature. The simulated val-
ues generallyagree with experimental data, especially
when considering the maximum delay spread case.
Also, simulated delay spread values are lower for the
LoS case. This is an expected result, also verified by
the measurements, due to the fact that the LoScompo-
nent has associated a higher power when compared to
the other MPCs, thus leading to a lower delay spread
value. The median values obtained by simulation are
lower than the ones measured, which could mean a
lower multipath richness. Since lower multipath rich-
ness usually means lower MIMO gain (MIMO chan-
nels become more correlated), one could say that this
implementation simulates a worst case scenario.
3.2.2 Azimuth Spread
The computation of the angular spread is performed
as mentioned in (3GPP, 2011), which is somewhat
similar to the delay spread computation but takes into
account the ambiguity of the modulo 2π operation.
Table 14 presents some azimuth spread values ob-
tained by the simulator along with measurements val-
ues available in the literature. Once again, the val-
WINSYS2012-InternationalConferenceonWirelessInformationNetworksandSystems
354

Table 14: Azimuth spread comparison.
Statistical measure
Azimuth spread [deg]
[1] [5] [6] [7]
median – BS, LoS 3.7 5.0 7.5 5.1
median – MS, LoS 30.8 62.5 — 28.9
median – BS, NLoS 9.4 19.0 20.0 12.6
median – MS, NLoS 73.3 68.0 — 40.3
Table 15: Elevation spread comparison.
Statistical measure
Elevation spread [deg]
[1] [7]
median – BS, LoS 3.3 1.3
median – MS, LoS 18.4 2.5
median – BS, NLoS 8.3 2.5
median – MS, NLoS 48.4 4.7
ues obtained by the simulator generally agree with
the measurements available in the literature. As can
be seen, azimuth spread values are lower for the LoS
cases. This result is again expected for the same rea-
sons as in the delay spread case, and once more it is
verified by the experimental data. Also, the simulated
azimuth spread is higher at the MSs than at the BS,
which is consistent with the measurements. This can
be justified by the always existent local cluster around
a MS and the use of the twin cluster concept.
3.2.3 Elevation Spread
Unlike the previous case, there is no modulo 2π oper-
ation ambiguitywhen computingthe elevation spread,
so an equivalent formula of the delay spread is used.
Table 15 presents some elevation spread values
obtained by the simulator along with measurements
values available in the literature. In this case, the
simulated values do not agree quantitatively with the
experimental data from the only reference available.
Besides the fact that the reference scenario physical
characteristics are not fully know and might be a lot
different, also the influence of the antennas’ opera-
tional angular ranges was not taken into account in
the measurements, as stated in (Correia, 2006). This
could mean a reduction in the elevation spread if the
measurement antennas were not omnidirectional in
the elevation plane. From a qualitative point of view,
the simulated results agreewith the experimentaldata,
because elevation spread values are lower for the LoS
case for both spreads observed at the BS and at the
MSs; also, the elevation spread generated by the sim-
ulator is higher at the MSs than at the BS.
4 FINAL REMARKS
This paper presentsa tutorial on how to implementthe
COST 273 DCM for microcell scenarios. All parame-
ters models and values requiredby the DCM that were
disperse in the literature are gathered in this work, as
well as some values that were proposed here, because
they were missing in the literature and are essential.
An implementation example of this COST 273 DCM
proved that its results agree with experimental data,
hence it is suitable for MIMO systems development.
ACKNOWLEDGEMENTS
This work was partially funded by Instituto de
Telecomunicac¸˜oes/LA and by Fundac¸˜ao para a
Ciˆencia e a Tecnologia (FCT) under a Doctoral grant
(SFRH/BD/62003/2009).
REFERENCES
3GPP (2011). Spatial Channel Model for Multiple Input
Multiple Output (MIMO) Simulations. Technical Re-
port 25.996, v. 10.0.0. www.3gpp.org/specifications.
Almers, P., Bonek, E., Burr, A., et al. (2007). Survey of
Channel and Radio Propagation Models for Wireless
MIMO Systems. EURASIP Journal on Wireless Com-
munications and Networking, 2007.
Correia, L. (2001). Wireless Flexible Personalized Com-
munications, COST 259: European Co-operation in
Mobile Radio Research. John Wiley & Sons, Inc.
Correia, L. (2006). Mobile Broadband Multimedia Net-
works: Techniques, Models and Tools for 4G. Aca-
demic Press.
Devasirvatham, D. (1988). Radio Propagation Studies in
a Small City for Universal Portable Communications.
In IEEE 38th VTC, pages 100–104.
Feuerstein, M., Blackard, K., et al. (1994). Path Loss, Delay
Spread, and Outage Models as Functions of Antenna
Height for Microcellular System Design. IEEE Trans-
actions on Vehicular Technology, 43(3):487–498.
Foschini, G. and Gans, M. (1998). On Limits of Wireless
Communications in a Fading Environment when Us-
ing Multiple Antennas. Wireless Personal Communi-
cations, 6:311–335.
Kozono, S. and Taguchi, A. (1993). Mobile Propagation
Loss and Delay Spread Characteristics with a Low
Base Station Antenna on an Urban Road. IEEE Trans-
actions on Vehicular Technology, 42(1):103–109.
Laurila, J., Molisch, A., and Bonek, E. (1998). Influence of
the Scatterer Distribution on Power Delay Profiles and
Azimuthal Power Spectra of Mobile Radio Channels.
In IEEE 5th ISSSTA, volume 1, pages 267–271.
Molisch, A., Asplund, H., et al. (2006). The COST259
Directional Channel Model–Part I: Overview and
ImplementationoftheCOST273DirectionalChannelModelinMicrocellScenarios
355

Methodology. IEEE Transactions on Wireless Com-
munications, 5(12):3421–3433.
Pajusco, P. (1998). Experimental Characterization of DOA
at the Base Station in Rural and Urban Area. In IEEE
48th VTC, volume 2, pages 993–997.
Telatar, E. (1999). Capacity of Multi-antenna Gaussian
Channels. European Transactions on Telecommuni-
cations, 10:585–595.
Verdone, R. and Zanella, A. (2012). Pervasive Mobile
and Ambient Wireless Communications: COST Action
2100. Springer.
Zhao, X., Kivinen, J., Vainikainen, P., and Skog, K. (2002).
Propagation Characteristics for Wideband Outdoor
Mobile Communications at 5.3 GHz. IEEE Journal on
Selected Areas in Communications, 20(3):507–514.
WINSYS2012-InternationalConferenceonWirelessInformationNetworksandSystems
356