Comparison of SCM, SCME, and WINNER
Channel Models
Milan Narandžić, Christian Schneider, Reiner Thomä
Technische Universität Ilmenau,
PSF 100 565, D-98694 Ilmenau, Germany
milan.narandzic@tu-ilmenau.de
Abstract—This paper is summarizing and comparing
properties of channel models used for Beyond-3G (B3G) MIMO
simulations: 3GPP Spatial Channel Model (SCM), its extension
(SCME), and models developed by WINNER. Compared models
are offering complete channel model description in a sense of
large-scale as well as small-scale effects in MIMO radio-channel.
WINNER targeted model was supposed to provide reliable tool
for estimation of system performance, covering frequencies up to
5 GHz and bandwidths of 100 MHz in different types of
environment. Since SCM was originally proposed for 2 GHz
range and 5 MHz bandwidth, certain extensions (SCME) were
necessary. However, SCME performance was restricted since it
has been design as backward compatible with SCM. That was the
motivation to start using the new WINNER generic channel
model, where model parameters are extracted from channel-
sounding measurements covering targeted frequency range and
bandwidth. This paper describes all important differences and
compares features and performances of the models.
Index Terms—Generic multipath MIMO channel, spatial-
channel-modelling, measurement-based parameterization,
system-level correlation.
I. INTRODUCTION (SHORT HISTORIC OVERVIEW)
W INNER project [1] has begun in 2004, and 3GPP SCM
(Spatial-Channel-Model) was adopted for its initial
channel model. SCM was developed in 3GPP/3GPP2 ad hoc
group for spatial channel models and released in September
2003 [2]. In the beginning of 2005 the first extension of the
original model was proposed by WINNER under the name
SCME (SCM Extension [3]). SCME was later modified for
3GPP LTE purposes in [4] – [7]. In the end of 2005 further
improvements and extensions are resulted in the model with
the new name: WINNER channel Model – Phase I (WIM1).
WIM1 model is described in the deliverable D5.4 [8] and
published in [9]. SCM, SCME, and WIM1 models were briefly
compared in [10]. The comparison tables of this paper also
include WINNER – Phase II (WIM2) interim model that was
recently published in the deliverable D1.1.1 [11]. The final
WIM2 model will be available in autumn 2007 (D1.1.2). In
this paper, WIM refers to WINNER model in general, WIM1
specifically to Phase I, and WIM2 specifically to Phase II.
In the course of the WINNER project all encountered
Tommi Jämsä, Pekka Kyösti, Xiongwen Zhao
Elektrobit,
Tutkijantie 7, FIN-90570 Oulu, Finland
tommi.jamsa@elektrobit.com
models are implemented in MATLAB/C and made available
through the official web site [1].
The paper is organized as follows: In Section II the
modeling methodology applied in SCM/SCME/WIM is
briefly summarized. In Section III different features of the
compared models are analyzed. In Section IV the performance
figures are given, and finally Section V concludes the paper.
II. MODELING METHODOLOGY
The basic modelling philosophy behind all
SCM/SCME/WIM is the same: the sum of specular
components is used to describe the changes in the channel
impulse response (CIR) between each transmitting and
receiving antenna element (so called sum-of-sinusoids (SOS)
method). Due to the different spatial position of elements
inside Tx/Rx antenna array, different channel characteristics
are obtained and MIMO concept is supported.
All compared models can be classified as stochastically
controlled spatial channel models what means that parameters
of each specular component (sinusoid) are related to a spatial
propagation of the single multipath-component (MPC). To
fully describe a MPC following (low-level) parameters are
used: departure (from Tx) and arriving (to Rx) angles,
propagation delay and power. However, an evolution of these
parameters is not based on the ray-tracing since positions of
scattering clusters are not known. Instead, MPC parameters are
chosen randomly from the appropriate probability
distributions.
Since the modeled radio-channel shows non-stationary
behaviour, distributions of low-level parameters are changing
in time. In order to support this phenomenon probability
distributions are parameterized and their changes are modeled
through the change of the control parameters. Since these
parameters are controlling probability distributions of other
(low-level) parameters in WINNER terminology they are
called Large-Scale-Parameters (LSPs). The large-scale control
parameters themselves also represent random variables that are
governed by the appropriate probability distributions. These
distributions are related to and measured in certain radio-
propagation environments, called scenarios (Section III.A.1).
 III. FEATURES
A. Applicability Range
1) Environment dependence: scenarios
To characterize different environments (scenarios)
WINNER models are using (temporal and spatial) parameters
obtained from measured CIRs. For each scenario measured
data is analyzed and processed to obtain scenario-specific
parameters. After this point, same generic channel is used to
model all scenarios, just by using different values of control
parameters.
SCM was originally dedicated to outdoor propagation, and
defines three environments: Suburban macro, Urban macro,
Urban micro. In SCME number of scenarios is not extended.
WINNER models are supporting considerably larger number
of scenarios then SCM/SCME. WIM1 defines the following
scenarios: A1–Indoor (small office/residential), B1–Typical
urban micro-cell, B3–Indoor hotspot, B5–Stationary feeder
with sub-scenarios a, b, and d having different deployment
assumptions, C1– Suburban, C2–Typical urban macro-cell and
D1–Rural macro-cell. The covered number of scenarios is
further extended for the WINNER Phase II model: A2–Indoor-
to-outdoor, B4– Outdoor-to-indoor, B5–Stationary feeder sub-
scenarios c, and f, D2a–Moving networks, and Bad Urban
extensions for micro-cell (B2) and macro-cell (C3) scenarios.
Usually, even for the same scenario, existence of LOS
component substantially influences values of model
parameters. Regarding to this property, each WINNER
scenario is differentiating between LOS and NLOS conditions.
Originally this was not the case with SCM where LOS
condition was analyzed only in the context of Urban Micro
scenario, but SCME provided the extensions for the other
SCM scenarios.
2) System dependence
Carrier frequency: Dependence on carrier frequency in
SCM/SCME/WIM is found in path-loss models. In SCM,
COST 231 Urban Hata (macro-cell) and COST 231 Walfish-
Ikegami (micro-cell) path-loss models are adopted and
adjusted for frequencies of 1.9 GHz. Applicability of proposed
models in different frequency ranges is not analyzed. In SCME
frequency range for SCM scenarios is extended based on
correction of the frequency dependant free space loss:
⎛f ⎞
ΔPL( f c ) = 20 log⎜ c ⎟ , (1)
⎝2⎠
where f c is carrier frequency in [GHz]. Additionally, in
SCME COST 231 Walfish-Ikegami path-loss model was
applied also for macro-cell environments in 5GHz range since
usage of higher frequencies decreases coverage area.
Following SCME approach, all scenarios defined by WIM
support frequency dependant path-loss models valid for the
ranges of 2 – 6 GHz. WIM path-loss models are based on
measurements that are mainly conducted in 5.2 GHz frequency
range.
From WINNER measurement results and literature survey it
was found that model parameters delay-spread (DS), angular
spread (AS) and Rician K-factor do not show significant
frequency dependence [3]. From that reason these parameters
show only dependence on environment (scenario) in
SCM/SCME/WIM.
For modeling systems with time-division-duplex (TDD) all
models (SCM/SCME/WIM) are using same (large-scale as
small-scale) parameters for both uplink (UL) and downlink
(DL). If system is using different carriers for duplexing
(FDD), than (additionally to path loss) random phases between
UL and DL are independent.
Bandwidth: In 3GPP-SCM document there is a note that
usage of 6 paths (clusters) may not be suitable for bandwidths
higher than 5MHz. This reflects that influence of system
bandwidth to the complexity of model structure (number of
clusters, number of MPCs per cluster) is not completely
investigated.
For the WINNER purposes it is required that channel model
supports bandwidths up to 100 MHz. Following SCM note and
approach described in [12] (for indoor propagation modeling)
SCME used intra-cluster delay spread as a mean to support
bandwidth extension. However, since SCM forced backward-
compatibility (to attain comparability with SCM) number of
clusters and total number of MPCs are not increased! Instead,
SCM cluster with 20 MPCs is subdivided into 3-4 zero-delay
sub-clusters (“mid-paths”), keeping total number of MPCs
constant. Introduced delay spread per cluster (10ns) in not
based on targeted 100 MHz bandwidth, but on PDP matching
since the equal power of all MPCs belonging to the same
cluster is enforced. From the above discussion it is not
apparent that intra-cluster spread concept (unchanged number
of MPCs) better reflects bandwidth dependence than
introduction of the new clusters. However, the latter also
increases complexity since total number of MPCs would be
increased.
In WIM different philosophy is applied: since measurement
systems were supporting 100 MHz bandwidth, during WIM
parameterization number of clusters was traced in delay and
angular domains from measured CIR. In this way number of
clusters reflects both system bandwidth and scenario
dependence.
B. System Level Description
In order to calculate path-loss, information about distance
between transmitting (Tx) and receiving (Rx) station is
necessary for all compared models. This means that system
layout information about positions of all stations is necessary.
Note that system layout usually does not contain any
information about environment characterization. Only
exception is related to the positions of far scatterers in the
SCM and WIM2.
Correlation at system-layout level: In all models all LSPs
are fully correlated for links between MS and different sectors represented by 3 or 4 zero-delay sub-clusters.
of the same BS. In this way, influences of selective antenna Predefined angular offsets from the cluster center: This
gain pattern or different LOS conditions between certain property is used by SCM/SCME/WIM to avoid random
sectors to the level of LSP correlation are not regarded. generation of MPC angles in each drop. Difference is that
Additionally, WIM uses positions of MS to introduce WIM scales initial offsets with standard deviation of angles
scenario-specific correlation of link LSPs for MSs being per cluster, while SCM uses predefined values for intra-cluster
connected to the same BS. In SCM/SCME correlation of LSPs (sub-path) departure angle spreads of 2 (macro) and 5 (micro)
between different MS is not supported. degrees and 35 degrees for spread of arrival angles. If separate
In SCM, standard deviation of shadowing fading for links drifting (continuous evolution) of MPCs inside cluster in used
from one MS to different BSs (“site-to-site”) has constant by SCME, angular offsets are not constant any more.
correlation coefficient equal to 0.5. Introduced correlation
does not depend on distances between BSs or their relative D. Comparison Tables
angular positions as seen from MS and therefore it is not TABLE I
FEATURE COMPARISON
layout dependant. Currently WIM1 does not support this type WINNER I WINNER
of correlation and same applies to SCM/SCME Feature SCM SCME
II
implementations supported by WINNER. Bandwidth > 100 MHz No Yes Yes Yes
Indoor scenarios No No Yes Yes
Antenna arrays: SCM/SCME/WIM1 introduce additional Outdoor-to-indoor and
indoor-to-outdoor No No No Yes
support for cross-polarized antenna arrays because used scenarios
representation for antenna arrays was not general enough (the AoA/AoD elevation No No Yes Yes
reference polarizations of antennas and environments are not Intra-cluster delay
No Yes No Yes
spread
modeled separately). The full polarimetric antenna description TDL model based on
No Yes Yes* Yes
will be developed in WIM2 [13]. the generic model
Cross-correlation
No No Yes Yes
C. Complexity Issues between LSPs
Time evolution of
No Yes** No Yes
WINNER channel models are intended for system level model parameters
*TDL model is based on the same measurements as generic model, but analyzed
evaluation. In order to keep complexity reasonably low, separately.
SCM/SCME/WIM models are relaying on some quasi-realistic **Continuous time evolution.
assumptions and concepts:
TABLE II
Channel segment/drop: Channel segment (drop) represents NUMERICAL COMPARISON
period of quasi-stationarity in which probability distributions WINNER WINNER
Parameter Unit SCM SCME
of low-level parameters are not changed. During this period all I II
Max. bandwidth MHz 5 100* 100** 100**
large-scale control parameters, as well as velocity and Frequency range GHz 2 2–6 2–6 2–6
direction-of-travel for mobile station (MS), are held constant No. of scenarios 3 3 7 12
No. of clusters 6 6 4-24 4-20
in SCM/WIM models. To the opposite, SCME has allowed
No. of mid-paths
simultaneous drifting of arriving angles and delays for every 1 3–4 1 1–3
per cluster
MPC at each simulation step inside a drop. In respect to this No. of sub-paths
20 20 10 20
per cluster
property SCME can be classified as model with continuous No. of taps 6 18 – 24 4-24 4-24
evolution (in discrete steps, being smaller than drop). BS angle spread º 5 – 19 4.7 – 18.2 3.0 – 38.0 2.5 – 53.7
MS angle spread º 68 62.2 – 67.8 9.5 – 53.0 11.7 – 52.5
Consequence was substantial increase of complexity and Delay spread ns 170 – 650 231 – 841 1.6 – 313.0 16 – 630
simulation time length in comparison to SCM/WIM1. The new Shadow fading
approach to the model evolution will be included in WIM2. standard dB 4 – 10 4 – 10 1.4 – 8.0 2–8
deviation
Zero-Delay-Spread-Cluster (ZDSC): MPCs belonging to the * artificial extension from 5 MHz bandwidth
** based on 100 MHz measurements
same cluster (path in SCM terminology) have “close” values
of both delay and angular parameters. If all MPCs belonging DS values in WIM2 scenarios are generally lower then for
to the same cluster have exactly the same delay, it is possible WIM1 due to the new estimation procedure that regards only
to define extremely simple relation between SCM/WIM and upper 20 dB of the observed signal range in CIR. The higher
tapped-delay-line (TDL) model. Due to mentioned similarity, DS range in Table II corresponds to the bad urban scenarios
the latter is also called cluster-delay-line (CDL) model in that are modelled with an additional far-cluster option.
WINNER terminology. This type of reduced-complexity Additionally, other parameters of WIM1-supported scenarios
model is offered in SCM/SCME/WIM for link-level are also slightly tuned for WIM2 according to the newly
simulations and calibration purposes. The support of intra- performed measurements.
cluster delay spread concept in SCME increases number of
“taps” in TDL/CDL model 3 or 4 times, since cluster is
 IV. COMPARISON: PERFORMANCE FIGURES SCME by removing the intra-cluster delay spread. This figure
shows that the correlation of SCM model for bandwidths
A. Fading Distribution and Autocorrelation exceeding 10 MHz is considerably higher than correlation of
The amplitude fading, autocorrelation, level crossing rate SCME and WINNER models.
and Doppler spread are important measures of the channel C. Outage Capacity
model performance. However, statistical analysis of these
Both spatial and polarization characteristics of the models
parameters have shown very similar results in all compared
are investigated by the means of a channel capacity. Based on
models. Figure 1 illustrates as example the temporal
the assumption that the channel state information is unknown
autocorrelation. to the transmitter, the narrowband capacity is calculated by
1 SCME urban macro ⎛ P ⎞
SCME urban micro C = log 2 det⎜ I + 2 HH H ⎟ , (3)
SCM urban macro
WIM C2
⎝ σ ⎠
where I is the identity matrix, P σ is the average SNR, H
Classical Doppler 2
Auto-correlation (d/λ)

0.5
is the narrowband channel matrix and (.)H denotes the
Hermitian transpose operation.
Simulated capacity curves for all (three) scenarios supported
0 by SCM/SCME and for signal to noise power ratio of 14 dB
are shown in Figures 4 to 6. The complementary cumulative
distribution function of capacity is plotted for a several
-0.5
independent channel snapshots and compared to Gaussian
0 1 2 3 4 5 6 7 8 i.i.d. channel matrix. The reference antenna configuration is
Distance/Wavelength, d/λ
4x4 MIMO with two sub-groups of ±45° slanted single
Figure 1. Temporal auto-correlation functions of the
polarized elements whose spacing is 4 wavelength on BS and
equivalent narrowband channels.
½ wavelength on MS, see Figure 3.
B. Frequency Correlation
Under the wide sense stationary uncorrelated scattering
(WSSUS) assumption frequency correlation function (FCF) is
related to the average power delay profile (PDP) through a Figure 3. MIMO antenna configuration for capacity
Fourier transform. Since the compared models are using only calculation.
specular components, FCF is estimated by the sum equation:
R (Δf ) = ∑ Pi exp(− j 2πτ i Δf ) ∑P , i (2) For the all three scenarios we can observe, that the median
i i outage capacity is about 7 bits/s/Hz lower than Gaussian i.i.d.
where Δf is the frequency difference, τi is the delay and Pi is reference curve. As expected, SCME gives equal outage
the power of the ith path. capacities to SCM because its spatial and polarization
characteristics are not changed in the respect to the basic
1
SCM SCM. The outage capacity of the WINNER model (WIM1)
0.9 SCME
WIM1
deviates slightly from SCM/SCME in Urban Micro scenario,
0.8
while in Urban Macro scenario WINNER model shows higher
0.7 capacity than SCM/SCME.
0.6
FCF(Δf)

0.5

0.4

0.3

0.2

0.1

0
0 10 20 30 40 50 60 70 80 90 100
Frequency difference, Δf [MHz]

Figure 2. Frequency correlation function of SCM, SCME,

and WIM1 for Suburban scenario.

Figure 2 depicts frequency correlation of SCM, SCME, and

WIM1 models in 100 MHz bandwidth [4], [14]. For this
comparison, we selected suburban NLOS scenarios (TDL Figure 4. Outage capacity of SCM, SCME, and WINNER
model) from each of them. The TDL of SCM was taken from Suburban Macro scenario.

Figure 5. Outage capacity of SCM, SCME, and WINNER
Urban Macro scenario.
Figure 6. Outage capacity of SCM, SCME, and WINNER
Urban Micro scenario.
V. CONCLUSION
At the beginning of the WINNER project, before having a
full insight into the complete measurement analysis results, the
mainly theoretical extension of SCM (SCME) is supported. In
SCME possibility of using intra-cluster delay spread to model
extended bandwidths is investigated. It was found that the
main limitations of the proposed approach come from the
decision to attain backward compatibility with SCM. Also the
complexity increase, as a consequence quasi-deterministic
drifting of small-scale parameters inside drop, was evaluated.
As could be seen from previous sections, modelling
methodology and certain SCM concepts are still part of WIM.
The similarity is observed in the narrowband capacity
measures that reflect the spatial and polarization properties of
the models. However, without extensions provided in
WINNER it would not be possible to apply SCM-based
methodology to the frequency ranges and extended system
bandwidths. The main strength of WIM is based in scenario-
customized parameters that are making possible to apply
generic spatial-channel-model to the numerous of different
environments. Up to now 12 different scenarios are supported
by WINNER models. Special WIM quality also comes from
the fact that parameterization is based upon the real channel-
sounding measurements.
The WINNER models reproduce correlations of LSPs at
link and system level (intra-site). This property is observed in
measured datasets and has not been modeled in SCM/SCME.
Additionally, WIM2 is proposing a reduced complexity time
evolution of the low-level model parameters in comparison to
SCME.
In this paper SCM/SCME/WIM models are briefly
compared against each others. The future validation work
should show how well these models are representing targeted
system level aspects (when compared to measured channels).
ACKNOWLEDGMENT
This work has been performed in the framework of the IST
project IST-4-027756 WINNER II, which is partly funded by
the European Union. The authors would like to acknowledge
the contributions of their colleagues.
REFERENCES
[1] https://www.ist-winner.org
[2] 3rd Generation Partnership Project; technical specification group radio
access network; spatial channel model for MIMO simulations, 3GPP TR
25.996 V6.1.0, 2003. Available: http://www.3gpp.org
[3] D. S. Baum, J. Salo, G. Del Galdo, M. Milojevic, P. Kyösti, and J.
Hansen, An Interim Channel Model for Beyond-3G Systems, in Proc.
IEEE VTC 2005 Spring, Stockholm, May 2005.
[4] Elektrobit, Nokia, Siemens, Philips, Alcatel, Telefonica, Lucent,
Ericsson, Spatial Radio Channel Models for Systems Beyond 3G,
contribution to 3GPP RAN4, R4-050854, London, September 2005.
[5] 3GPP, E-UTRA MIMO channel model – text proposal, R1-061002
(agreed)
[6] 3GPP, System level channel models for E-UTRA evaluations, R1-061594
(agreed)
[7] 3GPP, Update of polarization description in link level channel model,
R1-061371 (noted)
[8] IST-2003-507581 WINNER D5.4 Final report on Link Level and System
Level Channel Models v1.4, https://www.ist-winner.org.
[9] H. El-Sallabi, D. S. Baum, P. Zetterberg, P. Kyösti, T. Rautiainen, and C.
Schneider, Wideband Spatial Channel Model for MIMO Systems at 5
GHz in Indoor and Outdoor Environments, in Proc. IEEE Veh. Technol.
Conf. VTC’06 Spring, Melbourne, Australia, May 7-10, 2006.
[10] Elektrobit, Telefonica, Siemens, Comparison of SCM, SCME, and
WINNER models, contribution to 3GPP RAN4, R4-060521, Shanghai,
China, May 2006.
[11] IST-4-027756 WINNER II D1.1.1 WINNER II interim channel models,
v1.0, December 2006.
[12] A. Saleh, and R. A. Valenzuela, A statistical model for indoor multipath
propagation, IEEE J. Select. Areas Commun., vol. SAC-5, no. 2, Feb.
1987, pp. 128–137.
[13] M. Narandžić, M. Käske, C. Schneider, M. Milojević, M. Landmann, G.
Sommerkorn, R. S. Thoma, 3D-Antenna Array Model for IST-WINNER
Channel Simulations, IEEE VTC2007-Spring, April 23 - 25, Dublin, IR,
April 2007.
[14] Elektrobit, Siemens, Ericsson, Lucent, Telefonica, Alcatel, France
Telecom, Spatial Radio Channel Models for Systems Beyond 3G,
contribution to 3GPP RAN4 meeting #38, Denver, USA, 13 – 17
February 2006.