EUROPEAN COOPERATION COST 273 TD(02) 164
IN THE FIELD OF SCIENTIFIC Lisbon, Portugal
AND TECHNICAL RESEARCH 2002/Sep/19-20
SOURCE: University of Oulu,
ON THE USE OF MULTI-DIMENSIONAL CHANNEL
SOUNDING FIELD MEASUREMENT DATA FOR SYSTEM-
LEVEL PERFORMANCE EVALUATIONS
University of Oulu
ON THE USE OF MULTI-
DIMENSIONAL CHANNEL SOUNDING
FIELD MEASUREMENT DATA FOR
Tad Matsumoto1 , Uwe Trautwein2 , Christian Schneider3 , Reiner Thomä3
University of Oulu, Oulu, Finland
Medav / Tewisoft GmbH, Ilmenau, Germany
Technische Universität Ilmenau, Ilmenau, Germany
Abstract: Current advances in multi-dimensional channel-sounding techniques make it possible to
evaluate performances of radio multiple access and signal processing schemes in realistic propagation
conditions. This paper focuses on how recorded impulse response data gathered through multi-
dimensional channel sounding field measurements can be used to evaluate system-level performances of
the radio access schemes. Methods for link- and system-level simulations are presented. Performance
curves of a space-time equalizer, obtained as the results of the simulations, are also presented as an
example to demonstrate the effectiveness of the evaluation scheme using the field measurement data.
Evaluating system level performance figures such as outage probability and/or blocking probability is
crucial when determining new air interface specifications for future mobile communication systems.
The system level performance figures can be estimated from the geographical distributions of link-
level performance figures of the radio access schemes and signal processing algorithms such as bit
error rate (BER) and signal to interference plus noise power ratio (SINR). Previously, such system
level performances have been evaluated through field experiments using system prototype. T he
prototyping, however, has imposed heavy burden on those who don’t have enough budget or human
It is quite natural to set the bit rate targets for post-third generation (3G) systems higher than 3G’s
maximum speed, with the aim of supporting real-time multimedia communications. Given that the
system bandwidth needed to realize broadband communication may not be fully available, the post-3G
networks will have to have greater resistance to co-channel interference (CCI). It is obvious that
broadband signaling over mobile radio channels imposes severe inter-symbol interference (ISI) upon
the received signals due mainly to the effect of multipath propagation. Hence, a technological
breakthrough that can reduce the effects of CCI and ISI while taking full advantage of the benefits of
multipath transmission is needed.
Furthermore, unlike the 3G and prior systems, which mainly exploit the received signal’s temporal
structure, post-3G systems must jointly exploit the received signal’s temporal and spatial structures.
Such systems having resistance against CCI and ISI should be rather complex, and hence huge risk has
to be tolerated even for its prototyping.
A lot of efforts have been made to create a new propagation model  that can well express the two
dimensional (spatial and temporal) channel structure. The model-based simulation is quite effective in
evaluating fundamental performances of radio access schemes. However, when performances have to
be evaluated in more realistic conditions, field measurements using system prototype is still required.
Current advances ,  in multi-dimensional channel-sounding techniques make it possible to bridge
the gap between the model-based simulations and filed measurements using system prototype. Multi-
dimensiona l channel sounder systems have multiple receive antenna elements. Received channel
sounding waveform transmitted periodically from the transmitter is analyzed at the receiver, and the
channel impulse response (CIR) from the transmitter to each of the antenna elements is recorded. This
data is then used for off-line simulations , . Since the CIR data represents realistic propagation
scenario, results of the off-line simulations accurately express performances of the radio access
schemes in real fields.
Furthermore, if performances of different radio access and signal processing schemes are evaluated
using the same CIR data through off-line simulations, fair comparison can be made among the
candidate schemes. This should avoid the risk of prototyping, thereby it is made possible for those
who don’t have enough budgetary and human resources to make substantial contributions to the post-
3G system concept creation.
This paper focuses on how recorded impulse response data gathered through multi-dimensional
channel sounding field measurements is used to evaluate system-level performances of the radio
access schemes. Methods for link- and system-level simulations are presented. Performance curves of
a space-time equalizer, obtained as the results of the simula tions, are presented as an example to
demonstrate the effectiveness of the evaluation scheme using the field measurement data.
2 Link- and System-Level Simulations
We define two terms used in this article, link-level simulation and system-level simulation. Both are
off-line simulations using multidimensional channel sounding field measurement data, but their goals
(1) Link-level simulation aims to evaluate ST-equalizer link-performance in terms of bit
error rate (BER) in various propagation environments. It focuses on
analyzing/evaluating the impact of ST-equalizer configurations, parameters, and
algorithms on performance.
(2) System-level simulation aims to obtain system-level performance figures such as
signal-to-interference plus noise power ratio (SINR) and BER in terms of either
geographical distributions in the area of interest or outage probability. Main focus is
on interference scenarios, which depend on cell design including frequency reuse and
2.1 Field Measurements
Prior to the simulations, multi-dimensional channel sounding field measurements have to be
conducted based on the business scenario considered: e.g., indoor or outdoor, micro-cell or macro-cell,
frequency band, and/or duplex method. The field measurement campaign then takes place to collect
enough sets of CIR data. Figure 1 shows as an example the locations of CIR measurements conducted
in downtown Tokyo, where a typical sectored macro-cell scenario was assumed: There are 4 60-
degree sectors, A1 , B1 , B2 , and B3 , in their combined triangle . The business scenario in this example is
that each sector has one user using the same frequency and time slots. The antenna array of the
channel-sounder was located on top of a building with a height of approximately 40 meters. Other
surrounding buildings had almost the same height. The transmitter was moved along a street. There
was no line of sight between the transmitter and receiver in all propagation scenarios investigated. The
output of the field measurement is a series of data sets indicating the measured impulse responses of
the channels between the transmitter’s omni-directional antenna and each of the 8 elements of the
receiver antenna array.
2.2 Link-Level Simulation
The basic concept of link-level simulation is to replace CIR generated according to a model by the
measured CIR data. Since the CIR data between the omni-directional transmitter antenna and each of
the L antenna elements is available, each element’s received signal waveform can be calculated by
convoluting the transmitted desired waveform, corresponding to the radio access scheme to be
evaluated, with the measured CIR data. Received waveforms of co-channel interference signal can
also be calculated in the same way from the CIR data measured at different transmitter positions.
2.3 System-Level Simulation
Assume that the user in A1 is desired, and those in B1 , B2 , and B3 are interference users. Also assume
that there are i, j, k, and l sets of measured CIR data in the sector A1 , B1 , B2 , and B3 , respectively.
Then, there are ijkl combinations of locations. By performing the link-level simulations for each of the
ijkl locations, cumulative distribution of the link-level performance figures can be obtained. From the
obtained cumulative distribution curve, it can easily be estimated that at what percent of the
geographical area, acceptable communication quality can be achieved.
Planning of a field measurement campaign should reflect the assumed interference scenario in terms of
reuse distance, and cell layout. Obviously, the larger the ijkl value, the better the obtained cumulative
distribution curve reflects the real scenario .
This section describes as an example the results of system-level simulations conducted to evaluate a
space-time (S/T-) equalizer . Figure 2 shows a block diagram of the S/T-equalizer. It consists of a
cascaded connection of an adaptive array antenna and the maximum likelihood sequence estimator
(MLSE) , : each of the adaptive array antenna elements is equipped with a fractiona lly spaced
tapped delay line (FTDL), and the MLSE has taps covering a portion of the channel delay profile (This
S/T-equalizer configuration is referred to as FTDL/MLSE for convenience). The MLSE estimates the
sequence considered most likely to have been transmitted.
Key parameters are the numbers L, M, and N of the antenna elements, the FTDL taps, and the taps in
MLSE, respectively , which are expressed as (L, M, N) for notation convenience. The N taps in MLSE
are used to replicate the signal at the array output corresponding to the symbol sequence selected by
MLSE. There are LM+N taps that have to be adaptively determined according to the channel
conditions such as incident angles and strengths of the desired and interference users’ multipath
components. MLSE uses the Viterbi algorithm, for which the number of the states is QN-1 for Q-level
signaling. For quaternary phase shifted keying (QPSK), Q=4.
The LM+N tap weights are determined so that
WH X → min, (1)
where vectors W and X are the weight and sampled data vectors, respectively. The first LM entries of
W are the tap weights on the L antenna elements, and the following N entries are the MLSE taps, as
W = [Wa , Wt ]T with
Wa = [Wa11, Wa12 ,L ,Wa1M ,Wa 21,Wa22 ,L ,Wa2 M , L ,WaL1,WaL 2 ,L ,WaLM ] (2)
Wt = [Wt1 ,Wt 2 ,L ,WtN ] . (3)
The data vector X has the same structure as W, as X t = [ X t1 , X t2,L , X tN ] with
X a = [ X a11 , X a12 L , X a1M , X a21, X a22 L , X a2M ,L , X aL1 ,X aL2 L , X aLM ] (4)
X t = [ X t1 , X t2 ,L , X tN ] . (5)
Elements of Xa are of the samples taken at the outputs of FTDL on the L antenna elements. During the
train ing period, elements of Xt are the sequence of signal points corresponding to the signal reference
(= unique word sequence) transmitted at the head of the frame. The tap weights can be determined
recursively by using the recursive least square (RLS) algorithm.
As shown in Fig. 1, four users were considered in the simulations. The user in A1 is desired, and those
in B1 , B2 , and B3 are interference users. Assuming that the attenuation in signal strength due to the
path-loss is proportiona l to the 3.7th power of distance, and that shadowing is an independent random
variable distributed over a Log-Normal distribution with standard deviation of 10.0 dB, overall loss
with the channel between each user and its home base station was calculated. Pass-losses due to the
distances between each interferer’s location and desired user’s base station were also calculated, and
their corresponding shadowing losses were computer-generated, and multiplied by the pass-losses.
The desired and interference users were assumed to be power-controlled by their home base stations
so that the received signal-to-noise power ratio (SNR) becomes 15 dB. The effect of the power control
on strengths of the interferers’ signals received by the desired user’s base station was then taken into
There are 11, 8, 11, and 11 measurement points in A1 , B1 , B2 , and B3 , respectively. 10648 (=11 x 8 x
11 x 11) combinations were taken by randomly choosing the desired and three interference users’
locations from among the 41 field measurement points. For each of the 10648 combinations, the
overall channel losses due to the distance, shadowing, and power-control were calculated, and the
channel losses were multiplied by the field measurement CIR data. The signal processing for the S/T-
equalizer was then performed, and the cumulative distribution function of the MLSE input SINR
Total of Combinable Path Energies
SINR = (7)
Mean Squared Error after Convergence of the RLS Algorithm
was evaluated using the 10648 combinations of the field measurement data.
Figure 3 shows cumulative distribution functions of the MLSE input SINR for (L, M, N)=(8, 7, 5), (8,
1, 1), (1, 7, 5), and (1, 1, 1). Obviously, the performance with the (1, 1, 1) configuration is the worst
among them. Systems using the (1, 1, 1) configuration (=non-adaptive omni-directional antenna) may
need spreading of the signals in the frequency domain to achieve the process gain at the receiver side.
Neither the (1, 7, 5) configuration (=omni-directional antenna with FTDL/MLSE temporal-equalizer)
nor the (8, 1, 1) configuration (=8-element adaptive array antenna) achieves reasonable performance if
the requirement for the system outage is 5%. Even if the outage requirement is 10%, they need
auxiliary technique such as error correcting coding that has to work properly with 1..2 dB SINR.
The (8, 7, 1) configuration (S/T-equalizer without MLSE) achieves much better performance than the
(1, 1, 1), (1, 7, 5), and (8, 1, 1) configurations: if the requirement for the system outage is 5%, the
receiver has to work properly with 6 dB input SINR. Obviously, the (8, 7, 5) configuration achieves
the best performance: the cumulative distribution curve for (8, 7, 5) indicates that at 95% of the sector
A1 , the MLSE input SINR can be made larger than or equal to 8 dB. In fact, 8 dB input SINR is
sufficient for MLSE to achieve acceptable level of BER, say, 10-6 , that would support multi-media
communications , with a help of simple error correction coding.
The main focus of this paper has been on how recorded impulse response data gathered through multi-
dimensional channel sounding field measurements can be used to evaluate system-level performances
of the radio access schemes. Methods for link- and system-level simulations were presented. As an
example of the link-level simulation, methodology for evaluating up-link system-level performance of
a system utilizing a space-time equalizer is briefly described.
 Correia, L. M. (Ed) “Wireless Flexible Personalised Communications: COST 259
European co-operation in Mobile Radio Research”, Wiley, 2001
 Elektrobit AG, www.propsound.com
 MEDAV GmbH, www.channelsounder.de
 R. S. Thomä, D. Hampicke, A. Richter, G. Sommerkorn, U. Trautwein, “MIMO Vector
Channel Sounder Measurement for Smart Antenna System Evaluation”, European Trans.
on Telecomm., ETT Vol. 12, No. 5, Sept./Oct. 2001
 T. Jämsä, J. Kolu, Tsukamoto, T. Matsumoto, “Real-Time Simulation of Adaptive Array
Antenna using Broadband Vector Channel Simulator”, to be presented on WPMC 2002,
 T. Matsumoto, T. Yamada, and S. Tomisato, ”System- Level Performance Evaluations of
Spatio-Temporal Equalizers Using Two-Dimensional Channel Sounding Field
Measurement Data”, Conf. Record, 27th General Assembly of International Union of
Radio Science Maastricht, The Netherlands, 19 August, 2002
 T. Yamada, S. Tomisato, T. Matsumoto, and U. Trautwein, “Results of Link-Level
Simulations Using Field Measurement Data for an FTDL-Spatial/ MLSE-Temporal
Equalizer“, IEICE Trans. Commun., Vol. E84-B, No. 7 July 2001, pp. 1956-1960
 T. Yamada, S. Tomisato, T. Matsumoto, and U. Trautwein, “Performance Evaluation of
FTDL-Spatial/MLSE-Temporal Equalizers in the Presence of Co-Channel Interference –
Link-Level Simulation Results Using Field Measurement Data - “, IEICE Trans.
Commun., Vol. E84-B, No. 7 July 2001, pp. 1961-1964
BSA Sector A1
Sector B 2
Array Receiver There are:
11 Points in Sector A 1,
Sector B 3
8 Points in Sector B1 ,
11 Points in Sector B2 , and
11 Points in Sector B3 .
: Measurement Point
10648 (=11 x 8 x 11 x 11)
Combinations in Total
Fig. 1 Field Measurements for System-Level Simulations
X a11 X a12 Xa1(M-1) X a1M
T/2 … T/2
#1 + 2 Viterbi
Wa11 Wa12 Wa1(M-1) Wa1M -
+ X tN Xt(N-1) Xt2 Xt1
WtN Wt(N-1) Wt2 Wt1 =1
Key Parameters (L, M, N)are the Numbers of:
Antenna Elements L,
Feed-Forward Filter Taps M on Each Antenna Element,
MLSE Taps N.
There are LM+N Taps in this Configuration.
Fig. 2 S/T-Equalizer Configuration
Cumulative Distribution Function (%)
(1,7,5) w/ MLSE
1 Desired User,
0 5 10 15 20
MLSE Input SINR (dB)
Fig. 3 Results of System-Level Simulations for a S/T-Equalizer Configuration