A Multipath Fading Channel Model Applied to Fast

UMTS Link Simulation with Channel Estimation

~

Pierre Gelpi, Ahmed Saadani, Stefan Wendt

France TMcom R&D - 38, rue du Gntral-Leclerc
92794 lssy Moulineaux Cedex 9 - France
pierre.gelpi, ahmedsaadani, stefan.wendt@francetelecom.com
Abslracr-In this paper, a fast link simulation method with
channel estimation is presented. The conditions of uncorrelated
channel paths and large spreading factor are assumed. The
Rayleigh channel model used is based on a first order Markov
chain, valid for all fading speeds. The implementation of this
model is simple. The results presented in this paper show a
good fit between the performances obtained with the traditional
simulator and the fast one. for Universal Mobile Telecommunication System (UMTS) Frequency Division Duplex (FDD) uplink at
12.2 kbps for Third Generation Partnership Project (SGPP) and
International Telecommunications Union (ITU) channels. The
high computing time gain that has been measured opens a wide
range of perspectives for the method proposed in this paper.

1. FIRSTSIMPLIFICATION LEVEL: THE RECEIVER MODEL

When the spreading factor is large, the inter symbol into consider that
for a transmitted symbol s, the received.,symhol on path i
after depsreading is h , s n, where n, is an additive white
gaussian noise. If real channel estimation is assumed, the Rake
correlator for path i performs the product by h; and the output
of the Rake receiver can he written

terference (IsI) is low and it is

INTRODUCTION
For mobile communications systems using Code Division
Multiple Access (CDMA) such as UMTS, computing each
part of the communication chain is complex and leads to an
important simulation time. A fast simulator with a reduced
complexity is vety important to study and design such radiomobile systems. The multipath fading channel model is one
of the most crucial parts. Several fading channel models [I],
[21 allow an important computing time gain factor for slow
or fast fading. But for intermediate regimes, a more precise
model is needed [31.
In this paper, a fast link simulation method with channel
estimation is presented. It is based on two simplification levels.
First, the receiver model is based on perfect spreading which
consists in computing only the coded symbols and not the
chips. The process at the Rake receiver output is represented
by a quadratic form deduced from gains of the independent
channel paths. Non-correlated channel paths are assumed.
This is verified when paths are spaced by a multiple of the
chip duration because pulse shaping filters do not introduce
correlation in this case. The second simplification level is
brought by the simplified path generator. This channel model
is valid even for intermediate regimes. It is achieved through
a first order Markov chain based channel path generator
followed by an interpolation.
Section I describes the receiver model which consists in
a first simplification level and section I1 explains how the
simplified path generator brings an additional computing time
gain. Finally section I11 describes the performance results and
the computing time gain factors obtained for UMTS FDD
uplink at 12.2 khps for 3GPP and ITU channels with perfect
power control and real channel estimation.

a7803.7757.5/03BI7.W

BZW3 IEEE.

. .

This Rake receiver model is described'on figure 1

Figure 1. Rake receiver model

It is now clear that generating the h,'coeficients would

allow to simulate the soft output of the Rake receiver for a
system that verifies the above conditions:The h, coefficients
are complex gaussian distributed but channel convolutional or
turbo coding is very sensitive to time properties. Therefore
the auto-correlation function of the generated path has to be
very similar to the exact one which is R ~ ( T=)b o J o ( 2 x f o r )
[41_ where bo is the variance of the process, Jo is the zeroorder Bessel function of the first kind and fo is the maximum
Doppler frequency.
A solution is to generate these complex gaussian coefficients
with the usual channel simulation method. This method is
known as the Jakes method and consists in generating two
gaussian processes corresponding to the in-phase and quadra-

2548

ture components of the path:

are
where E: is the path energy, N is large, the (C,,)I<~<N
normalized, the ( W , ) ~ < ~ S , V are the Doppler shifts between
- 2 r f ~ and 2 n f ~ and
, the ( & , ) I ~ ~ S N
are uniformly distributed between 0 and 2 ~ .
Equivalently it is possible to filter two orthogonal gaussian
centered variables with the Doppler transfer function

2F1

Model Y = l
FSMC M=1

L.
. ,

200.

..

.~

Figure 3. Avlc-correlation functions of the usual path,ampliNde (ZFI),

the one obtained without interpolation (FSMC M=I) and the one from the
simplified generaror (Model M=l), M is the Markov chain order; fDTaums =
1.7810-a.

as illustrated on figure 2. The path energy is 207

adapt exactly the generated paths rate to the symbol rate of

the simulation. Precisely, if R is the symbol rate; the adequate
interpolation factor is K =

R
and the path coefficients hi
. . , .

2fD
are obtained from the outputs y; of the Markov.chain by
Figure 2. Complex gaussian dirfribution generation

It brings a first simplification level and the gain factor

is important, because only the symbols are computed, not
the chips. But it i s possible to increase this computing time
gain by generating the h , coefficients using a first order
Markov chain and an interpolation. Next section describes
this simplified generator.
11. SECONDSIMPLIFICATIONSTEP: THE PATH GENERATOR

The path generator is based on the flat fading model

introduced in [3]. A first order Markov chain is built from
usual simulation samples at a fixed sampling frequency of
2 f ~ This
. frequency satisfies the sampling theorem because
the channel is band limited. Figure 3 shows that the autocorrelation functions of the path amplitude generated by the
usual way and the one of the path amplitude from this
generator are similar. Even if there is a liltle difference on
the second peak value, the oscillations of the two curves are
simultaneous. On the contrary, the auto-correlation function
of the path amplitude generated without interpolation has no
oscillation in this case and the samples are very correlated.
Other sampling frequencies lead to the same result, but with
a higher order Markov chain. The advantage of 2 f D is tc
keep order 1, which is faster to compute than higher orders.
Because of the fixed sampling frequency, the generated paths
do not correspond yet to the symbol rate of the simulation.
An interpolation, justified by the sampling theorem, allows to
2549

sin(+
1
1
where sinc(z) = -, TS&a = - and T
.,,, = 2:
R
,
2fD'
The complete simplified simulator is shown on figure 4. :

Figure 4. Simulator with the simplified path generator

The first operation consists in generating the path samples

with the usual simulator at a frequencyof 2 f D . In fact, the
sampling frequency can be normalized so that it does not even
depend on the mobile speed. This makes the channel :path
generator very simple to implement. Then, the Markov chain
outputs have to be interpolated and the interpolation factor
is calculated as explained above. Finally the computation of
equation (I) leads to the soft outputs of the Rake receiver.
111. APPLICATION
TO FAST LINK SIMULATION: RESULTS

The results presented in this paper were obtained for UMTS

FDD uplink at 12.2 kbps with perfect power control and real
channel estimation. The Dedicated Traffic Channel (DTCH)
coded Bit Error Rate (BER) and Block Error Rate (BLER)
obtained with the usual simulator are compared to the ones
given by the fast link simulator described in the previous

sections. The computing time gain factor is 160, which is very

important. The results for 3GPP channel Case 2 at 3 kmh are
presented on figure 5 and the ones for ITU channel Vehicular
A at 3 kmlh on figure 6. The definition of these two channels is
given in table 1. A good fit, with parallelism and differences
of less than 0.3 dB, is observed between the performances
obtained with the two simulators for both channels.
The little differences between usual error probabilities and
simplified ones come from the small imperfection of the
model used at the second simplification level, for example the
peak value of the auto-correlation function first oscillation.
The correction code used is a 1/3 convolutional code. The
parallelism between the usual performance curve and the
simplified one is an important result because the performance
of this code is very good and very sensitive to the correlation
between samples.

TABLE I
3GPP C A S E 2

ITU VEHICULAR A C H A N N E L S DEFINITION

AND

I delay (ma) 1

attenuation (dB)
3GPP C a ~ e2

path #

I 1
2 1
3
I

976

zowo

ITU Vehicular A

710
4

1090

-15
-20

1730
2510

. . . . . . .. . .
. . . ? . CONCLUSION AND PERSPECTIVES

The Rake receiver model described in this paper corresponds to the conditions of uncorrelated paths, large spreading
factor and real channel estimation. The channel path coefficient generator used by ,the fast simulator is based on a first
o<der,Markov chain model. The fixed sampling frequency of
2 fD'followed by an interpolation keeps the model valid for
intermediate regimes with the Markov chain order limited to
one. It also makes the simulator implementation easier because
the Markov chain training is done only once for all simulation
parameters.
The high computing time gain obtained with this fast
simulator allows to evaluate the performance of a mobile
communications s'ystem .with real channel estimation for a
large number of different.parameters and services, much more
rapidly than with the traditional simulator. This opens a wide
range of perspectives for the simulation method presented in
this paper. Future work will introduce real power control,
trying to keep a large computing time gain factor.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..~. . . . . . . . .
.
.
.
.
.
6

I
EM19(dB)

*1

$0

13,

12

~
I4

Figure 5 . Case 2, 3 kmlh. 12.2 kbps

REFERENCES
[ I ] , Ch. C. Tan,N. C. Beaulieu, On Finf-Order Markov Modeling for fhe
R@leigh Foding C&d,
lEEE Transacdonf on Communications, vol.
48, na:I2. 2000:
[2] P. Kucrynski. A. Rigoll6, W. Gerstacker, J. Huber, Hidden Morkw
,Morleling of Enor Pollem mdS0~7Ourprirsfor Simulorion of Widebond
CDMA Tmnsmission Sysfem. Proceedings of 2000 European Confaence mi'Wmless Technology (2000 E W , pp. 135-138, 2000.
131 A. 'Saadani. P. Tortelicr, A Firsr Order Markov C h i n Based Model
for FLU fading Chmnel. IEEE Personal lndwr and Mobile Radio
Communications. 2W.
[4] 'W. C:Jakes:Micmw&e Mobile Com"icofionr. John Wiley & Sons,
1974.

. . ^..
.

. .. .

.. .~

. . . . j. ..... :..... .

...

. .. ,

.~

...... . . . . .%":...

.... : . . . . ... . . . ... . . . . . . . . . . . . . . . . . . . . .

"'t
10-

:..

.,

;...:
. ..

~1

..: ..............

.\:.....!

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .~
...................
I

10

?I

$2

13

EW(dB)

Figure 6. Vehicular A. 3 kmlh. 12.2 kbps

2550

14

1s