UPLINK VERSUS DOWNLINK WIRLESS MOBILE

POSITIONING IN UMTS CELLULAR RADIO SYSTEMS

Wrya. Muhammad1 , Emmanu`ele Grosicki 2 , Karim Abed-Meraim 2 , Jean Pierre Delmas1 ,
Francois Desbouvries1
1
Institut National des Telecommunications, Dept. CITI, Evry, France
2
Ecole Nationale Superieure des Telecommunications de Paris, Dept. TSI, France
e-mail: wrya.muhammad@int-evry.fr, grosiski@tsi.enst.fr
ABSTRACT
A performance comparison between radio location positioning (LP) in Wide band Code Division Multiple Access
(WCDMA), more precisely in UMTS FDD cellular phones,
using the Uplink (UL) and Downlink (DL) scenarios, is carried out. The radio location algorithm using the Time Difference Of Arrival (TDOA) is applied, where the TDOA is the
difference between two TOA estimated at two base stations.
We propose a practical algorithm to obtain high precision
PL under high level interference. It consists in a matched
filter followed by high resolution MUltiple SIgnal Classifiction (MUSIC) algorithm to estimate with a high resolution
the TOA. indeed, in practice nearfar resistant detection algorithms are not easy to implement directly on the received
data due to the large spreading factors. The powerup and
Idle Slot (IS) techniques together with Interference cancellation algorithms are used to combat the nearfar problem.
Simulations were carried out in the same observation conditions for UL and DL scenarios. The results show an enhancement in the LP using MUSIC algorithm after matched
filtering besides, a better performance in the DL scenario.

INTRODUCTION

Radio location is a growing practice in cellular mobile phone

systems because of its vast useful applications like: localizing
the traffic by the operators, emergency interventions, billing
tariffs depending on the mobile position and discouraging
criminal activities using the cellular systems. Techniques of
radio location were developed for the GSM cellular phone
and are now standardized. However, these standards are not
applicable to the wide band CDMA cellular systems since it
uses a different air interface technology. Studies on the radio
LP of cellular phones in the IS95 CDMA systems have been
performed [1] [2]. But this system is also different from the
Universal Mobile Telecommunications System (UMTS) wide
band CDMA standards. Few studies were carried out on the
radio LP in UMTS systems [3] [4].
One of the limiting factors of the TDOA radio location algorithms is the Non Line of Sight (NLOS) propagation paths,
since it introduces a bias error on the TOA. In [5] an essay
is introduced to discriminate the LOS from the NLOS paths.
In this work we consider the first multi path component as
the LOS path. The TDOA is deduced from the difference
beween two TOA estimated at 3 or more base stations (BS).

Each two TDOAs are then used to obtain unambiguous hyperbolic curves to position the cellular systems. A matched
filter is used to estimate the TOA at the BS or the mobile
station (MS) for UL or a DL radio LP scenarios, respectively.
However, the matched filter is unable to resolve multi path
components that arrive within Tc /Q, where Tc is the chip
duration and Q is an oversampling factor besides, the multiple users interference will be highly present at the matched
filter output. Hence, in the UL scenario, interference cancellation and high resolution estimation of TOA is needed to
obtain accurate PL, while in the DL the intracell interference
will not highly affect the matched filter output. Nevertheless,
the output of the matched filter, after interference cancellation, can be used to obtain high resolution TOA estimations:
the idea of using superresolution method in the context of
delay profile estimation of a channel has been proposed in [6].
In this work we study the application of a superresolution
method using MUSIC algorithm together with interference
cancellation on the PL.
The Signal to interference and Noise ratio (SINR), as in all
other signal processing problems, is another limiting factor
because of its direct relation with the variance of the TOA
estimation. Besides, since the CDMA systems is interference
limited, a power control algorithm must be used to minimize
the signal strength at the neighboring stations or mobiles (i.e.
to combat the nearfar problem). As a result, it is difficult
to hear, with a good SINR, a mobile phone at multiple BS,
or to hear multiple BS by the mobile phone; in DL and UL
location positioning scenarios, respectively. In practice, the
SINR is increased using two techniques: (1) the powerup in
the UL LP scenario (2) the Idle Slot (IS) in the DL PL scenario [3]. However, these techniques have the disadvantages
of reducing network capacity. In fact, theoretically nearfar
resistant estimation methods could be applied directly to the
received data to avoid the nearfar problem. Unfortunately,
in practice they are very expensive to implement due to the
large spreading factors.
Finally, many factors can be considered in the choice of the
LP scenario: cost, accuracy, communication interruptions
during localization etc. In this paper we are interested in
the accuracy performance of the two mentioned LP scenarios.

SIGNAL MODELS

In this paper we tackle Frequency Division Duplex (FDD)

UMTS system. The signal models for this system are con-

structed according to the standards proposed by the ETSI

[7].

2.1

Uplink Signals

The UL baseband signal referring to the scheme proposed by

[7] (see figure (1)) can be written as:
xk (t) = {

+
X

(l)

(l)

[dk bIk (t lT ) + jck bQ

k (t lT )]}sk (t)

(1)

l=

where T is the data and control bit duration, sk (t) is the

complex scrambling code of the k th user selected from the
Large Kasami family. bI (t) and bQ (t) are the channelization
(l)
(l)
codes of the dedicated data dk {1, 1} and control ck
{1, 1} channels. These codes are orthogonal and selected
from the Walsh family. sk (t), bIk (t) and bQ
k (t) can be written
as:
PL1
PLc 1 I
sk (t) =
s (l)(t lTc ); bIk (t) =
bk (l)(t
l=0Pk
l=0
Lc 1 Q
lTc ); bQ
(t)
=
b
(l)(t

lT
)
c
k
k
l=0
where L is the scrambling code length and Lc is the channelization code length or the spreading factor (SF), sk (l)
{a + jb : a {1, 1}, b {1, 1}}, bIk (l), bQ
k (l) {1, 1}
and is the chip waveform here considered as a rectangular
function of width Tc equal to the chip duration.
channelization code bI
(t)
k

Re{}

(i)
Data channel d
k

channelization code b

Q
(t)
k

Im{}

DPCCH

control channel c

(i)
k

Complex Scrambling code

sk (t) selected from large Kasami
codes of length 256

Figure 1: Uplink spreading and modulation of the k th

user
The LP can be carried out in the UL scenario either with
known or unknown signals in the control channel. Unknown
signals can be used provided that the scrambling code length
extend over one bit. This is possible in UMTS systems where
short scrambling code, such as the large Kasami or Gold code
families of length L = 256, could be allocated in UL and DL
[8]. However, an ambiguity problem will result using these
short scrambling codes, if the search radius is greater than
256 Tc c ' 18 km, where c is the electromagnetic wave
speed. Hence, this is not a drastic problem because normally
suburban cell radius have smaller dimensions.
The UL baseband signal of equation (1), with synchronous
rectangular functions of the scrambling and channelization,
can be rewritten for the th bit and equal data and control
rate, as: P
Lc 1 () I
()
()
xk (t) = l=0
[dk bk (l) + jck bQ
k (l)]sk (l + Lc )
(t (l + Lc )Tc )

2.2

common
pilot
channel

S
to
P

channelization
bp (t)

Scrambling code sBS (t)

Other physical channels

Ck (l)
of K users

to
Modulation

Gk

Figure 2: Downlink spreading and modulation for K

user
The baseband signal at the th bit of the pilot channel at
any BS is then:
PLc 1 ()
PK
()
xBS (t) = l=0
{[pe + jpo ]bp (l) + k=1 Gk Ck (l)}
sBS (l + Lc )(t (l + Lc )Tc )

Modulation

DPDCH

mon pilot channel with a known sequence. This pilot channel

is used for channel estimation by the mobile phone. It is normally transmitted continuously with a relatively high power
and no power control is applied. Estimating the TOA of these
signals is a good choice for LP. The channelization codes are
Variable Orthogonal SF codes (OVSF) Walsh codes.

Downlink Signals

Figure (2) shows the DL modulation and spreading for K

users [7]. It can be noted that the base station uses a com-

where the bp (l) is the Walsh code used for the common pilot
channelization, sBS is the scrambling code a base station and
()
()
pe , po are the th even and odd bits of the common pilot
channel, respectively. Ck (l) represent the sum of all physical
channels of a user k and Gk is the gain factor of the user
k. In the following we omit the superscript to simplify.
The complex baseband signal is then QPSK modulated and
transmitted.
The noise free bandpass signal y(t) received through a static
multi path channel with coefficients hi , where i = 1, .., M is
the channel length, is: P

M
y(t) = i=1 hi 2P x(t i )
where hi is a complex fading coefficient, i is the time
delays of the ith path and P is the carrier transmitted signal
power. The QPSK signals at the receiver are down converted
integrateanddumped with an integration time Te = Tc /Q
with Q . We will assume, without loss of generality, that
the i are multiples of Te (i.e. i = i Te where i )
this assumption is realistic if Q >> 1. Furthermore, we
assume that the integrateanddump is synchronous with the
received chips. These assumptions simplify to a great extent
the equation writings at the receiver without producing big
performance changes.
The sampled received signals at the BS for the UL case is:
y

UL

(nTe ) =

K Mk
X
X
k=1 i=1

L
k,i Te ) + w(nTe )
hU
k,i xk (nTe

(2)

L
ul
j(c k,i +k )
th
where hU
, hul
user comk,i = hk,i Pk e
k,i is the k
plex coefficient channel gain, k is a phase error between the
received and local carrier, Mk is the kth channel length, c
is the carrier angular frequency and w(nTe ) is a zero mean
complex additive white Gaussian noise with variance 0 /Te
where 0 is the noise power at the receiver input. In matrix

form this is written as:

QL(

yU L = XhU L + w
M )

(3)
M )

k
k
k
k
. The sampled
and hU L CI
where X CI
received signal at the mobile phone for the DL case is:

y DL (nTe ) =

M
X
i=1

hDL
xBS (nTe i Te ) + w(nTe )
i

(4)

where hDL
= hdl
PBS ej(c i +) . Or in matrix form:
i
i
yDL = XBS hDL + w

(5)

where XBS CI QLM and hDL CI M .

ESTIMATION OF TOA

Direct application of MUSIC algorithm to equations (3,5) is

very expensive to implement due to the large spreading codes
besides the oversampling factor Q. In practice, matched filtering is more convenient, especially since it can be followed
by a MUSIC algorithm to obtain higher resolution estimations. In fact, in this case, the MUSIC algorithm is carried
out on a data length equal to an order of magnitude of the
channel length and not equal to the code length. Nevertheless, matched filtering is not nearfar resistant. As we mentioned powerup and IS techniques are proposed to combat
the problem of hearability in the UL and DL scenarios, respectively. However, the power in the UL scenario cannot be
increased infinitely, this is why nearfar resistant algorithms
like Parallel Interference Cancellation (PIC) must be used in
the BS to improve the TOA estimation. On the other hand,
the IS are periods of silence of BSs positioned in a predictable
manner and communicated to the mobiles. This will allow
the mobiles, in the silent BS, to hear the far BS.
In the UL scenario a bank of matched filters matched to the
control channels of the users is used as the first stage of PIC
in the BS will give a first estimation of the channels coefficients:
L
L
U L() = SH X1 hU
+ SH w
+ S H Xt hU
h
t
1

(6)

where H represent transconjugate, S is the matrix of the

locally generated replicas of the scrambling and the control
channelization
codes of all the users, X1 CI LQM , Xt
P
(LQ

M )

k
UL
L
k6=
CI
are the searched user and all
and hU
1 , ht
other users channel coefficients, respectively. The second
term represents interference to the searched users channel
U L() for the interferestimation. PIC uses first estimation h
ence subtraction. Since the data and control bits are available
at the BS, then X can be reconstructed, and we get:

L UL
L
UL
U L() = SH XU
U L() ) + SH w
h
+ S H XU
h
1 h1
t (ht
t

The second term now is less important than that of equa0

tion (6). SH X1 = RU L CI L M is a crosscorrelation matrix between the global signal and the scrambling and control
channelization code.
A matched filter in the mobile phone (DL case) with the pilot
channel data known will give:
DL
DL
DL = SH
+ SH
= SH
h
pw
p XBS h
py
DL
DL
= Rp hDL + Rt hDL + yw
= RDL hDL + yw

where RDL
CI LQM . The second term is unknown for the
p
mobile which represents the intra-cell interference and it is
not negligible and depends on the orthogonality factor of the
link level.
In the two cases, an averaging of these matched filter outputs
over a number Jof independent snapshots is necessary to get
rid of the noise and interference components. A threshold

then should be chosen and the values of the arguments of h

where h > are relative to TOA estimations. It is clear that

if the paths i are close to each other (to less than one chip
period) then the matched filter is unable to resolve them and
a bias could be introduced. In order to have more accurate
results a hight resolution method must be applied.
The MUSIC algorithm can be applied to the output of the
after an interpolation of the latter by
delay profile vector h
N times to obtain higher resolution TOA estimations [6].
The correlation matrix of the delay profile vector can be estimated using the J independent snapshots:
=

PJ

j=

H
j h
H
h
+
j = RGR

R0
Te

where R0 is the Hermitian correlation matrix of the descrambling code and G is the correlation matrix of the
channel coefficients. Now, the generalized
eigenvalue probP
i = i R0 ei for i = 1, . . . ,
M
for the UL and
lem e
k
k
i = 1, . . . , M for DL, must be solved. The TOA estimations
are then the maximums of the criterion:
v( ) =

r rH

r EEH rH

(7)

where E is
Pthe matrix of the column eigenvectors from
i = K, . . . , k Mk or M and r are the inter-correlation
vector, at shift , between the baseband signal and the Descrambling and channelization sequences.

SIMULATION RESULTS

Simulations are carried out for the two scenarios using Walsh
codes for channelization and Large Kasami set of length 256
as scrambling code in the UL and DL. Propagation in microcell environment is simulated using the channel delay profile
proposed by the ETSI [9] with an average delay spread of 750
ns. The first path is taken always as LOS path. The path
loss is simulated using the ITU recommendations for Manhattan micro-cell deployment; the effect of shadowing with
a standard deviation of =4dB is also added to the path
loss. A bit rate of 16 kHz is chosen, Q = 3. A three point
interpolation N = 3 is carried out on the matched filter output before the application of MUSIC algorithm. A required
Eb /N0 = 3 dB, at the BS in the UL case is considered, with
a transmission power of 14 dBm at the MS. A pilot channel
power of 33 dBm and a power of 20 dBm [9] per user channel
are simulated in the DL case. Other parameters such as frequency are chosen according to the WCDMA FDD features.
The mobile movement is taken to be 3km/h, then according
to the channel simulations the coherence time was about 30
ms so the channel can be considered static over the observation period T =1/16000=62.5 s , 25 observation periods
were combined noncoherently to combat interference and fast
fading. Having one observation each 35 ms will give a LP in
0.875 s. Hence, the mobile phone can be considered static for
0.875s at a 3 km/h speed. The radio location algorithm used
is the one presented in [10] using 3 and 4 BS. A powerup

0.9

MUSIC DL
Matched DL
MUSIC UL
Matched UL

0.8

0.7

0.6

CDF

of 20dB is simulated to combat the nearfar problem in the

UL scenario; a constraint on the maximum UL transmitted
power of 24dBm is imposed. While, for the DL scenario silent
neighboring BS are assumed when TOA estimation from a
given BS is carried out.
During each observation a random deployment of the MS
interferers together with the static searched mobile in the
cell are simulated. Zero intercell interference is assumed and
background noise is added. Figure (3) depicts the circular
error cumulative probability density function (CDF) of the
estimated locations using 3 BS with 5 and 30 users at each
cell. The results of the two TOA estimation methods and the
two scenarios are ploted together. Figure (4) shows the same
results using 4 BS. Figure (5) shows a scatter plot of the position location using 4 BS in the DL scenario with matched
filter and MUSIC TOA estimation methods and having 30
users.

0.5

0.4

0.3

0.2

0.1

50

100

150

200

250

radius

Figure 4: Circular CDF of the location positioning error

using 4 BS and having 5 users
80

Matched DL
MUSIC DL

60

DISCUSSIONS

40

0.9

0.9

0.8

0.8

0.7

0.6
CDF

CDF

0.7

MUSIC DL
Matched DL
MUSIC UL
Matched UL

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

20

40

60

80

100

120

radius

(a) 5 users

140

160

180

MUSIC DL
Matched DL
Matched UL
MUSIC UL

50

100

150

200

250

radius

(b) 30 users

Figure 3: Circular CDF of the location positioning error

using 3 BS
References
[1] James J. Caffery,Wireless location in CDMA cellular radio systems Kluwer Academic Publishers Group,
2000.

20

y (m)

The results shows more accurate location positioning in the

DL scenario. This is due to the synchrounous sum of orthogonal users before convolution by a single channel (see
equation (4)). While, signals in the UL are transmitted from
different locations, which means asynchronous reception of
the users (see equation (2)). As seen from the equation, we
have more parameters to estimate from the same given number of observations as the DL case. Althoght the DL scenario
give better PL estimations, it requires modifications in the
mobile phone to make the required computations especially
the implementation of the MUSIC algorithm. Robust near
far resistant methods are not easy to implement in practice
due to the large spreading factors. However, the use of high
resolution algorithms on the delay profile improves the location positioning provided that the number of observations
are sufficiently high to give full rank channel coefficients correlation matrix. This is why the LP time in UL scenario may
increase drastically with the number of users. The LP is well
improved by increasing the number of involved BS.

20

40

60
150

100

50

50
x (m)

100

150

200

250

Figure 5: Scatter plot of the location positioning using

4 BS and 30 users in the DL scenario, (0,0) is the exact
position
[2] E. Hepsaydir,Mobile Positioning in CDMA Cellular
Networks 50th IEEE conference on Vehicular Technology VTC 1999. Vol. 2, pp 795-799.
[3] R. Prasad, W. Mohr and W. Konhuser,Third Generation Mobile Communication Systems, Artech House
Publishers 2000, pp 301-323.
[4] D. Porcino and A. Yule, Tutorial: Positioning Technology for mobile devices, the IEEE semiannual vehicular
technology conference VTC 2001 spring.
[5] Marilynn P. Wylie and Jack Holtzman,The Non-Line
of Sight Problem in Mobile Location Estimation, pp
827-831,1996.
[6] Takishi Manabe and Hitoshi Takai,Superresolution of
Multipath Delay Profiles Measured by PN Correlation
Method IEEE Transactions on Antennas and Propagation, Vol.40, No. 5, pp 500-509, May 1992.
[7] TS 25.213,Spreading and Modulation (FDD) 3GPP
Technical Specification, Version 3.2.0, March 2000.
[8] E. Nikula, A. Toskala, E. Dahlman, L. Girard and A.
Klein,FRAMES Multiple Access for UMTS and IMT2000 , IEEE Personal Communications, pp 16-24, April
1998.
[9] TR 101 112 Selection procedures for the choice of radio
transmission technologies of the UMTS, ETSI, v3.2.0,
April 1998.
[10] Y. T. Chan, K. C. Ho,A Simple and Efficient Estimator
for Hyperbolic Location, IEEE Transactions on Signal
Processing, VOL. 42, NO. 8, pp 1905-1915, AUGUST
1994.