Adjacent channel interference in WCDMA networks equipped with multiple antennas mobile stations

Abstract

The purpose of this work is to show to what extent multiple antenna terminals can combat inter-operator interference in WCDMA networks. To this end, a realistic modeling of adjacent channel interference-including filters imperfections and power amplifier nonlinearities is considered. Adjacent channel interference creates dead zones where the QoS target cannot be reached. We first characterize these zones in terms of radius and probability of finding a given amount of users in it. In this respect, an interesting result is that the probability of finding one user in a dead zone can reach 48% for a 50% user load. But the main contribution is to show that certain linear multiantenna reception schemes are able to cope with ACI. Compared to a conventional ID Rake, the dead zone radius is decreased from 80 m to 20 m for typical NLOS scenarios. The results are far more significant when the interfering adjacent band operator has a LOS channel, the bit error rates is decreased from 50% to 0.1% around the interfering BS. At last, the best choice of the linear receiver scheme is discussed regarding propagation conditions (power delay profile, power angle profile, antenna correlation).

Full text

Page 1

Adjacent Channel Interference in WCDMA Networks equipped with
Multiple Antennas Mobile Stations

Julien Dumont
France Telecom R&D
38-40, rue du Général Leclerc
92794 Issy-les-Moulineaux Cedex 9
France
Email: julien.dumont@francetelecom.fr


Abstract ─ The purpose of this paper is to show to what extent
multiple antenna terminals
interference in WCDMA networks. To this end, a realistic
modeling of adjacent channel interference - including filters
imperfections and power amplifier non-linearities – is considered.
Adjacent channel interference creates dead zones where the QoS
target cannot be reached. In this paper, we first characterize
these zones in terms of radius and probability of finding a given
amount of users in it. In this respect, an interesting result is that
the probability of finding one user in a dead zone can reach 48%
for a 50% user load. But the main contribution is to show that
certain linear multi-antenna reception schemes are able to cope
with ACI. Compared to a conventional 1D Rake, the dead zone
radius is decreased from 80 m to 20 m for typical NLOS
scenarios. The results are far more significant when the
interfering adjacent band operator has a LOS channel, the bit
error rates is decreased from 50% to 0.1% around the
interfering BS. At last, the best choice of the linear receiver
scheme is discussed regarding propagation conditions (power
delay profile, power angle profile, antenna correlation).

I. INTRODUCTION

In WCDMA cellular networks (e.g. UMTS – FDD) there are
many sources of interference that are likely to degrade the mobile
station (MS) performance. Interchip interference, interpath
interference [1], multiple access interference, intercell interference
and inter-system interference [2] are known to be the most influential
sources of interference. However the adjacent channel interference
(ACI) has not been deeply studied. In this paper we precisely focus
on ACI. In the literature there are few papers presenting WCDMA
inter-operator interference. In this respect [3] is a good reference.
Where does ACI come from? ACI is due to the presence of several
operators in the same geographical area. Adjacent band operators
generate out-of-band emissions because of transmit filter
imperfections and transmit power amplifier (PA) non-linearities.
Additionally, the receiver filter has a finite frequency selectivity,
which makes it sensitive to in-band emissions of adjacent band
operators. As 3GPP specifications have been designed with cost and
design constraints taken into account, there are situations where ACI
can have a considerable impact on the receiver performance. The
position of the mobile station within the main cell plays an important
role regarding the influence of ACI. ACI power can be stronger than
the useful signal when the mobile station of interest is very close to
an adjacent band operator base station (BS) and distant from the
useful BS. Note that ACI also exists in non-CDMA networks (e.g.
GSM) but is not influential because frequency reuse is high (e.g. 7)
while in CDMA networks it equals 1. Making a simple link budget
Samson Lasaulce
CNRS - LSS
5, rue Joliot Curie
Plateau du Moulon
91192 Gif-sur-Yvette
France
Email: lasaulce@lss.supelec.fr

Jean-Marie Chaufray
France Telecom R&D
38-40, rue du Général Leclerc
92794 Issy-les-Moulineaux Cedex 9
France
Email: jeanmarie.chaufray@francetelecom.fr
can combat inter-operator
analysis shows the existence of dead zones (DZ) in which the
adjacent band operator blinds the mobile station under consideration.
In this paper we assume a network deployment based on 3GPP
parameters. Because mobile stations are generally the limiting factor
of cellular network performance, we only consider the downlink case.
The proposed approach could be extended to the uplink. In this
context, the ultimate goal of this paper is to know to what extent
suited signal processing algorithms can compensate for transmit and
receive devices imperfections. More specifically, we focus on low
complexity multi-antenna reception algorithms. For the problem
under consideration in this paper, namely ACI suppression in
WCDMA networks, the most useful contributions on ACI
cancellation are [4] and [5]. Both of these papers still focus on filters
imperfections in TDMA systems and out-of-band emissions are taken
into account. Additionally, single-antenna algorithms proposed in [4]
rely on adjacent channels knowledge, which is not available in the
considered problem. In [5], a useful array processing technique is
proposed to combat ACI but would require major changes in order to
be used to our context (CDMA, multipath channels, limited MS
complexity).

Therefore the main purpose of this paper is fourfold:
•
to consider a realistic modeling of adjacent channel interference,
which essentially includes the PA non-linearity and filters
imperfections;
•
to characterize the importance of the dead zone problem (dead
zone area, probability of finding a given number of users in a
dead zone);
•
to assess the impact of ACI on a conventional MS receiver
(single-antenna, Rake receiver);
•
to choose good multiple antenna reception schemes (in terms of
the performance/complexity trade-off), evaluate the impact of
using them for decreasing the dead zone radius and discuss the
best scheme to be selected.

This paper is structured as follows. In section II we describe our
signal model. Then we discuss in section III the importance of the
dead zone problem. Section IV provides the considered multi-antenna
reception algorithms. Corresponding simulation results are provided
and discussed in section V. At last, in section VI we review the main
results and contributions of this paper.

II. PROBLEM STATEMENT

In the paper we always consider two base stations: one useful base
station (the subscriber’s operator) and another one belonging to an
adjacent band operator. As mentioned in the introduction, ACI stems
both from the adjacent band BS out-of-band emissions and finite

Page 2

receive filter frequency selectivity. In the following two subsections
we show how these imperfections can be taken into account in the
signal model. We will introduce notations associated with the useful
base station only and give the corresponding signal, since the
adjacent band signal is generated in the same way up to a frequency
offset (5 MHz in the UMTS - FDD mode).

A. Transmit signal model

In this paper the downlink case is considered. No assumption is
made on the number of antennas used by the useful base station. But
we assume that the mobile station of interest is equipped with “Q”
sensors. Four sources of reception performance degradation are taken
into account: the thermal noise, the interchip interference (ICI), the
multiple access interference (MAI) and, of course, the adjacent
channel interference generated by the adjacent band operator. To be
more realistic, intercell should be accounted for. We will see that in
the area of interest, which is in and around the dead zone, neglecting
intercell interference does not prevent us from drawing useful
interpretations from the presence of ACI in cellular networks.

Transmit filter output: denoting by Pk the power allocated to user “k”,
ck(n) its spreading code, bk(n) its QPSK symbols, s(n) the useful base
station scrambling code and g(t) the equivalent transmit filter allows
us to express the baseband signal at the output of the transmit filter
(fig. 1):
∑∑
∈=
Znk
d
k
(
where “K” is the number of active users in the cell covered by the
considered base station, “i” is the symbol index, “n” is the chip index
and Tc is the chip duration. According to the UMTS-FDD mode
specifications, the transmit filter is a root raised-cosine (RRC) filter
with roll-off 0.22 and truncated to 8 chip durations (the purpose of
this operation is to trade spectral performance against computational
burden).

Power amplifier input: power amplification is the last stage of the
transmitter, it has to amplify the in-phase plus in-quadrature signal
(say
)(x
), which writes as follows:
)(
)()(
txtxtx
QI
ω=
πω
2/
0
f
is the useful carrier frequency.

Power amplifier output: a good approximation of the power amplifier
non-linearity is the polynomial model given in [6]. This
approximation is often used by mobile and base stations
manufacturers. Under this approximation, the PA output expresses as:
∑
=
i
0

()
−=
C
K
n
k
4
kk
nTtgn
3
sncibP
4
tx
1
)
)(
4
)()
2
(
44
)(
41
(1)
~t

[][]
)sin()(Im)cos()(Re
~~~
00
ttxttx
ω+
+=
(2)
where
0=

=
D
i
i
txatxty
2
)(
~
)(
~
)(
(3)
where “D” is the approximation degree and a0 = 1.

B. Receive signal model

Reception filter: the baseband filter is a discrete-time finite impulse
response filter and is often chosen to be a RRC filter (roll-off 0.22).
In WCDMA networks receiving mobile stations are sensitive to in-
band emissions of adjacent band operators because the latter is
truncated (as in the transmitter) to 8 chip durations. Figure 2 shows
the importance of truncating the RRC filter. Assuming a white input,
it depicts the power spectral density (PSD) of the RRC output when
the filter is truncated to 8 chip durations (top curve) and 32 chip
durations (bottom curve). The out-of-band PSD levels differs by
about 30 dB.

Figure 1: Transmitter model
−8−6−4−2 02468
6
x 10
−100
−80
−60
−40
−20
0
20
Frequency
Power (dB)
Truncated RRC (32 chips)
Truncated RRC (8 chips)

Figure 2: Influence of RRC truncation

Received signal model: although the received signal is generated
through a non-linear device (PA), the receiver considers, for the
baseband reception algorithms, a linear model of the transmission
chain in the sense that the useful signal is separated from the ACI in
an additive way. The discrete-time model that is used to the
interference cancellation purpose is the following:
KL
()(
10

==
44441

(
1
)
{
AWGN
ACI
nvn
3
in
44
i
nd
4
hny
outin
4
i
k
ku
)()(
4
)(
)(
4
)
)()(
1
MAI
+
signal
+
42
useful
++





−
4
=
∑ ∑
−=
4
432
ll
l
l
α
α
(4)
where y(n)=[ y1(n) … yQ(n) ]T is the signal received by the Q-sensor
MS antenna, “L” is the number of paths of the overall channel
impulse response h(.), dk(.) is the useful chip sequence defined in (1),
αu corresponds to propagation losses over the useful link, αi
corresponds to propagation losses over the ACI link. At last,
notations i(in)(.) and i(out)(.) stands for ACI received in and out of the
theoretical MS frequency band [ f0 - (1+ρ) / Tc , f0 + (1+ρ) / Tc ], ρ

Page 3

being the RRC roll-off factor. Figure 3 clearly shows what we mean
by in-band and out-of-band interference.

Figure 3: two origins of ACI

III. DISCUSSION ON THE IMPORTANCE OF THE DEAD ZONE
EFFECT

In the previous section we have seen how ACI can be modelled.
The purpose of this section is to discuss the importance of ACI in
cellular networks. We want to know how often the dead zone (DZ)
phenomenon occurs. To this end we first review 3GPP’s definition
regarding ACI, then we give the propagation losses model we used in
order to evaluate a typical dead zone radius. Making a simple
analysis on the probability of the DZ phenomenon to occur concludes
this section.

A. Review of 3GGP’s definitions on ACI

Out-of-band-emissions maximum power: in the UMTS – FDD mode
[7], the maximum level of base stations out-of-band emissions has
been specified through the adjacent channel leakage ratio (ACLR). It
is defined by ACLRBS=PTX(in band)/PTX(out-of-band) and has to be
greater than 45 dB.

Maximum of adjacent channel power: in the same way, the minimum
frequency selectivity of receive mobile stations has been specified
through the adjacent channel selectivity (ACS). It is defined by
ACSMS=PRX(in band)/PRX(out-of-band) and has to be greater than 33
dB.

In order to measure the combined effects of transmission and
reception in the adjacent band, the downlink adjacent carrier-to-
interference ratio has been defined as ACIR-1=ACLRBS
The dominant part of ACI is due to MS frequency selectivity since
ACIR ≈ 32.7 dB.

B. Dead zone radius evaluation for a given SIR

We first need to define more precisely what a dead zone is. In the
downlink, it is an area around the adjacent band operator base station
in which the QoS (quality of service) target cannot be reached. A
dead zone can be roughly thought of as a circle centred on the
adjacent band BS. This is why we allow ourselves to use the term
“dead zone radius”. Now, if we want to evaluate a dead zone radius
we need to know the relation between the desired QoS and needed
receive signal-to-total interference ratio (SIR), which is receiver-
dependent. In this subsection we assume that the required SIR is
given. In practice the SIR target is obtained from desired QoS thanks
to look-up tables. The chosen path loss model for the useful and ACI
links is defined by:
-1+ACSMS
-1.

( )
( )
i
r
[
[
α
]
( )
( )
i
r
]



+=
+=
iii
uu
log
uuu
KL
rKLr
1010
1010
log10
loglog10
α
(5)
where ru and ri are the distance between the mobile station and the
two base stations (useful and adjacent band BSs). If we assume that
the ratio of the useful BS transmitted power to the transmitted ACI
power is close to the standardized ACIR (target value), it is possible
to easily relate the receive SIR condition to the condition on ri. It can
be checked that:
P
≥
,

(
2
DZ
R
)
434444444441
dBdBiuiiu
ACIRSIRLLKKK
ui
dB
iRX
u
,
RX
rr
SIR
P
−+−
−
×≥⇒
1
10
/
(6)
Assuming a 3GPP typical NLOS urban micro-cell environment [8]
for both the useful and ACI channels we find that the dead zone
radius equals about 70 meters for a 0 dB SIR (voice service) and a
MS at 500 m from the useful BS. In fact, this is not the worse case
scenario since it can happen that the ACI propagation channel is less
severe than the useful one and the SIR target can be higher than that
(data service). In any case, compared to a typical micro-cell radius
(about 600 m), the DZ radius is significant even for low rates
transmissions.

C. Probability for active users to be in a dead zone

Given a dead zone radius, we want to know what are the chances
of one user (or more) to be in a dead zone. To this end we assume
that the active users can be anywhere in the useful cell with an equal
probability. We always denote by K the number of users in the cell
under consideration. Let C be the number of users in a dead zone.
We can show that the probability of the event k≥K0 to occur is
roughly given by:
()


=
/ )()(
cell DZ
RKRK
ε
In particular, one can check that when the number of active users
goes to the infinity, the probability of finding one user in the DZ
tends to 1. Note that the dead zone radius depends on the user load
because the QoS is MAI-dependent. For a 1D Rake, a 32 spreading
factor and a 600 m cell radius we have obtained the following results:

K 4 8
RDZ(m) 60 70
P(k≥1) 4% 10%

[]
()



−






−=≥
−
−
=∑
i
2
1
0
0
0
)(1)(1
0
iK
i
K
KK
K
K
KkP
εε
(7)
12
80
19%
16
120
48%
Table 1: Probability of finding 1 user in a DZ
Now we are convinced that what could be at a first glance considered
as second-order effects can have a significant impact on the network
performance, we propose studying different multi-antenna reception
schemes in order to know to what extent signal processing can be of
help to compensate for Tx and Rx imperfections.

IV. MULTIPLE ANTENNAS RECEPTION SCHEMES

We want to know to what extent using several antennas at the
mobile station can help to combat ACI. In order to keep the receiver
complexity reasonable we only study linear reception schemes. In
this case the receiver consists of a linear space-time equalizer,
followed by descrambling and despreading operations. One can show

Page 4

that for these reception strategies, the symbol estimate of the user of
interest (index 1) has a generic expression:
∑ ∑
==
0
0
nl
where N is the spreading factor and w(.) is the Q-dimensional
weighting vector depending on the reception scheme.

The best linear scheme is the chip-rate space-time Wiener filtering
(STWF). This filter is designed in order to minimize the mean square
error between its output and a reference sequence (the pilot sequence
d0(n) for instance); it is given by

[
)()(nYnYEw
=

where w and Y(.) are obtained by stacking the L successive chip-rate
samples w(ℓ) and y(n+l), ℓ =0, …, L-1, in a vector. In practice, this
filter is implemented by replacing the expectation operator in (9) with
a discrete sum over n. Hence, implementing the STWF requires the
inversion of a QL × QL matrix. The complexity of this operation is
too high regarding the fact that RRC filters have been truncated to
decrease complexity. Additionally, the STWF performance can be
severely degraded with respect to the strategies proposed below
because of ill-conditioning of this matrix. In this context (limited
complexity, continuous PAP, presence of a low-rank interference) the
following reception schemes have been selected: conventional 2D
RAKE, spatial matched filter (SMF), generalized 2D Rake and
whitened 2D Rake.

A. Conventional 2D Rake

In this case there is one Rake per antenna and antenna outputs are
simply added. It follows that:

)(
l
hw
It is known that this strategy is optimum when the noise-plus-
interference term is Gaussian, temporally and spatially white.

B. Generalized 2D Rake (G2D-Rake)

The difference with the strategy above is that the outputs of each
Rake are combined in order to minimize the chip estimate MMSE:

=
l

−−




+=
1
*
1
1
1
)()()()()(
ˆ
NL
H
nsnclnylwib
(8)

]
()
[] )n()(
*
1
1
dnYE
H
−
(9)
)(
l
=
(10)

()
[] )n()()()(
)()( min
w
arg)(
*
1
1
2
1
dnzEnznzE
ndnzwEw
H
H




=



−
−
(11)
where the vector z(n)=[z1(n) ... zQ(n) ]T represents the outputs of the
Q Rake receivers.

C. Spatial matched filter (SMF)

A spatial filter matched to the pilot sequence is first applied to the
received signal, and it is followed by a 1D RAKE receiver. The
coefficients of the RAKE receiver are the spatially filtered versions
of channel coefficients. For a given delay denoted by ℓ0, the
weighting vector writes:
)()(
0
ll
aaw
with

−=

)()(
0
ll
h
H
=
(12)

)()()(min
a
arg)(
0
1
2
010
lll
hndnyaEa
y
H
−
=



−
R
(13)
and
[] )n()(
1
ynyE
H
y=
−
R
. The delay
0 l is chosen in order to
maximize:

)()(
0
1
0
ll
hRh
y
H
−
(14)

D. Whitened 2D Rake (W2D-Rake)

The idea here is to approximate space-time Wiener filtering by
assuming the space-time covariance matrix to be block diagonal,
which is equivalent to assume temporally white the noise-plus-
interference term. It is easy to show that under this assumption the
STWF boils down to:

)(
l
w

V. SIMULATION RESULTS

Simulation setup: in all the simulations provided here, the number of
active users is 8, the spreading factor equals 32 and the useful
channel is estimated thanks to the common pilot channel (2560 chips,
10% of the BS transmit power). The PA non-linearity is modelled by
a 3-degree polynomial (a1 = 4/27) and works at its 1 dB compression
point. By default, the useful and ACI channels correspond to a NLOS
Vehicular A microcell environment [8], which corresponds to
LU=LI=34.53, KU=KI=38. The corresponding power angle profiles
are continuous and flat if nothing is specified. At last, the distance
between useful and adjacent band operator base stations is 500 m by
default and the mobile station is moving along a line between the
useful and adjacent band base stations.
)(
1
l
h
y
−
= R
(15)
250300 350 400450 500
10
−3
10
−2
10
−1
10
0
Distance to the useful station
BER
ACLR: 63 dB, ACS: 54 dB
ACLR: 45 dB, ACS: 54 dB
ACLR: 63 dB,ACS: 34 dB
ACLR: 45 dB, ACS: 34 dB

Figure 4 : Influence of ACLR and ACS on performance

Dead zone existence: figure 4 represents the performance of a 1D
Rake (raw BER) as a function of the distance in meter (ru) between
the MS and serving BS. The top curve corresponds to what is
obtained when using 3GPP specifications (ACLRBS=45 dB, ACSMS=
34 dB). If the BER target is 1%, the corresponding dead zone radius
is about 80 m. The other curves show what would be obtained if the
Tx or/and Rx were improved in terms of ACLR or/and ACS. We
also see that increasing the ACSMS significantly reduces the dead
zone radius. We can notice that the receiver performance is the same
between 250 m and 400 m: this is due to the 1D Rake error floor in
presence of MAI, which makes it less sensitive to the link budget (or
SNR).

Benefits of using multiple antennas at the MS: figure 5 depicts (from
left to right) the performance of the 1D Rake (single antenna), 2D

Page 5

Rake, generalized 2D Rake, SMF, whitened 2D Rake. This time the
MS is equipped with a 4-sensor linear array antenna (this assumption
is made for simplicity but is not restrictive). Using the 2D Rake
allows the DZ radius (BER target = 1%) to be decreased from 80 m
to 40 m. The whitened 2D Rake achieves the best performance by
reducing the DZ radius to 22 m, which is remarkable. However one
may ask about our assumption on the propagation environment
(NLOS microcell) when the MS is very close to the interfering BS:
what does the receiver performance become if a line-of-sight appears
between the MS and interfering BS? The answer is given in the
following subsection.

350400450500
10
−3
10
−2
10
−1
10
0
Distance to the useful station
Figure 5 : Comparison of the receivers
BER
Conventional 1D−RAKE
Conventional 2D−RAKE
Generalized 2D−RAKE
Spatial matched filter
Whitened 2D−RAKE


Influence of the propagation environment: in order to compare the
multi-antenna reception schemes between themselves, we provide in
the table below raw bit error rates for different power delay profiles
[8]. The mobile position is fixed and is close to the interfering station
(20 m). In this case, the existence of a line-of-sight is possible. We
first notice that the 1D-, 2D- and G2D-Rake performance are totally
degraded by the presence of a line-of-sight, while SMF and W2D-
Rake still get excellent performance. This can be explained by the
fact that in the LOS case, there is only one dominant path coming
from the interfering station. Indeed the presence of a LOS changes
the interfering channel power angle profile, the latter becomes sparse
and mobile sensors become correlated. It turns out that only SMF and
W2D Rake are able to exploit this correlation to spatially reject the
dominant interfering path whatever its power is. As for the
comparison between SMF and W2D Rake we can see in the table that
SMF is the best scheme when the interfering channel has few
significant paths (like the Pedestrian A environment). On the other
hand when the interfering channel becomes richer in terms of path
(like Vehicular A), the W2D Rake is more suited.

Useful
Station Station Rake
Vehicular
A A
Vehicular
A Sight
Pedestrian
A A
Pedestrian
A Sight
Interfering 1D 2D
Rake
21.8 %
G 2D
Rake
9.5 %
SMF W 2D
Rake
1.7 % Vehicular 32.6 % 2.2 %
Line-of-49.7 % 49.4 % 28.9 % <0.1 % 0.2 %
Pedestrian 35.3 % 29.4 % 6.6 % <0.1 % <0.1 %
Line-of-50 % 49.8 % 29.6 % <0.1 % <0.1 %
Table 2: Influence of the PDP on the receiver performance

VI. CONCLUSIONS

In this paper we have focused on a special kind of interference,
namely ACI, not famous in the signal processing literature. Assuming
a network deployment based on 3GPP parameters, we have provided
simple but useful results to characterize the importance of the dead
zone effect. We have showed how a dead zone radius can be found
and the probability of finding a given number of users in it. In this
respect having a 120 m dead zone and a 48% probability of finding
one user in it is possible. Based on a realistic transmit and receive
signals modeling we have showed that the dead zone effect plays an
important role. We have seen that in the vicinity of the adjacent band
base station, ACI is clearly the dominant interference and multiple
access interference is no longer the limiting factor of the MS
performance. In order to reduce this dramatic effect we have
proposed to study adapted multi-antenna reception schemes taking
into account MS complexity constraints. In this respect the spatial
matched filter and whitened 2D Rake achieve the best performance.
For typical scenarios using a 4-antenna SMF or W2D Rake allow the
dead zone radius to be reduced from 80m to 20 m. We have also
shown the influence of the propagation environment by evaluating
the receiver performance for different power delay profiles and
power angle profiles. An important result is that SMF and W2D Rake
achieve very good performance (0.1% bit error rates at 20 meters
from the ACI base station) whatever the power angle (continuous or
sparse) and power delay profiles are, which means that the proposed
strategies work in any place of the network area (and then both for
correlated and uncorrelated antennas). Several extensions of this
study could be done. The uplink case could be considered. But what
seems the most important work to be done is to make the general
space-time Wiener filtering robust to ill-conditioning of the space-
time matrix to be inverted, which is important in the presence of low-
rank interferers (like adjacent band interferers). The recent work of
Honig and Goldstein [9] seems to be a good way of solving this
problem.

REFERENCES

[1]
An optimum combining and concatenated-Rake for dual-antenna
mobile terminals in UMTS, S. Tantikovit and M. Wang, IEEE Comm.
Letters, Vol 6, June 2002, Pages 231-233
[2]
The effect of inter-system interference in UMTS at 1920 MHz, H.
Hass, S. MCLaughlin and G. Povey, IEE Conf. Publ. No 471, 2000,
Pages 130-107.
[3]
WCDMA inter-operator interference and dead zones, G. Povey, L.
Gatzoulis, L. Stewart and I. Band, 5th European Personal Mobile
Communications Conference, April 2003, Glasgow, Scotland.
[4]
Adjacent channel interference suppression in FDMA/TDMA mobile
radio systems using joint demodulation, H. Arslan, S. Gupta, G.
Bottomley and S. Cheunakeshu, ICC 1998, Vol 2, Pages 723-727.
[5]
Adjacent channel interference rejection for land mobile radio
systems, J. Ching Guey, A. Khayrallah and G. Bottomley, VTC 1998,
Vol 3, Pages 1715-1719.
[6]
Digital predistortion of wideband signals based on power amplifier
model with memory, J. Kim and K. Konstantinou, Electronic Letters,
November 2001, Vol 37, No 23.
[7]
Evaluation of up- and downlink adjacent channel performance”,
Turin, 15-19 February 1999, TSGR4#2(99)048, ,3GPP Technical
Specification Group, Radio Access Network, Working Group 4
[8]
3GPP TS 25.996 v6.1.0 Technical Specification Group Radio Access
Network; Spatial Channel Model for MIMO Simulations
[9]
Adaptive reduced-rank interference suppression based on the
multistage Wiener filter, M. Honig and J. Goldstein, IEEE Trans. on
Communications, Vol 50, No 6, June 2002, Pages 986-994.