A Novel Uplink Receiver Ior GSM/EDGE Systems

with Orthogonal Sub Channel Feature

Daniele Molteni and Monica Nicoli
Dip. di Elettronica e InIormazione, Politecnico di Milano, Piazza L. da Vinci 32, I-20133 Milano, Italy
e-mail: molteni,nicoli}elet.polimi.it
$EVWUDFW-The increasing demand for trafc capacity of the
GSM/EDGE Radio Access Networks (GERAN) induced the
Third Generation Partnership Project (3GPP) to examine the
Orthogonal Sub Channel (OSC) transmission scheme. This new
feature aims at doubling the cell capacity by multiplexing two
co-cell users on the same radio resource. This work presents
a new two-stage receiver specically designed for the uplink
GSM-OSC system. The proposed strategy relies on two optimum
interference mitigation lters jointly derived for the multiplexed
users. The ltered signals are then processed by a multi user
detector that recovers the transmitted sequences. The proposed
algorithm shows a signicant performance gain with respect to
existing interference cancellation receivers.
I. INTRODUCTION
Despite oI many previsions, the Global System Ior Mobile
Communications (GSM) is Iar Irom being a dead technology.
Voice traIfc, in particular, is expected to grow nearly threeIold
all over the world by 2012 |1|. Mobile operators are pushed
to fnd new solutions Ior improving the eIfciency oI current
GSM networks with low impact on existing hardware equip-
ments, to reduce as much as possible the upgrading costs.
A well known approach Ior improving the traIfc capacity
is reducing the cellular reuse Iactor; the drawback, however,
is the increase oI co-channel interIerence and the consequent
perIormance degradation. To Iace this problem, the Third Gen-
eration Partnership Project (3GPP) has recently investigated
the use oI Single/Double Antenna InterIerence Cancellation
(SAIC/DAIC) algorithms |2|-|7|. PerIormance specifcations
Ior mobile stations (MS) equipped with interIerence cancella-
tion have been provided by the Downlink Advanced Receiver
PerIormance (DARP) phase I-II requirements. New DARP
handsets are already on the market.
Relying on the improved co-channel interIerence robustness
oI the new GSM mobiles, the 3GPP is now examining the
Ieasibility oI allocating diIIerent users to a same radio resource
in the cell. Among the proposed solutions, the Orthogonal Sub
Channel (OSC) multiplexing scheme |8| aims at doubling the
voice capacity with negligible impact on the existing handsets
by enabling the transmission oI two GMSK streams on the
same time slot. The increased capacity oI GSM-OSC networks
can be exploited in the Iuture either to serve more users in the
given bandwidth (e.g. in highly loaded markets) or to group the
voice traIfc into a portion oI the available bandwidth leaving
This work has been supported by Nokia Siemens Networks and the MIUR-
FIRB Integrated System Ior Emergency (InSyEme) Project under the grant
RBIP063BPH.
Out-of-cell
nterferer
OSC
0
OSC
1
h
1
h
I
h
0
Intra-Cell
) (
1
N D
) (
0
N D
) (N E
5
1
1

Prefilter
MUD
OSC
0
OSC
1

) ( `
1
N D
) ( `
0
N D
Fig. 1. UL GSM layout with OSC Ieature. Two MSs are sharing the same
Irequency: the proposed Joint OSC receiver applies a preflter Ior each OSC
user to mitigate the out-oI-cell interIerer and perIorms a MUD to recover the
desired sequences.
Iree resources Ior services oI data transmission. In the OSC
downlink (DL), in order to minimize the mutual interIerence,
the two GMSK signals are generated with a phase oIIset (using
two QPSK subsets oI the 8-PSK EDGE constellation) and
they are recovered at the two MSs through DARP receivers.
On the other hand, in the uplink (UL), the two signals are
transmitted independently by two normal GMSK MSs and
have to be detected at the BS by multi-user detection (MUD).
New processing techniques are required at the BS to handle
the interIerence generated both Irom out-oI-cell and in-cell
users.
The FRQWULEXWLRQ oI this work is the design oI a new two-
stage UL receiver: the Joint OSC Receiver (JOR) combines
a Iront-end prefltering stage Ior the mitigation oI the out-
oI-cell interIerence, with a MUD that optimally handles the
mutual interIerence between the two in-cell users (Fig. 1).
DiIIerently Irom conventional SAIC/DAIC, here preflters are
optimized MRLQWO\ Ior the two OSC users, so as to minimize
the interIerence Irom neighboring cells while preserving the
two multiplexed signals. The second stage is a MUD that
jointly estimates the two transmitted sequences: joint maxi-
mum likelihood sequence estimation (JMLSE) is considered
as optimal non-linear solution, while linear detection (L-MUD)
is proposed as a good trade-oII between perIormance and
computational complexity. The proposed scheme is shown to
outperIorm all the existing receivers, in particular the interIer-
ence fltering schemes based on conventional SAIC/DAIC and
977 978-1-4244-5827-1/09/$26.00 2009 IEEE Asilomar 2009
the Serial/Parallel InterIerence Cancellation (SIC/PIC) MUD
algorithms suggested Ior the OSC Ieature |8|.
This work is organized as Iollows: Sec. II introduces the
adopted GSM with OSC system model, Sec. III recalls the
state oI the art receivers applied to the OSC context whereas
Sec. IV describes the proposed receiver. Numerical simulations
are discussed in Sec. V where a realistic GSM system has been
designed. Finally the conclusions are drawn in Sec. VI.
II. SYSTEM MODEL
The uplink oI a synchronous multicell GERAN environment
is modelled adopting the OSC Ieature as prescribed in |8|: two
in-cell mobile stations (MSs) share the same radio resource
transmitting two GMSK modulated signals |9| while an out-oI-
cell interIerer is impairing the transmissions (see the example
in Fig. 1).
At each MS a sequence oI inIormation bits is convolu-
tionally encoded, mapped over diIIerent bursts according to
the adopted channel coding scheme and fnally diIIerentially
encoded. Each burst is composed by N @ 47; symbols made
up oI: 447 data (divided into two semibursts oI P @ 8: each),
; guards and 59 training symbols placed in the midamble. The
training sequences (TSC) oI the in-cell OSC users are known
at the BS, while no inIormation is available about the out-oI-
cell MS. We denote the nth transmitted symbol as d
l
+n, Ior
the lth OSC user and e+n, Ior the out-oI-cell interIerer, with
id
l
+n,> e+n,j 5 i4j, n @ 4> = = = > N, and l @ 3> 4=
We suppose that all the MSs employ a single antenna (see
Fig. 1) whereas the BS is equipped with an array oI Q
U
antennas that are suIfciently spaced apart so that the received
signals can be considered uncorrelated. The Q
U
4 complex-
valued signals at the output oI the receiver`s flter (beIore
sampling) can be written as:
|+w,@
4

l@3
N

n@4

k
l
+wnW,m
n
d
l
+n,.
N

n@4

k
L
+wnW,m
n
e+n,. q
eq
+w,>
(1)
where 4@W is the baud rate, the Q
U
4 vector

k
l
+w, gathers
the Q
U
channel impulse responses oI the lth OSC user, while

k
L
+w, reIers to the out-oI-cell interIerer. Channel responses
comprise the pulse shaping flter (i.e., the linearized JPVN
pulse waveIorm), the multipath Iading channel and the re-
ceiver`s flter. The term m
n
represents the periodic symbol
rotation prescribed by the GMSK linearized model |9|. Finally,
q
eq
+w, models the additive white Gaussian noise (AWGN)
fltered by the receiver`s flter, with zero mean and covariance
H^ q
eq
+w, q
K
eq
+w . ,` @
5
eq
L
Q
U
J+,, where
5
eq
is the noise
variance and J+, is the autocorrelation oI the flter impulse
response.
A. Real-valued discrete-time signal model
According to |4|-|5|, in order to boost the perIormance
oI linear interIerence cancellation, the received signals (1)
are sampled with time interval W@5 and re-organized into an
equivalent real-valued discrete-time domain, through the real-
valued processing procedure summarized in the sequel (see
|3| Ior details).
The two signal samples associated with the nth symbol time
are frst stacked into the 5Q
U
4 hybrid space-time (S-T)
vector |
v
+n, @
_
|
W
+nW,> |
W
+nW .W@5,

W
. This signal is
then derotated in order to transIorm each GMSK stream into a
BPSK-like modulated signal. The derotated signal is obtained
as |
g
+n, @ M+n, |
v
+n,, using the 5Q
U
5Q
U
diagonal matrix
M+n, @ gldj
_
m
n
> m
+n.4@5,
_
L
QU
where denotes the
Kronecker product. Finally, real and imaginary parts are sepa-
rated leading to the 7Q
U
4 vector representation oI the signal
sampled at the baud rate: |+n, @ ^Uhi |
W
g
+n,j> Lpi |
W
g
+n,j`
W
.
The signal vector |+n, can be written as a Iunction oI
the 7Q
U
4 discrete-time real-valued channel responses
ik
3
+n,> k
4
+n,> k
L
+n,j that are obtained Irom the continuous-
time complex-valued responses i

k
3
+w,>

k
4
+w,>

k
L
+w,j by ap-
plying the same procedure oI sampling, derotation, and re-
al/imaginary separation as described above. Assuming a chan-
nel response length oI O symbol times, we can arrange all the
resulting channel samples into the 7Q
U
+O . 4, matrices
iK
3
> K
4
> K
L
j, with K
l
@^k
l
+3, k
l
+O,`, Ior l @ 3> 4, and
K
L
defned accordingly.
The overall 7Q
U
+N .O, received samples are:
\ @ ^|+4, |+N .O,` @ K
3
D
3
.K
4
D
4
.K
L
E.Q
eq
. .
Q
>
(2)
where D
l
and Eare the +O.4,+N.O, convolution matrices
Ior, respectively, the lth OSC user and the out-oI-cell interIerer
symbols. The convolution matrices have elements ^D
l
`
p>q
@
d
l
+p q, and ^E`
p>q
@ e+p q,, Ior p @ 4> = = = > O . 4,
and q @ 4> = = = > N .O. The matrix Q collects the samples oI
all the impairments due to the out-oI-cell interIerence and the
background noise Q
eq
@ ^q
eq
+4, q
eq
+P .O,`.
The flters Ior interIerence mitigation and the channel im-
pulse responses Ior data detection are estimated within each
burst using the training signals received during the midamble.
Let Q
w
. O be the training sequence length, the 7Q
U
Q
w
submatrix oI (2) containing the midamble signals can be
represented as:
\
w
@ ^K
3
> K
4
`
_
D
w>3
D
w>4
_
.Q
w
@ KD
w
.Q
w
. (3)
where D
w>l
is the +O . 4, Q
w
convolution matrix oI the
training symbols oI the lth OSC user, and Q
w
denotes the
noise-plus-inter-cell-interIerence samples. Notice that the frst
O symbols oI the received midamble are discarded due to the
ISI generated by the tail oI the frst semi-burst oI data.
III. OVERVIEW OF EXISTING RECEIVERS
The OSC Ieature oI GSM has been designed so as to ensure
compatibility with standard GSM receivers that are equipped
with interIerence cancellation capabilities (here reIerred to
as single-user interIerence cancellation, SU-IC). In the UL,
the receiver at the BS may employ well-known linear flters
Ior SU-IC (e.g., conventional SAIC/DAIC algorithms) and/or
MUD schemes Ior a more eIIective in-cell interIerence can-
cellation (e.g., the SIC solution suggested within 3GPP Ior
978
MLSE
0 , w
K
Max
SNR
0
) (k \
) ( `
0
k a
68,&
) (
0 ,
k v
w
MLSE
1 , w
K
Max
SNR
1
) ( `
1
k a ) (
1 ,
k v
w
R
N 4
1 L
R
N 4
Z

68,&
Fig. 2. SU-IC receiver block diagram. A preflter is applyed to the received
signal maximizing the SINR oI a single OSC channel.
the OSC Ieature or other linear/no-linear solutions widely
studied Ior CDMA system). In this section, some oI these
existing algorithms are recalled and combined into possible
receiving architectures Ior the specifc GSM-OSC Iramework.
These solutions will be used as perIormance reIerence in the
numerical analysis in Sec. V.
JMLSE Equali:er. Joint detection oI the OSC sequences
id
3
+n,> d
4
+n,j is obtained by the optimal multi-antenna
JMLSE equalizer, without any prefltering stage. JMLSE can
be implemented by a Viterbi Algorithm (VA) with 5
5O
states
using as input the complex-valued signals (1) sampled at baud-
rate (see |11| Ior details). Signals incoming Irom the Q
U
antennas are optimally combined in the branch metric oI the
trellis.
SU-IC. The BS detects one OSC user at a time (Fig. 2)
by means oI conventional SU-IC algorithms developed Ior
the DARP I-II GSM evolution, such as SAIC Ior Q
U
@ 4,
DAIC Ior Q
U
@ 5 or their direct extension to any number oI
antennas (see, e.g., |6|, |7|). We remark that no standardized
versions oI these receivers are available, only diIIerent vendor
solutions. For the lth OSC sequence, a linear preflter is
designed by maximizing the Signal to InterIerence plus Noise
Ratio (SINR):
VLQU
l
@
nk
z>l
D
w>l
n
5

z
W
l
\
w
k
z>l
D
w>l

5
> (4)
obtaining the flter z
l
and the relative fltered channel k
z>l
.
Notice that the interIerence term at the denominator includes
both the out-oI-cell and the other OSC user. The two fltered
signals |
z>l
+n, @ z
W
l
\+n,, l @ 3> 4, are then processed by
separate MLSE equalizers.
SIC. We consider a successive cancellation receiver (see
Fig. 3-a) as suggested within 3GPP Ior the OSC Ieature |8|:
the strongest OSC signal is detected frst, then the detect-
ed sequence is re-modulated, convolved with the estimated
channel response and subtracted Irom the received signals Ior
detection oI the other OSC sequence. A variety oI interIerence
mitigation receivers can be used Ior the SIC implementation,
SU-C
) (
`
1
k +

SU-C
) ( `
0
k a
) ( `
1
k a ) (k \
SC
) ( `
0
k a
6,&
) (k \
SC
) ( `
1
k a
3,&
D E
Fig. 3. Block diagram oI the SIC (a) and PIC (b) algorithm.
here we have used the SU-IC algorithm applied twice.
PIC. A parallel cancellation approach is designed as the
combination oI two SIC receivers (Fig. 3-b).
IV. JOINT OSC RECEIVER
We propose a new receiving strategy specifcally designed
Ior the OSC-aware system, as the combination oI a Iront-
end fltering stage with a MUD, as sketched in Fig. 4. Two
linear flters iz
3
> z
4
j are applied to the received signal (2)
generating the Iollowing fltered streams Ior the two users:

|
W
z>3
@ z
W
3
\ @ k
W
z>33
D
3
.k
W
z>34
D
4
.q
W
3
|
W
z>4
@ z
W
4
\ @ k
W
z>43
D
3
.k
W
z>44
D
4
.q
W
4
= (5)
The coeIfcients oI the flters iz
3
> z
4
j and the relative fltered
channel responses ik
z>33
> k
z>44
j are jointly optimized during
the training phase so as to minimize the out-oI-cell interIerence
at the flter output, as described in Sec. IV-A. These signals
are then used in the second stage Ior joint detection oI the
symbol sequences id
3
+n,> d
4
+n,j, taking into account the
mutual interIerence due to the cross-channels k
W
z>34
@ z
W
3
K
4
and k
W
z>43
@ z
W
4
K
3
, and the residual noise q
W
l
@ z
W
l
Q (see
Sec. IV-B).
A. Optimal design of JOR prehltering
The 7Q
U
4 flters iz
3
> z
4
j and the corresponding fltered
channels ik
z>33
> k
z>44
j are the parameters to be optimized.
They are calculated jointly Ior the two OSC mobiles using
the midamble signals \
w
in (3) and the knowledge oI the
training sequences (TSC) iD
w>3
> D
w>4
j. The optimization is
carried out by maximizing the signal-to-noise-plus-out-oI-cell-
interIerence ratios SINR
3
,SINR
4
} at the flters` output. The
SINR Ior user l @ 3 (and similarly Ior user l @ 4) can be
written as:
VLQU
3
@

k
W
z>33
D
w>3

z
W
3
\
w
z
W
3
a
K
4
D
w>4
k
W
z>33
D
w>3

5
> (6)
where
a
K
4
denotes the least square estimate oI the user-1
channel, obtained as ^
a
K
3
>
a
K
4
` @ \
w
D
W
w
+D
w
D
W
w
,
4
. It is
important to notice that, diIIerently Irom conventional SU-IC,
the power on the denominator includes onlv the out-of-cell
interIerence, not the in-cell one, as the aim oI this frst stage
is to exploit all the available space dimensions to cancel as
much as possible the interIerence Irom other cells.
979
MUD
Z

10 , 11 ,
,
w w
K K
Max
SNR
) (k \
) (
0 ,
k v
w
) (
1 ,
k v
w
R
N 4
ChEst
R
N 4 2
1 0
`
,
`
+ +
) ( `
1
k a
) ( `
0
k a
01 , 00 ,
,
w w
K K
) 1 ( 2 L
Fig. 4. JOR block diagram: the flters are designed to provide the maximum
SINR Ior each OSC user.
The solutions can be carried out as described in |10| Ior the
rank-one receiver derivation accounting Ior ]
3
@ \
w

a
K
4
D
w>4
as the received signal, so that:
z
3
U
4
}}>3
U
}d>3
k
z>33
(7)
k
z>33
U
4
dd>3
hlj
pd{

U
W@5
dd>3
U
d}>3
U
4
}}>3
U
}d>3
U
4@5
dd>3

(8)
where the covariance matrices are defned as U
}}>3
@
]
3
]
W
3
@Q
w
, U
d}>3
@ D
w>3
]
W
3
@Q
w
, and similarly Ior U
dd>3
and
U
}d>3
. For the other user (i.e. l @ 4) the solutions iz
4
> k
z>44
j
are dually defned.
Intuitively the application oI the optimal flter z
l
perIorms
a whitening through the covariance U
4
}}>l
oI the signal ]
l
(where the other OSC user is excluded) and then extracts the
dominant component oI the channel by projecting the signal
over the eigenvector associated with the maximum SINR.
B. MUD for foint detection of the two OSC users data
We can rewrite the 5 +N . O, signal (5) at the input oI
the detection stage as:
\
z
@ ^|
z>3
> |
z>4
`
W
@ K
z>3
D
3
.K
z>4
D
4
.Q
z
> (9)
where the 5 +O . 4, matrices K
z>3
@ ^k
z>33
> k
z>43
`
W
,
K
z>4
@ ^k
z>44
> k
z>34
`
W
gather the fltered channels Ior
the two in-cell users while the residual interIerence can be
expressed as Q
z
@ ^q
3
> q
4
`
W
=
MUD is the best choice to recover the two sequences
Irom (9). In this section we consider both the optimal non-
linear JMLSE and the linear detection, as a good trade-
oII between perIormance and computational complexity |12|.
Detection is carried out over halI a burst at a time, estimat-
ing the 5P symbols associated with each semi-burst. For
optimal detection, a pre-whitening oI the noise Q
z
has to
be perIormed on the signal (9) using the noise covariance
U
q
@ H^Q
z
Q
W
z
`@+N . O,. This matrix can be estimated
Irom the training signals \
z>w
(i.e., the submatrix oI \
z
relative to the midamble) as
a
U
q
@
4
Qw
a
Q
z>w
a
Q
W
z>w
with
a
Q
z>w
@
\
z>w
K
z>3
D
w>3
K
z>4
D
w>4
. The pre-whitening is given by

\
z
@ ^ |
z>3
> |
z>4
`
W
@
a
U
W@5
q
\ @

K
z>3
D
3
.

K
z>4
D
4
.

Q
z
=
(10)
The corresponding whitened channel responses have been
denoted as

K
z>l
@ U
W@5
q
K
z>l
.
1) Non-linear MUD. Joint maximum likelihood estimation
oI the two users` sequences can be obtained by a VA (see,
e.g., |11|). The estimate is:
a d @duj
ad
plq
N
[
n@4

\
z
+n,

K
z>3
d
3
+n,

K
z>4
d
4
+n,

5
 ~}
M
n
+ad+n,,
 ~}
Path metric
> (11)
where d
3
+n, @ ^d
3
+n,> ===> d
3
+n O,`
W
and d
4
+n, @
^d
4
+n,> ===> d
4
+n O,`
W
collect the symbols associated to the
selected trellis state. The path metric is the overall cost Ior
a given sequence ad while M
n
+ad+n,, is the branch metric,
i.e. the distance between the nth noisy observation and the
possible transmitted symbols ad+n, @ ^a d
3
+n,> a d
4
+n,`. Using
the VA, the path metric can be calculated recursively as:
M
n
+a d+n,, @ M
n4
+a d+n 4,, . M
n
+ad+n,,= The resulting
number oI states in the JMLSE trellis is 5
5O
.
2) Linear MUD. Let us re-arrange the samples oI (10)
associated with a single semi-burst into the 5+P . O, 4
vector:
|
z
@

K
z
d . q
z
> (12)
where d @ ^d
3
+4,> ===> d
3
+P,> d
4
+4,> ===> d
4
+P,`
W
collects the
5P transmitted symbols,

K
z
is the 5+P .O,5P convolu-
tion matrix obtained Irom the fltered channels i

K
z>3
>

K
z>4
j
and q
z
@ ^ q
W
3
> q
W
4
`
W
is the overall residual noise/interIerence.
Minimum mean square error (MMSE) estimation oI the data
vector d is obtained as |13|:
ad @ +

K
W
z

K
z
.L
P
,
4

K
W
z
|
z
> (13)
Iollowed by a threshold detector.
V. NUMERICAL RESULTS
The perIormance oI the proposed receiver is assessed
through Montecarlo simulations oI a standard compliant GSM-
OSC uplink scenario with Q
U
@ 5 antennas and subcarrier
Irequency <33 MHz. Two in-cell and one out-oI-cell MSs are
simulated. One oI the in-cell mobiles employs a conventional
TSC, the other one uses a new TSC as specifed in |14|. Ac-
cording to the ideal Irequency hopping and receiving diversity
assumptions, independent realizations oI the Typical Urban
(TU) channel model |12| are generated Ior each burst, user
and receiving antenna. The vehicular speed is fxed to 6 km/h
Ior all the MSs. The estimated channel has length O @ 7 taps.
The receiver employs a root raised-cosine flter with roll-oII
@ 3=6 and 6dB bandwidth equal to 573 kHz.
The average perIormance oI the generic OSC user is evaluat-
ed in terms oI raw Bit Error Rate (BER) at the detection output
and the Frame Error Rate (FER) Ior an Adaptive Full Speech
(AFS) coded transmission with 8=< kbit/s. The perIormance is
980
Z

-0/6(
Z

-0/6(
Z

/08'
Z

/08'
3,& 3,&
-0/6( -0/6(
-45 -40 -35 -30 -25 -20 -15 -10 -5 0 5 10
10
-3
10
-2
10
-1
10
0
C/
R
a
w

B
E
R
05& 05& 6,& 6,& DAIC DAIC
MRC (no OSC)
SC
PC
JOR L-MUD
JOR JMLSE
JMLSE
DAC
MRC (no OSC)
SC
PC
JOR L-MUD
JOR JMLSE
JMLSE
DAC
Fig. 5. Raw BER vs C/I Ior all the uplink receivers.
-45 -40 -35 -30 -25 -20 -15 -10 -5 0 5
10
-3
10
-2
10
-1
10
0
C/
F
E
R
Z

-0/6(
Z

-0/6(
Z

/08'
Z

/08'
6,& 6,& 05& 05&
3,& 3,& DAIC DAIC
MRC (no OSC)
SC
PC
JOR L-MUD
JOR JMLSE
DAC
MRC (no OSC)
SC
PC
JOR L-MUD
JOR JMLSE
DAC
Fig. 6. FER as a Iunction oI C/I Ior all the uplink receivers.
drawn as a Iunction oI the carrier-to-interIerence ratio defned
as F@L @ S@S
L
, where S @ H^nk
l
+n,n
5
`, l 5 i3> 4j,
denotes the average received power oI each OSC user and
S
L
@ H^nk
L
+n,n
5
` reIers to the interIerer. The proposed JOR
algorithm is compared with all the conventional receivers
presented in Sec. III. A standard uplink GSM receiver, with
Maximum Ratio Combining (MRC) oI signals at diIIerent
antennas, is considered as well as perIormance reIerence,
assuming a conventional cellular scenario where only one user
is active in the cell (no OSC).
Fig. 5 and 6 give, respectively, the raw BER and FER
results. From the comparison between the two fgures we
can see the gain introduced by the coded transmission with
the diversity provided by the block Iading model. Among
all receivers, the JMLSE equalizer experiences the worst
perIormance, as this receiver optimally handles the mutual
interIerence between the two OSC users, but no prefltering is
perIormed to mitigate the out-oI-cell impairments. Both SIC
and PIC oIIer good perIormance, even though the SIC algo-
rithm shows perIormance comparable with the conventional
DAIC approach. The JOR flter with the JMLSE detection
outperIorms all the other solutions, but the 589 states oI the VA
can be a strong limitation Ior the hardware implementation. A
better trade-oII between computational cost and perIormance
is reached by the JOR with linear MUD which guarantees
good perIormance with reasonable computational complexity.
VI. CONCLUSION
The 3GPP is evaluating the new OSC Ieature as the next
solution to improve the cell capacity in the existing GERAN
networks. In this work we have developed a new uplink
receiver with a prefltering stage that has been optimized to
mitigate the out-oI-cell interIerence and a MUD stage that has
been designed to jointly decode the two in-cell users` streams.
Simulation results corroborated the perIormance gain provided
by the proposed receiver with respect to other conventional
receivers.
ACKNOWLEDGMENT
The authors would like to thank Carlo Masseroni and Sergio
Parolari Ior their Ieedbacks and the important guidelines and
the Iormer student M. Minelli Ior his contribution to the
simulation oI GSM systems.
REFERENCES
|1| Nokia Siemens Networks, 'Doubling GSM voice capacity with
Orthogonal Sub Channel, Technology brieI, http://www.nsn.com.
|2| P.A. Hoeher, S. Badri-Hoeher, W. Xu, C. Krakowski, 'Single-Antenna
Co-Channel InterIerence Cancellation Ior TDMA Cellular Radio
Systems, IEEE Wireless Comm., Vol. 12, pp. 30-37, April 2005.
|3| P. Chevalier, PF. Pipon, 'New Insights Into Optimal Widely Linear Array
Receivers Ior the Demodulation oI BPSK, MSK, and GMSK Signlas
Corrupted by Noncircular InterIerencesApplication to SAIC, IEEE
Trans. Signal Processing, Vol. 54, No. 3, pp. 870-883, March. 2006.
|4| R. Meyer, W. H. Gerstacker, R. Schober, J. B. Huber, 'A Single Antenna
InterIerence Cancellation Algorithm Ior Increased GSM Capacity, IEEE
Trans. on Wireless Comm., Vol. 5, No. 4, pp. 1-6, April 2006.
|5| S. Badri-Hoeher, P.A. Hoeher, W. Xu, 'Single antenna interIerence
cancellation (SAIC) Ior cellular TDMA networks by means oI decoupled
linear fltering/nonlinear detection, IEEE Proc PIMRC `06, pp. 1-5,
Sept. 2006.
|6| L. Mattellini, 'Single Antenna InterIerence Cancellation Ior GSM by
Enhanced Maximum SNR Filter, IEEE Proc ISWCS `05, pp. 332-336,
Sept. 2005.
|7| V. Bril, P.S. R. Niniz, R.D. Vieira, 'Adanced Downlink Receivers Ior
GERAN, Proc ITS `06, pp. 801-806, Sept. 2006.
|8| GP-071792, 'Voice Capacity Evolution with Orthogonal Sub Channels,
Nokia Siemens Networks, 3GPP GERAN#36, Canada, Nov 2007.
|9| 3GPP TR 45.004 v7.0.0, 'Modulation, 3GPP GERAN, Aug. 2007.
|10| U. Spagnolini, 'Adaptive Rank-One Reciver Ior GSM/DCS System,
IEEE Trans. Vehicular Tech., Vol. 51, n. 5, pp.1264-1271, Sept. 2002.
|11| P.A. Ranta, A. Hottinen, Z. Honkasalo, 'Co-channel InterIerence Can-
celling Receiver Ior TDMA Mobile Systems, IEEE Proc. ICC `95, Vol.
1, pp. 17-21, June 1995.
|12| S. Verdu, 'Multiuser detection, Cambridge University Press, 1998.
|13| S.M. Kay, 'Fundamentals oI Statistical Signal Processing: Estimation
Theory, Prentice Hall, 1993.
|14| X. Chen, Z. Fei, J. Kuang, L. Liu, G. Yang, 'A scheme oI Multi-User
Reusing One Slot on Enhancing Capacity oI GSM/EDGE Networks,
IEEE Proc. ICCS `08, pp. 1574-1578, Jan 2008.
981