Sie sind auf Seite 1von 6

Enhanced Receiver Structure for Cellular Multiple–Access Systems

with Two Service Classes


Michael Mecking and Zaher Dawy Alexander Seeger
Institute for Communications Engineering Information and Communication Mobile
Munich University of Technology Siemens AG
Arcisstr. 21, D–80290 Munich D–81379 Munich
email: fMecking,Dawyg@ei.tum.de Alexander.Seeger@icn.siemens.de
phone: +49 (0)89 289–23492 +49 (0)89 722–49666

Abstra t | Information theoretic considerations tures and cell geometries. Section IV explains the en-
for the uplink of a cellular mobile communication hanced receiver structure applicable for the two service
system with two service classes are presented. Dif- classes to comply with practical constraints on complex-
ferent receiver structures are compared on the ba- ity and delay. Finally, Section V discusses results on
sis of achievable rates, required received powers, and the user capacity increase due to more sophisticated re-
supportable number of users. The channels consid- ceiver schemes for the power–limited as well as power–
ered are restricted to AWGN channels. Particular unlimited cases.
emphasis is on interference modeling via spill–over
factors for large number of users and different de- II. T HE C ELLULAR M ULTIPLE –ACCESS M ODEL
coding schemes including stripping receivers. This
work offers theoretical bounds for practical cellular The uplink of a wireless mobile communication sys-
systems and gives an insight into the effects of the dif- tem is characterised by a set of multiple transmitters that
ferent system parameters on performance. correspond to mobile users distributed in the coverage
area and a set of multiple receivers that correspond to
I. I NTRODUCTION base stations. The general framework to analyse such a
system is a multiple–input multiple–output channel [2],
A wideband accessing scheme has been selected where each base station acts as a receiver for all the
for the third generation universal mobile telecommu- users dissipated in the coverage area. A practical ap-
nications system (UMTS) due to its service flexibility proach that is implemented in current systems assigns
and improved performance over second generation sys- users to specific receivers to form cells, thus, the cov-
tems [1]. UMTS will provide access to a comprehensive erage area is partitioned into several cells with a single
set of services, ranging from simple voice connections receiver each. A cell receiver processes only signals
to high quality multimedia applications. Data services received from users allocated to the corresponding cell,
in UMTS are characterised by high transmission rates and treats signals received from neighbouring cells as
and non–stringent delay requirements, thus, enabling interference. This approach falls within the class of
long block length coding schemes. On the other hand, multiple–access and interference channels, the optimum
speech services are characterised by low transmission capacity region of which is still an open problem [3].
rates and real time delay constraints, thus, requiring The multiple–access is from users within the cell and
shorter block lengths and low complexity decoding. the interference is from users in other cells which are
transmitting to their own base stations. In the sequel,
the other cell interference is modeled as Gaussian noise
In this work, information theoretic considerations
which can be viewed as a worst case assumption [2].
for the uplink of a cellular system with two service
classes, e.g., voice and data, are presented. The chan-
Consider the uplink scenario of a cellular mobile
nels considered are restricted to AWGN channels. An
communication system composed of M cells with K
information theoretic analysis of different multiuser
users per cell. It is assumed that each user is allocated
receiver structures for a cellular environment is per-
only to one cell at a time and, thus, multiple cell site pro-
formed which offers theoretical bounds for practical
cessing is not considered in this work. All users follow
systems and gives an insight into the effects of the
a wideband accessing strategy using the whole available
different system parameters on performance. In cellular
resource all the time resembling the accessing scheme in
systems, performance is mainly limited by the amount
UMTS. The received signal Ym at cell site m is given by
of interference from neighbouring cells. Therefore, it is
X X X q (m) (m0 )
Ym = Xk(m) + + Zm ;
important to understand the implications of the receiver
structures to the power consumption within the cell as k;m0 Xk (1)
well as the interference into adjacent cells. k2K m0 2Am k2K
(m)
where Xk is the received signal of user k in cell m,
Section II introduces the cellular multiple–access
model. In Section III, particular emphasis is on the spill– Am is the set of indicesqof all cells that contain users
m)
over factor computation for different receiver struc- interfering with cell m, (k;m 0 is an attenuation factor
0
multiplied by Xk
(m )
to give the received signal of user from cell m0 to cell site m tends to the statistical mean
k in cell m0 at cell site m, and Zm is additive Gaussian by the law of large numbers
noise of zero mean and variance  2 . We assume that
P X m
the users’ signals are statistically independent, Gaussian
(m)
 0
K k2K k;m
! P E fmm0 (w)g as K ! 1: (3)
distributed, and symbol and block synchronized. k;m0
depends on the location of user k with respect to the cell
sites m and m0 . Under the assumption of an attenuation Thus, can be calculated as
(m) X
in power proportional to a power a of the distance, k;m0 = m0 (w)g :
E fm (4)
m0 2Am
is given by
0 m0 a
(
= m 0 )a ;
(m )
k;m 0 rk;m
( )
The validity of the spill–over factor approach is justified
(m)
k;m0 = (2)
(m)
k;m0 (rk;m0 )
( ) by the following argument. Let
!
X X X
where
m)
(k;m 0 is the geometrical path loss from user k Im = PS k;m0 + PD
(m) (m)
k;m0 (5)
m0 2Am k2KS k2KD
in cell m0 to the site of cell m, rk;m0 is the distance
(m)

from user k in cell m0 to the site of cell m, and a is the denote the total intercell interference at cell m from all
(m)
attenuation exponent. Shadowing and fast fading are not neighbouring cells . Mind that k;m0 merely depends on
considered. 0
the position of user k in cell m . Let

=  + (K S P+SI+m KD PD )
2
Assume that the K users are divided into KS speech T (6)
users and KD data users with transmission rates RS and 2

RD , respectively. The decoding of the K users in the


denote the ratio of actual to average interference and
same cell is based only on information received at that
noise power. Then, for a fixed ratio KS =KD and =
cell site. Moreover, each cell decoder is ignorant about
the codebooks used by users assigned to other cells and, +
KS KD K , =
thus, treats signals received from other cells as Gaussian !
P P P
noise. 2 + PS k;m0 + PD
(m) (m)
k;m0
m0 2Am k2KS k2KD
III. I NTERFERENCE M ODELING VIA S PILL –OVER T = 2 + (KS PS + KD PD )
FACTORS
(7)
!
The model presented in (1) can be simplified by mod- P
+ PS P (m) P (m)
+
eling the intercell interference via a spill–over or –
2
K K k;m 0 PKD k;m0
m0 2Am k2KS k2KD
factor defined by the ratio of the intercell interference
to the intracell total received power [2] [4]. The corre-
= 2 + (
KKS PS KKD PD + )
K
sponding system model is depicted in Fig. 1, where PS (8)
and PD are the average received powers of speech and
data users, respectively.
=1 as K!1 (9)

with probability 1 by (4) and the law of large num-


Xk ; k 2 K S bers [5]. By continuity,

Xk ; k 2 K D
Y log T = 0 as K ! 1 (10)

with probability 1, so that for all capacity calculations,


the actual intercell interference can be replaced by the
average intercell interference.

Z  N (0; 2 + (KS PS + KD PD )) –factors have been evaluated for a linear and hexag-
onal array of cells and different attenuation coefficients a
Fig. 1: Cellular GMAC with 2 service classes. based upon a uniform user distribution within each cell.
Results can be found in Table I. Results for a two di-
mensional hexagonal array of cells have been calculated
The modeling of the intercell interference via the – by Monte–Carlo integration due to the more complex
factor approach can be applied for all decoding schemes cell geometry. For both cases, the intercell interference
that require that users within a decoding group should be is dominated by interference from the first tier of cells.
received with the same power P at the receiver. In this Note that is decreasing with increasing a due to de-
case, the transmit power of user k depends only on his creased intercell interference. Moreover, the value of
location and is independent of the locations of the other is independent of the number of users per cell and the
users. Thus, a snapshot average intercell interference required received power per user.
TABLE I x 2 [r; 2r℄, respectively, for an average user distribu-
–factor for a network with equal received powers tion. Im;i
(1)
(a) and Im;i (a) can be evaluated as
(2)
a linear hexagonal
4 0:1271 0:4149 Z r
2 ( Kr 1 x)R (x +jx(2i rj1)r)a dx;
a
3:75 0:1369 0:4659 Im;i
(1)
(a)= K;R 2

3:5 0:1487 0:5307 0

3:25 0:1629 0:6179 Z r (14)


3 0:1806 0:7246 Im;i
(2)
(a) = K;R 22(K Kr 1 x)R xa
(x + 2ir)a dx;
0
(15)
Stripping receivers are known to achieve maximum where
sum rate for a multiple–access channel [6]. Furthermore, K2 (22R 1)
K;R = :
stripping receivers appear as the solution to minimise
sum transmit power for a multiple–access channel with
2r (16)

attenuations [7]. If the attenuations are in increasing or- The intracell received signal, Iin , at cell site m is equal
der, control the transmit power of user k such that he is to KP where P is the average received power of each
received with power user in cell m. Since these powers correspond to a ver-
Pj>k Rj tex in the power region, the sum power KP can be cal-
Pk = 2 2 Rk 1 2
2 2
: (11) culated as

Then, decode the users in increasing order of attenua- Iin = KP =  (2 KR 1):


2 2
(17)
tions followed by a cancellation or stripping of the users’
contribution to the received signal at the cell site. This Thus, the –factor for the case with stripping, S , is
insures a signal–to–noise ratio at the input of decoder k given by
P
Pk
P = 2 Rk 1;
2
S
2
= i> Im;i (a)
;
2 +
(12) 0
j>k Pj
(18)
Iin
which, under the assumption of Gaussian signaling, al- where the factor of 2 takes care of the interference from
lows the reliable communication of user k at rate Rk . + 0
cells m i; i < . Note that the value of S depends
Due to the ordering of users according to the attenua- on the input parameters R and K . The integrals are
tions, the user having the highest attenuation is received calculated numerically. Table II presents the value of
with the least possible power. As users close to cell S for a linear cellular model with stripping receivers
boundaries tentatively have higher attenuations, strip- with different values of the attenuation exponent a for
ping receivers prove beneficial in cellular environments different parameters R and K . Mind that the underly-
and reduce intercell interference. However, the trans- ing idea is to compute the interference for an average
mit power of a user also depends on his decoding posi- scenario and that the results are approximations to the
tion, and thus, transmit powers are not independent of average of all scenarios.
each other precluding a law of large numbers argument.
Consider a linear cellular array of cells as depicted in
TABLE II
m 2 m 1 m m+1 m+2 S –factor for a single service linear network with
stripping
a R = 0:0417 bits R = 0:0026 bits
r x K = 10 K = 80
4 0:1031 0:1137
Fig. 2: Linear cellular network. 3:75 0:1116 0:1227
3:5 0:1218 0:1336
3:25 0:1344 0:1468
Fig. 2 with a fixed number K of users per cell requiring 3 0:1499 0:1633
a service characterised by a rate R. Assuming a uniform
user distribution, the fraction of users in an interval of
  (2 )
width x converges towards x= r as the number K These results can also be extended to account for a
of users increases. This motivates the evaluation of the
interference at cell m emanating from cell m i, with + linear network with two service classes such that all
0
i > , as
users are decoded jointly via stripping. This case is not
practical due to the different delay constraints of the ser-
Im;i (a) = Im;i (a) + Im;i (a);
(1) (2) vice classes, but it is analysed for comparison purposes.
(13) Furthermore, results for a hexagonal array of cells have
been derived by a Monte–Carlo approach with an itera-
(1)
() (2)
()
where Im;i a and Im;i a are the interference from tive procedure of updating transmit powers in the cells to
users in cell m + 0
i; i > ; having x 2 ; r and [0 ℄ overcome intercell interference from surrounding cells
until convergence is reached [8]. The results for a hexag- able rates is readily given by
onal cellular array are presented in Table III where it is  
PS
shown that intercell interference increases in a hexag-
onal model due to the existence of more neighbouring
RS  C  2 PS + (1 + )Iin
; (22)
cells and different cell geometry. It is apparent that S  
PD
increases as the attenuation exponent a decreases, and is RD  C  2 PD + (1 + )Iin
: (23)
decreasing in both R and K .
The corresponding region of powers supporting given
TABLE III rates can be obtained for all schemes by fixing the rates
S –factor for a single service network with stripping RS and RD and solving for the powers PS and PD . In
(R = 0 0417
: bits, K = 16) the sequel, this receiver will be denoted as the conven-
a linear hexagonal
4 0:0887 0:3020
tional scheme.

3:75 0:0963 0:3406 B. Enhanced Receiver Structure


3:5 0:1055 0:3916
3:25 0:1167 0:4540 The system performance can be improved by taking
3 0:1309 0:5380 advantage of the properties of the two service classes.
Due to the delay constraints, speech users are decoded
first in parallel according to the conventional scheme.
Then, they are inter–group stripped or canceled. Data
Comparing the results of Table II and III with those users can tolerate longer delays and afford more decod-
of Table I, it is obvious that the spill–over factor is sig- ing complexity. Thus, they are decoded via stripping.
nificantly less when all users are decoded via stripping The effect of inter–group stripping (IS) is that speech
than when users are decoded as one group with equal users do not act as interference to the data users. Hence,
received powers, i.e., the stripping scheme allows the data users to lower their
transmit powers, which in turn supports a reduction of
S < : (19) the speech users’ transmit powers as they see less in-
terference than for the conventional scheme. The set of
IV. P ERFORMANCE OF D IFFERENT R ECEIVER achievable rates is given by
S TRUCTURES WITH T WO S ERVICE C LASSES  
PS
RS C ;
In this section, different receiver structures for the 2 PS + KS PS + S KD PD + Iin
uplink of a cellular mobile communication system with (24)
two service classes are compared using information the-  
KD PD
KD RD  C :
2 + KS PS + S KD PD
oretic arguments. The different considered schemes of- (25)
fer a trade–off between increased complexity and im-
proved system performance, and take into account the
different properties of the service classes. With intercell Note that > S due to the modified power con-
interference and decoding of the users only in their re- trol (11). Applying IS has a low complexity and im-
spective cells, all schemes are interference limited, thus, proves the system performance significantly. This pro-
there is an upper limit on the achievable user capacity ir- posed receiver structure is denoted as the enhanced
respective of the used powers. In the sequel, denote the scheme.
total intracell received power at each base station by C. Joint Receiver Structure
Iin = KS PS + KD PD ; (20) This scheme can achieve the best performance among
the presented schemes, but it cannot be implemented in
and let a practical system since all users, speech and data, are
C (x) =
1 log(1 + x) decoded jointly as one group neglecting the delay and
2 (21) complexity constraints of the service classes. Therefore,
it is presented for comparison purposes. In general, the
denote the capacity in bits per channel use of a
single–user AWGN channel with signal–to–noise ratio
2
rate region is determined by KS +KD 1 constraints [3],
but it can be proved that the set of achievable rates is
(SNR) x. given by the following three class constraints:
 
A. Conventional Receiver Structure K P
KS RS  C 2 S S ;
 + Iin
(26)
The first proposed scheme corresponds to the basic  
K P
receiver structure proposed in UMTS system specifica- KD RD  C 2D D ;
 + Iin
(27)
tions where each user is independently decoded treat-  
KS PS +KD PD
KS RS + KD RD 
ing all other users as interference. The simplicity of this
C : (28)
scheme makes it the least complex. The set of achiev- 2 + Iin
For a subset of k speech users, attenuation exponent a decreases due to the increase in
1 
kP
 the intercell interference. For a = 3 75
: , the enhanced
RS  C 2 S ; 168
 + Iin
(29) scheme can handle a maximum of users while the
k conventional scheme can handle only users, so nearly 66
which is decreasing in k . Thus, k = KS gives the tight- a three fold increase in user capacity can be achieved.
est constraint (26). Similar arguments yield (27) as the Very high gains can be achieved by (the non–practical)
tightest constraint on RD for a group of data users. Fi- joint/stripping scheme where all users are decoded via
nally, (28) gives the tightest constraint on the sum rate of stripping as one group. In some cases, the enhanced
all users in the cell. It can be shown that no other mixed scheme operates even better than the joint/equal scheme
constraint of the form (where users are received with equal powers per group
 
kS PS + kD PD
and jointly decoded) due to the lower intercell interfer-
kS RS + kD RD  C ence for stripping schemes.
2 + Iin
(30)

for kS ; kD  1 can provide tighter bounds on RS or RD . The second performance measure is based on com-
paring the power regions of the different receiver struc-
The value of the –factor varies depending if decod- tures. The power region gives the set of received powers
ing is done jointly with users within each service class required for reliable communication. The power region
received with the same powers PS and PD , or is done via of each scheme is bounded by a set of constraints. It can
stripping with users within each service class received be plotted in the PD –PS plane, where PS and PD are
with different powers of average PS and PD . In the se- the average received powers of the speech and data ser-
vices, respectively. Fig. 3 depicts the power regions for
quel, the scheme with users decoded jointly and received
with the same power within each service class is denoted a hexagonal network with KS = 40
users and KD =8
joint/equal while the scheme where users are decoded users.
via stripping is denoted joint/stripping. Note that the 0.04
joint/stripping scheme is an optimal scheme with users conventional scheme
enhanced scheme
decoded and stripped at the cell site in increasing order 0.035
joint/equal scheme
joint/stripping scheme

of attenuations irrespective of their service classes.


0.03

V. R ESULTS AND D ISCUSSION


0.025

Cellular system operators care most for the user ca-


2
P /σ

pacity that can be handled in the network. Having the 0.02


S

set of rates fixed and operating at minimum required


0.015
powers, a measure is to compare how many users each
receiver structure can support. The different receiver 0.01
structures are compared on the basis of maximum user
capacity. Due to the effect of intercell interference, all 0.005

the considered schemes are interference limited in the


sense that for a given RS , RD , and = KS =KD , 0
0 0.05 0.1 0.15 0.2 0.25
P /σ2
0.3 0.35 0.4 0.45 0.5

there exists an upper bound on the number of users, D

which depends on and , that can be handled in


each cell irrespective of the users’ powers. For RS = Fig. 3: Power regions for a hexagonal network with K =
1 3 1 128 = 0 0026
=  = : bits, RD = 1 3 1 8 = 0 0417
=  = : 48 users (RS = 1=3  1=128 = 0:0026 bits, RD = 1=3 
=5
bits, and , the maximum user capacity of the dif- 1=8 = 0:0417 bits, = 5, a = 3:75).
ferent multi–cellular receiver structures is presented in
Table IV. The network model used is linear and the re-
sults are presented for different values of the attenuation The set of minimum received powers is given by
exponent a. the dominant face for schemes with joint decoding and
by the intersection point of the bounding lines for the
TABLE IV conventional and enhanced schemes. The joint/stripping
Maximum user capacity of the different receiver scheme gives a lower bound on the set of required
structures for a linear network ( =5
) received powers. The proposed enhanced scheme per-
decoding scheme a =4 = 3 75
a : a : =35 forms very near to the optimum scheme and, thus, offers
conventional 72 66 66 a very attractive trade–off between complexity and
enhanced 168 168 162 performance. As expected, the conventional scheme has
joint/equal 168 162 156 the poorest performance and requires much more power
joint/stripping 318 294 276 consumption.

Finally, the different receiver structures are compared


These results show that the maximum user capacity with respect to achievable user capacity with fixed power
of the different receiver structures decreases as the consumption. While in the previous paragraph an un-
limited power of the users was assumed, the total power network. The comparison of the different schemes was
consumption is now fixed to the one required for the based mainly on information theoretic considerations
conventional scheme to support the required number of that provide fundamental performance bounds and give
users. Then, the maximum supportable number of users insight into the effects of different system parameters
for the remaining schemes with the same overall power on the achievable performance. The basic comparison
consumption is evaluated. Fig. 4 depicts the user ca- measures used were achievable rate regions, required
pacity increase achievable with the proposed structures. power regions, required transmit powers, and maximum
Note that the straight line for the conventional scheme user capacities. The system parameters used for the
has a slope of 1 and the total number of users is a multi- comparisons comply with the UMTS specifications.
ple of . 1+
Comparing the different proposed schemes, nearly a
160
conventional scheme
three fold increase in user capacity over the conventional
enhanced scheme
joint/equal scheme
scheme based on independently decoding all users can
140 joint/stripping scheme
be achieved by independently decoding speech users
first, then stripping them from the data users, and finally
120
decoding the data users via stripping.
D
total no. of users K = K + K

100
S

This work is being extended to a multi–cellular sys-


80 tem with soft handover, where outer bounds on the set
of achievable rates can be found. Moreover, the influ-
60 ence of asynchronous users with overlapping codewords
on capacity under a strict delay constraint can be in-
40
vestigated. Through this work, coding in the Shannon
sense was used to achieve the optimal performance. In
20
practice, CDMA systems apply spread spectrum meth-
0
ods based on direct sequence spreading. Further work
0 10 20 30 40 50 60 70
conventional scheme total no. of users K = K + K
S D
can be done by investigating the effect of spreading on
system performance and comparing different detection
Fig. 4: Capacity increase with fixed power consumption schemes.
( RS= 1 3 1 128 = 0 0026
=  = : bits, RD =  = =1 3 1 8=
0 0417
: = 5 = 3 75
bits, ,a : ). VII. R EFERENCES

[1] H. Holma and A. Toskala, WCDMA for UMTS. Eng-


land: Wiley, 2000.
Consider a scenario with the conventional scheme
employed as a decoding scheme to achieve a user ca- [2] S. Shamai (Shitz) and A. D. Wyner, “Information-
pacity of 48
users. There are two ways to increase the theoretic considerations for symmetric, cellular,
user capacity from to users. Either increase the 48 66 multiple-access fading channels - part i,” IEEE
power extensively or keep on using the same amount Trans. Info. Theory, vol. 43, pp. 1877–1894,
of power but employ the enhanced scheme as the de- November 1997.
coding scheme. Moreover, the performance of the en- [3] T. M. Cover and J. A. Thomas, Elements of Infor-
hanced scheme is very near to that of the joint/stripping mation Theory. New York: Wiley, 1991.
scheme for an interesting range of users, and congru- [4] B. Rimoldi and Q. Li, “Potential impact of rate–
ent to decoding all users jointly but without stripping splitting multiple access on cellular communica-
(joint/equal scheme). Thus, with an enhanced but subop- tions,” in IEEE GLOBECOM’96, pp. 92–96, 1996.
timal receiver structure, user capacity can be increased
substantially without increasing overall power consump- [5] P. Billingsley, Probability and Measure. New York:
tion. It is worth spending more complexity on the minor- Wiley, 1995.
ity of high rate users where the effort is affordable, and [6] B. Rimoldi and R. Urbanke, “A rate–splitting ap-
stripping proves to be beneficial within the cell (maxi- proach to the gaussian multiple–access channel,”
mum sum rate with low complexity) and outside the cell IEEE Trans. Info. Theory, vol. 42, pp. 364–375,
(lower intercell interference). March 1996.
[7] D. N. C. Tse and S. V. Hanly, “Multiaccess fad-
VI. C ONCLUSIONS
ing channels - part I: Polymatroidal structure, opti-
The present work has given a comparison of opti- mal resource allocation and throughput capacities,”
mum and suboptimum receiver structures for the uplink IEEE Trans. Info. Theory, vol. 44, pp. 2796–2815,
of a CDMA cellular mobile communication system November 1998.
with two service classes. The two service classes are [8] R. D. Yates, “A framework for uplink power control
characterised by different required transmission rates, in cellular radio systems,” IEEE Journal on Sel. Ar-
delay constraints, power constraints, and user loads, and eas in Comm., vol. 13, pp. 1341–1347, September
they correspond to speech and data services in a real 1995.

Das könnte Ihnen auch gefallen