Sie sind auf Seite 1von 12

Admission control in UMTS in the presence of shared channels

S.-E. Elayoubi
*
, T. Chahed, G. Hebuterne
GET/Institut National des Telecommunications, 9 rue C. Fourier, Evry, Cedex 91011, France
Abstract
In this article, we focus on Call Admission Control (CAC) in the presence of both real-time and elastic ows in a multicell UMTS
environment where elastic ows share a common, shared channel. We rst study the capacity of a multicell UMTS system. We determine an
upper bound on the other-cell interference and obtain novel expressions for the SIR and powers for both the uplink and the downlink. The
former is an asynchronous CDMA system, often using one of three types of receivers: matched lter, minimum mean-square error and
decorrelator. In the latter, the SIR depends on the distance between the user and the base station. Based on these expressions, we develop a
novel CAC algorithm that takes into account the effects of mobility, coverage as well as the wired capacity in the UMTS Terrestrial Radio
Access Network (UTRAN), for the uplink, and the maximal transmission power of the base station, for the downlink. Our algorithm
outperforms classical ones as it achieves lower dropping rates. We eventually use this framework to manage priorities between real-time
calls, transmitted over dedicated channels, and elastic calls, transmitted over common or shared ones. Real-time calls are given an admission
priority if the channel is constrained, while decreasing the ongoing elastic calls share.
q 2004 Elsevier B.V. All rights reserved.
Keywords: Universal mobile telecommunication system; Call admission control; Multiuser receivers; Elastic trafc
1. Introduction
Third generation radio communication systems are
designed to offer multimedia services, including voice and
video telephony and high-speed Internet access. In Uni-
versal Mobile Telecommunication System (UMTS), the
radio interface is based on the Wideband Code-Division
Multiple Access (W-CDMA) technology. In such a context,
QoS is tightly dependent on the Signal to Interference Ratio
(SIR) level for each user. A Connection Admission Control
(CAC) algorithm is then needed to limit the interference by
controlling the number of accepted ows.
Several CAC algorithms have been developed in UMTS.
Most commonly, CAC algorithms are based on a pre-
determined maximal number of users in the system [7,8].
This approach, called NCAC, is not adequate in a CDMA-
based system, rst because the system is interference-
limited and second, because multimedia calls often have
variable bit rates. Other algorithms are more CDMA-
oriented and consider the SIR as the determinant parameter
in accepting or not a new call; the idea being mainly that a
new call is accepted if its contribution to the overall
interference does not make the latter exceed a given value
[811]. This choice is based on the fact that the mutual
information achieved for each user under an independent
Gaussian input distribution is equal to 1=2 ln(1 -SIR) bits
per symbol time. Meeting a target SIR is hence equivalent to
meeting a target rate and satisfying a QoS requirement.
Those algorithms are commonly called InterferenceCAC
(ICAC).
Our aim in this work is to develop a novel CAC
algorithm that combines the efciency of ICAC and the
simplicity of the NCAC, and that functions on both
dedicated and shared channels. To do so, we rst produce
a novel analytical model for a multicell UMTS system that
takes into account other-cell interference and distinguishes
between uplink and downlink [17].
For the uplink, we base our work on the model developed
in Ref. [1], where effective bandwidth characterizations
were formulated for different kinds of receivers, namely the
matched lter receiver, the Minimum Mean-Square Error
(MMSE) one and the decorrelator [2]. However, their study
considered a single-cell system, and it is well known that
other-cell interference, i.e. inter-cell interference, is very
important in CDMA systems. Hence, in this present work,
we extend this model to a multicell setting and specically
0140-3664/$ - see front matter q 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.comcom.2004.01.023
Computer Communications 27 (2004) 11151126
www.elsevier.com/locate/comcom
* Corresponding author. Tel.: -33-1-60-76-47-16; fax: -33-1-60-76-42-
91.
E-mail addresses: salah_eddine.elayoubi@int-evry.fr (S.-E. Elayoubi);
tijani.chahed@int-evry.fr (T. Chahed); gerard.hebuterne@int-evry.fr
(G. Hebuterne).
investigate a Gaussian versus log-normal distribution for
modeling other-cell interference [3].
As of the downlink, we obtain an upper bound on the inter-
cell interference as a function of the distance between the user
and the base station and use it to obtain the relationship
between the SIR and the power emitted to each user.
We next use this analytical model to produce a novel
CAC algorithm [18]. For the uplink, this algorithm takes
into account the mobility and coverage as well as the wired
capacity of the UMTS Terrestrial Radio Access Network
(UTRAN), beyond the Node B. Indeed, air and wired
resources are complementary; the former is preponderant in
large cells, used in rural environments where we have a
small load factor and the system is coverage-limited, while
the wired capacity is more signicant in urban environments
where small cells with high load are used and the system is
capacity-limited [6]. As of the downlink, in addition to the
mobility and coverage constraints, the transmission capacity
of the base station is considered as a limiting factor.
The obtained CAC algorithm works well for dedicated
channels, but to attain a global level of QoS, we
shall enhance it to deal with elastic calls on common/shared
channels. Indeed, real-time and non-real-time trafc are
transported differently. Real-time trafc, such as voice, is
carried over so-called Dedicated CHannels (DCHs),
wherein each call is allocated a specic code. Elastic trafc
calls, however, can share the same code on a Common
Packet CHannel (CPCH) in the uplink or a Shared CHannel
(DSCH) in the downlink. This is preferable as it preserves
code resources (especially in the downlink) and facilitates
the support of variable bit rates. We then produce a
squeezing mechanism that, combined with the CAC
algorithm, distributes the resources between calls of
different classes, and insures priority for real-time calls
over elastic ones.
The remainder of this work is organized as follows. In
Section 2, we develop an analytical model for the other-cell
interference for both the uplink and the downlink. In Section
3, SIR and power expressions are derived as a function of
several system parameters, including inter-cell interference.
This is done for the most commonly used types of receivers,
namely the matched lter, MMSE and the decorrelator.
Based on these expressions, we detail, in Section 4, our
CAC algorithm considering wired as well as wireless
system constraints, applied to both dedicated and shared
channels. The efciency of this algorithm is further
illustrated by simulations in Section 5. Section 6 eventually
concludes the paper.
2. Modeling other-cell interference
2.1. Case of the uplink
We consider a homogeneous DS/CDMA cellular system
with hexagonal cells, of radius R; uniformly deployed. Each
cell is sectorized using directional antenna into three
sectors, each antenna having 120 degrees effective beam-
widths. This decreases by 3 the overall interference [3]. In
such a system, the path loss between the subscriber i and the
cell site is given by q
i
= r
m
i
10
j
i
=10
; where r
i
is the distance
from the base station and j
i
is a random variable due to
shadowing, and m = 4:
For a perfect power control, each user is received by its
own base station with equal power P: However, the other-
cell interference power P
I
at the target base station is caused
by users that are not power controlled by this latter, and
whose received powers cannot be determined individually.
The solution is then to derive distributions for the other-cell
interference in capacity bounds calculations.
2.1.1. Gaussian approximation
Several works as in Refs. [3,4] modeled the inter-cell
interference as a Gaussian random variable, whose rst and
second order moments are upper bounded by Ref. [3]
E(P
I
=P) # 0:247N
S
(1)
Var(P
I
=P) # 0:078N
S
(2)
N
S
being the average number of users by sector. This choice
is based on the idea that this interference is the sum of a
large number of individual powers, and that the central limit
theorem allows us to consider this sum as Gaussian-
distributed.
In this case, the other-cell interference is upper bounded
as a Gaussian variable by
P
I
# E(P
I
) -2

Var(P
I
)
_
= I
u
with Pr(P
I
. I
u
) = 0:025: Using Eqs. (1) and (2), the other-
cell interference bound becomes
I
u
. (0:247N
S
-2

0:078N
S
_
)P = dP (3)
where d; function of N
S
; can be interpreted as an equivalent
number of local users.
2.1.2. Log-normal approximation
Alternatively, some studies considered that Gaussian
distribution is not very accurate in modeling the other-cell
interference in inhomogeneous UMTS. Experimental
studies in Ref. [5] showed that the log-normal distribution
models better the other-cell interference. An iterative
algorithm is proposed to compute the mean E(P
I
) and
variance Var(P
I
) of this interference.
In such a case, ln(P
I
) is normal distributed with
parameters
E(ln(P
I
)) = ln(E(P
I
)) 2
1
2
ln
Var(P
I
)
E(P
I
)
2
-1
_ _
(4)
Var(ln(P
I
)) = ln
Var(P
I
)
E(P
I
)
2
-1
_ _
(5)
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1116
and we have as before
ln(P
I
) # E(log(P
I
)) -2

Var(log(P
I
))
_
(6)
This leads to the upper limit of the other-cell interference in
the uplink:
I
u
=
E[P
I
]

Var[P
I
]
E[P
I
]
2
-1
_ exp 2

ln
Var[P
I
]
E[P
I
]
2
-1
_ _
_ _ _
(7)
By numerical comparison (please refer to Section 5.1), the
Gaussian and log-normal distributions are very close in
the calculation of the SIR, both in mean value and
convergence. We hereafter use the Gaussian one for the
uplink. It is log-normal in the downlink as shall be seen
next.
2.2. Case of the downlink
The calculation of the other-cell interference is now
different, because of the fact that the interference generated
by users in cell l 0 (0 being the target cell) is emitted by
the base station of cell l; and not distributed on the
cell surface as in the uplink. On the other hand, this
interference depends on the position of the mobile station in
the cell, i.e. on its angle (u
i
) with the reference axis and
distance (r
0i
) from its own cell Node B (Fig. 1). For user i;
the shadowing can be modeled by considering the random
component of the dB loss as the sum of a component (j
1li
)
common to all base stations and another one (j
2li
)
independent from one Node B to another: j
li
= aj
1li
-
bj
2li
where a
2
-b
2
= 1 with E(j
li
) = E(j
1li
) = E(j
2li
) = 0
and Var(j
li
) = Var(j
1li
) = Var(j
2li
) = 6
2
: A reasonable
assumption is that a
2
= b
2
= 1=2:
The power of other-cell interference then veries
P
d
i
q
0i
=P
tot;0
=

1
l=1
P
tot;l
P
tot;0
r
0i
r
li
_ _
m
10
(j
li
2j
0i
)=10
(8)
where P
tot;l
is the total power emitted by base station l:
If we consider all these powers to be equal to P
d
; we have
E[P
d
i
q
0i
] = P
d

1
l=1
r
0i
r
li
_ _
m
E[10
(j
li
2j
0i
)=10
lP
r0i
. P
rli
]
where P
r0i
and P
rli
are the powers received for user i from
base stations 0 and l; respectively; note that
P
r0i
. P
rli
if b(j
2li
2j
20i
) , 10m log
r
li
r
0i
_ _
This condition infers that a mobile station belongs to a
certain BS with the smallest attenuation rather than the
shortest distance. This can be written as [14]
E[P
d
i
q
0i
] =P
d
e
(b6V)
2

1
l=1
r
0i
r
li
_ _
m
1 2Q
10m log
r
li
r
0i
_ _

2
_
b6
2

2
_
6bV
_
_
_
_
_
_
_
_
_
_
_
_
_
_

_
_
_

_
_
_
(9)
where V = ln(10)=10 and Q() is the complementary
cumulative distribution function of a standard (zero mean,
unit variance) Gaussian process.
However, in the literature, only the mean value of the
other-cell interference is used. To nd an upper limit of the
interference, we calculate its variance and we nd out after
tedious calculations that
Var[P
d
i
q
0i
]=P
d
2

1
l=1
r
0i
r
li
_ _
2m
e
4(b6V)
2
12Q
10mln
r
li
r
0i
_ _

2
_
b6
22

2
_
6bV
_
_
_
_
_
_
_
_
_
_
_
_
_
_

_
_
_

_
_
_
2e
2(b6V)
2
12Q
10mln
r
li
r
0i
_ _

2
_
b6
2

2
_
6bV
_
_
_
_
_
_
_
_
_
_
_
_
_
_

_
_
_

_
_
_
2
(10)
This variance can be relatively large compared to the case of
the uplink, due to the large number of uplink interferers (that
have different path losses), which makes the uplink
interference converge to a Gaussian variable (central limit
theorem); this is not the case in the downlink because the
interferers (Node Bs) are localized in few points in
the space. However, the other cell interference for the
downlink can be considered as log-normal ([16] page 1005).
As the path loss q
0i
for each position in the cell is also Fig. 1. Cellular system in the downlink.
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1117
log-normal
y
i
= ln(P
d
i
q
0i
) = ln(P
d
i
) -ln(q
0i
)
is a Gaussian variable having the following parameters
E(y
i
) = ln(E(P
d
i
q
0i
)) 2
1
2
ln
Var(P
d
i
q
0i
)
E(P
d
i
q
0i
)
2
-1
_ _
(11)
Var(y
i
) = ln
Var(P
d
i
q
0i
)
E(P
d
i
q
0i
)
2
-1
_ _
(12)
We then have, in as much as 90% of the cases
y
i
, E(y
i
) -1:25

Var(y
i
)
_
which gives P
d
i
q
0i
, I
sup
i
where
I
sup
i
=
E[P
d
i
q
0i
]

Var[P
d
i
q
0i
]
E[P
d
i
q
0i
]
2
-1
_ exp 1:25

ln
Var[P
d
i
q
0i
]
E[P
d
i
q
0i
]
2
-1
_ _

_
_
_
_
_
This upper limit is drawn in Fig. 2 as a function of the
normalized distance r
0
=D; D being the distance between
two adjacent base stations, and for different values of the
angle.
One can see that the difference between the curves is
small for different values of u
i
: We can then consider the
maximal value of I
sup
i
; obtained in most of the cases for
u = p=2; as the upper limit of the interference. Let us call
this variable I
d
i
: The latter is thus an upper bound on the
upper bound of the downlink other-cell interference.
3. SIR and capacity in multicell UMTS
3.1. SIR in the uplink
In the UMTS, the uplink is an asynchronous spread
spectrum system (see Ref. [15] page 15). The sampled
discrete-time model for the received signal is then given by
~
Y = X
01
s
01
--
-

K
0
i=2
(X
0i
u
0i
--
-Y
0i
n
0i
--
)
-

1
l=1

K
l
k=1
(X
lk
u
lk
--
-Y
lk
n
lk
--
) - ~ w (13)
where s
lk
--
is the code attributed to user k of cell l containing
K
l
users. X
lk
; Y
lk
are two consecutive transmitted symbols of
user (l; k) which overlap with the target user (0; 1) in the
observation window. These have effective signature
sequences u
lk
[ R
N
and v
lk
[ R
N
; respectively. The
effective signature sequences are completely determined
by the original signature sequences s
lk
--
and the delays
relative to target user, d
lk
[ Z
-
(u
lk
)
n
=
1

N
_ V
lk(N2d
lk
-n)
; n # d
lk
0; d
lk
, n # N
_
_
_
_
_
(v
lk
)
n
=
0; n # d
lk
1

N
_ V
lk(n2d
lk
)
; d
lk
, n # N
_
_
_
_
_
where s
lk
--
=
1

N
_
(V
lk1
; ; V
lkN
)
t
and the random variables
V
lkn
are independent and identically distributed with zero
mean and variance 1. In general, V
lkn
s are equal to 21 or
-1. Note that the terms s
lk
--
; l = 01; k = 1K
l
; are
different because of the use of scrambling codes after
spreading with channelization codes in UMTS systems.
Moreover, E(X
lk
) = E(Y
lk
) = 0 and E(X
2
lk
) = E(Y
2
lk
) = P
lk
where P
lk
is the power received from user l in cell k at the
target cell base station. ~ w is N(0; s
2
Id
N
); the background
Gaussian noise, and Id
N
is an N N identity matrix.
~
Y and
s
lk
--
are of length N; N being the length of the spreading
code. To extract the symbol X
01
from the quantity
~
Y; the
demodulator c
1
-
generates a soft decision

X
01
X
01
; c
1
-t
~
Y: The
SIR is dened as the ratio of the desired signal power to
the sum of the powers due to noise and multiple access
interference at the output of the lter c
1
-
[12]. For the case of
a single cell studied in Ref. [1], the SIR b
1
is
b
1
=
P
1
(c
1
-t
s
1
-
)
2
s
2
(c
1
-t
s
1
-
)
2
-

K
i=2
P
i
[(c
1
-t
u
i
-
)
2
-(c
1
-t
n
i
-
)
2
]
Our aim is to introduce in the expression of SIR the effect of
other-cell interference
~
I =

1
l=1

K
k=1
(X
lk
u
lk
--
-Y
lk
n
lk
--
)
~
I is not white like the Gaussian noise ~ w and hence
~
I and ~ w
are not trivially additive. In the following, we will consider
the limiting regime where the number of users per cell is
large, i.e. K
l
!1 for l = 01. N is typically large which
yields a xed number of users per degree of freedom Fig. 2. Other-cell interference for the downlink.
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1118
(equivalently, per unit bandwidth) a
l
=
K
l
N
: We assume that
as the system is scaled up, the empirical distribution of the
received powers converges to a xed distribution F
l
(P) [1].
The empirical distribution of the delays of the different
users, relative to the observation window is also assumed to
converge to a xed distribution G(t); where the delay d
relative to the reference user is given by d = tN: In a typical
asynchronous system, we can assume that the arrivals are
equally probable to be anywhere in [0; N) and hence
uniform delay distribution is a good model for the delays
relative to a particular user.
3.1.1. The matched lter receiver
The matched lter receiver demodulates each signal by a
single-user detector which consists of a matched lter
followed by a threshold detector [2]. The demodulator for
user 1 is then c
1
-
= s
01
--
:
Theorem 1. Let b
(N)
1;MF
be the (random) SIR of the
conventional matched lter receiver for user 1 for a
spreading length N: As N; K
l
!1 and
K
l
N
!a
l
; b
(N)
1;MF
converges in probability to
b
MF
1
<
P
1
s
2
-

1
l=0
a
l
E
P
l
[P]
where the expectation is taken with respect to the limiting
empirical distribution F
l
of the received powers.
Proof. As the demodulator is s
01
--
; the expression for SIR
b
1;MF
is given by replacing c
1
-
by s
01
--
in the Eq. (14)
where G = s
2
Id
N
is the covariance matrix of ~ w: By the weak
law of large numbers, ( s
01
--t
s
01
--
) converges to 1. As of the
other terms, we consider the cross correlation z
lk
between
the signature sequences of target user and user (l; k); l =
01 and k = 1K
l
given by
z
(u)
lk
;
1

N
_

d
lk
n=1
V
01n
V
lk(N2d
lk
-n)
=

N
_
(s
1
-t
u
lk
--
)
z
(v)
lk
;
1

N
_

N
n=d
lk
-1
V
01n
V
lk(n2d
lk
)
=

N
_
(s
1
-t
n
lk
--
)
It can be shown, using the codes properties only, that Ref.
[1], for the target cell
1
K

K
i=2
P
i
[(z
(u)
0i
)
2
-(z
(n)
0i
)
2
] !
_1
0
PdF
0
(P)
This result is easily extended to include other cells, as there is
no difference between the properties of the other-cell codes
and those of the same cell, in term of correlation with s
01
--

K
l
k=1
P
lk
[( s
01
--t
u
lk
--
)
2
-( s
01
--t
n
lk
--
)
2
]
=
1
N

K
l
k=1
P
lk
[(z
(u)
lk
)
2
-(z
(n)
lk
)
2
] !a
l
_1
0
PdF
l
(P) (15)
Hence
b
MF
1
!
P
01
s
2
-

1
l=0
a
l
_1
0
PdF
l
(P)
A
Thus, for large N; the performance of the conventional
receiver is approximately
b
MF
1
=
P
01
s
2
-
1
N

1
l=1

K
l
k=1
P
lk
-
1
N

K
i=2
P
0i
(16)
We now study the term

1
l=1

K
l
k=1
P
lk
in Eq. (16). For the
target cell, all users are power controlled and their powers P
i
are well known; there is, however, no power control in target
cell 0 on other-cell signals. One can nevertheless model the
total interference they produce as a Gaussian random
variable P
I
; as stated in Section 2.1. The
quantity

1
l=1

K
l
k=1
P
lk
is then the instant value of P
I
;
upper bounded by I
u
in Eq. (3). The SIR expression is
then upper bounded by
b
MF
1
$ g
MF
1
=
P
01
s
2
-
I
u
N
-
1
N

K
0
i=2
P
0i
(17)
3.1.2. The MMSE receiver
Eq. (13) can be written as
~
Y = X
01
s
01
--
-
~
Z where
~
Z =

K
0
i=2
(X
0i
u
0i
--
-Y
0i
n
0i
--
) - ~ w -
~
I
is the total interference, including that from other cells. If
~
Z were white, the optimal demodulator would have been
c
1
-
= s
01
--
; the matched lter demodulator, and the value of
X
01
would have become the projection of
~
Y onto s
01
--
:
b
1
=
P
01
(c
1
-t
s
01
--
)
2
c
1
-t
Gc
1
-
-

1
l=1

K
l
k=1
P
lk
[(c
1
-
t
u
lk
--
)
2
-(c
1
-
t
n
lk
--
)
2
] -

K
0
i=2
P
0i
[(c
1
-
t
u
0i
--
)
2
-(c
1
-
t
n
0i
--
)
2
]
(14)
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1119
However,
~
Z is not white and has a well-known component

K
0
i=2
(X
0i
u
0i
--
-Y
0i
n
0i
--
): The MMSE receiver exploits the
structure of this component to maximize the SIR in Eq.
(14).
In general, this can be done by whitening the interference
~
Z followed by a projection (see Refs. [1,12] for the single
cell system case). In our case, however, whitening
~
Z is very
difcult, because it is basically impossible to know all codes
used in the overall system and, even if we limit ourselves to
neighboring cells only, the implementation will be extre-
mely complex. On the other hand, exploiting the structure of
the term

K
0
i=2
(X
0i
u
0i
--
-Y
0i
n
0i
--
) is still possible, at the
expense of some approximations to determine the optimal
demodulator c
1
-
: It is worthwhile to note that the MMSE
receiver is, for other-cell signals, equivalent to a matched
lter receiver because the demodulator c
1
-
is independent
from their spreading-scrambling codes s
lk
--
: The code c
1
-
is
in fact a linear combination of the codes s
0i
--
; i = 1; ; K
0
;
and has the same characteristics as them [12]. Hence, the
approximation of Section 3.1.1 is still valid for users of
other cells, and for a large system where N; K
l
!1; l =
1; ; 1; the lower bound of the SIR b
MMSE
1;N
is then given by
Eq. (18)
g
MMSE
1;N
.
P
01
(c
1
-t
s
01
--
)
2
c
1
-t
Gc
1
-
-
I
u
N
-

K
0
i=2
P
0i
[(c
1
-t
u
0i
--
)
2
-(c
1
-t
n
0i
--
)
2
]
(18)
Introducing the matrix

G = (s
2
-
I
u
N
)Id
N
and consider-
ing that c
1
-t
c
1
-
. 1 for large N; the lower bound of SIR in Eq.
(18) can be approximated by the Eq. (19)
g
MMSE
1;N
<
P
01
(c
1
-t
s
01
--
)
2
c
1
-t
Gc
1
-
-

K
0
i=2
P
0i
[(c
1
-t
u
0i
--
)
2
-(c
1
-t
n
0i
--
)
2
]
(19)
Let A =

K
0
i=2
P
0i
[ u
0i
--
u
0i
--t
] - n
0i
--
n
0i
--t
] -

G; then the
denominator of Eq. (19) can be written as c
1
-t
Ac
1
-
and the
MMSE solution c
1
-
satises Ref. [12]
Ac
1
-
= P
01
(1 2 c
1
-
t
s
01
--
) s
01
--
If now the symbol of interest is estimated by observing over
T symbol intervals, T $ 1 being an odd integer, we have:
Theorem 2. In an asynchronous system, if N !1 with K
l
/
N !a
l
, l = 0 1, and the observation window is of T $ 1
(T being an odd integer) symmetric about the symbol to be
demodulated, the SIR attained by user 1 b
1,N
MMSE
converges in
probability to a deterministic constant b
1
MMSE
, where
b
MMSE
1
=
_
T-1
2
T21
2
w(x)dx
and the function w(x) $ 0 in [0,T] is the solution of the
Eq. (20)
w(x) =
P
01
s
2
-
I
u
N
-aE
P
E
t
[In(P; P
01
;
_
C(x;t)
w(z)dz)]
(20)
where E
t
denote the expectation with respect to the power
distribution F(P) and the delay distribution G(t), respect-
ively
In(P; P
01
; b) =
PP
01
P
01
-Pb
is the effective interference function.
The region of integration C(x,t) is given by (i = 1T 2 1)
C(x; t) =
[0; t]; x [ [0; t]
[t -i 21; t -i]; x [ [t -i 21; t -i];
[t -T 21; T]; x [ [t -T 21; T]
_
_
_
_
_
_
_
The solution to w(x) exists and is unique in a class of
functions w(x) $ 0:
Eq. (20) can be solved numerically for a given delay
distribution G(t): However, the complexity of the
expression makes it difcult to draw interesting analytical
insights. The following theorem (derived by means of a
weak symmetry assumption on the delay distribution,
satised in most cases of interest) gives a lower bound on
the SIR achieved.
Theorem 3. In an asynchronous system, if the relative delay
distribution satises the condition G(t) = 1 2 G(1 2 t)
and the observation window is an odd integer T, symmetric
about the symbol to be demodulated, the asymptotic SIR
attained is lower bounded by g
1
p
, which is the unique
solution of the xed-point Eq. (21).
The proofs of the above theorems are similar to the
single-cell case found in Ref. [1] and are based on the
limiting eigenvalue distribution of large matrices whose
elements are random variables, and on the properties of w(x)
deduced from Eq. (20). For large systems, the SIR bound
has the expression (22)
g
p
1
=
P
01
s
2
-
I
u
N
-
a
T
E
P
E
t
[In(tP; P
01
; g
p
1
) -(T 21)In(P; P
01
; g
p
1
) -In((1 2t)P; P
01
; g
p
1
)]
(21)
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1120
where the quantity
1
T
[In(t
i
P
i
; P
01
; g
p
1
) -(T 21)
In(P
i
; P
01
; g
p
1
) -In((1 2t
i
)P
i
; P
01
; g
p
1
)] can be called the
effective interference of user i on user 1. It depends on the
received power of user i; the received power of user 1 and
the SIR. The other interferers are decoupled and their
inuence is only through the SIR.
Comparing Eqs. (17) and (22) for conventional and
MMSE receivers, respectively, we note that the SIR is better
in the case of MMSE receiver due to the fact that the
effective interference is less than P
i
: However, as expected,
the other-cell interference
I
u
N
remains the same.
For the special case of uniform relative delay distribution
and equal received powers, the xed-point equation (21) for
the lower bound g
p
1
gives, by calculating the mean over t;
the expression (23)
3.1.3. The decorrelator
Another important linear receiver is the decorrelator, its
an overall lter given by (S
t
S)
21
S
t
which is applied on the
received signal; S being the N (2K 21) matrix whose
columns are the signature sequences ( s
01
--
; u
02
--
; ;
u
0K
0
---
; n
02
--
; ; n
0K
0
---
): This is then a matched lter (represented
by the matrix S
t
) used to extract sufcient statistics for the
signals, and followed by a decorrelating lter (S
t
S)
21
aiming to minimize the bit error probability (here the matrix
inversion is the generalized inversion, so if the real inverse
does not exist, one can use the pseudo-inverse in its place).
In the absence of external noise, this would give perfect
estimates of the information symbols and hence is a zero-
forcing linear lter [1]. But in reality, there is a noise given
by (S
t
S)
21
S
t
(
~
I - ~ w): This causes an extra power interference
given by the matrix
S = (S
t
S)
21
s
2
-
1
N

1
l=1

K
l
k=1
P
lk
_ _
where the term
1
N

1
l=1

K
l
k=1
P
lk
is the power of the other-
cell interference on the output of the matched lter, upper
bounded as before by
I
u
N
: The SIR attained by user 1 is then
given by b
1
=
P
1
S
11
: Again, solutions with multiple symbol
observation windows can be considered, giving, in the same
conditions as in Section 3.1.2, the following asymptotic
bound
g
p
1;dec
=
P
01
s
2
-
I
u
N
1 2
T -1
T
_ _
K
0
N
_ _
;
K
0
N
,
T
T -1
0;
K
0
N
$
T
T -1
_
_
_
_
_
_
_
_
_
_
_
(24)
The proof is based on the fact that the decorrelator can be
obtained from the MMSE receiver when the Signal-to-Noise
Ratio (P
k
=s
2
) !1 for each user [1], and then Eq. (24) is
derived from Eq. (21), by doing P
k
!1:
The decorrelator cannot then be used when the number of
users is greater than
T
T-1
N:
3.1.4. Powers at the base station
We now calculate the powers at the base station,
supposing that the SIR is equal to its asymptotic bound in
Eqs. (17), (22) and (24). In fact, we show by numerical
applications in Section 5.1 that the SIR converges to this
bound when the number of users, i.e. the load, increases, and
this for both Gaussian and log-normal assumptions of the
other-cell interference.
For J classes of users, with a SIR b
j
and a power P
j
are
required for each of the L
j
calls of class j; the SIR equations
give the power expressions for class i calls
P
mf
(i) =
b
i
(s
2
-I
u
=N)
1 2
1
N

J
j=1
L
j
e(b
j
)
(25)
where the quantities
e(b
j
) =
b
j
; for MF
T 21
T
b
j
1-b
j
-
2
T
12
ln(1-b
j
)
b
j
_ _
; for MMSE
T -1
T
; for dec:
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
can be considered as the effective bandwidth for class j users
as in the single-cell case [1].
g
p
1
<
P
01
s
2
-
I
u
N
-
1
N

K
0
k=2
2P
01
Tg
p
1
1 2
P
01
P
0k
g
p
1
ln 1 -
P
0k
g
p
1
P
01
_ _
_ _
-
T 21
T
In(P
0k
; P
01
; g
p
1
_ _ (23)
g
p
1
<
P
01
s
2
-
I
u
N
-
1
NT

K
i=2
[In(t
i
P
i
; P
01
; g
p
1
) -(T 21)In(P
i
; P
01
; g
p
1
) -In((1 2t
i
)P
i
; P
01
; g
p
1
)]
(22)
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1121
3.2. SIR in the downlink
In the downlink, the transmission is governed by the base
station: the system is then synchronous, and the signal
emitted by base station 0 is
~
Y =

K
0
i=1
X
0i
s
0i
--
: For user i; this
signal arrives with the path loss q
0i
; common to all users
signals. To this signal are added the Gaussian thermal noise
~
W and the other-cell interference that arrives asynchro-
nously from other base stations. As the design of mobile
stations is limited by size and cost, multiuser detection is
very difcult and orthogonal codes eliminate the intra-cell
interference ([15] page 18). The latter is then caused entirely
by the multipath propagation channel which reduces the
orthogonality by introducing the orthogonality factor e [
[0; 1]: The interference at mobile j is then equivalent to that
generated at the input of the matched lter by users having
powers of (1 2e)P
(j)
i
; P
(j)
i
being the power received at
mobile j for user i of the same cell. For a surrounding cell l;
the power relative to user k arrives to the mobile station
under study with power P
(j)
l;k
: The equation of the SIR is then
given from Eq. (16) by
b
j
=
P
(j)
j
s
2
-
1
N

1
l=1

K
l
k=1
P
(j)
l;k
-
1 2e
N

K
i=1
P
(j)
i
The quantity

1
l=1

K
l
k=1
P
(j)
l;k
is again the other-cell inter-
ference. The SIR for user j is then lower bounded by
b
j
$ g
d
j
=
P
(B)
j
s
2
q
0j
-
I
d
j
N
-
1 2e
N

K
0
i=1
P
(B)
i
(26)
with P
(B)
i
the signal emitted by the base station for user i:
4. Connection admission control
4.1. Case of the uplink
4.1.1. Effect of mobility and coverage
The power that a mobile station can emit is not innite,
but limited by a maximal value P
i;e;max
for class i: Using the
power expression (25), and taking into account the users
mobility, affecting their path losses, we obtain the CAC
equation

J
j=1
L
j
e(b
j
) # min
1#i#J
min
1#l#L
i
N 2
b
i
(Ns
2
-I
u
)q
li
P
i;e;max
_ _ _ _
(27)
where the L
i
users of class i; i = 1J have a SIR
requirement of b
i
; q
li
being the path loss of user l of class i:
Let us discuss this equation. Let
f (i; l) = N 2
b
i
(Ns
2
-I
u
)q
l
P
i;e;max
Each element of the set
S = {f (i; l); i = 1; ; J; l = 1; ; L
i
}
corresponds to a user in the system. The condition

J
j=1
L
j
e(b
j
) # f (i; l)
insures, if veried, that the call of user (i; l) will not be
dropped if the new call is accepted. Condition (27)
corresponds then to a zero-dropping CAC, but with the
price of a possibly large blocking rate for new calls. This
policy is, in general, preferable because premature termin-
ation of connected calls is more annoying than rejection of a
new call request.
4.1.2. Effect of the UTRANs wired capacity
Very often, UTRANs wired capacity has been neglected
in CAC decisions, as air resources seemed so scarce
compared to wired ones. However, with rst real exper-
imentation of UMTS systems, it appears that the wired
capacity, in todays systems, is rather low. Indeed, air and
wired resources are complementary; the former is pre-
ponderant in large cells, used in rural environments where
we have a small load factor and the system is coverage-
limited, while the wired capacity is more signicant in
urban environments where small cells with high load are
used and the system is capacity-limited. To take this
resource into account, we use the noise rise
l = I
total
=s
2
= g(D)
to determine the relationship between the total received
power I
total
; and the global throughput D; g being a
continuous, strictly increasing function determined, by
simulations ([6] page 175). We then obtain the second
CAC condition

J
j=1
L
j
b
j
-
W
u
Ns
2
-I
u
e(b
j
)
_ _
#
NW
u
Ns
2
-I
u
(28)
where W
u
= l
max
s
2
2s
2
2I
u
:
4.2. Case of the downlink
The downlink is limited by the maximum power W
B
that
could be emitted by the base station

K
0
i=1
P
(B)
i
# W
B
Using the power expression and the other-cell interference,
we obtain the downlink CAC condition

K
i=1
b
i
(1 2e)W
B
N
-s
2
q
0i
-
I
d
i
N
_ _
# W
B
(29)
This equation introduces in the CACdecision both the effects
of mobility and coverage (included in the path losses q
0i
and
the other-cell interference I
d
i
) and capacity (in terms of W
B
).
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1122
The inuence of both air resources and wired ones on the
CAC decision in a one-cell system is illustrated by
numerical applications in Ref. [18].
4.3. CAC in the presence of elastic ows
Let us see how to apply the obtained CAC Eqs. (27)
(29) on the different channels of a UMTS system. In the case
of real-time trafc only, each call is assigned a new code and
transported over DCHs. In this case, ows keep a constant
bit rate and hence a constant SIR gure over their duration.
Elastic ows, however, may be assigned variable bit rates
and hence SIRs, and may be sent over shared channels. We
make use of this exibility so as to maximize the number of
accepted calls as follows.
Real-time ows are carried on DCHs whereas elastic
trafc over CPCHs in the uplink and Shared Channels
(DSCHs) in the downlink. They nevertheless interfere with
each other, inuencing each others SIR. We suggest to give
real-time trafc, typically voice, its required SIR level, as
agreed upon in the already established Service Level
Agreement (SLA). In the presence of a large amount of
resources, the same thing is done to elastic ows, typically
data, i.e. they are granted their SIR requirement. When
resources get scarce and in the presence of a large number of
ows of both types, the bit rates of data ows, or a subset of
them, are decreased, and hence their SIR, to a minimum as
specied in the SLA, granting access to higher priority, new
voice calls. When all data ows are set to their minimum,
and the remaining resources cannot, however, accommodate
the new incoming voice call, the latter is simply rejected. On
the other hand, if, because of the users mobility or the
termination of ongoing calls, resources are freed, we
increase the transmission of some squeezed elastic ows
to maximize the resource utilization.
Fig. 3. Realized SIR for the matched lter and MMSE.
Fig. 4. Effect of the other-cell interference on the sector capacity for the
three kinds of receivers.
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1123
5. Numerical applications
5.1. Convergence of the SIR
We rst study the convergence of the SIR towards the
asymptotic lower bound for the matched lter and
MMSE, given by Eqs. (17) and (22).
Fig. 3(a) and (c) show the realized SIR for randomly
generated spreading codes for the matched lter and the
MMSE receivers, respectively, for the case of Gaussian-
distributed other-cell interference. We observe that the SIR
realizations (dots in the gure) converge to the asymptotic
bound (continuous line) as the number of users by degree of
freedom increases.
Fig. 3(b) and (d) shows the same performance measures
for the case when the other-cell interference is considered as
log-normal. The same convergence towards the asymptotic
bound is observed. The assumption that the SIR is equal to
its asymptotic bound in Eqs. (17), (22) and (24) is then
justied.
To illustrate the effect of the other-cell interference on
the capacity, we now calculate the user capacity from the
power expression (25)
C(b) =
1
e(b)
N 2
(Ns
2
-I
u
)b
P
_ _
This capacity is shown in Fig. 4 for different kinds of
receivers. We observe that the system capacity is lower
when the other-cell interference is taken into account and
this for all receivers. This is expected as a part of the air
resources is used by the inter-cell interference. However, in
all cases, the MMSE receiver has a better performance than
the matched lter receiver and the decorrelator. This latter
outperforms the matched lter for large values of the SIR.
5.2. Performance of our CAC algorithm
To show the added value of our CAC algorithm, we
compare its performance with respect to the most known
Fig. 6. Blocking probabilities for the two classes. Fig. 8. Cell throughput.
Fig. 7. Number of accepted ows.
Fig. 5. Comparison between our CAC and the classical CAC.
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1124
classical ones, where a new call is accepted if the total
interference is below a threshold [13]
I
0
N
0
,
1
h
Fig. 5 draws the blocking and dropping rates (outage
probability) for both methods (one class of users and the
ICAC threshold is determined empirically to maximize its
efciency). One can see that the dropping rate is better in
our CAC, at the expense of a larger blocking rate. This is
preferable, because premature termination of connected
calls is usually less desirable than rejection of a new call
request. Indeed, in our algorithm, interference is the
essential parameter, expressed in the SIR requirements of
the different calls. Yet, this is not the sole parameter;
mobility, coverage and UTRANs wired capacity are also
taken into account and do make a difference in the CAC
decision.
5.3. Performance of our CAC in the presence of elastic
trafc
In order to investigate the performance of our CAC
algorithm, we built a UMTS activity simulator. Call arrivals
are supposed Poisson with two SIR requirements: 5 dB for
voice calls and 10 dB for data ones. Each call has an
exponentially distributed life time with mean values
1=m
n
= 3 min and 1=m
d
= 9 min. Note that r
n
and r
d
; the
workload of voice and data trafc, are taken as equal. Users
move at a randomly generated speed ranging between 0 and
100 km/h. These values are not necessarily typical but
merely illustrative. The new call is directly accepted if CAC
equations are veried. Otherwise, if the new call is a real-
time one, we try to nd out if, by decreasing the
transmission of ongoing elastic calls (SIR of data calls
squeezed to 5 dB), we can liberate resources to accept it.
Fig. 6 shows the blocking rates for the two classes of
calls. One can realize that a real-time call has more chance
to be accepted than an elastic one in all cases, and the failing
probabilities increase with the number of users. However,
the squeezing strategy decreases the blocking rates for both
classes of trafc.
Fig. 7 shows the number of sessions that are accepted
in the system for the case when the elastic ows keep a
constant rate versus the case where they are squeezed,
i.e. their rates and hence SIR are degraded to a minimal
value. As it can be seen from the gure, our squeezing
strategy makes us accept more calls. However, Fig. (8)
shows that the overall throughput decreases.
We now turn to the impact of our CAC algorithm on
elastic ows. We draw in Fig. 9 the percentage of class 2
users squeezed to the minimal SIR. One can see that elastic
calls can emit with their demanded SIR if the system is not
heavily loaded, but when the number of users in the system
increases, some class 2 users must decrease their trans-
mission to the minimal SIR.
6. Conclusion
In this article, we rst studied the impact of other-cell
interference on the capacity of a DS-CDMA cellular system.
For the uplink, we studied linear receivers, especially the
matched lter, the MMSE receiver and the decorrelator. We
developed an analytical model introducing the inter-cell
interference in the SIR expressions. We showed that the
effective bandwidth and effective interference are the same
that in the single-cell case; the overall capacity, however,
decreases. For the downlink, we focused on the conven-
tional receiver because multiuser detection is not usually
used in the mobile station; we obtained the other-cell
interference and the SIR.
Using these expressions, we developed a CAC algorithm
for UMTS, based on the notion of effective bandwidth. In
addition to the classical power constraint, these equations
take into consideration the effect of mobility and coverage,
preponderant in large cells used in rural environments, as
well as the wired uplink capacity, and the Node B maximal
transmission power, which are important in small, highly-
loaded cells in urban environments. Compared to classical
ICAC algorithms, our CAC algorithm achieves better
performance, by decreasing the dropping rate, and is simple
to implement.
To take advantage of the existence of two types of
channels, namely dedicated and shared ones for the
transport of real-time and non-real-time trafc, respectively,
we developed a strategy that insures the priority of real-time
calls on elastic ones and maximizes the resource utilization
by squeezing elastic ows in the available resources.
Another major alternative is the measurement-based
approach that we explore in our work in Ref. [19].
Fig. 9. Percentage of squeezed elastic calls.
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1125
A comparison between those two approaches is part of our
future work perspectives.
References
[1] Kiran, D.N.C. Tse, Effective interference and effective bandwidth of
linear multiuser receivers in asynchronous CDMA systems, IEEE
Transactions on Information Theory 46 (4) (2000).
[2] P.B. Rapajic, B.S. Vucetic, Adaptive receiver structures for
asynchronous CDMA systems, IEEE Journal on Selected Areas in
Telecommunications 12 (4) (1994).
[3] K.S. Gilhousen, I.M. Jacobs, R. Padovani, A. Viterbi, L.A. Weaver,
C.E. Weatley, On the capacity of a cellular CDMA system, IEEE
Transactions on Vehicular Technolog 40 (2) (1991).
[4] G. Karmani, K.N. Sivarajan, Capacity evaluation for CDMA cellular
systems, IEEE Infocom, Anchorage, Alaska, 2001.
[5] D. Staehle, K. Leibnitz, K. Heck, B. Schroder, A. Weller, P. Tran-Gia,
Approximating the Othercell Interference in Inhomogeneous UMTS
Networks, IEEE VTC, 2002.
[6] H. Holma, A. Toskala, WCDMA for UMTS, Radio Access for Third
Generation Mobile Communications, Wiley, England, 2000.
[7] Y. Ma, J.J. Han, K.S. Trivedi, Call admission control for reducing
dropped calls in CDMA cellular systems, Computer Communication,
Elsevier, 2001.
[8] Y. Ishikawa, N. Umeda, Capacity design and performance of call
admission control in cellular CDMA systems, IEEE Journal on
Selected Areas in Communications 15 (8) (1997).
[9] N. Dimitriou, G. Skas, R. Tafazolli, Call admission policies for
UMTS, Vehicular Technology Conference Proceedings, VTC 2000-
Spring Tokyo. 2000 IEEE 51st, vol. 2, 2000.
[10] N. Dimitriou, G. Skas, R. Tafazolli, Quality of service for
multimedia CDMA, IEEE Communications Magazine 38 (7) (2000).
[11] Z. Liu, M. Elzarki, SIR-based call admission control for DS-CDMA
cellular systems, IEEE Journal on Selected Areas in Telecommunica-
tions May (1994).
[12] U. Madhow, M.L. Honig, MMSE inference suppression for direct-
sequence spread sprectrum CDMA, IEEE Transactions on Communi-
cations 42 (12) (1994).
[13] A.J. Viterbi, CDMA principles of spread spectrum communication,
Wireless Communications Series, Addison-Wesley, 1997.
[14] C. Mihailescu, X. Lagrange, Ph. Godlefski, Radio resource manage-
ment for packet transmission in UMTS WCDMA system,Vehicular
Technology Conference, 1999. VTC 1999-Fall, IEEE VTS 50th, vol.
1, 1999.
[15] J. Blogh, L. Hanzo, Third Generation Systems and Intelligent
Wireless Networking, Wiley, England, 2002.
[16] J.S. Lee, L.E. Miller, CDMA systems Engineering Handbook, Mobile
Communication Series, Artech House Publishers, BostonLondon,
1998.
[17] S.-E. Elayoubi, T. Chahed, G. Hebuterne, On the capacity of multi-
cell UMTS, in: Proceedings of IEEE Globecom, San Fransisco
(2003).
[18] S.-E. Elayoubi, T. Chahed, Measurement-Based Admission Control in
UMTS: Single-Cell Case, in: Proceedings of International Teletrafc
Congress ITC18, Berlin, 2003.
[19] S.-E. Elayoubi, T. Chahed, G. Hebuterne, Measurement-Based
Admission Control in UMTS: Multiple-Cell Case, in: Proceedings
of IEEE PIMRC, Beijing, 2003.
S.-E. Elayoubi et al. / Computer Communications 27 (2004) 11151126 1126

Das könnte Ihnen auch gefallen