Impact of Intelligent Access Selection Algorithms in Cooperative Wireless Networks

aroocarmineAI and Robotics

Oct 29, 2013 (3 years and 7 months ago)

118 views

VI INTERNATIONAL TELECOMMUNICATIONS SYMPOSIUM (ITS2006),SEPTEMBER 3-6,2006,FORTALEZA-CE,BRAZIL
Impact of Intelligent Access Selection Algorithms
in Cooperative Wireless Networks
F.R.P.Cavalcanti,V.A.de Sousa Jr.,R.A.de O.Neto and F.de S.Chaves
Abstract—Future generation of wireless networks will solicit
the cooperation of heterogeneous Radio Access Networks (RANs)
in order to provide multimedia traffic to different classes of
users with varying Quality of Service (QoS) requirements over
regions and time zones.The use of such RANs as subsystems of
a larger wireless network,i.e,a multi-access network,claims for
an Access Selection (AS) scheme to coordinate the load sharing
so as to meet the optimum combined network performance.
Depending on the individual subsystem capacities,the traffic
load conditions and the subsystem channel quality,an intelligent
AS algorithm can provide capacities beyond the sum of the
subsystems capacities.This paper addresses the issue of the AS
in the context of multi-access networks.We intend to provide
an analytical formulation which indicates the suitable/unsuitable
scenarios for employment of an intelligent AS strategy.The
usability of the proposed qualitative studies are demonstrated
by means of simulations.
Index Terms—Cooperative Beyond 3G Systems,Multi-Access
Scenarios,Access Selection.
I.I
NTRODUCTION
Overall wireless connectivity is an essential aspect of
upcoming wireless systems.Mobile users wish to have access
to voice and data services everywhere (homes,vehicles,
workplaces or public areas).There is a large variety of
radio-enabled equipments at the disposal of the users -
smartphones,PDAs,laptops,and desktop computers - as well
as the choice of RANs (3G cellular networks,wireless local
area networks,and satellite systems).One important question
that arises for network operators is how different RANs may
be compared in terms of coverage,capacity and performance,
and how these differences could be used for increasing the
user perceived QoS.Then,the basic motivation for studying
this problemis to determine scenarios where an intelligent AS
approach yields in performance gains.
The definition of multi-access scenarios is a task that
consists of situating the social context of the possible
applications,as well as providing technical aspects of the
models and parameters that may have a meaningful impact
over the proposed multi-access solutions.
Social aspects of the interworking between two or more
RANs may be essentially characterized by how the subsystems
are deployed.As an example,a specific RAN can be
distributed in public areas such as restaurants,shopping centers
and airports,enabling the users to select the desired network
and seamlessly maintaining their connectivity while passing
through the different RANs.
The authors are with the GTEL-UFC:Wireless Telecom Research Group,
Teleinformatics Engineering Department,Federal University of Ceara,Brazil.
URL:www.gtel.ufc.br.E-mail:{rodrigo,vicente,neto,fabiano}@gtel.ufc.br.
Technical aspects include identification of parameters that
influence the multi-access networks performance.For a useful
qualitative result on multi-access scenarios,we have to identify
general parameters and rules that cover a large set of
scenarios.This is addressed in this paper.Several research
projects have been developing a similar investigation on
the multi-access subject [1]–[8].Besides papers and public
deliverables produced in these projects,there are two academic
works directly related to our investigation [9]–[11].Results
show that there is no significant cost in adding hot spot
like subsystems (e.g.WLAN) to legacy cellular (e.g.UMTS).
In such a hybrid network,a dual-mode mobile equipment
will access the network via a hot spot like subsystem when
inside the hot spot,or cellular subsystem otherwise.The
financial advantages of the multi-access networks have been
investigated in [9].Results show that there is no significant
cost in adding hot spot like subsystems (e.g.WLAN) to
legacy cellular (e.g.UMTS).In such a hybrid network,a
dual-mode mobile equipment will access the network via a
hot spot like subsystem when inside the hot spot,or cellular
subsystem otherwise.This AS strategy is also investigated in
[11],[12].For different system-level aspects,the gain of the
“coverage threshold algorithm” criterion is investigated and the
possible gains when taking load traffic into account during the
AS process.However,there is no detailed studies about the
attractiveness of intelligent AS and its analytical formulation.
This paper presents studies of attractiveness of AS in
the multi-access wireless networks.Initially,we intend
to provide an analytical formulation which indicates the
suitable/unsuitable scenarios for employment of an intelligent
AS strategy.After that,the validity of these “rules” is
illustrated in a number of more detailed case studies.
II.A
CCESS
S
ELECTION
P
ROBLEM
In multi-access heterogeneous networks the concept of
Common Radio Resource Management (CRRM) is essential
for assuring an efficient handling of the radio resources.It
extends the traditional RRM techniques and has a much
broader scope,which certainly makes a difference in terms
of resource optimization and performance gains.Nevertheless,
one is not intended to replace the other,but rather work in a
complementary fashion.
Some of the existing RRM techniques may be extended to
the CRRMcontext.For instance,the traditional call admission
procedure determines whether a user may be admitted into the
system.In a multi-access scenario,however,the call admission
control (here called AS algorithm) is additionally responsible
for deciding which is the best suited RAN to accommodate
the incoming connections.
SBrT © 1020
VI INTERNATIONAL TELECOMMUNICATIONS SYMPOSIUM (ITS2006),SEPTEMBER 3-6,2006,FORTALEZA-CE,BRAZIL
The upcoming sections will look at the multi-access
scenarios in order to define a general rule to determine a
suitable combination of RANs where an intelligent AS strategy
provides performance gains.
III.A
CCESS
S
ELECTION
A
TTRACTIVENESS
E
VALUATION
We study the attractiveness of the employment of an
intelligent AS algorithm considering the Coverage Threshold
Algorithm (CTA) as a reference.Such algorithm does not use
any kind of intelligence in the process of allocation.In such
a procedure,a dual-mode mobile equipment will access the
network via a hot spot like subsystem when inside the hot
spot,and use cellular subsystem otherwise.
A.Multi-Access System Model
We use a system model which represents the access of the
shared channel by the link utilization concept.This model
is originally presented in [12] and described below.The
multi-access environment is illustrated in Fig.1.
RAN 1
RAN 2
HS
Fig.1:Multi-Access Environment.
We consider a single-cell including a hot spot (HS) and two
generic RANs.Our main performance indicator is the CSE
(Circuit Switched Equivalent) bit rate which is determined
based on a queue theory formulation.As defined in [13],
user perceived data quality could be represented by CSE bit
rate.Roughly,the CSE bit rate measures the rate at which
transmission buffers of users are emptied and hence,includes
effects of queuing.
Firstly,we define the desired link utilization factor as the
ratio of the traffic generated per user and the mapped radio
link bit rate,
ρ
Des
i
=
t
i
R
i
(1)
For high loads,the remaining capacity is shared by users
with worst link qualities following a maximumrate scheduling
policy.Then,the highest possible link utilization (ρ
Res
) for
these users is evaluated as:
N

i=1
min

ρ
Des
i

Res

= 1 (2)
where p users are connected with their correspondent ρ
Des
and
(N−p) users with ρ
Res
matching the RAN capacity limitation.
According to equation (2),the effective link utilization factor
of the i
th
user can be written as
ρ
i
= min

ρ
Des
i

Res

(3)
Then,the CSE bit rate of the i
th
user is calculated
as follows:
CSE
i
= R
i
·


1 −

j
=i
ρ
j


(4)
Prior to exposing our studies,we present some general
definitions which specify the above formulations for the
multi-access model in focus:

Set of users outside (Ω
Out
) and inside (Ω
In
) the HS;

Set of users connected to the RAN
j
(Ω
RAN
j
);

Capacity limitation model:

i∈Ω
RAN
j
ρ
i
￿ 1,(j = 1,2).
One can note that Ω
RAN1
= Ω
Out
and Ω
RAN2
=

In
considering the CTA.Next section presents the general
formulation of scenario attractiveness.
B.Theoretical Analysis of the Scenario Attractiveness
The theoretical study of scenario attractiveness starts
determining a criterion that indicates if an intelligent AS
algorithm outperforms the CTA.Equation (5) shows a simple
way to measure the free “resources” considering that all users
inside the HS area are connected to the RAN 2,i.e.,the CTA.


1 −

j∈Ω
Out
ρ
j




1 −

i∈Ω
In
ρ
i


> 0


i∈Ω
In
ρ
i
>

j∈Ω
Out
ρ
j
(5)
Considering the following approximation:

1￿k￿n
ρ
k

n · E[ρ],where E[·] is the expectation operator,we can
rewrite (5) as:
n
In
.E[ρ
In
] > n
Out
.E[ρ
Out
] ⇒
n
In
n
Out
>
E[ρ
Out
]
E[ρ
In
]
(6)
Here,ρ
Out
and ρ
In
are random variables which represent
the link utilization factors of users connected to the RAN 1
(outside HS) and RAN 2 (inside HS),respectively.
One can note that the formulation presented in (5) is a
necessary condition but not sufficient to an AS algorithmbased
on load balancing to provide performance gain over the simple
CTA.Therefore,it is possible that,in some scenarios,the
criterion affirms a low attractiveness,but the utilization of an
algorithm more intelligent than the load balancing one results
in good performance.However,if the criterion indicates high
attractiveness,a more intelligent algorithm will also provide
gain.
Now,the mathematical model of the RANs is stated.
Each RAN is characterized by their radio link quality (SNR
definition) and link adaptation (specific link capacity) models.
SBrT © 1021
VI INTERNATIONAL TELECOMMUNICATIONS SYMPOSIUM (ITS2006),SEPTEMBER 3-6,2006,FORTALEZA-CE,BRAZIL
The radio link quality is expressed by the SNR
(Signal-to-Noise Ratio).It is modeled as a gaussian distributed
random variable with mean SNR
mean
and standard deviation
σ,resulting from the modeling of path loss and shadowing.
Fig.2 illustrates this SNR model.Herein,we approximate this
gaussian probability density function to a triangular one.This
approach was chosen due to its simplicity for mathematical
manipulation.Then,the input variables related to our radio
link quality are SNR
min
,SNR
max
and SNR
mean
.
Gaussian distributed SNR
Mathematical simplification
SNR
min

∆∆
∆SNR
SNR
mean
SNR
max
Fig.2:PDF of the Radio Link Quality model (f
SNR
).
The triangular probability density function is
mathematically determined by the equation (7):
f
SNR
(s) =









0,if s < SNR
min
4.(s−SNR
min
)
(∆SNR)
2
,if SNR
min
< s ≤ SNR
mean
4.(SNR
max
−s)
(∆SNR)
2
,if SNR
mean
< s ≤ SNR
max
0,if s > SNR
max
(7)
where ∆SNR = SNR
max
−SNR
min
.
The link adaptation is idealized,as seen in Fig.3.Then,
the input variables related to the ideal link adaptation are
SNR
knee
,SNR
sat
,R
min
and R
max
.In order to facilitate the
future notation,we use ∆R = R
max
−R
min
.
SNR
knee
R
max
R
min
SNR
RATE
slope = k
SNR
sat
Fig.3:Ideal Link Adaptation model.
Equation (5) establishes that the AS gains depends on
the relation between the number of users and the resources
required/available of both RANs.Considering users requiring
the same amount of resource (t
i
= t),equation (5) becomes:
n
In
n
Out
>
E[ρ
Out
]
E[ρ
In
]
=
E

1
R
Out

E

1
R
In

(8)
where R
Out
and R
In
are the random variables that represent
the bit rate of users in RAN 1 and RAN 2,respectively.
Now,one can note that the quotient 1/R is a random
variable since,R is a function of a random variable (SNR),
whose probability density function is expressed in (7).From
the stochastic process theory,
E[g(x)] =
+∞

−∞
g(x)f
X
(x)dx (9)
where f
X
(x) is the probability density function of the
random variable X.In our approach,the SNR is the random
variable X and g(x) = 1/R(SNR).
The values of the parameters SNR
sat
,SNR
min
,SNR
max
and SNR
mean
determine different cases regarding the
expression above.Equation (9) shows a case where
SNR
min
< 0,SNR
max
> 0,SNR
knee
= 0 and
SNR
sat
> 0.The other cases can be obtained in an
analogous manner.
Therefore,we can conclude that the degree of attractiveness
depends on three parameters:the proportion of users inside to
outside the hot spot n
In
/n
Out
;the capacity of the systems,
expressed by R
max
and SNR
sat
;and the SNR distribution,
characterized by SNR
min
and SNR
max
.
IV.D
EMONSTRATIVE
P
ERFORMANCE
R
ESULTS AND
G
ENERAL
D
ISCUSSIONS
In this section,we present the performance results aiming
to demonstrate the validity of the proposed attractiveness
formulation.We simulate two AS strategies.The first one is
the CTA.The second strategy is the Load Balancing Algorithm
(LBA),which tries to balance the offered load in both RANs,
assigning a new user to RAN1 or RAN2 so that the normalized
load in both RANs becomes closer to each other [14]:
Used Capacity at RAN
1
Total Capacity of RAN
1
Used Capacity at RAN
2
Total Capacity of RAN
2
(10)
For the presented study,we focus on a best-effort service
over shared channels where users are assumed to generate
an average traffic of 128 kbps,regardless of position.The
multi-access scenario is a composition of relevant parameters
in each RAN.These parameters are the expected offered
load (proportion of users inside to outside HS),the link
quality (SNR distribution),the bit rate capacity (SNR to
rate mapping with link-adaptation if any).The last column
of Table I expresses the scenario attractiveness from our
proposed analytical study (see (5),section III-B),which will be
confirmed by the system-level simulations.Column 4 shows
the SNR
min
and SNR
max
which represents the triangular
approximation of the gaussian distributed SNR only for the
verification of the attractiveness criterion (see Fig.2).
SBrT © 1022
VI INTERNATIONAL TELECOMMUNICATIONS SYMPOSIUM (ITS2006),SEPTEMBER 3-6,2006,FORTALEZA-CE,BRAZIL
if SNR
sat
≤ SNR
mean
E[1/R] =
4
(SNR)
2
·

SNR
2
sat
R

R
min
SNR
2
sat
(R)
2
· ln(
R
max
R
min
)+
(SNR
mean
−SNR
sat
) · (SNR
mean
+SNR
sat
−2SNR
min
)
2R
max
+
(SNR
max
−SNR
mean
)
2
2R
max

if SNR
mean
< SNR
sat
≤ SNR
max
E[1/R] =
4SNR
sat
R· (SNR)
2
·

2SNR
mean
−SNR
sat


R
min
SNR
sat
R
+SNR
min

·
ln

R· SNR
mean
R
min
SNR
sat
+1

+

R
min
SNR
sat
R
+SNR
max

· ln

R
max
SNR
sat
R· SNR
mean
+R
min
SNR
sat

if SNR
max
< SNR
sat
E[1/R] =
4SNR
sat
R· (SNR)
2
·

2SNR
mean
−SNR
max


R
min
SNR
sat
R
+SNR
min

·
ln

R· SNR
mean
R
min
SNR
sat
+1

+

R
min
SNR
sat
R
+SNR
max

· ln

R· SNR
max
+R
min
SNR
sat
R· SNR
mean
+R
min
SNR
sat

(9)
TABLE I:Evaluated scenarios.
Scenario
Expected Offered Load
(n
In
/n
Out
)
Bit Rate
Capacity [Mbps]
Link Quality [dB]
Attractiveness
Scenario 1
4/1
R
max
1
= 6
R
max
2
= 6
SNR
mean
1
= 10,SNR
min
1
= -7,SNR
max
1
= 27
SNR
mean
2
= 11,SNR
min
2
= -7.5,SNR
max
2
= 29.5
high
Scenario 2
1/4
R
max
1
= 6
R
max
2
= 6
SNR
mean
1
= 10,SNR
min
1
= -7,SNR
max
1
= 27
SNR
mean
2
= 11,SNR
min
2
= -7.5,SNR
max
2
= 29.5
low
Scenario 3
1/1
R
max
1
= 54
R
max
2
= 6
SNR
mean
1
= 10,SNR
min
1
= -7,SNR
max
1
= 27
SNR
mean
2
= 26,SNR
min
2
= 8,SNR
max
2
= 44
high
Scenario 4
1/1
R
max
1
= 6
R
max
2
= 54
SNR
mean
1
= 10,SNR
min
1
= -7,SNR
max
1
= 27
SNR
mean
2
= 26,SNR
min
2
= 8,SNR
max
2
= 44
low
In simulations,the SNR of all scenarios is a gaussian
distributed random variable.The mean (SNR
mean
) is given
in the Table I and the standard deviation is 4 dB for users
inside the HS connected to the RAN 1 and 8 dB elsewhere.
Section IV-A demonstrates the influence of the expected
offered load distribution in the performance of AS strategies.
In section IV-B the impact of the system capacity and
radio link quality in the multi-access scenario definition is
demonstrated.We present the results in two steps.First,
we show the scenario configuration drawing CDFs of SNR
and rate of both generic RANs.Finally,we show the
results in term of the CSE bit rate averaged over 1000
snapshots.The proportion of users indicates the load difference
between inside and outside HS areas.For instance,considering
n
In
/n
Out
= 1/4 and a load of 20 users per cell,there are 16
users outside the HS and 4 users inside the HS.
A.Scenarios 1 and 2:evaluation of the proportion of users
inside-outside HS
Here,we present two scenarios where the influence of
the expected offered load distribution (n
In
/n
Out
) in the
performance of AS algorithms can be illustrated.Figure 4
and 5 draw the general configuration of the scenarios.The
maximum rate capacities of the RANs are equivalent and the
10th percentile of their SNR are similar (see Table I).Fig.5
shows that although the RANs have similar 10th percentile of
SNR,there is possibility of rate improvement in the CDF of
the bit rate spatial distribution.
same 10
th
percentile
Fig.4:Scenarios 1 and 2 - CDF of the Radio Link Quality model
(SNR).
Fig.6 shows the CSE bit rate for the CTA and LBA as
a function of the system offered traffic load.We conclude
that an intelligent AS (LBA) will be attractive,if for similar
RANs (SNR distributions and capacities),the number of users
inside the HS is higher than the corresponding one in the
macrocell.This performance enhancement corresponds to a
trunking gain provided by the admission of users inside the HS
in the macrocell site,decreasing the number of connections in
SBrT © 1023
VI INTERNATIONAL TELECOMMUNICATIONS SYMPOSIUM (ITS2006),SEPTEMBER 3-6,2006,FORTALEZA-CE,BRAZIL
Possibility of rate
improvement
Fig.5:Scenarios 1 and 2 - CDF of the Bit Rate Spatial Distribution.
the highly loaded RAN 2.One can note that in scenario 2,
the LBA has the same performance of the CTA one (low
attractiveness).
Fig.6:Average CSE bit rate regarding n
In
/n
Out
= 4/1
(Scenario 1:attractiveness is high) and n
In
/n
Out
= 1/4
(Scenario 2:attractiveness is low).
B.Scenarios 3 and 4:Bit Rate Capacity and link Quality
Evaluation
In the previous section,the performance results have been
evaluated considering scenarios with RANs which have the
same maximum capacities and similar radio link quality
characteristics.This time,the behavior of the AS algorithms
is illustrated as a function of the maximum RAN capacity and
radio link quality.
We consider a case where RAN 2 has better reception
properties than RAN 1,i.e.,better SNR (see Fig.7).Figures
8 and 9 show the rate configurations of scenarios 3 and 4.
In scenario 3,the maximum rate capacities are 54 Mbps and
6 Mbps for RAN 1 and RAN 2,respectively.In scenario 4,the
maximum rate capacities are 6 Mbps and 54 Mbps for RAN 1
and RAN 2,respectively.
As expected,the possibility of rate improvement is strongly
dependent on the maximum system capacities.This can be
confirmed in Fig.10:although link quality is favorable,there
is no gain with the LBA,when the maximum rate capacity of
RAN 2 is much higher than that of the RAN 1 (scenario 4).
RAN 2 has
better SNR
than RAN 1
Fig.7:Scenario 3 - CDF of the Radio Link Quality model (SNR).
Possibility of rate
improvement
Fig.8:Scenario 3 - CDF of the Bit Rate Spatial Distribution
(R
max
1
= 54 Mbps and R
max
2
= 6 Mbps).
Low possibility of
rate improvement
Fig.9:Scenario 4 - CDF of the Bit Rate Spatial Distribution
(R
max
1
= 6 Mbps and R
max
2
= 54 Mbps).
V.C
ONCLUSIONS
In a multi-access network composed of two RANs,one of
them with a “hot spot” like coverage,the simplest admission
strategy is the CTA.Such a strategy does not require more
than the signal level measure for deciding in which RAN the
new user will be admitted.
The attractiveness of intelligent AS strategies in a
multi-access network depends on three parameters:the
proportion of users inside to outside hot spot;the maximum
SBrT © 1024
VI INTERNATIONAL TELECOMMUNICATIONS SYMPOSIUM (ITS2006),SEPTEMBER 3-6,2006,FORTALEZA-CE,BRAZIL
Fig.10:Performance comparison regarding the influence of the bit
rate capacity.Scenario 3:R
max
1
= 54 Mbps and R
max
2
= 6 Mbps
(Attractiveness is high).Scenario 4:R
max
1
= 6 Mbps and R
max
2
= 54 Mbps (attractiveness is low).
capacity of both systems,represented by the link adaptation
model;and their radio link qualities,expressed by the SNR
distributions.
An intelligent AS algorithm will not be attractive when
the expected offered load inside the HS is lower than the
corresponding one in the macrocell.Another unattractive
scenario is the one where the capacity of the RAN serving
the HS only,is much higher than the correspondent one of the
RAN covering the whole macrocell.Furthermore,the scenario
characterized by a RAN covering only the HS with a much
better bit rate distribution than the other RAN,also indicates
that the intelligent AS algorithms will achieve performances
similar to that of the CTA.
Further studies include the generalization of the
attractiveness criterion considering other user-related features
like mobility,user service classes,traffic model and access
selection criteria.
A
CKNOWLEDGEMENT
This work is supported by a grant from Ericsson of Brazil -
Research Branch under ERBB/UFC.10 Technical Cooperation
Contract.The authors V.A.de Sousa Jr.and R.A.de O.
Neto are scholarships supported by FUNCAP and Francisco
R.P.Cavalcanti was partly funded by CNPq - Conselho
Nacional de Desenvolvimento Cient´ıfico e Tecnol´ogico,grant
no.304477/2002-8.
R
EFERENCES
[1] WINNER I and II IST projects,“Wireless world initiative new radio
- winner project ist 2004-507581 and winner ii project ist-4-027756.”
[Online].Available:http://www.ist-winner.org
[2] Ambient Network Project,“Ambient network project.” [Online].
Available:http://www.ambient-networks.org/
[3] Everest IST Project,“Evolutionary strategies for radio resource
management in cellular heterogeneous networks - IST 2002-001858.”
[Online].Available:http://www.everest-ist.upc.es/
[4] MONASIDRE IST Project,“Management of networks and services
in a diversified radio environment - IST programme - action line
IV.5.2 terrestrial wireless systems and networks.” [Online].Available:
http://www.monasidre.com/
[5] CAUTION++ IST project,“Capacity and network management
platform for increased utilisation of wireless systems of
next generation++ - IST 2001-38229.” [Online].Available:
http://www.telecom.ece.ntua.gr/CautionPlus/
[6] BRAIN IST Project,“Broadband radio access for IP based networks -
IST 1999-10050.” [Online].Available:http://www.ist-brain.org/
[7] DRIVE IST Project,“Dynamic radio for IP-services in vehicular
environments.” [Online].Available:http://www.ist-drive.org
[8] OVERDRIVE IST Project,“Spectrum efficient uni- and multicast
services over dynamic multi-radio networks in vehicular environments.”
[Online].Available:http://www.ist-overdrive.org
[9] E.Mitjana,D.Wisely,S.Canu,and F.Loizillon,“Seamless IP service
provision:techno-economic study by the MIND ans TONIC projects,”
IST Mobile and Wireless Telecommunication,June 2002.
[10] A.Furuskar,“Radio resource sharing and bearer service allocation
for multi-bearer service,multi-access wireless networks - methods to
improve capacity,” Ph.D.dissertation,Royal Institute of Technology -
KTH,2003.
[11] O.Yilmaz,A.Furuskar,J.Pettersson,and A.Simonsson,“Access
selection in WCDMA and WLAN multi-access networks,” IEEE
Vehicular Technology Conference,2005.
[12] O.Yilmaz,“Access selection in multi-access cellular and WLAN
networks,” Master’s thesis,Royal Institute of Technology,Sweden,
Stockholm,February 2005.
[13] A.Furuskar,“Can 3G services be offered in existing spectrum?”
Licentiate Thesis,Royal Institute of Technologie,Stockholm,Sweden,
May 2001.
[14] A.T¨olli,P.Hakalin,and H.Holma,“Performance evaluation of Common
Radio Resource Ranagement (CRRM),” IEEE International Conference
on Communications,vol.5,pp.3429–3433,April 2002.
SBrT © 1025