Вы находитесь на странице: 1из 14

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO.

4, JULY 2008 2509


Call Admission Control Optimization
in WiMAX Networks
Bo Rong, Member, IEEE, Yi Qian, Senior Member, IEEE, Kejie Lu, Senior Member, IEEE,
Hsiao-Hwa Chen, Senior Member, IEEE, and Mohsen Guizani, Senior Member, IEEE
AbstractWorldwide interoperability for microwave access
(WiMAX) is a promising technology for last-mile Internet access,
particularly in the areas where wired infrastructures are not
available. In a WiMAX network, call admission control (CAC) is
deployed to effectively control different trafc loads and prevent
the network from being overloaded. In this paper, we propose
a framework of a 2-D CAC to accommodate various features of
WiMAX networks. Specically, we decompose the 2-D uplink and
downlink WiMAX CAC problem into two independent 1-D CAC
problems and formulate the 1-D CAC optimization, in which the
demands of service providers and subscribers are jointly taken
into account. To solve the optimization problem, we develop a
utility- and fairness-constrained optimal revenue policy, as well
as its corresponding approximation algorithm. Simulation results
are presented to demonstrate the effectiveness of the proposed
WiMAX CAC approach.
Index TermsCall admission control (CAC), optimization,
orthogonal frequency-division multiple access (OFDMA), time-
division duplex (TDD), worldwide interoperability for microwave
access (WiMAX).
I. INTRODUCTION
T
HERE exist many regions in the world where wired
infrastructures (i.e., T1, DSL, cables, etc.) are difcult
to deploy for geographical or economic reasons. To pro-
vide broadband wireless access to these regions, many re-
searchers advocate worldwide interoperability for microwave
access (WiMAX) [1], which is an IEEE 802.16 standardized
wireless technology based on an orthogonal frequency-division
multiplexing (OFDM) physical-layer architecture.
Manuscript received April 17, 2007; revised August 12, 2007 and
September 24, 2007. This work was supported in part by the National Science
Foundation (NSF) under Award 0424546 and an NSF Experimental Program
to Stimulate Competitive Research start-up grant in Puerto Rico and in part by
the National Science Council of Taiwan under Grant NSC96-2221-E-110-035
and Grant NSC96-2221-E-110-050. The review of this paper was coordinated
by Dr. E. Hossain.
B. Rong and K. Lu are with the Department of Electrical and Computer
Engineering, University of Puerto Rico at Mayagez, Mayagez, PR 00681
USA (e-mail: bo.rong@ece.uprm.edu; lukejie@ece.uprm.edu).
Y. Qian is with the National Institute of Standards and Technology,
Gaithersburg, MD 20899 USA (e-mail: yqian@nist.gov).
H.-H. Chen is with the Department of Engineering Science, National Cheng
Kung University, Tainan 701, Taiwan, R.O.C. (e-mail: hshwchen@ieee.org).
M. Guizani is with the Department of Computer Science, Western Michigan
University, Kalamazoo, MI 49008-5201 USA (e-mail: mguizani@gmail.com).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TVT.2007.912595
To support a variety of applications, IEEE 802.16 has dened
four types of service [2]: 1) unsolicited grant service (UGS);
2) real-time polling service (rtPS); 3) non-real-time polling
service (nrtPS); and 4) best effort (BE) service. In a WiMAX
network with heterogeneous trafc loads, it is essential to nd
a call admission control (CAC) solution that can effectively
allocate bandwidth resources to different applications. Moti-
vated by this, in this paper, we propose a WiMAX CAC frame-
work, which effectively meets all operational requirements of
WiMAX networks. In this CAC framework, we decompose
the 2-D uplink (UL) and downlink (DL) WiMAX CAC prob-
lem into two independent 1-D CAC problems. We further
formulate the 1-D CAC as an optimization problem under a
certain objective function, which should be chosen to maximize
either the revenue of service providers or the satisfaction of
subscribers.
With respect to 1-D CAC optimization problems, most pre-
vious studies were focused only on two approaches: 1) the
optimal revenue strategy (also known as the stochastic knap-
sack problem) [3][9] and 2) the minimum weighted sum
of blocking strategy [10]. In this paper, we will show that
these two strategies are, in fact, equivalent. Therefore, we
can mainly concentrate on the investigation of the optimal
revenue strategy and view the minimum weighted sum of
blocking strategy as the basis for fast calculation algorithms.
Clearly, the optimal revenue policy only considers the prot
of service providers. As an effort to conduct a multiobjec-
tive study, in this paper, we will also take into account the
requirements from WiMAX subscribers and develop a policy
with a satisfactory tradeoff between service providers and
subscribers.
The major contributions of this paper include the following:
1) the development of a framework of CAC for WiMAX
networks; 2) the investigation on various CAC optimization
strategies; and 3) the proposal of a series of constrained greedy
revenue algorithms for fast calculation. Through detailed per-
formance evaluation, the study carried out in this paper will
show that the proposed CAC solution can meet the expectations
of both service providers and subscribers.
The rest of the paper is outlined as follows: We rst introduce
the CAC model for WiMAX networks in Section II. We will
then calculate the UL and DL capacity in Section III. In
Sections IV and V, we will introduce different 1-D CAC
optimization strategies and develop their corresponding ap-
proximation algorithms. In Section VI, the performance of
the proposed CAC optimization approach is demonstrated by
simulation results. Finally, Section VII concludes this paper.
0018-9545/$25.00 2008 IEEE
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
2510 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 4, JULY 2008
Fig. 1. CAC deployment in a WiMAX PMP network.
II. CAC DEPLOYMENT IN WIMAX NETWORKS
A. WiMAX Networks
WiMAX technology promises to support both mesh and
point-to-multipoint (PMP) networks. A WiMAX mesh network
usually suits for constructing wide-area wireless backhaul net-
works such as citywide wireless coverage. On the other hand,
a WiMAX PMP network is used for providing the last-mile
access to broadband ISPs. In this paper, we will only discuss
the issues related to the WiMAX PMP network, which consists
of one base station and N subscribers.
As specied in the IEEE 802.16 standards [11], [12],
WiMAX employs OFDM in its physical-layer architecture.
In particular, IEEE 802.16 standards have dened a special
avor of the OFDM system, namely, orthogonal frequency-
division multiple access (OFDMA), which employs a larger
fast-Fourier-transform space (2048 and 4096 subcarriers) that
can be further divided into subchannels. The subchannels
are introduced to separate the data into logical streams on
DL transmission. Those logical streams may employ different
modulation/coding schemes and amplitude levels to address
subscribers with different channel characteristics. The subchan-
nels are also used for multiple-access purposes on UL. In prac-
tice, subscribers are assigned with subchannels through media
access control messages sent in the downstreams. Without the
loss of generality and for discussion brevity, in this paper,
we mainly concern ourselves with the scenario where each
WiMAX subscriber occupies exactly one subchannel.
To satisfy different operational environments, WiMAX
OFDMA supports ve types of subcarrier allocation schemes to
formulate subchannels: 1) partial usage of subchannels (PUSC)
on UL and DL; 2) optional PUSC on UL; 3) full usage of
subchannels (FUSC) on DL; 4) optional FUSC on DL; and
5) adaptive modulation and coding (AMC) on UL and DL.
The rst four subcarrier allocation schemes employ distributed
subcarrier permutation, whereas AMC employs adjacent sub-
carrier permutation. Since distributed subcarrier permutation
performs well in both xed and mobile environments, it
becomes the dominant subcarrier permutation strategy for
WiMAX applications.
Once the subcarrier allocation is determined, the power
assigned to each subcarrier becomes another important issue
Fig. 2. Decomposition of UL CAC and DL CAC.
that affects the data transmission rate of a certain subscriber. For
UL, the transmission power of each subscriber depends on its
own transmitter. For DL, all subscribers share the total power on
the base station. To simplify the implementation, most WiMAX
base stations employ an equal power assignment scheme, which
grants every subcarrier the same DL power.
Frequency-division duplex (FDD) and time-division duplex
(TDD) are two of the most prevalent duplexing schemes used
in wireless systems. WiMAX can employ either of them to
separate UL and DL communication signals. FDD is usually
deployed in symmetric communication scenarios, where the
applications require equal bandwidth on UL and DL. On the
contrary, TDD is usually deployed in asymmetric communica-
tion scenarios.
In this paper, we assume that WiMAX is to provide the
last-mile Internet access, which is a typical asymmetric
communication scenario. Accordingly, we will study WiMAX
OFDMA with TDD duplexing. For the kth WiMAX subscriber,
we assume that its UL and DL data transmission rates are tr
U
k
and tr
D
k
, respectively. If % time slots on the subchannel are
allocated to DL trafc, then the DL bandwidth capacity is given
by B
D
k
= %tr
D
k
(thus, % becomes the TDD DL bandwidth
proportional factor). Similarly, the UL bandwidth capacity can
be calculated by B
U
k
= (1 %)tr
U
k
.
B. CAC Deployment
To handle a multiservice WiMAX network, it is very impor-
tant to implement the CAC mechanism [13], [14]. First, CAC
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
RONG et al.: CALL ADMISSION CONTROL OPTIMIZATION IN WiMAX NETWORKS 2511
TABLE I
PARAMETER DECOMPOSING MODEL (rer
avg
i
= (rer
U
i
b
U
i
+rer
D
i
b
D
i
)/(b
U
i
+b
D
i
))
is a critical step for the provision of quality-of-service (QoS)-
guaranteed services, because it can prevent the system from
being overloaded. Second, CAC can help a WiMAX network
provide different classes of trafc loads with different priorities
by manipulating their blocking probabilities.
As illustrated in Fig. 1, we propose to add a CAC manager
to the WiMAX base station. This CAC manager provides CAC
functions to all N subscribers under the base station. Moreover,
the access bandwidth capacity and the trafc load prole of
different subscribers are distinct and independent. Accordingly,
the CAC manager is composed of N CAC modules, in which
the kth CAC module only takes care of the kth subscribers
local network. In the rest of this paper, our study will be
concentrated on the design of an individual CAC module.
When an application in the kth subscribers local network
initiates a connection to access the Internet, it sends a connec-
tion request to the kth CAC module with upstream bandwidth
requirement b
U
and downstream bandwidth requirement b
D
.
Then, the CAC manager simultaneously performs an admission
control check on the kth subscribers UL and DL. In this re-
spect, the CAC in a WiMAX network can naturally be modeled
by a 2-D CAC problem.
C. Decomposition of WiMAX CAC
A 2-D model for the WiMAX CAC can be formulated as fol-
lows. For the local network of a given WiMAX subscriber, sup-
pose that M classes of bidirectional trafc loads share B
U
units
of UL bandwidth resource and B
D
units of DL bandwidth
resource, respectively. With regard to class i trafc, we assume
the following: 1) The requests arrive from a random process
with an average rate
i
; 2) the average connection holding
time is 1/
i
s; 3) the UL and DL bandwidth requirements
of a connection are xed to b
U
i
and b
D
i
, respectively; and
4) the UL and DL revenue rates of a connection are rer
U
i
and rer
D
i
, respectively. Then, the WiMAX CAC is responsible
for accepting or rejecting a connection request according to a
certain policy.
For investigation simplicity, we decouple the 2-D WiMAX
CAC problem into two independent 1-D CAC problems in this
paper. As shown in Fig. 2, the CAC module employs a UL CAC
policy and a DL CAC policy to separately make an admission
test on UL and DL, and only the connection requests that passes
both admission tests can be eventually accepted.
Either UL CAC or DL CAC can be modeled as a 1-D
CAC problem as follows. For the local network of a given
WiMAX subscriber, suppose that M classes of trafc loads
share B units of access bandwidth resource. With respect
to class i trafc, we assume the following: 1) The requests
arrive from a random process with an average rate
i
; 2) the
average connection holding time is 1/
i
s; 3) the bandwidth
requirement of a connection is xed to b
i
; and 4) the revenue
Fig. 3. Framework for UL CAC or DL CAC.
rate of a connection is rer
i
. We can then dene the bandwidth
requirement vector as

b = (b
1
, b
2
, . . . , b
M
) and the system state
vector as n = (n
1
, n
2
, . . . , n
M
), where n
i
is the number of
class-i connections in the system. Based on these parameters,
we can further dene
CS
as the set of all possible system
states, which can be expressed as
CS
= {n|n

b B}. Under
this denition, the subscript CS stands for complete sharing,
which means that an incoming connection will be accepted if
sufcient bandwidth resources are available in the system. We
can now dene a CAC policy, which is denoted by , as an
arbitrary subset of
CS
. Given , a connection request will be
accepted if and only if the system state vector remains in after
the connection is accepted.
In this paper, we decompose the WiMAX CAC into UL
CAC and DL CAC with the parameter setup, as shown in
Table I. Although UL CAC and DL CAC are running as two
independent 1-D processes, their parameters are still related.
On one hand, UL CAC and DL CAC have their own special
characteristics such as distinct bandwidth capacity and band-
width requirements. On the other hand, they do share some
common features such as the number of trafc classes M,
arrival rate
i
, service time 1/
i
, and revenue rate r
avg
i
. It is
noted that in Table I, we confer the same revenue rate r
avg
i
to
UL CAC and DL CAC, despite the fact that originally, the UL
revenue rate is r
U
i
and the DL revenue rate is r
D
i
. The reason
is that the revenue could be achieved from a connection if and
only if it can pass both UL and DL CAC tests. Therefore, from
the perspective of 2-D WiMAX CAC, UL CAC and DL CAC
should have the same revenue rate. In this paper, we name this
mechanism as UL/DL revenue rate binding.
D. Framework for WiMAX CAC
The aforementioned CAC manager is composed of N CAC
modules, in which the kth CAC module contains two inde-
pendent parts, i.e., UL CAC and DL CAC. To t the special
features of a WiMAX environment, we present a WiMAX CAC
framework in Fig. 3, which can be applied to either UL CAC or
DL CAC.
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
2512 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 4, JULY 2008
As illustrated in Fig. 3, the proposed framework consists of
three major modules: 1) CAC policy; 2) trafc load estimation;
and 3) bandwidth capacity estimation. Clearly, the CAC policy
is the most important module of all. To construct a successful
WiMAX network, the CAC policy has to consider the expec-
tations of both service providers and subscribers. Here, we
suppose that the UL and DL CAC policies independently work
with their parameter setups, as shown in Table I, and utilize the
same optimization strategy.
In a WiMAX network, the trafc load and the subcarrier
channel condition are changing from time to time. Conse-
quently, UL and DL CAC policies have to be adaptive to these
dynamics. The trafc load estimation module should record
every connection request, regardless of whether it is eventually
accepted or not. Then, the UL and DL trafc load proles
are periodically captured based on this record and sent to the
CAC policy module. As for the dynamic of channel condition,
the bandwidth estimation module retrieves the channel state
information (CSI) from the physical layer at regular intervals
of time. Then, the UL and DL bandwidth capacities of a given
subscriber are calculated from the CSI and sent to the CAC
policy module. To operate adaptively, the CAC policy module
should adjust its parameters based on the updated information
from the other two modules.
To mitigate the channel errors in the physical layer, it is ap-
propriate to apply link-layer fragmentation and retransmission
to WiMAX [15]. In this respect, the CAC policy module also
needs to interact with the WiMAX link layer. In particular, after
a connection request is accepted, the CAC policy module has to
transfer the bit error rate (BER) requirement of this connection
to the fragmentation and retransmission module in the link
layer.
For the proposed framework in Fig. 3, we need to consider
two major issues: 1) how to calculate UL and DL bandwidth ca-
pacity and 2) what the design of a 1-D CAC policy is. Although
it is also important to capture the trafc load prole, we do not
particularly discuss this issue here as it has been extensively ad-
dressed in many previous works on wired networks [16], [17].
In Section III, we will investigate the UL and DL bandwidth ca-
pacity of a given subscriber. In Sections IV and V, we will study
the design of a 1-D CAC policy as an optimization problem.
III. BANDWIDTH CAPACITY
In this section, we calculate the UL and DL bandwidth
capacity of the kth subscriber based on CSI. We begin with the
approach for DL capacity, and then, this approach will be easily
extended to UL.
Fig. 4 illustrates the N-subscriber WiMAX DL channel
model, in which H
k
(f) and N
k
(f) denote the channel fre-
quency response function and noise power density function of
the kth subscriber, respectively. The quality of each subscribers
subchannel is indicated by the signal-to-noise ratio (SNR)
function SNR
k
(f), which is dened as
SNR
k
(f) = |H
k
(f)|
2
/N
k
(f) (1)
where SNR
k
(f)(1 k N) is the CSI that the CAC man-
ager needs. Since the CAC manager is placed in the base
Fig. 4. WiMAX DL channel model.
station, it can collect SNR
k
(f) easily from the physical
layer.
For analytical simplicity, in this paper, we are only interested
in the investigation of the most common scenario, where each
WiMAX subscriber occupies exactly one subchannel. Once the
subcarrier allocation and power assignment of the subchannels
are known, the DL bandwidth capacity of the kth WiMAX
subscriber can be calculated with the following notations:
N number of subscribers;
J number of subcarriers;
S
J
set of all subcarriers, dened as {1, 2, . . . ,
j, . . . , J};
f physical bandwidth of each subcarrier;
D
k
subcarrier set assigned to subscriber k;
SNR
k
[j] the SNRfunction of subscriber k on the frequency
of subcarrier j;
p[j] transmit power on subcarrier j;
c
k
[j] achievable transmission efciency (data rate per
hertz) of subscriber k on the frequency of
subcarrier j, assuming that subcarrier j is allo-
cated to subscriber k(j D
k
).
To assign subcarrier set D
k
to subscriber k, we assume that
a nonoverlapped partition is used such that
_
D
k

D
l
= , k = l

N
k=1
D
k
S
J
(2)
where is a null set.
According to [18][20], c
k
[j] can be expressed as
c
k
[j] = f [log
2
(1 +p[j]SNR
k
[j])] (3)
where = 1.5/ ln(5BER) is a constant, and f(.) depends
on the adaptation scheme. For example, if the continuous-rate
adaptation is concerned, we have f(x) = x; if a variable M-ary
quadrature amplitude modulation (M-QAM) with modulation
levels 0, 2, 4, 6,. . . is employed, we have f(x) = 2(1/2)x
instead.
Then, the DL data transmission rate of subscriber k is
given by
tr
D
k
=

jD
k
c
k
[j]f. (4)
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
RONG et al.: CALL ADMISSION CONTROL OPTIMIZATION IN WiMAX NETWORKS 2513
Consequently, the DL bandwidth capacity of the kth subscriber
can be calculated by B
D
k
= % tr
D
k
. Equations (1)(4) can
also be used to compute the UL data transmission rate of
subscriber k, denoted as tr
U
k
, if the variables in these equa-
tions are replaced by UL parameters. Then, the UL bandwidth
capacity of the kth subscriber can be obtained by B
U
k
= (1
%) tr
U
k
.
IV. ONE-DIMENSIONAL CAC OPTIMIZATION
In this section, we rst review the existing CAC optimization
strategies in the literature. We then propose a utility- and
fairness-constrained optimal revenue strategy as an effort to
benet both service providers and subscribers.
A. Existing 1-D CAC Optimization Strategies
The previous study on 1-D CAC optimization was mainly
focused on two approaches, i.e., the optimal revenue strategy
and the minimum weighted sum of blocking strategy.
1) Optimal Revenue Strategy: In general, service providers
expect a CACpolicy that can produce high revenues. To achieve
this goal, [3][9] studied the CAC optimization problems with
the optimal revenue strategy, which is also known as the sto-
chastic knapsack problem. The optimal revenue strategy can be
introduced in the context of WiMAX as follows.
Let rer
UGS
, rer
rtPS
, rer
nrtPS
, and rer
BE
be the revenue
rates of UGS, rtPS, nrtPS, and BE service, respectively. Then,
we use rer
i
to denote the revenue rate of a class-i connection,
which can be one of rer
UGS
, rer
rtPS
, rer
nrtPS
, and rer
BE
, de-
pending on which type of service is required. Correspondingly,
we can calculate the long-term average revenue of a CAC
policy by
R() =

n
(n r)P

(n) (5)
where P

(n) is the steady-state probability that the system is


in state n, r = (r
1
, r
2
, . . . , r
M
) is the reward vector, and r
i
=
rer
i
b
i
is the average revenue generated by accepting a class-i
connection. For a given WiMAX system, (5) can be separately
applied to UL and DL. In the most common case, UL and
DL have different parameter setups but have the same revenue
rate rer
i
.
In this paper, we use

to denote the optimal revenue policy.


Intuitively,

prefers the trafc load with a high-revenue


output.
2) Minimum Weighted Sum of Blocking Strategy: A CAC
optimization strategy was proposed in [10] to minimize the
weighted sum of blocking, which is dened as
W
B
= weighted sum of blocking =
M

i=1
w
i
Pb
i
(6)
where Pb
i
stands for the blocking probability of class-i trafc,
and w
i
stands for the weight of class-i trafc. By adjusting the
value of w
i
, the minimum weighted sum of blocking strategy
can give different trafc classes different priorities.
B. Equivalence of Two CAC Optimization Strategies
It is noted that if a policy satises the coordinate convex
condition and the arrival and service processes are both memo-
ryless, then P

(n) can be calculated by (as shown in [21])


P

(n) =
1
G()
M

i=1

n
i
i
n
i
!
, n (7)
where G() =

M
i=1

n
i
i
/n
i
!, and
i
=
i
/
i
.
Moreover, the blocking probability of class-i trafc is
Pb
i
() =
G
_

b
i
_
G()
(8)
where
b
i
= {n|n & n +e
i
/ }, and e
i
is an
M-dimension vector of all zeros, except for its ith element.
Based on the above equations, we can have the following
lemma.
1) Lemma 1: For any CAC policy that satises the coordi-
nate convex condition, the long-term average revenue dened
in (5) can always be calculated by
R() = AW
B
(9)
where A =

M
i=1
r
i

i
is a constant, W
B
=

M
i=1
w
i
Pb
i
, and
w
i
= r
i

i
.
In (9), R() is the revenue that policy can achieve or
the revenue of accepted trafc, A =

M
i=1
r
i

i
is the revenue
of arriving trafc, and W
B
=

M
i=1
w
i
Pb
i
=

M
i=1
r
i

i
Pb
i
is the revenue of rejected trafc. Obviously, the revenue of
accepted trafc can be calculated by subtracting the revenue of
rejected trafc from the revenue of arriving trafc. Therefore,
Lemma 1 holds for any policy that satises the coordinate
convex condition.
Lemma 1 also implies that the revenue can be maximized
when the weighted sum of blocking is minimized. Therefore,
the minimum weighted sum of blocking strategy is equivalent
to the optimal revenue strategy if the condition w
i
= r
i

i
holds. In this paper, we mainly consider the optimal revenue
strategy, since it has more explicit meaning in practice. As for
the minimum weighted sum of blocking strategy, we consider
it only as an alternative way to express the optimal revenue
strategy, which will be employed to develop fast calculation
algorithms in Section V.
C. Utility- and Fairness-Constrained Optimal
Revenue Strategy
The optimal revenue strategy highlights only the demand of
service providers. Nevertheless, in a practical WiMAX system,
we also need to consider the requirements of subscribers.
1) Utility Requirement: Generally speaking, subscribers
prefer a CAC policy that can achieve maximal utility or, equiva-
lently, the maximum access bandwidth [22], [23]. Let B denote
the physical access bandwidth and SB denote the statistical
bandwidth that the subscriber can achieve after the CAC policy
takes effect. Then, the utility function is dened as SB/B.
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
2514 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 4, JULY 2008
Since B is a constant, the maximization of the utility function
leads to the maximization of the statistical bandwidth SB. In
this paper, we use
+
to denote the optimal utility CAC policy.
Different from

, which denotes the optimal revenue policy,

+
allocates more bandwidth resources to the trafc load that
can yield high utility.
Following (5), we derive the statistical bandwidth that
policy can achieve as
SB() =

n
(n

b)P

(n) (10)
where we have replaced the reward vector r given in (5) with
the bandwidth requirement vector

b.
Based on (10), the utility of policy is given by
U() =
1
B
SB() =
1
B

n
(n

b)P

(n). (11)
It is noted that if all trafc classes have the same revenue rate,
i.e., rer
1
= rer
2
= = rer
M
,

turns into
+
.
2) Fairness Requirement: When the optimal revenue or op-
timal utility strategy is employed, there may exist a great bias
among the blocking probabilities of different trafc classes.
This bias can result in unfairness such that some trafc classes
is severely blocked, whereas others can easily access the net-
work. Therefore, in addition to the utility requirement, the
fairness among different trafc classes becomes another major
issue that subscribers might be concerned about [24], [25].
Fairness requirement guarantees that the blocking probabilities
of all trafc classes will be kept relatively uniform such that no
particular trafc class will be unfairly treated.
In this paper, we mainly address the fairness at the CAC
level, which means that the applications of every service class
have a chance to achieve access bandwidth. Therefore, different
fairness policies can be applied in the underlying service model,
such as a service policy of packet scheduling. For instance,
in the packet scheduling level, unfairness may be necessary to
guarantee the QoS provisioning.
To achieve absolute fairness (AF), each trafc class is given
the same blocking probability, whereas the utility of the CAC
policy is maximized. This paper mainly considers the scenario
of a stressful network where the arriving trafc load is higher
than the bandwidth capacity. Then, the blocking probability of
each trafc class is given by
Pb
AF
= 1
U(
AF
)B
M

i=1
b
i

i
Pb
AF
lb
(12)
where Pb
AF
lb
= 1 (B/

M
i=1
b
i

i
/
i
) is the lower bound of
Pb
AF
. As dened, the AF policy has the highest utility among
all policies that can offer equal blocking probabilities. There-
fore, in practice, Pb
AF
can have a value very close to Pb
AF
lb
.
3) Constrained Optimal Revenue Strategy: The objective
of CAC optimization can be chosen to satisfy either service
providers or subscribers. Usually, a contradiction exists be-
tween the expectations of service providers and subscribers.
Therefore, we have to develop a constrained CAC optimization
strategy that can give a good tradeoff. In other words, we need
a CAC policy that can balance the optimal revenue requirement
from the service providers and the optimal utility and fairness
requirements from the subscribers. This leads to a concept of a
utility- and fairness-constrained optimal revenue policy, which
is proposed in this paper.
The fairness constraint requires that the highest blocking
probability among all trafc classes (or, in short, the highest
blocking probability) is lower than the threshold PB
th
. In this
paper, we use
F
to represent the fairness-constrained opti-
mal revenue policy. Clearly, the blocking probability threshold
PB
th
must be higher than Pb
AF
lb
; otherwise, no CAC policy
can be found to meet the fairness constraint. Therefore, PB
th
is subject to the relationship Pb
AF
lb
< PB
th
< 1. Correspond-
ingly, we dene the normalized blocking probability thresh-
old as pb
th
, which ranges from zero to one (0 < pb
th
< 1).
The relationship between pb
th
and PB
th
can be formulated
by pb
th
= (PB
th
Pb
AF
lb
)/(1 Pb
AF
lb
) or PB
th
= (1
Pb
AF
lb
)pb
th
+Pb
AF
lb
.
The utility constraint requires that the utility of a CAC policy
must be higher than the threshold U
th
. In this paper, we use

UF
to represent the utility and fairness constrained optimal
revenue policy. If the fairness constraint is already known, it
is clear that the utility threshold U
th
must satisfy 0 < U
th
<
U(
F+
), where U(
F+
) denotes the utility of the fairness-
constrained optimal utility strategy. Otherwise, no
UF
can
be found to satisfy both utility and fairness constraints. Corre-
spondingly, we dene the normalized utility threshold as u
th
,
which satises U
th
= u
thth
U(
F+
) (0 < u
th
< 1). It is also
noted that when PB
th
= 1 or pb
th
= 1,
UF
degenerates into
the utility-constrained optimal revenue policy
U
.
Practically, it is convenient to use u
th
and pb
th
, instead of
U
th
and PB
th
, to describe the utility and fairness constraints,
since the valid ranges of U
th
and PB
th
depend on the band-
width capacity and trafc load, whereas the valid ranges of u
th
and pb
th
are always from zero to one.
V. ONE-DIMENSIONAL CAC OPTIMIZATION ALGORITHMS
In this paper, we mainly study the optimal revenue policy

and constrained optimal revenue policies


U
,
F
, and
UF
,
which are likely to be used by service providers to construct a
commercial WiMAX network. In addition, we are interested in

+
as well, since it can serve as a benchmark for utility perfor-
mance. To specify the above policies, brute-force searching is a
straightforward method. For example, to achieve

, one needs
to calculate the long-term average revenue of each possible pol-
icy using (5). To achieve
UF
, one needs to calculate the long-
term average revenue, the blocking probabilities, and the utility
of each possible policy using (5), (8), and (11). Nevertheless,
the previous studies pointed out that the brute-force searching
has an unbearable complexity [3], [5]. Even if the optimal
solution could be found, it is usually very complicated and,
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
RONG et al.: CALL ADMISSION CONTROL OPTIMIZATION IN WiMAX NETWORKS 2515
therefore, extremely difcult to implement due to excessively
high storage and table lookup time requirements. As a result,
many researchers endeavored to develop some simply struc-
tured approximate solutions. For example, to nd the optimal
revenue policy of a complete partition (CP) structure, Ross
and Tsang [3] proposed a nite-stage dynamic programming
algorithm, which has a complexity of O(B
2
M). Here, the
algorithm complexity is measured by the size of the searching
space that a CAC optimization algorithm has to explore [3]. In
the rest of this section, we will develop a series of CP-structured
heuristic algorithms, which have a complexity of O(BM).
A. CP-Structured Admission Control Policy
A CP policy allocates each trafc class a certain amount of
nonoverlapped bandwidth resource. In this manner, the block-
ing rate of one trafc class will not inuence that of others.
Because of this partitioning characteristics, a CP policy can
be decomposed into M independent subpolicies, and the ith
subpolicy takes care of class-i trafc. In other words, a CP
policy separates the overall bandwidth resource B into M
nonoverlapped parts, denoted by B
1
CP
, B
2
CP
, . . . , B
M
CP
, where
B
i
CP
belongs to class-i trafc.
For a given CP policy, the ith subpolicy can be modeled by an
M/M/N/N queuing system, in which the number of servers is
s
i
= B
i
CP
/b
i
. Therefore, according to (5) and (7), the long-term
average revenue obtained from class-i trafc is given by
R
i
(CP) =
s
i

j=0
r
i
j

j
i
j!
s
i

k=0

k
i
k!
(13)
where the overall long-term average revenue of the CP policy
is dened as R(CP) =

M
i=1
R
i
(CP). Moreover, the statistical
bandwidth of class-i trafc is given by
SB
i
(CP) =
s
i

j=0
b
i
j

j
i
j!
s
i

k=0

k
i
k!
. (14)
Correspondingly, the overall statistical bandwidth and
the utility of the CP policy can be calculated by
SB(CP)=

M
i=1
SB
i
(CP) and U(CP)=(1/B)

M
i=1
SB
i
(CP),
respectively.
According to the Erlang B formula, the blocking probability
of class-i trafc is [26]
Pb
i
(CP) = B(s
i
,
i
) =

s
i
i
s
i
!
s
i

k=0

k
i
k!
. (15)
It is noted that the Erlang B formula can be calculated by the
following recursion [26]:
B(s
i
+ 1,
i
) =

i
B(s
i
,
i
)
s
i
+ 1 +
i
B(s
i
,
i
)
(16)
where we have B(0,
i
) = 1.
The above discussions have shown that the optimal CP
problem is to nd the best bandwidth partitioning scheme.
In this paper, we dene the CP policy of optimal revenue,
optimal utility, utility-constrained optimal revenue,
fairness-constrained optimal revenue, and utility- and
fairness-constrained optimal revenue as CP

, CP
+
,
CP
U
, CP
F
, and CP
UF
, respectively, which can be viewed
as the approximate solution for

,
+
,
U
,
F
, and

UF
. In the following sections, we will address the issues on
developing heuristic algorithms for CP
F
and CP
UF
. With
some appropriate adjustment, these heuristic algorithms can be
used to approximate CP

, CP
+
, and CP
U
.
B. Fairness-Constrained Greedy Revenue Algorithm to
Approximate CP
F
To achieve a heuristic algorithm for CP
F
, we have to deal
with the issues on the optimal revenue strategy and the fairness
constraint. In the following text, we rst develop a greedy
revenue approximation method for the optimal revenue strategy,
and then, we will add the fairness constraint to it.
Supposing that class-i trafc is assigned j (j = B
i
CP
/b
i
)
servers, from Lemma 1, we can derive the corresponding rev-
enue as
R
i
(CP)|
B
i
CP
=jb
i
= r
i

i
r
i

i
B(j,
i
). (17)
If one server is withdrawn or the server number is reduced to
(j 1), the following equation holds:
R
i
(CP)|
B
i
CP
=(j1)b
i
= r
i

i
r
i

i
B(j 1,
i
). (18)
Accordingly, we dene the revenue brought by the jth server as
R
i
S
(j,
i
) =R
i
(CP)

B
i
CP
=jb
i
R
i
(CP)

B
i
CP
=(j1)b
i
=r
i

i
[B(j 1,
i
) B(j,
i
)]
=r
i
F
i
S
(j,
i
) (19)
where F
i
S
(j,
i
) =
i
[B(j 1,
i
) B(j,
i
)] is the load car-
ried by the jth server in the M/M/N/N queuing system,
and

s
i
j=1
R
i
S
(j,
i
) = R
i
(CP). Then, the revenue rate of the
jth server is
r
i
S
(j,
i
) =
R
i
S
(j,
i
)
b
i
=
r
i

i
[B(j 1,
i
) B(j,
i
)]
b
i
.
(20)
Initially, (19) was introduced in the theory of marginal eco-
nomic analysis [27] by intuition. In this paper, we use Lemma 1
to give it a detailed explanation. Since F
i
S
(j,
i
) is decreas-
ing with j [28], R
i
S
(j,
i
) and r
i
S
(j,
i
) are decreasing with
j as well.
Based on the above analysis, we propose a greedy revenue
approximation method to obtain the optimal revenue. The
greedy revenue approximation method always allocates the
bandwidth resource to the trafc class of the highest revenue
rate in each iteration. Taking into account the fairness constraint
as well, a fairness-constrained greedy revenue algorithm to
approximate CP
F
is presented in Algorithm 1.
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
2516 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 4, JULY 2008
Algorithm 1: Fairness-Constrained Greedy Revenue Algo-
rithm for CP
F
1) Input pb
th
;
2) Capture the trafc load prole in the kth subscribers
local network;
3) Collect the CSI from the physical layer, calculate B
U
k
for UL CAC or B
D
k
for DL CAC, and let B = B
U
k
or
B = B
D
k
;
4) / PHASE 1: Allocate bandwidth resources to satisfy
the fairness constraint /
5) B
free
=B; Pb
AF
lb
=1(B/

M
i=1
b
i

i
/
i
); PB
th
=
pb
th
Pb
AF
lb
;
6) for i = 1 to M do
7) B
i
CP
= 0; s
i
= 0; B(s
i
,
i
) = 1; Pb
i
= B(s
i
,
i
);
8) end for
9) for i = 1 to M do
10) while Pb
i
> PB
th
do
11) B
free
=B
free
b
i
; B
i
CP
= B
i
CP
+b
i
; B(s
i
+ 1,

i
) = (
i
B(s
i
,
i
)/s
i
+ 1 +
i
B(s
i
,
i
));
12) Pb
i
= B(s
i
+ 1,
i
); SB
i
= SB
i
+b
i

i
[B(s
i
,

i
) B(s
i
+ 1,
i
)]; s
i
= s
i
+ 1;
13) end while
14) B
i
CP
(F) = B
i
CP
;
15) end for
16) / PHASE 2: Allocate the remaining free bandwidth
resources according to optimal revenue strategy /
17) I = M + 1;
18) for all i(0 i M): B(s
i
+ 1,
i
) = (
i
B(s
i
,
i
)/
s
i
+1+
i
B(s
i
,
i
)); r
i
S
=(r
i

i
[B(s
i
,
i
)B(s
i
+1,

i
)]/b
i
);
19) while I > 0 do
20) I = arg max
1iM
{r
i
S
};
21) if b
I
B
free
then
22) / allocate b
I
bandwidth resource to class I
trafc /
23) B
free
=B
free
b
I
; B
I
CP
=B
I
CP
+b
I
; s
I
=s
I
+1;
24) B(s
I
+1,
I
)=(
I
B(s
I
,
I
)/s
I
+1+
I
B(s
I
,
I
));
r
I
S
=(r
I

I
[B(s
I
,
I
)B(s
I
+1,
I
)]/b
I
);
25) else
26) r
I
S
= 0;
27) if

M
i=1
r
i
S
= 0 then
28) I = 1; // bandwidth allocation is nished.
29) end if
30) end if
31) end while
32) return {B
i
CP
(F), 1 i M} as the bandwidth allo-
cation to satisfy fairness constraint;
33) return {B
i
CP
, 1 i M} as the nal bandwidth al-
location decision for CP
F
;
As for the dynamic of trafc load, the DL trafc load
estimation module records every connection request in its
database, regardless of whether it is accepted or not. Then,
the trafc load prole is periodically captured based on this
record and sent to the CAC policy module. As for the dynamic
of channel condition, the DL bandwidth estimation module
retrieves the CSI from the physical layer at regular intervals
of time. Then, the DL bandwidth capacity of a given sub-
scriber is calculated from the CSI and sent to the CAC policy
module.
Algorithm 1 proceeds in two phases. In the rst phase, it
allocates class-i trafc B
i
CP
(F) bandwidth resource so that the
fairness constraint can be guaranteed. In the second phase, it
iteratively employs the greedy method to implement the optimal
revenue strategy. For instance, steps 18 and 24 of the algorithm
initialize and update the revenue rates of different trafc classes,
respectively, and then, step 20 picks up one of the maximum
revenue rates.
Algorithm 1 can exactly locate CP
F
before step 26 is
reached. The iterations may incur approximation errors if and
only if the bandwidth capacity boundary condition B
free
<
max
1iM
{b
i
} becomes true. If so, the greedy method cannot
be thoroughly implemented because the capacity limit can
interfere with the process of locating the trafc class of the
maximal revenue rate. If compared to CP
F
, the revenue error
generated by Algorithm 1 can be strictly bounded by the
following inequality:
R
err
< R(CP
F
, B) R
_
CP
F
, B max
1iM
{b
i
}
_
(21)
where R
err
stands for the revenue approximation error, and
R(, C) stands for the revenue obtained from policy with
bandwidth capacity C.
Next, we investigate the complexity of the fairness-
constrained greedy revenue algorithm. Clearly, the rst phase of
this algorithm has a complexity of O(B). Moreover, there are
O(B) iterations in the second phase of this algorithm. During
each iteration, this algorithm searches through M possible
system states to locate the trafc class that yields the maximal
revenue rate. Therefore, the size of the whole searching space
or the complexity of the second phase is O(BM). Combining
phases 1 and 2, we can conclude that this algorithm has a
complexity of O(BM).
It is noted that if we let pb
th
= 1, Algorithm 1 degenerates
into the pure greedy revenue algorithmfor CP

. If we let rer
1
=
rer
2
= = rer
M
= 1, Algorithm 1 degenerates into the
fairness-constrained greedy utility algorithm for CP
F+
. If we
let pb
th
= 1 and rer
1
= rer
2
= = rer
M
= 1, Algorithm 1
degenerates into the pure greedy utility algorithm for CP
+
.
C. Utility- and Fairness-Constrained Greedy
Revenue Algorithm
To develop a CP-structured heuristic algorithm for
UF
, we
have to address three issues: 1) optimal revenue; 2) fairness
constraint; and 3) utility constraint. Since the optimal revenue
and the utility constraint have been discussed in the previous
sections, we focus only on the utility constraint in this section.
Similar to the revenue, for class-i trafc, we dene the
statistical bandwidth brought by the jth server as
SB
i
S
(j,
i
) = b
i
F
i
S
(j,
i
) = b
i

i
[B(j 1,
i
) B(j,
i
)]
(22)
where the relation SB
i
(CP) =

s
i
j=1
SB
i
S
(j,
i
) holds.
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
RONG et al.: CALL ADMISSION CONTROL OPTIMIZATION IN WiMAX NETWORKS 2517
Then, the utility of the jth server is given by
U
i
S
(j,
i
) =
SB
i
S
(j,
i
)
b
i
=F
i
S
(j,
i
)
=
i
[B(j 1,
i
) B(j,
i
)] . (23)
To deal with the utility constraint, we can use (23) to cal-
culate the utility in each iteration step. Combining the utility
constraint with the CP-structured heuristic algorithm for
F
,
we obtain a utility- and fairness-constrained greedy revenue
algorithm to approximate CP
UF
, as presented in Algorithm 2.
Algorithm2: Utility- and Fairness-Constrained Greedy Rev-
enue Algorithm for CP
UF
1) Input u
th
, pb
th
;
2) Capture the trafc load prole in the kth subscribers
local network;
3) Collect the CSI from the physical layer, calculate B
U
k
for UL CAC or B
D
k
for DL CAC, and let B = B
U
k
or
B = B
D
k
;
4) / PHASE 1: Allocate bandwidth resource for fairness
constraint and calculate U
th
/
5) Achieve {B
i
CP
(F), 1 i M} and locate CP
F+
by
running Algorithm 1 with parameter setup rer
1
=
rer
2
= = rer
M
= 1;
6) B
free
= B

M
i=1
B
i
CP
(F); U
th
= u
th
U(CP
F+
);
7) for i = 1 to M do
8) B
i
CP
= B
i
CP
(F); s
i
= B
i
CP
(F)/b
i
; B(s
i
,
i
) =
(
s
i
i
/s
i
!/

s
i
k=0

k
i
/k!); B(s
i
+ 1,
i
) = (
i
B(s
i
,

i
)/s
i
+ 1 +
i
B(s
i
,
i
)); r
i
S
= (r
i

i
[B(s
i
,
i
)
B(s
i
+ 1,
i
)]/b
i
); U
i
S
=
i
[B(s
i
,
i
) B(s
i
+
1,
i
)]; SB
i
=

s
i
j=0
b
i
j(
j
i
/j!/

s
i
k=0

k
i
/k!);
9) end for
10) / PHASE 2: Allocate the remaining free bandwidth
resources according to utility constrained optimal rev-
enue strategy. /
11) T = M + 1;
12) while T > 0 do
13) Set
U
= {i|i satises (U
i
S
b
i
+

M
i=1
SB
i
/(B
B
free
+b
i
)) > U
th
}; / the set of trafc classes
qualied for utility constraint. /
14) T = arg max
iSet
U
{r
i
S
};
15) if b
i
|
i=T
B
free
then
16) B
free
= B
free
b
i
|
i=T
; B
i
CP
|
i=T
= B
i
CP
|
i=T
+
b
i
|
i=T
;
17) SB
i
|
i=T
= SB
i
|
i=T
+U
i
S
b
i
|
i=T
; s
i
|
i=T
=
s
i
|
i=T
+ 1;
18) B(s
i
+ 1,
i
)|
i=T
= (
i
B(s
i
,
i
)/s
i
+1+
i
B(s
i
,

i
))|
i=T
;
19) r
i
S
|
i=T
=(r
i

i
[B(s
i
,
i
)B(s
i
+1,
i
)]/b
i
)|
i=T
;
20) U
i
S
|
i=T
=
i
[B(s
i
,
i
) B(s
i
+ 1,
i
)]|
i=T
;
21) else
22) / Capacity limit begins to take effect. /
23) r
i
S
|
i=T
= 0; U
i
S
|
i=T
= 0;
24) U
th
= 0; / change to use the pure greedy rev-
enue algorithm. /
25) if

M
i=1
r
i
S
= 0 then
26) T = 1; / the algorithm is completed. /
27) end if
28) end if
29) end while
30) Return {B
i
CP
, 1 i M} as the nal bandwidth al-
location decision;
There are two phases in Algorithm 2. In the rst phase,
we employ Algorithm 1 to allocate each trafc class a certain
amount of bandwidth resource such that the fairness constraint
can be guaranteed. In the meantime, Algorithm1 can also locate
CP
F+
, from which U
th
is calculated.
Then, in the second phase, we employ the utility-constrained
optimal revenue strategy to allocate the remaining bandwidth.
To meet the utility constraint, in each iteration of phase 2, only
the trafc classes that can make the utility higher than U
th
are chosen as qualied candidates for bandwidth allocation (as
shown in step 13). Moreover, this algorithm can achieve high
revenues as well, since it always selects the candidate with the
highest revenue rate to assign bandwidth resource (as shown in
step 14). It is noted that if we let pb
th
= 1, Algorithm 2 de-
generates into the utility-constrained greedy revenue algorithm
for CP
U
.
Next, we will address the issues on complexity of the utility-
and fairness-constrained greedy revenue algorithm. Clearly, in
the rst phase, Algorithm 1 is employed, and its complexity is
O(BM). In the second phase, there are O(B) iterations. During
each iteration, this algorithm searches through O(M) possible
system states in Set
U
to locate the trafc class of maximal
revenue rate. Therefore, the size of the whole searching space
or the complexity of the second phase is O(BM). Combining
phases 1 and 2, we can conclude that this algorithm has a
complexity of O(BM).
VI. SIMULATION RESULTS
In this section, we conduct a simulation study to evaluate
the performance of our proposed WiMAX CAC optimization
scheme. The simulations are programmed on the Matlab plat-
form, using the analytical results developed in the previous
sections. In particular, we rst compare the capability of dif-
ferent 1-D CAC policies, then study the 2-D WiMAX CAC in
a subscribers local network, and nally, we show the overall
benet of our proposed WiMAX CAC scheme in a WiMAX
PMP network.
For the 1-D CAC optimization, we present the numerical
results as shown in Figs. 57 to demonstrate the performance
of different CAC policies with three metrics (revenue, utility,
and fairness) while varying the arrival rate of class-3 trafc.
Here,

,
+
,
U
,
F
, and
UF
are obtained by brute-
force searching. In this simulation scenario, the total bandwidth
capacity B is set to be 75 Mb/s, the revenue rate is priced
as rer
UGS
= 5, rer
rtPS
= 2, rer
nrtPS
= 1, and rer
BE
= 0.5,
and the trafc load is congured as those shown in Table II.
Moreover, for the utility constraint, we set u
th
= 90%, and for
the fairness constraint, we set pb
th
= 65%. In Figs. 5 and 6,
the revenue and utility are normalized by R(

) and U(
+
),
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
2518 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 4, JULY 2008
Fig. 5. Revenue of different CAC optimization policies while varying the
arrival rate of class-3 trafc.
Fig. 6. Utility of different CAC optimization policies while varying the arrival
rate of class-3 trafc.
respectively. In Fig. 7, the highest blocking probability remains
unchanged.
As shown in Figs. 57,

/
+
gives a good performance
only in terms of revenue/utility. On the other hand,
U
/
F
performs well in terms of both revenue and utility/fairness.
Finally,
UF
satisfactorily performs in terms of all three
metrics, including revenue, utility, and fairness.
To reduce the algorithm complexity, we have developed
a utility- and fairness-constrained greedy revenue algorithm
for
UF
. This approximation algorithm can also be utilized
to approximate

,
+
,
U
, and
F
if we appropriately
degenerate the utility or fairness constraint. Numerical results
are shown in Table III to illustrate the average error of the
approximation algorithm, whereas the trafc load is congured
the same as that listed in Table II. For convenience, the revenue
and utility approximation errors are normalized by R(

) and
U(
+
), respectively. We can conclude from Table III that our
approximation algorithm can provide an adequate approxima-
tion to the exact solution. Here, the CP error means the er-
ror generated from the CP-structured solution if compared with
Fig. 7. Highest blocking probability of different CAC optimization policies
while varying the arrival rate of class-3 trafc.
the exact solution, the CP approximation error means the error
generated from the approximation algorithm if compared with
the CP-structured solution, and the total approximation error
means the sum of the CP error and the CP approximation
error.
Next, we evaluate the performance of 2-D WiMAX CAC in
a subscribers local network using the decomposing approach
proposed in Section II. In the simulation scenario, we em-
phasize the utility- and fairness-constrained optimal revenue
policy
UF
and its approximation algorithm, because they
take into account all the requirements of service providers and
subscribers. Considering the parameter decomposing model
given in Table I, we suppose that UL CAC and DL CAC employ
the same policy, i.e.,
UF
or its approximation algorithm, with
normalized utility threshold u
th
and normalized blocking prob-
ability threshold pb
th
identically congured for UL and DL.
Numerical results are shown in Figs. 8 and 9 to demonstrate the
performance of
UF
and its approximation algorithm based
2-D WiMAX CAC while varying u
th
and keeping pb
th
con-
stant at 65%. In the simulation, we set the UL/DL bandwidth
capacity in a subscribers local network to be 60/75 Mb/s.
Moreover, the UL trafc load is congured the same as that
given in Table IV, whereas the DL trafc load is congured in
Table II, except that the arrival rate of class-3 trafc is xed
to be 64 calls/h. The revenue rates of both UL CAC and DL
CAC are priced as rer
UGS
= 5, rer
rtPS
= 2, rer
nrtPS
= 1,
and rer
BE
= 0.5.
Fig. 8 illustrates the revenue and utility of 2-D WiMAX
CAC, which can be derived from the UL/DL revenue and utility
as follows: 1) The revenue of 2-D CAC is dened as the sum
of the UL revenue and the DL revenue, and 2) the utility of
2-D CAC is dened as the average of the UL utility and the
DL utility. For analytical simplicity, in Fig. 8, the revenue is
normalized by that of the
F
based 2-D CAC, and the utility
is normalized by that of the
F+
based 2-D CAC.
As illustrated in Fig. 8, when u
th
= 0,
UF
turns to be
F
,
yielding a solution of high revenue but low utility. Similarly,
when u
th
= 1,
UF
turns to be
F+
, yielding a solution
with high utility but low revenue. Thus, we should choose an
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
RONG et al.: CALL ADMISSION CONTROL OPTIMIZATION IN WiMAX NETWORKS 2519
TABLE II
TRAFFIC LOAD CONFIGURATION
TABLE III
AVERAGE ERROR OF THE APPROXIMATION ALGORITHM (REVENUE/UTILITY/HIGHEST BLOCKING PROBABILITY)
Fig. 8. Revenue and utility of
UF
and its approximation algorithm in
2-D CAC.
Fig. 9. Highest blocking probability of
UF
and its approximation algo-
rithm in 2-D CAC.
appropriate value for u
th
(i.e., 90% in this case) to give
UF
balanced revenue and utility. In addition, Fig. 8 also shows that
the approximation algorithm has similar performance as
UF
.
The fairness feature of 2-D WiMAX CAC is illustrated in
Fig. 9, which indicates that the highest blocking probability
of
UF
and its approximation-algorithm-based 2-D CAC is
strictly bounded by the blocking probability threshold derived
from pb
th
.
Finally, we investigate the performance of
UF
and its
approximation-algorithm-based 2-D CAC in a WiMAX PMP
network, which employs a OFDMATDD mode of 32 sub-
scribers with PUSC on the UL and FUSC on the DL. In our
simulations, the UL and DL channels are assumed to have
a bad-urban delay prole [29] and suffer from shadowing
with 8-dB standard deviation. Let the amount of available
subcarriers be 1024 and each subcarrier occupy 10 kHz of
physical bandwidth. The distances between the subscribers and
the base station are randomly chosen from 2 to 10 km, and the
acceptable BER is set to be 10
6
. We assign each subcarrier the
same UL and DL power resource and congure the TDD DL
proportion factor % as 60%.
As for the trafc pattern, 32 subscribers are programmed to
have different UL and DL trafc loads, which are uniformly
distributed in [40 Mb/s, 100 Mb/s] and [60 Mb/s, 140 Mb/s],
respectively. To emulate an environment of broadband wireless
access, we suppose that the WiMAX network is dominated
by multimedia applications. Correspondingly, among the UL
and DL trafc loads of a certain subscriber, the proportions of
UGS, rtPS, nrtPS, and BE trafc, which are denoted by PP
UGS
,
PP
rtPS
, PP
nrtPS
, and PP
BE
, are set to be random variables,
which are dened as follows:
PP
BE
uniformly distributed in [10%, 30%];
PP
UGS
uniformly distributed in [10%(1 PP
BE
),
30%(1 PP
BE
)];
PP
rtPS
uniformly distributed in [20%(1 PP
BE
),
60%(1 PP
BE
)];
PP
nrtPS
dened as (1 PP
BE
PP
UGS
PP
rtPS
).
As for the revenue rate, we assume that rer
UGS
= 5,
rer
rtPS
= 2, rer
nrtPS
= 1, and rer
BE
= 0.5. As for the utility
constraint and the fairness constraint, we set u
th
= 90% and
pb
th
= 65% for both UL CAC and DL CAC.
Simulation results are presented in Figs. 10 and 11 to demon-
strate the average performance of the 2-D CAC using
UF
or its approximation algorithm among 32 subscribers. As men-
tioned earlier,

and
+
have the best performance in terms
of revenue and utility, respectively. To facilitate the analysis,
in Fig. 10, we use the average revenue/utility of

/
+
to
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
2520 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 4, JULY 2008
TABLE IV
UPLINK TRAFFIC LOAD CONFIGURATION
Fig. 10. Average revenue and utility of
UF
and its approximation algo-
rithm in 2-D WiMAX CAC of 32 subscribers.
Fig. 11. Average highest blocking probability of
UF
and its approximation
algorithm in 2-D WiMAX CAC of 32 subscribers.
normalize the average revenue/utility. Figs. 10 and 11 show that
the 2-D CAC with
UF
, or its approximation algorithm, can
achieve high revenue and utility while still meeting the fairness
constraint.
VII. CONCLUSION
In this paper, we have proposed a framework for WiMAX
CAC, in which the 2-D CAC problem is decomposed into
two independent 1-D CAC problems. Then, we make the
1-D CAC an optimization problem and evaluate its different
strategies. From the perspective of service providers, optimal
revenue is the major concern. However, from the perspective of
subscribers, optimal utility and fairness are the requirements.
To successfully deploy a WiMAX system, we have to take
into account the expectations of both service providers and
subscribers. Accordingly, we develop a utility- and fairness-
constrained optimal revenue policy, as well as its approximation
algorithm. Simulation results verify that our proposed WiMAX
CAC approach can meet the requirements of both service
providers and subscribers.
REFERENCES
[1] S. J. Vaughan-Nichols, Achieving wireless broadband with WiMAX,
Computer, vol. 37, no. 6, pp. 1013, Jun. 2004.
[2] R. Marks, C. Eklund, K. Stanwood, and S. Wang, The 802.16 Wireless-
MAN MAC: Its Done, but What Is It?, Nov. 2001. Tutorial on the IEEE
802.16-01/58r1.
[3] K. W. Ross and D. H. K. Tsang, The stochastic knapsack problem, IEEE
Trans. Commun., vol. 37, no. 7, pp. 884895, Jul. 1989.
[4] K. W. Ross and D. D. Yao, Monotonicity properties for the stochas-
tic knapsack, IEEE Trans. Inf. Theory, vol. 36, no. 5, pp. 11731179,
Sep. 1990.
[5] A. Gavious and Z. Rosberg, A restricted complete sharing policy for a
stochastic knapsack problem in B-ISDN, IEEE Trans. Commun., vol. 42,
no. 7, pp. 23752379, Jul. 1994.
[6] E. Altman, T. Jimenez, and G. Koole, On optimal call admission con-
trol in resource-sharing system, IEEE Trans. Commun., vol. 49, no. 9,
pp. 16591668, Sep. 2001.
[7] E. L. Ormeci, Dynamic admission control in a call center with one shared
and two dedicated service facilities, IEEE Trans. Autom. Control, vol. 49,
no. 7, pp. 11571161, Jul. 2004.
[8] B. C. Dean, M. X. Goemans, and J. Vondrdk, Approximating the sto-
chastic knapsack problem: The benet of adaptivity, in Proc. 45th Annu.
IEEE Symp. Found. Comput. Sci., Oct. 2004, pp. 208217.
[9] V. Sarangan, D. Ghosh, N. Gautam, and R. Acharya, Steady state distri-
bution for stochastic knapsack with bursty arrivals, IEEE Commun. Lett.,
vol. 9, no. 2, pp. 187189, Feb. 2005.
[10] C. C. Beard and V. S. Frost, Prioritized resource allocation for stressed
networks, IEEE/ACM Trans. Netw., vol. 6, no. 5, pp. 618633, Oct. 2001.
[11] IEEE Standard for Local and Metropolitan Area NetworksPart 16:
Air Interface for Fixed Broadband Wireless Access Systems, IEEE 802.
16-2004, Oct. 2004.
[12] I. Koffman and V. Roman, Broadband wireless access solutions based
on OFDM access in IEEE 802.16, IEEE Commun. Mag., vol. 40, no. 4,
pp. 96103, Apr. 2002.
[13] B. Rong, Y. Qian, and K. Lu, Integrated downlink resource management
for multiservice WiMAX networks, IEEE Trans. Mobile Comput., vol. 6,
no. 6, pp. 621632, Jun. 2007.
[14] B. Rong, Y. Qian, and K. Lu, Downlink call admission control in
multiservice WiMAX networks, in Proc. IEEE ICC, Jun. 2007,
pp. 50825087.
[15] S. Sengupta, M. Chatterjee, S. Ganguly, and R. Izmailov, Improving
R-score of VoIP streams over WiMax, in Proc. IEEE ICC, Jun. 2006,
vol. 2, pp. 866871.
[16] A. Sang and S. Li, A predictability analysis of network trafc, in Proc.
IEEE Infocom, Mar. 2630, 2000, pp. 342351.
[17] B. Rong, B. Tremblay, M. Bennani, and M. Kadoch, Integrating trafc
aggregation mechanism into SIP based IP telephony over MPLS net-
work, in Proc. IEEE Globecom. St. Louis, MO, Nov. 28Dec. 2, 2005,
pp. 797801.
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
RONG et al.: CALL ADMISSION CONTROL OPTIMIZATION IN WiMAX NETWORKS 2521
[18] X. Qiu and K. Chawla, On the performance of adaptive modulation in
cellular systems, IEEE Trans. Commun., vol. 47, no. 6, pp. 884895,
Jun. 1999.
[19] G. Song and Y. G. Li, Cross-layer optimization for OFDM wire-
less networksPart I: Theoretical framework, IEEE Trans. Wireless
Commun., vol. 4, no. 2, pp. 614624, Mar. 2005.
[20] G. Song and Y. G. Li, Cross-layer optimization for OFDM wire-
less networksPart II: Algorithm development, IEEE Trans. Wireless
Commun., vol. 4, no. 2, pp. 625634, Mar. 2005.
[21] J. S. Kaufman, Blocking in a shared resource environment, IEEE Trans.
Commun., vol. COM-29, no. 10, pp. 14741481, Oct. 1981.
[22] S. Shenker, Fundamental design issues for the future Internet, IEEE J.
Sel. Areas Commun., vol. 13, no. 7, pp. 11761188, Sep. 1995.
[23] Z. Jiang, Y. Ge, and Y. Li, Max-utility wireless resource management for
best-effort trafc, IEEE Trans. Wireless Commun., vol. 4, no. 1, pp. 100
111, Jan. 2005.
[24] Y.-C. Lai and Y.-D. Lin, A novel admission control for fairly admitting
wideband and narrowband calls, IEEE Commun. Lett., vol. 7, no. 4,
pp. 186188, Apr. 2003.
[25] R.-H. Hwang and C.-F. Chi, Fairness in QoS guaranteed networks com-
munications, in Proc. IEEE ICC, May 2003, vol. 1, pp. 218222.
[26] D. Jagerman, Some properties of the Erlang loss function, Bell Syst.
Tech. J., vol. 53, no. 3, pp. 525551, Mar. 1974.
[27] A. Jensen, Moes Principle. Copenhagen, Denmark: Copenhagen Tele-
phone Co. (K.T.A.S.), 1950.
[28] R. W. Wolff and C.-L. Wang, On the convexity of loss probabilities,
J. Appl. Probab., vol. 39, no. 2, pp. 402406, Jun. 2002.
[29] G. L. Stuber, Principles of Mobile Communication, 2nd ed. Norwell,
MA: Kluwer, 2000.
Bo Rong (M07) received the B.S. degree from
Shandong University, Jinan, China, in 1993, the M.S.
degree from the Beijing University of Aeronautics
and Astronautics, Beijing, China, in 1997, and the
Ph.D. degree from the Beijing University of Posts
and Telecommunications in 2001.
After receiving the Ph.D. degree, he was a Soft-
ware Engineer with a startup company in Beijing
for one year. Then, he was a Postdoctoral Fellow
with the Department of Electrical Engineering, Ecole
de Technologie Superieure, Universite du Quebec,
Quebec City, QC, Canada, for three years. He is currently a Postdoctoral Fellow
with the Department of Electrical and Computer Engineering, University of
Puerto Rico at Mayagez. His current research interests focus on modeling,
simulation, and performance analysis for next-generation wireless networks.
Yi Qian (M95SM07) received the Ph.D. degree
in electrical engineering with a concentration in
telecommunication networks from Clemson Univer-
sity, Clemson, SC.
He is currently with the National Institute of Stan-
dards and Technology, Gaithersburg, MD. He was an
Assistant Professor with the Department of Electri-
cal and Computer Engineering, University of Puerto
Rico at Mayagez (UPRM), from July 2003 to July
2007. At UPRM, he regularly taught courses on
wireless networks, network design, network manage-
ment, and network performance analysis. Prior to joining UPRM in July 2003,
he worked for several start-up companies and consulting rms in the areas of
voice over IP, ber optical switching, Internet packet video, network optimiza-
tions, and network planning as a Technical Advisor and a Senior Consultant. He
was also with the Wireless Systems Engineering Department, Nortel Networks,
Richardson, TX, as a Senior Member of Scientic Staff and as a Technical
Advisor for several years. While with Nortel Networks, he was a Project Leader
for various wireless and satellite network product design projects, customer
consulting projects, and advanced technology research projects. He was also in
charge of wireless standard development and evaluations. He has publications
and is a holder of patents in all these areas. He is a coauthor of the book
Information Assurance: Dependability and Security in Networked Systems
(Morgan Kaufmann, 2008). His current research interests include network secu-
rity, network design, network modeling, simulations and performance analysis
for next-generation wireless networks, wireless sensor networks, broadband
satellite networks, optical networks, high-speed networks, and the Internet.
Dr. Qian is a member of the Association for Computing Machinery.
Kejie Lu (S01M04SM07) received the B.S.
and M.S. degrees in telecommunications engineering
from the Beijing University of Posts and Telecom-
munications, Beijing, China, in 1994 and 1997,
respectively, and the Ph.D. degree in electrical en-
gineering from the University of Texas at Dallas,
Richardson, in 2003.
From 2004 and 2005, he was a Postdoctoral Re-
search Associate with the Department of Electrical
and Computer Engineering, University of Florida,
Gainesville. Since July 2005, he has been an As-
sistant Professor with the Department of Electrical and Computer Engineer-
ing, University of Puerto Rico at Mayagez. His research interests include
architecture and protocol design for computer and communication networks,
performance analysis, network security, and wireless communications.
Hsiao-Hwa Chen (S89M91SM00) received the
B.Sc. and M.Sc. degrees from Zhejiang University,
Zhejiang, China, in 1982 and 1985, respectively, and
the Ph.D. degree from the University of Oulu, Oulu,
Finland, in 1991.
He is currently a Full Professor with the Depart-
ment of Engineering Science, National Cheng Kung
University, Tainan, Taiwan, R.O.C. He is the author
or a coauthor of more than 250 technical papers in
major international journals and conference proceed-
ings and ve books and three book chapters in the
areas of communications. He served or is currently serving as an Editorial
Board Member and/or Guest Editor of the Wireless Communications and
Mobile Computing Journal and the International Journal of Communication
Systems. He is the founding Editor-in-Chief of the Security and Communication
Networks Journal.
Dr. Chen has served as the Symposium Cochair of major international
conferences, including the IEEE Vehicular Technology Conference (VTC), the
IEEE International Conference on Communications (ICC), the IEEE Global
Communications Conference (Globecom), and the IEEE Wireless Communica-
tions and Networking Conference (WCNC). He served or is currently serving
as an Editorial Board Member and/or Guest Editor of IEEE Communications
Magazine, the IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS,
IEEE Wireless Communications Magazine, the IEEE TRANSACTIONS ON
WIRELESS COMMUNICATIONS, and IEEE Vehicular Technology Magazine.
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.
2522 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 4, JULY 2008
Mohsen Guizani (S87M90SM98) received the
B.S. (with distinction) and M.S. degrees in elec-
trical engineering and the M.S. and Ph.D. degrees
in computer engineering from Syracuse University,
Syracuse, NY, in 1984, 1986, 1987, and 1990,
respectively,
He is currently a Full Professor and Chair of the
Department of Computer Science, Western Michigan
University, Kalamazoo. He was the Chair of the De-
partment of Computer Science, University of West
Florida, Pensacola, from 1999 to 2003. He was an
Associate Professor of electrical and computer engineering and the Director of
graduate studies with the University of Missouri, Columbia, from 1997 to 1999.
Prior to joining the University of Missouri, he was a Research Fellow with
the University of Colorado at Boulder. From 1989 to 1996, he held academic
positions at the Computer Engineering Department, University of Petroleum
and Minerals, Dhahran, Saudi Arabia. He has more than 140 publications
in refereed journals and conference proceedings in the areas of high-speed
networking, optical networking, and wireless networking and communications.
He currently serves on the editorial boards of many national and international
journals, such as the Journal of Parallel and Distributed Systems and Networks
and the International Journal of Computer Research. He was a Guest Editor for
the Journal of Communications and Networks and several other publications.
He is the Founder and Editor-in-Chief of Wireless Communications and Mobile
Computing.
Dr. Guizani currently serves on the editorial boards of the IEEE
TRANSACTIONS ON VEHICULAR TECHNOLOGY and IEEE Communications
Magazine. He has served as a Guest Editor for the IEEE Communications Mag-
azine and the IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS.
He served as the General Chair of the International Conference Parallel and
Distributed Computing Systems (PDCS) in 2002 and 2003, the IEEE Vehicular
Technology Conference (VTC) in 2003, and the International Conference on
Wireless Networks, Communications, and Mobile Computing (WirelessCom)
in 2005. He has also served as the Program Chair for many conferences. He was
designated by the IEEE Computer Society as a Distinguished National Speaker
through 2005.
Authorized licensed use limited to: GMR Institute of Technology. Downloaded on December 7, 2009 at 04:21 from IEEE Xplore. Restrictions apply.

Вам также может понравиться