Você está na página 1de 10

Permission to make digital or hard copies of all or part of this work for

personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
QShine06 The Third International Conference on Quality of Service in
Heterogeneous Wired/Wireless Networks
August 7-9 2006, Waterloo, ON, Canada
2006 ACM 1-59593-472-3/06/08$5.00
A Game-Theoretic Approach to Bandwidth Allocation and Admission Control
for Polling Services in IEEE 802.16 Broadband Wireless Networks

Dusit Niyato and Ekram Hossain
Dept. of Elect. and Comp. Engg., University of Manitoba
Winnipeg, MB R3T 5V6, Canada
{tao, ekram}@ee.umanitoba.ca
Abstract
In this paper, we propose an adaptive bandwidth alloca-
tion (BA) and connection admission control (CAC) mech-
anism based on game theory for polling services in IEEE
802.16 broadband wireless networks. A noncooperative
two-person general-sumgame is formulated where the base
station and a new connection are the players of this game.
The game formulation provides not only the decision on ac-
cepting or rejecting a connection, but also the amount of
bandwidth allocated to a new connection (if admitted). A
queueing model considering adaptive modulation and cod-
ing (AMC) in the physical layer is used to analyze quality of
service (QoS) performances, namely, delay and throughput
performances, respectively, for real-time and non-real-time
polling services. This queueing model is used by the pro-
posed bandwidth allocation and admission control game to
ensure that the payoffs for both the base station and the
new connection are maximized. The performance of the
proposed scheme is evaluated by simulation and compared
with that of traditional admission control with static and
adaptive bandwidth allocation.
Keywords- IEEE 802.16, bandwidth allocation, admis-
sion control, game theory.
1 Introduction
Broadband wireless access (BWA) based on the IEEE
802.16 technology [1] is a promising technique for last-mile
access. IEEE 802.16 standard has been proposed to provide
high-speed broadband wireless connectivity through a pre-
dened quality of service (QoS) framework for multimedia
trafc. Even though the physical layer specications and
the medium access control (MAC) protocol signaling are

This work was supported by a scholarship from the TRLabs, Win-


nipeg, Canada, and in part by a grant from the Natural Sciences and Engi-
neering Research Council (NSERC) of Canada.
well dened in the standard, the resource allocation and ad-
mission control policies for the IEEE 802.16 air-interface
remain as open issues.
Bandwidth allocation and connection admission control
are classical problems in wireless systems. While band-
width allocation is responsible for allocating scarce wire-
less resource to ongoing and incoming connections such
that their QoS requirements are satised, admission con-
trol is applied to prevent overwhelming of limited radio re-
source by accepting too many connections. The resource
allocation and admission control problems for the cellular
wireless networks such as TDMA and FDMA-based cellu-
lar wireless systems [2] were studied in the literature.
Game-theoretic models were used for admission control
in wireless systems. In [3], a game-theoretic framework for
guard bandwidth reservation in a cellular wireless network
was proposed. In this work, channel allocation for handoff
call is chosen based on equilibriumamong different types of
trafc. In [4], an integrated admission control and rate con-
trol method using noncooperative game was proposed for
CDMA wireless networks in which user churning among
the different service providers was considered. In [5], the
admission control problem in a IEEE 802.11 wireless LAN
was posed as a noncooperative game. The game is played
by the access point and an arriving mobile station. The
objective is to maximize service providers revenue while
maintaining the QoS of the ongoing ows at the desired
level. However, these game-theoretic models for admission
provide only the decision on accepting or rejecting a new
connection while ignoring the bandwidth allocation issue.
The polling service classes dened in the IEEE 802.16
(i.e., real-time and non-real-time polling services) require
strict delay and throughput guarantee. However, most of
the bandwidth allocation and admission control schemes in
the literature did not consider the delay requirement explic-
itly and also did not consider the important physical layer
aspects such as adaptive modulation and coding as well as
channel fading.
Packet delay performance largely depends on the radio
link level queue management, scheduling and retransmis-
sion policies. The problem of analyzing radio link level
queueing (and hence link level delay) under wireless packet
transmission was addressed in the literature. In [6], a
Markov-based model was presented to analyze the radio
link level packet dropping process under automatic repeat
request (ARQ)-based error control. In [7], a cross-layer an-
alytical model was presented to derive packet loss rate, av-
erage throughput and delay under adaptive modulation and
coding (AMC). However, these works did not consider the
radio link level queueing performances in a resource alloca-
tion and admission control problem framework.
In this paper, we propose a bandwidth allocation and ad-
mission control method based on queueing analysis and a
game-theoretic formulation. The queueing model is used to
calculate the throughput and the delay performances to be
used by the game-theoretic model to decide whether a new
connection can be accepted or not. For the queueing analy-
sis, adaptive modulation and coding scheme as dened in
the IEEE 802.16 standard is assumed in the physical layer.
For the radio-link level queues, realistic trafc parameters
(i.e., peak rate and probability of peak rate) are considered.
The objective of the proposed game-theoretic model is
to nd the equilibrium point between the base station and a
new connection. The conict in this game arises due to the
fact that constrained by limited radio resources (i.e., band-
width), the base station wants to maximize its utility (e.g.,
revenue) from the ongoing connections by providing higher
level of QoS to these connections, while a new connection
wants to achieve the highest possible QoS performance as
well. Among the available strategies of both base station
and new connection, the Nash equilibrium is determined
by using the best response function and the decision on ad-
mission control is made based on admissible strategy pair
from the Nash equilibrium. The connection-level and the
packet-level performances for the proposed bandwidth allo-
cation and admission control scheme are investigated. Also,
the performances are compared with those of the traditional
bandwidth allocation and admission control schemes.
2 Overview of the IEEE 802.16 Air-interface
and the System Model
2.1 IEEE 802.16 Air Interface
In the IEEE 802.16 architecture [1], there are two
types of stationary stations: subscriber station (SS) and
base station (BS). The BS governs all communications
in the network. The physical layer of IEEE 802.16 op-
erates at 10-66 GHz (IEEE 802.16) or 2-11 GHz (IEEE
802.16a) band and support data rate in the range of 32-
130 Mbps depending on the operational bandwidth as well
as the modulation and coding schemes. In the 10-66
GHz band, the signal propagation between a BS and an
SS should be line-of-sight and single-carrier modulation is
used. WirelessMAN-SC is the air interface specication for
IEEE 802.16 operating in this frequency band. In contrast,
IEEE 802.16a operates in the 2-11 GHz band and supports
non-line-of-sight communication. The air-interface spec-
ication for 802.16a are: WirelessMAN-SC2 for single-
carrier modulation, WirelessMAN-OFDM for orthogonal
frequency-division multiplexing (OFDM) with TDMA ac-
cess scheme, and WirelessMAN-OFDMA for orthogonal
frequency-division multiple access (OFDMA) scheme.
The system uses a transmission frame size of 0.5, 1, or
2 ms and a frame is divided into subframes for downlink
and uplink transmissions. While time-division multiplex-
ing (TDM) is used for downlink transmission, time-division
multiple access (TDMA) is used for uplink transmission.
These subframes are composed of transmission bursts (i.e.,
uplink and downlink bursts) which carry medium access
control (MAC) information and users data. The transmis-
sion bursts corresponding to a particular SS are separated
from each other by a preamble eld and contains several
MAC protocol data units (PDUs). The lengths of uplink and
downlink subframes are determined dynamically by the BS,
and the synchronization between SSs and BS is done based
on pre-dened time slot.
On the downlink, the BS broadcasts data to all SSs in
the same network. Each SS processes only the MAC PDUs
containing its own connection identier (CID) and discards
the other PDUs. IEEE 802.16 MAC supports two classes of
SS: grant per connection (GPC) and grant per SS (GPSS).
In case of GPC, bandwidth is granted to a connection indi-
vidually. In contrast, for GPSS, a portion of the available
bandwidth is granted to each of the SSs, and each SS is re-
sponsible for allocating bandwidth among the correspond-
ing connections. In the uplink direction, each SS can re-
quest bandwidth to BS by using BW-request PDU.
2.2 Service Types in IEEE 802.16
IEEE 802.16 standard denes the following four types of
services each of which has different QoS requirements:
Unsolicited grant service (UGS) supports constant-bit-
rate (CBR) trafc.
Real-time polling service (rtPS) supports real-time
trafc in which delay is an important QoS requirement.
The amount of bandwidth required for this type of ser-
vice is determined based on the required QoS perfor-
mances (i.e., delay), the channel quality and the trafc
arrival rates at the sources.
Non-real-time polling service (nrtPS) requires QoS
guarantee which is not as tight as that for rtPS. This
Table 1. Modulation and coding schemes for
IEEE 802.16.
Rate ID Modulation
level (coding)
Information
bits/symbol
Required
SNR
0 BPSK (1/2) 0.5 6.4
1 QPSK (1/2) 1 9.4
2 QPSK (3/4) 1.5 11.2
3 16QAM (1/2) 2 16.4
4 16QAM (3/4) 3 18.2
5 64QAM (2/3) 4 22.7
6 64QAM (3/4) 4.5 24.4
is suitable for applications such as le transfer with
guaranteed throughput. The bandwidth allocation is
also adaptive as in the case of rtPS.
Best-effort service (BE) is for best-effort trafc with no
QoS guarantee.
2.3 System Model
We consider a single BS serving multiple connections
from different SSs through the TDMA/TDD access mode
using single carrier modulation (e.g., as in WirelessMAN-
SC). For each connection, a separate queue (with size of X
PDUs) is maintained for buffering the PDUs fromthe corre-
sponding application. We consider GPC type of SSs; that is,
a certain amount of bandwidth is reserved for each connec-
tion when the bandwidth allocation and connection admis-
sion control is performed. Adaptive modulation and cod-
ing is used to adjust the transmission rate adaptively in each
frame according to the channel quality. The modulation and
coding schemes for the IEEE 802.16 air-interface are shown
in Table 1. We consider the performance measures for up-
link transmission. However, the same model can be used
for analyzing the performance of downlink transmission.
3 Queueing Formulation
3.1 Trac Source and Arrival Probability
Matrix
To capture the burstiness in trafc arrival we use Markov
modulated Poisson process (MMPP) to model a trafc
source. MMPP is able to model multimedia trafc as
well as Internet trafc [8]. With MMPP, the PDU arrival
rate
s
is determined by the phase s of the corresponding
Markov chain, where the total number of phases is S (i.e.,
s = 1, 2, . . . , S). An MMPP can be represented by matrices
M and , in which the former is the transition probability
matrix of the modulating Markov chain, and the latter is
the matrix corresponding to the Poisson arrival rates. Since
we consider a trafc source with normal and peak rates, the
number of phases of the MMPP is two (i.e., S = 2) and the
matrices can be dened as follows:
M =

1 P
peak
P
peak
1 P
peak
P
peak

, =

(1)
where P
peak
is the probability of peak arrival rate, is the
normal PDU arrival rate and
p
is the peak rate.
A discrete-time MMPP (dMMPP) [9] is equivalent to
MMPP in the continuous time. In this case, the rate ma-
trix is represented by the diagonal probability matrix
a
when the number of PDUs arriving in one frame is a. Each
element of
a
can be obtained as follows:

a
=

f
a
()
f
a
(
p
)

(2)
where the probability that a PDUs arrive during time inter-
val t (i.e., frame length) with mean rate is given by
f
a
() =
e
t
(t)
a
a!
(3)
where a {0, 1, . . . , A 1} and A is the maximum batch
size for PDU arrival. The complementary cumulative prob-
ability mass function for this arrival process is given by
f
A
(, t) =

a=A
f
a
(, t).
3.2 Channel Model and Transmission
Probability Matrix
We consider a Nakagami-m channel model in which the
channel quality is determined by instantaneous received
SNR . With adaptive modulation, the SNR at the re-
ceiver is divided into N +1 non-overlapping intervals (i.e.,
N = 7 in the IEEE 802.16 specications) by thresholds
n
(n {0, 1, . . . , N}) where
0
<
1
< . . . <
N+1
= .
The channel is said to be in state n (i.e., rate ID n will
be used) if
n
<
n+1
. To avoid possible trans-
mission error, no PDU is transmitted when <
0
. Note
that, these thresholds correspond to the required SNR spec-
ied in IEEE 802.16 standard as shown in Table 1. With
Nakagami-m fading, the probability of using rate ID n (i.e.,
Pr(n)) is given by
Pr(n) =
(m, m
n
/) (m, m
n+1
/)
(m)
(4)
where is the average SNR, m is the Nakagami fading
parameter (m 0.5), (m) is the Gamma function, and
(m, ) is the complementary incomplete Gamma func-
tion.
In this paper, bandwidth b is dened as the number of
PDUs that can be transmitted in one frame using rate ID =
0. For a given amount of bandwidth and the transmission
rate ID, the number of transmitted PDUs can be calculated
from the information bits per symbol. For example, with
b = 1, if rate ID = 1, two PDUs can be transmitted in one
frame. Similarly, with rate ID = 6, 9 PDUs (i.e., 2 4.5)
will be transmitted in one frame. We assume that the chan-
nel condition for a connection remains static over a frame
interval ( 2 ms) and all PDUs corresponding to a connec-
tion transmitted during one frame interval use the same rate
ID.
We can dene row matrix D
b
whose elements d
k+1
cor-
respond to the probability of transmitting k PDUs in one
frame, with b units of bandwidth, as follows:
D
b
=

d
0
d
k
d
9b

(5)
where
d
(In2b)
= Pr(n) (6)
in which I
n
denotes the number of information bits per
symbol corresponding to transmission rate ID n, and d
0
=
1

9b
k=1
d
k
. With b units of bandwidth, the throughput for
a connection can be obtained as follows:
=
9b

k=1
k d
k
. (7)
3.3 State Space and Transition Matrix
For each connection, a separate queue with size of X
PDUs is used for buffering data from the corresponding ap-
plication. The state of a queue (i.e., the number of PDUs
in the queue and the phase of arrival) is observed at the be-
ginning of each frame. A PDU arriving in frame f will
not be transmitted until the next frame f + 1 at the earli-
est. The state space of a queue can be dened as follows:
= {(X , M ), 0 X X, M {1, 2}} where X and M
indicate the number of PDUs in the queue and the phase of
MMPP arrival. The transition matrix P of a queue can be
expressed as follows: P =

p
0,0
p
0,A
.
.
.
.
.
.
.
.
.
.
.
.
p
U,0
p
U,U
p
U,U+A
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
p
x,xU
p
x,x
p
x,x+A
.
.
.
.
.
.
.
.
.
.
.
.
p
X,XU
p
X,X

.
The element p
x,x
inside this matrix denotes the probability
matrix for the case when the number of PDUs in the queue
changes from x in the current frame to x

in the next frame.


Since in one frame several PDUs can arrive and be trans-
mitted, this matrix P is divided into three parts. The rst
part, from row 0 to U 1, indicates the case that the max-
imum total transmission rate is greater than the number of
PDUs in the tagged queue and none of the incoming PDUs
is dropped. The second part, from row U to X A, repre-
sents the case in which the maximumPDUtransmission rate
is equal to or less than the number of PDUs in the queue and
none of the incoming PDUs is dropped. Since the size of
queue is nite, some of the arriving PDUs will be dropped
due to the lack of buffer space. The third part, from row
X A+1 to X, indicates the case that some of the incom-
ing PDUs are dropped due to the lack of space in the queue.
Let D
(x)
denote transmission probability when there are x
PDUs in the queue and it can be obtained from
D
(x)
=

d
0
d
U

(8)
where U

= min (x, U) and


d
U
=

d
U
, U

= U

U
k=x
d
k
, U

= x.
(9)
Note that, the maximum total PDU transmission rate can
be greater than the number of PDUs in the queue, while
the maximumnumber of transmitted PDUs cannot be larger
than the number of PDUs in the queue.
The elements in the rst and the second parts of matrix
P can be obtained as follows:
p
x,xu
= M


kj=u

D
(x)

k+1
I
2

(10)
for u = 1, . . . , U

p
x,x+v
= M


jk=v

D
(x)

k+1
I
2

(11)
for v = 1, . . . , A
p
x,x
= M

k=j

D
(x)

k+1
I
2

(12)
where k {0, 1, 2, . . . , U

} and j {0, 1, 2, . . . , A} rep-


resent the number of departed and arriving PDUs, respec-
tively, and I
2
is an identity matrix with size 2 2. Note
that,

D
(x)

k+1
indicates the element at column k + 1 of
row matrix D
(x)
.
Considering both the PDU arrival and the PDU departure
events, (10), (11) and (12) above represent the transition
probability matrices for the cases when the number of PDUs
in the queue decreases by u, increases by v, and does not
change, respectively.
The third part of matrix P ({x = X A + 1, X A +
2, . . . , X}) has to capture the PDU dropping effect. There-
fore, for x + v X, (11) becomes
p
x,x+v
=
A

i=v
p
x,x+i
for x + v X (13)
and for x = X, (12) becomes
p
x,x
= p
x1,x1
+
A

i=1
p
x,x+i
for x = X. (14)
Eqs. (13) and (14) indicate the case that the queue will be
full if the number of incoming PDUs is greater than the
available space in the queue.
3.4 QoS Measures
To obtain the performance measures, the steady state
probabilities for the queue would be required. Since the size
of the queue is nite (i.e., X < ), the probability matrix
is obtained by solving the equations P = and 1 = 1,
where 1 is a column matrix of ones. The matrix contains
the steady state probabilities for the number of PDUs in the
queue and the phases of the MMPP trafc source. This ma-
trix can be decomposed into (x, s) (i.e., the steady state
probability that there are x PDUs in the queue and the phase
of the MMPP arrival is s) as follows:
(x, s) = []
2x+s
. (15)
Here, s = 1 and s = 2 indicate that the trafc source trans-
mits with normal and peak rate, respectively.
3.4.1 Average Number of PDUs in a Tagged queue
The average number of PDUs in a tagged queue is obtained
as follows:
x =
X

x=1
x

s=1
(x, s)

. (16)
3.4.2 PDU Dropping Probability
It refers to the probability that an incoming PDU will be
dropped due to the unavailability of buffer space. It can
be derived from the average number of dropped PDUs per
frame following the method used in [6]. Given that there are
x PDUs in the queue and the number of PDUs in the queue
increases by v, the number of dropped PDUs is v (Xx)
for v > X x, and zero otherwise. The average number of
dropped PDUs per frame is obtained as follows:
x
drop
=
2

s=1
X

x=0
A

v=Xx+1
(x, s)

j=1
[p
x,x+v
]
s,j

(v (X x)) (17)
where the term

2
j=1
[p
x,x+v
]
s,j

indicates the total


probability that the number of PDUs in the queue increases
by v at every arrival phase. Note that, we consider the prob-
ability p
x,x+v
rather than the probability of PDU arrival
here since we have to consider the PDU transmission in the
same frame as well. After calculating the average number
of dropped PDUs per frame, we can obtain the probability
that an incoming PDU is dropped as follows:
P
drop
=
x
drop

(18)
where is the average number of PDU arrivals per frame
and it can be obtained from
= (1 P
peak
) +
p
P
peak
. (19)
3.4.3 Queue Throughput
It measures the number of PDUs transmitted during one
frame and it can be obtained from
= (1 P
drop
). (20)
3.4.4 Average Delay
The average delay is dened as the number of frames that
a PDU waits in the queue since its arrival before it is trans-
mitted to the BS. We use Littles law to obtain the average
delay as follows:
d =
x

(21)
where is the throughput (same as the effective arrival rate
at the queue) and x is the average number of PDUs in the
queue.
4 Game-Theoretic Formulation of Band-
width Allocation and Connection Admis-
sion Control
We formulate a noncooperative two-person general-sum
game for bandwidth allocation and admission control. The
players in this game are the base station (i.e., the service
provider) and a newconnection which receives service from
the base station. The game here is a noncooperative one
since the decisions are made by these two players indepen-
dently and it is a general-sumgame since the payoffs for the
players can be arbitrary numbers and do not need to sum
to zero. This game of bandwidth allocation and admission
control is based on the fact that the base station can admit
a new connection if the performances of the ongoing con-
nections do not degrade below the desired level and the new
connection accepts the services offered by the base station
only if its delay and/or throughput requirements can be met.
The proposed bandwidth allocation and admission con-
trol method works as follows. When a new connection is
initiated, the base station invokes the bandwidth allocation
and admission control algorithm. In this case, the new con-
nection informs the base station of the connection type (i.e.,
rtPS or nrtPS), parameters of trafc sources (i.e., normal ar-
rival rate, peak rate and probability of peak rate) and the de-
lay or the throughput requirement. Then, the base station es-
tablishes a set of strategies and computes the expected pay-
off corresponding to each strategy. Next, the game is solved
and the decision on accepting or rejecting the connection is
made as well as the amount of bandwidth assigned to new
connection (if accepted) is determined.
4.1 Utility Function
To describe users level of satisfaction quantitatively as
a function of the QoS metrics, we use the concept of utility
functions and the utility (i.e., level of satisfaction) is used
to determine the payoff for the game. In the system under
consideration, the utility for connection i depends on aver-
age delay d(b(i)) and throughput (b(i)) for rtPS and nrtPS,
respectively, when b(i) units of bandwidth are allocated to
connection i. We use the modied sigmoid function [10]
to obtain utility as a function of these performance mea-
sures. If d
req
and
req
denote the delay and the throughput
requirements, the utility for rtPS and nrtPS connections can
be expressed as a function of allocated bandwidth as fol-
lows:
U
i
(b(i)) =

1
1
1+exp(grt(d(b(i))dreqhrt))
, rtPS
1
1+exp(gnrt((b(i))reqhnrt))
, nrtPS
(22)
where g
rt
, g
nrt
, h
rt
and h
nrt
are the parameters of the sig-
moid function. Note that, while g
rt
and g
nrt
indicate the
steepness (i.e., sensitivity of delay or throughput require-
ment), h
rt
and h
nrt
represent the center of the curve (i.e.,
satisfaction on performances perceived by users and their
requirements). From the base stations point of view, the
utility can represent the revenue from the ongoing connec-
tions in which the revenue for each connection depends on
the perceived performance.
4.2 Game Strategies
4.2.1 Base Station
The strategies of the base station correspond to accepting
a new connection with different amount of allocated band-
width. In this case, if the base station allocates zero unit of
bandwidth, the new connection is not admitted. The proce-
dures of establishing the BS strategies are as follows:
In order to allocate bandwidth to a new connection,
some amount of bandwidth need to be taken away from
the ongoing rtPS (C
rt
) and nrtPS (C
nrt
) connections.
The maximum amount of bandwidth that can be taken
from rtPS (b

rt
) and nrtPS (b

nrt
) can be obtained as
follows:
b

rt
=


(rt)
1
,

iCrt
b(i) < B
(rt)
TH

(rt)
2
,

iCrt
b(i) B
(rt)
TH
(23)
b

nrt
=


(nrt)
1
,

jCnrt
b(j) < B
(nrt)
TH

(nrt)
2
,

jCnrt
b(j) B
(nrt)
TH
.
(24)
Here,
(rt)
1
,
(rt)
2
,
(nrt)
1
,
(nrt)
2
, B
(rt)
TH
and B
(nrt)
TH
are
system design parameters. In particular, if the amount
of allocated bandwidth to rtPS connections is higher
than threshold B
(rt)
TH
, the maximum amount of band-
width taken from C
rt
is
(rt)
2
, and
(rt)
1
otherwise.
Similar is the case for nrtPS. By using these thresholds,
we can prioritize bandwidth allocation between rtPS
and nrtPS. For example, if the amount of bandwidth
used by rtPS is larger than that of nrtPS, the larger por-
tion of the bandwidth for the new connection should
come from C
rt
compared to that from C
nrt
. Note that,

(rt)
1
,
(rt)
2
,
(nrt)
1
and
(nrt)
2
can be chosen according
to the type of new connection. For instance, if the new
connection is of type rtPS, larger amount of bandwidth
should come from C
rt
.
The set of base station strategies can be dened as
BS = {bs
0,0
, bs
0,1
, . . . , bs
r,n
, . . . , bs
b

rt
,b

nrt
} where r
and n units of bandwidth (i.e., r {0, 1, . . . , b

rt
} and
n {0, 1, . . . , b

nrt
}) are taken from rtPS and nrtPS
connections and allocated to a new connection. The
total number of strategies is (b

rt
+ 1) (b

nrt
+ 1).
After obtaining the amount of bandwidth, next we
need to decide which connections in C
rt
and C
nrt
the
bandwidth will be taken from. In this case, we iter-
atively search for the ongoing connections with the
highest utility. Let row matrix B
r+n
denote the list
of the remaining amount of bandwidth allocated to on-
going connections if the base station allocates r + n
units of bandwidth to the new connection. Note that,
B
0
denotes the initial amount of bandwidth of the
ongoing connections before the game is played and

i
[B
k
]
i
= B, k = {0, 1, . . . , b

rt
+ b

nrt
}, where
B is the total amount of bandwidth for rtPS and nrtPS
connections.
This matrix B
r+n
is obtained as follows:
i
h
= max
i
(U([B
r+n1
]
i
)) for r + n > 0
[B
r+n
]
i
h
= [B
r+n1
]
i
h
1 (25)
where i
h
is the index of the connection with the highest
utility in B
r+n1
.
4.2.2 New Connection
There are two possible strategies for a new connection - ei-
ther accept or deny the service offered by the base station.
The set of strategies can be dened as NC = {nc
1
, nc
2
},
where nc
1
and nc
2
denote the strategies that the new con-
nection accepts or denies the service, respectively.
4.3 Payo for the Base Station and the
New Connection
The payoff matrix for the base station is dened in (26)
where column j of matrix
bs
corresponds to strategy bs
r,n
where r =
j1
b

rt
+1
and n = j
j1
b

rt
+1
(b

rt
+1) 1. The
elements of this payoff matrix can be obtained from

bs
(bs
r,n
, nc
1
) =

i
U
i
([B
r+n
]
i
) + U
c
(n)
(r + n) (27)

bs
(bs
r,n
, nc
2
) =
bs
(bs
0,0
, nc
2
) r, n (28)
where the subscript c
(n)
denotes newconnection. (27) is ob-
tained from the fact that if the new connection is accepted
and allocated r + n units of bandwidth, the payoff for the
base station becomes the total utility of the ongoing con-
nections after some portion of bandwidth have been taken
away (which results in decrease in the total utility for the
ongoing connections) plus the utility of the new connection.
However, if the new connection chooses to deny the offered
service, the payoff for the base station is identical to that
when the base station rejects the new connection.
The payoff matrix for a new connection is dened in (29)
where

nc
(bs
r,n
, nc
1
) = U
c
(n)
(r + n) (30)

nc
(bs
r,n
, nc
2
) = 1 U
c
(n)
(r + n). (31)
In the above, (30) gives the payoff (i.e., utility) for a new
connection when it accepts the service with r + n units
of bandwidth offered by the base station. Conversely, (31)
represents the payoff in the case that the new connection
denies to accept the service. Therefore, this payoff is neg-
ative to that of accepting service. For example, if r + n is
large enough to achieve the QoS requirements, the payoff
for denying the service is low (i.e., it is better for the new
connection to accept the offered service).
4.4 Solution of the Game
For the proposed bandwidth allocation and admission
control scheme, we rst obtain the Nash equilibrium, and
then the decision of bandwidth allocation and admission
control is made based on the Nash equilibrium along with
the admissible property.
The pure strategy pair (bs

r,n
, nc

i
) is a Nash equilibrium
to the game (
bs
,
nc
) if

bs
(bs

r,n
, nc

i
)
bs
(bs
r,n
, nc

i
) bs
r,n
(32)

nc
(bs

r,n
, nc

i
)
nc
(bs

r,n
, nc
i
) nc
i
. (33)
Namely, the Nash equilibriumof a game is a strategy prole
(list of strategies, one for each player) with the property that
no player can increase his payoff by choosing a different
action, given the other players actions [11].
To determine the Nash equilibrium, we use the best re-
sponse function. The best response function of the base
station BR
bs
(nc

i
), given that the new connection chooses
strategy nc

i
, is dened as follows:
BR
bs
(nc

i
) = max
bsr,n
(
bs
(bs
r,n
, nc

i
)) . (34)
Similarly, the best response function of new connection
BR
nc
(bs

r,n
), given that the base station chooses strategy
bs

r,n
, is expressed as follows:
BR
nc
(bs

r,n
) = max
nci

nc
(bs

r,n
, nc
i
)

. (35)
A pair (bs

r,n
, nc

i
) of strategies is a Nash equilibrium if
and only if bs

r,n
= BR
bs
(nc

i
) and nc

i
= BR
nc
(bs

r,n
).
One of the properties of Nash equilibrium that we use for
bandwidth allocation and admission control is the admissi-
ble strategy pair [12].
The pure strategy pair (bs
r,n
, nc
i
) weakly dominates
(bs

r,n
, nc

i
) if and only if

bs
(bs
r,n
, nc
i
)
bs
(bs

r,n
, nc

i
) (36)

bs
(nc
r,n
, nc
i
)
nc
(bs

r,n
, nc

i
). (37)
The pure strategy pair (bs
r,n
, nc
i
) is inadmissible if there
exists a strategy pair (bs

r,n
, nc

i
) that weakly dominates
(bs
r,n
, nc
i
). Alternatively, the pure strategy pair (bs
r,n
, nc
i
)
is admissible if it is not inadmissible.
The proposed bandwidth allocation and admission con-
trol is based on the presence of an admissible strategy pair.
That is, for admission control, a new connection is accepted
by the base station if there exists an admissible strategy
pair, and rejected otherwise. For bandwidth allocation, if
the new connection is accepted, the base station takes r and
n units of bandwidth fromthe groups of rtPS and nrtPS con-
nections, respectively, and allocates these to the new con-
nection.
5 Performance Evaluation
5.1 Parameter Setting
We consider a TDMA/TDD-based uplink transmission
scenario from the SSs to a BS. The SSs work in the GPC

bs
=


bs
(bs
0,0
, nc
1
)
bs
(bs
0,1
, nc
1
)
bs
(bs
b

rt
,b

nrt
, nc
1
)

bs
(bs
0,0
, nc
2
)
bs
(bs
0,1
, nc
2
)
bs
(bs
b

rt
,b

nrt
, nc
2
)

. (26)

nc
=


nc
(bs
0,0
, nc
1
)
nc
(bs
0,1
, nc
1
)
nc
(bs
b

rt
,b

nrt
, nc
1
)

nc
(bs
0,0
, nc
2
)
nc
(bs
0,1
, nc
2
)
nc
(bs
b

rt
,b

nrt
, nc
2
)

. (29)
mode. The PDU arrival process for each polling service
connection follows a Poisson process and the maximum
batch size of PDU arrival is 100 (i.e., A = 100). The PDUs
from PS connections are buffered into a single queue and
the queue size for each of the connections is assumed to be
200 PDUs (i.e., X = 200).
The transmission bandwidth is 25 MHz. The transmis-
sion frame size is 1 ms and the length of a MAC PDU is
xed at 80 bits. AMC is used in which the modulation level
and the coding rate are adjusted based on the channel qual-
ity. In one frame period the maximumnumber of PDUs that
can be transmitted (i.e., total amount of bandwidth) is 125.
We reserve 25 units of bandwidth for UGS and BE trafc
and 100 units of bandwidth for both rtPS and nrtPS trafc
(i.e., B = 100). For queueing performance evaluation of
a particular connection, we assume that the average SNR
is 15 dB, and the MMPP parameters are given as follows:
normal rate is 10 and peak rate is 30 PDUs per frame and
probability of peak rate is 0.4 (i.e., = 10,
p
= 30, and
P
peak
= 0.4). Note that, we vary some of these parameters
according to the evaluation scenarios, while the rest of the
parameters remain xed.
The QoS constraints for the rtPS and the nrtPS con-
nections are assumed as follows: d
req
(i) = 5 frames and

req
(i) = 20 PDUs per frame. We set g
rt
= 2, g
nrt
= 5,
h
rt
= h
nrt
= 0,
(rt)
1
=
(nrt)
1
= 5,
(rt)
2
=
(nrt)
2
= 2,
and B
(rt)
TH
= B
(nrt)
TH
= 50. We compare the performance
of the proposed scheme with that of admission control with
static and adaptive bandwidth allocation. For the adaptive
scheme, we consider bandwidth allocation and admission
control similar to that in [2] in which the amount of allo-
cated bandwidth is dynamically adjusted according to the
number of ongoing connections. For the static scheme, an
incoming connection is statically allocated with the mini-
mum amount of bandwidth so that it is able to maintain de-
lay and throughput requirements (for the rtPS and the nrtPS
cases, respectively).
5.2 Numerical and Simulation Results
5.2.1 Queueing Performances and Observations
We rst examine average queueing delay and throughput
corresponding to rtPS and nrtPS connections (Fig. 1(a) and
(b)), respectively. As expected, when the channel quality
becomes better, since the transmitter can use higher modu-
lation and coding level, the queueing delay decreases while
the throughput increases. We observe that the relation-
ship between the average delay and the amount of allocated
bandwidth as well as that between the average SNR and the
throughput are non-linear.
8 10 12 14 16 18 20 22 24
0
10
20
30
40
50
60
70
Average SNR (dB)
A
v
e
r
a
g
e

d
e
l
a
y

(
m
s
)
b=2
b=4
b=6
b=8
(a)
10 15 20 25 30
0
1
2
3
4
5
6
7
x 10
4
Average SNR (dB)
T
r
a
n
s
m
i
s
s
i
o
n

r
a
t
e

(
P
D
U
s
/
s
e
c
o
n
d
)
b=2
b=4
b=6
b=8
(b)
Figure 1. (a) Average delay and (b) through-
put under different average SNR.
5.2.2 Performance of the Admission Control and
Bandwidth Allocation Scheme
We assume that the connection arrival process in the cell fol-
lows a Poisson distribution and the mean arrival rate varies
from 0.05 to 0.9 connections per minute. Connection hold-
ing time is exponentially distributed with mean 20 minutes.
The normal and the peak rates of incoming connection is
assumed to be uniformly distributed in the range of 1-20
and 21-40 PDUs par frame, respectively. The probability of
peak rate is also uniformly distributed between 0.1 to 0.5.
Fig. 2 shows average delay and throughput of rtPS and
nrtPS connections under different requirement scenarios.
As is evident, the proposed bandwidth allocation and ad-
mission control successfully maintains the delay and the
throughput performances at the desired level. Moreover,
the proposed scheme can adaptively allocate the available
bandwidth according to the trafc load in the cell. Speci-
cally, when the load is low (i.e., 0.05-0.2), the ongoing con-
nections receive larger amount of bandwidth, and therefore,
the average delay is minimized and the throughput is max-
imized. However, when the trafc load in the cell becomes
higher, the proposed scheme adjusts the bandwidth alloca-
tion accordingly.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
2
4
6
8
10
Connection arrival rate
A
v
e
r
a
g
e

d
e
l
a
y

(
m
s
)
d
req
=5,
req
=20
d
req
=10,
req
=20
d
req
=10,
req
=30
d
req
=5,
req
=30
(a)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
2.5
3
3.5
4
4.5
5
5.5
x 10
4
Connection arrival rate
T
h
r
o
u
g
h
p
u
t

(
P
D
U
s
/
s
e
c
)
d
req
=5,
req
=20
d
req
=10,
req
=20
d
req
=10,
req
=30
d
req
=5,
req
=30
(b)
Figure 2. (a) Average delay and (b) through-
put under different requirements.
5.2.3 Performance Comparison
Next, we compare the performance of the proposed game-
theoretic bandwidth allocation and admission control with
that of each of the two other traditional schemes. Figs. 3(a)
and (b) show connection blocking probability and average
delay, respectively, for an rtPS connection. Figs. 4(a) and
(b) show throughput for nrtPS connections and connection
utility for both types of connections, respectively. With the
adaptive scheme, even though the blocking probability is
minimized (Fig. 3(a)), the delay and the throughput require-
ments can not be met (Figs. 3(b) and Fig. 4(a)) when the
trafc load in the cell becomes higher. Therefore, users
satisfaction represented by connection utility decreases sub-
stantially (Fig. 4(b)).
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
0.1
0.2
0.3
0.4
Connection arrival rate
B
l
o
c
k
i
n
g

p
r
o
b
a
b
i
l
i
t
y
rtPS (game)
nrtPS (game)
rtPS (stat)
nrtPS (stat)
rtPS (adap)
nrtPS (adap)
(a)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
5
10
15
20
Connection arrival rate
A
v
e
r
a
g
e

d
e
l
a
y

(
m
s
)
rtPS (game)
rtPS (stat)
rtPS (adap)
(b)
Figure 3. (a) Connection blocking probability
and (b) average delay for different admission
control schemes.
On the other hand, the static bandwidth allocation
scheme uses the queueing analytical model and can satisfy
the delay and the throughput requirements (Fig. 3(b) and
Fig. 4(a)). As expected, when the trafc load in a cell in-
creases, the connection blocking probability increases for
both the static and the proposed scheme. Since the base
station must provide QoS guarantee, some of the incom-
ing connections must be blocked so that the performances
of the ongoing connections do not deteriorate. Moreover,
by utilizing game theory to obtain the best decision on
bandwidth allocation and admission control, the proposed
scheme can provide smaller delay and larger throughput for
rtPS and nrtPS connections compared to those for the static
scheme. Therefore, the connection utility becomes higher
(Fig. 4(b)).
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
2
3
4
5
x 10
4
Connection arrival rate
T
h
r
o
u
g
h
p
u
t

(
P
D
U
s
/
s
e
c
o
n
d
)
rtPS (game)
rtPS (stat)
rtPS (adap)
(a)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0.2
0.4
0.6
0.8
1
Connection arrival rate
C
o
n
n
e
c
t
i
o
n

u
t
i
l
i
t
y
rtPS (game)
rtPS (stat)
rtPS (adap)
(b)
Figure 4. (a) Throughput and (b) connec-
tion utility for different admission control
schemes.
6 Conclusions
We have proposed a game-theoretic adaptive bandwidth
allocation and admission control scheme for IEEE 802.16
networks. For real-time and non-real-time polling services
in IEEE 802.16, delay and throughput are the QoS per-
formances, respectively, that need to be guaranteed. We
have formulated a noncooperative two-person general-sum
game. Nash equilibrium has been used to obtain the candi-
date strategies and the decision of the game is made based
on the presence of admissible strategy pair. The game-
theoretic formulation provides not only the decision on ac-
cepting or rejecting a connection, but also the amount of
bandwidth allocated to a new connection. The performance
measures used by the game-theoretic formulation have been
obtained from a queueing model in which adaptive modu-
lation and coding in physical layer and peak rate of traf-
c source have been considered. We have evaluated the
proposed bandwidth allocation and admission control al-
gorithm by extensive simulations and the proposed scheme
has been observed to successfully maintain the QoS perfor-
mances at the required level. Also, comparison with the
traditional schemes has been presented.
References
[1] IEEE 802.16 Standard - Local and Metropolitan Area
Networks - Part 16, IEEE Std 802.16a-2003.
[2] C. T. Chou and K. G. Shin, Analysis of adaptive band-
width allocation in wireless networks with multilevel
degradable quality of service, IEEE Trans. Mobile
Comput., vol. 3, no. 1, pp. 5-17, Jan.-Mar. 2004.
[3] J. Virapanicharoen and W. Benjapolakul, Fair-efcient
guard bandwidth coefcients selection in call admis-
sion control for mobile multimedia communications us-
ing game theoretic framework, in Proc. IEEE ICC04,
vol. 1, pp. 80-84, June 2004.
[4] H. Lin, M. Chatterjee, S. K. Das, and K. Basu, ARC:
An integrated admission and rate control framework for
competitive wireless CDMA data networks using non-
cooperative games, IEEE Trans. Mobile Comput., vol.
4, no. 3, pp. 243-258, May-June 2005.
[5] Y. L. Kuo, E. H. K. Wu, and G. H. Chen, Nonco-
operative admission control for differentiated services
in IEEE 802.11 WLANs, in Proc. IEEE GLOBE-
COM04, vol. 5, pp. 2981-2986, 29 Nov.-3 Dec. 2004.
[6] M. Zorzi, Packet dropping statistics of a data-link pro-
tocol for wireless local communications, IEEE Trans.
Veh. Tech., vol. 52, no. 1, pp. 71-79, Jan. 2003.
[7] Q. Liu, S. Zhou, and G. B. Giannakis, Queuing With
adaptive modulation and coding over wireless links:
Cross-layer analysis and design, IEEE Trans. Wireless
Comm., vol. 4, no. 3, pp. 1142-1153, May 2005.
[8] L. Muscariello, M. Meillia, M. Meo, M. A. Marsan,
and R. L. Cigno, An MMPP-based hierarchical model
of Internet trafc, in Proc. IEEE ICC04, vol. 4, pp.
2143-2147, June 2004.
[9] P. Salvador, R. Valadas, and A. Pacheco. Multi-
scale tting procedure using Markov modulated Pois-
son processes, Telecommunication Systems, vol. 23,
pp. 123-148, 2003.
[10] M. Xiao, N. B. Shroff, and E. K. P. Chong, Utility-
based power control in cellular wireless systems, in
Proc. IEEE INFOCOM01, vol. 1, pp. 412-421, 2001.
[11] M. J. Osborne, An Introduction to Game Theory, Ox-
ford University Press, 2003.
[12] L. Weiss, Statistical decision theory, McGrawHill,
1961.

Você também pode gostar