IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO.

10, OCTOBER 2008 3951

Queueing Analysis for OFDM Subcarrier Allocation

in Broadband Wireless Multiservice Networks
Yan Zhang, Member, IEEE, Yang Xiao, Senior Member, IEEE, and Hsiao-Hwa Chen, Senior Member, IEEE

Abstract—In this paper, we perform a tele-traffic queueing algorithms to accommodate users under certain constraints.
analysis for OFDM subcarrier allocation in wireless multiservice However, the system model and tele-traffic analysis for OFDM
networks. For this purpose two call admission control algorithms, subcarrier allocation schemes have not been fully studied.
the batch blocking scheme and the partial blocking scheme,
are proposed. Call connection requests are classified into two Tele-traffic modeling is significant for designing efficient call
different types, narrow-band and wide-band. For either class of admission control mechanisms and is equally important for
calls, the traffic process is characterized as batch arrival, as each service operators to implement and deploy OFDM wireless
call may request multiple subcarriers to satisfy its Quality-of- networks. The recent work reported in [4] performed a sys-
Service (QoS) requirements. The batch size is a random variable tem queueing model study for subcarrier allocation issue in
which obeys a probability mass function (pmf) with a realistic
maximum value. In addition, the service times for various call OFDM-based wireless networks with a single service class.
classes are different. Consequently, an OFDM-based broadband It is believed that supporting multiple services is an indis-
wireless multiservice network can be formulated as a multiclass pensable requirement in future-generation wireless networks.
multiserver batch arrival queueing system. The methodology and Hence, it becomes necessary and significant to investigate the
results are further generalized into a multiple-class scenario OFDM subcarrier allocation in wireless multiservice networks.
with service priority provision. Formulae are developed for
evaluation of the following performance metrics: the probability In this paper, we should concentrate on the system modeling
that a call will be blocked, the average number of subcarriers and tele-traffic analysis for OFDM subcarrier allocation in
used, and bandwidth utilization. Numerical results are presented wireless multiservice networks. To achieve this, a call ad-
to demonstrate the interactions between key parameters and
performance metrics. The analytical model was validated by the mission control mechanism is required. Two different call
simulation results, showing the fact that it can be used as an admission control schemes, i.e., a batch blocking scheme
efficient tool for design of future-generation broadband wireless and partial blocking scheme, are suggested upon subcarrier
access networks. requests. It is noteworthy that the role of call admission
Index Terms—Wireless network, multiservice, OFDM, subcar- control schemes is to help construct analysis model and
rier allocation, queueing system. may not be always efficient for transmitting all types of
traffics. However, the relative simplicity of call admission
I. I NTRODUCTION control mechanisms enables the primary objective in this paper
and also demonstrates the interactions between parameters
O RTHOGONAL Frequency Division Multiplexing
(OFDM) has been widely adopted in various wireless
standards, e.g., IEEE 802.11a/g for Wireless Local Area
and system performance without adding extra unnecessary
complication to analytical model construction. Furthermore,
in the last section, we point out several potential extensions
Networks (Wireless LANs), IEEE 802.16a/d/e for Wireless
in the context of call admission control design, which will
Metropolitan Area Networks (Wireless MANs), Digital Audio
be left as a future work. For subcarrier allocation in OFDM
Broadcasting/Digital Video Broadcasting (DAB/DVB) and
system, each call connection may request multiple subcarriers
satellite radio [3]. OFDM is also the basis of Orthogonal
to satisfy its Quality-of-Service (QoS) requirement. As a
Frequency Division Multiple Access (OFDMA), which
consequence, the call requests can be regarded as a batch
is believed to be a highly promising technology for the
arrival process. Based on the detailed analysis in Section II,
upcoming fourth-generation (4G) wireless networks.
the subcarrier allocation can be characterized as a multiclass
One of the most significant features of OFDM wireless
multiserver batch arrival queueing system. The methodology
systems is its flexibility with respect to subcarrier allocation
and results obtained with two types call connections are further
for satisfying various service requirements [1]-[4]. The studies
generalized into a multiple-class scenario with service priority
carried out in [5] [6] have focused on subcarrier allocation
provisioning, which is not available in the literature.
Manuscript received May 11, 2007; revised August 12, 2007; accepted Solely from the perspective of the queueing system, this
September 27, 2007. The associate editor coordinating the review of this
letter and approving it for publication was Y.-B. Lin. is not a trivial extension of previous works. In the context of
Y. Zhang is with Simula Research Laboratory, Martin Linges vei 17, PO the batch arrival process in queueing systems, [11] studied the
Box 134, 1325 Lysaker, Norway (e-mail: [email protected]). blocking and delay characteristics in a queueing system with
Y. Xiao is with the Department of Computer Science, University of
Alabama, Box 870290, Tuscaloosa, AL 35487, USA (e-mail: yangx- single batch arrival and infinite buffer. In addition, the number
[email protected]). of arrivals in each arrival epoch is fixed. The work reported in
H.-H. Chen is with the Department of Engineering Science, National Cheng [16] performed the analysis with a single batch arrival queue
Kung University, 1 Da-Hsueh Road, Tainan City, 70101 Taiwan (e-mail:
[email protected]). and an infinite buffer. The study in [10] conducted an analysis
Digital Object Identifier 10.1109/T-WC.2008.070495 with a single batch arrival queue and a finite buffer. The works
1536-1276/08$25.00 
c 2008 IEEE

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 3952 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 10, OCTOBER 2008

presented in [7]-[9] analyzed the relevant queueing model with

Narrow-band call λ n Narrow-band call service rate μn
a single batch arrival process. In all these related references,
only a single batch arrival process was taken into account.
C subcarriers
On the other hand, the analytical model characterized in this
paper differs from the previous studies in the literature. The Wide-band call λ w Wide-band call service rate μw
queueing model has the following properties: 1) it has multiple
batch arrival processes; 2) the processes have different service
times; 3) the batch size is not fixed but rather it is assumed Fig. 1. Queueing model for a cell
as a random variable following a distribution function; 4) the
batch size is not infinite but has a practically maximum value.
From the analysis and simulation given in this paper, it can where xn represents the expected value of the batch size.
be seen that the scenarios with multiple batch arrival traffic Equivalently, xn Rb represents the average data rate that a
processes could significantly complicate the system dynamics narrow-band call requests. The service time of a narrow-band
and exhibit different characteristics. call is exponentially distributed with mean 1/μn . Here, the
The remainder of this paper is outlined as follows. In exponential distributed service time is not exceptional. The
section II, we propose two call admission control algorithms. agreement between the simulation and analysis results given in
We will describe the system model and performance analysis Section IV indicates that different distribution functions lead to
methodology for two traffic processes. In this section we insignificant discrepancies in the performance metrics; and the
will also develop the following performance metrics: the exponential distribution for the service time is able to provide
probability that a call will be blocked (henceforth, blocking a sufficient accuracy.
probability), the average number of subcarriers used, and Similarly, assume that wide-band call requests follow a
bandwidth utilization. In section III, we present the analytical Poisson process with the mean batch arrival rate λw . At each
model for generalized multiple service classes. In Section arrival instant, the number of requested subcarriers is denoted
IV, we provide numerical results which will be validated by by xw . Let this discrete random variable xw follow the pmf
simulation results. Finally, Section V concludes the paper. xw,j (j = 1, 2, · · · Nw ) where Nw denotes the highest data
rate Nw Rb that a wide-band call requests. Then, we have
II. S YSTEM Q UEUEING M ODEL AND P ERFORMANCE
Nw

A NALYSIS
xw,j = 1; 1 ≤ j ≤ Nw ≤ C (3)
Let C denote the number of subcarriers in a cell, and j=1
Rb denote the average data rate per subcarrier. The value of
The average value of the group size in a wide-band call is
Rb can be calculated from the statistical value of Adaptive
given by
Modulation Coding (AMC) parameters for each subcarrier. As
a result, a cell has CRb rate resources in total [4]. In a mul- Nw

tiservice wireless network, we categorize the call connections xw = jxw,j (4)
into two types: narrow-band and wide-band calls. Each call j=1
connection may request multiple subcarriers to satisfy its data Equivalently, xw Rb represents the average data rate that a
rate requirement. As a result, the call requests can be seen as a wide-band call requests. The service time of a wide-band
batch arrival process. It is noteworthy that, for the narrow-band call is exponentially distributed with mean 1/μw . Hence, the
calls, it is reasonable to assume multiple subcarriers instead cell has two batch arrival processes. The batch size for each
of a single subcarrier, because one subcarrier may not be process is a random variable that has a maximum value. The
sufficient to support its data rate request in an OFDM wireless service times for the two processes are different. Thereafter,
system. Voice/data integrated wireless networks may serve as the system of concerned can be characterized as a multiclass
examples. multiserver batch arrival queueing system.
The narrow-band call connection is consequently character- Upon the arrival of an either narrow-band or wide-band
ized as follows. Narrow-band call requests follow a Poisson call, if the number of free subcarriers is not less than the
process with the mean batch (or group) arrival rate λn . At required number of subcarriers, the call is accepted. On
each arrival instant, the number of requested subcarriers is the other hand, if the number of free subcarriers is smaller
denoted by xn . Let the discrete random variable xn follow the than the required number of subcarriers, the call is blocked
probability mass function (pmf) xn,j (j = 1, 2, · · · Nn ) where directly; this mechanism is called a batch blocking scheme
Nn denotes the highest data rate Nn Rb of narrow-band call and is exploited in Section II-A. Alternatively, the call can be
requests. Then, we have still accepted with degraded QoS by employing the available
Nn
 subcarriers instead of the number of required subcarriers. This
xn,j = 1; 1 ≤ j ≤ Nn ≤ C (1) alternative mechanism is called a partial blocking scheme and
j=1 is studied in Section II-B.
The average value of the group size is given by
Nn
 A. Batch Blocking Scheme
xn = jxn,j , (2) In the batch blocking scheme, at the time when a narrow-
j=1 band call arrives, if the number of available subcarriers is less

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 ZHANG et al.: QUEUEING ANALYSIS FOR OFDM SUBCARRIER ALLOCATION IN BROADBAND WIRELESS MULTISERVICE NETWORKS 3953

Fig. 2. State transit rate diagram in the batch blocking scheme.

than the required number of subcarrier, the call is blocked. state (i, j) ∈ Γ. On the basis of the transit rate diagram, we
In other words, if an arriving narrow-band call requests k develop the set of global balance equations as follows.
(1 ≤ k ≤ Nn ) subcarriers, while the number of unoccupied 1) For the state (i, j) = (0, 0):
subcarriers is smaller than the batch size k, then the narrow-
b0,0 π(0, 0) = μn π(1, 0) + μw π(0, 1) (6)
band call is blocked. This policy is also applicable to wide-
band calls. where b0,0 = λn + λw .
Fig. 1 shows the queueing model for a cell. It can be seen 2) For the states 1 ≤ i < C; j = 0:
that the cell is characterized as two batch arrival processes. bi,0 π(i, 0) = (i + 1)μn π(i + 1, 0)
The batch size is a random variable that has a maximum min (i,Nn )
value. The service times for the processes are different. Define 
+ xn,k λn π(i − k, 0) + μw π(i, 1) (7)
the system state as (i, j) with i representing the number of
k=1
subcarriers used by narrow-band calls and j the number of
where
subcarriers used by wide-band calls. Then, a two dimensional
min (C−i,Nn ) (C−i,Nw )
state space Γ can be defined as  
  bi,0 = iμn + xn,k λn + xw,k λw (8)

 k=1 k=1
Γ = (i, j)  0 ≤ i ≤ C, 0 ≤ j ≤ C, 0 ≤ i + j ≤ C (5)
3) For the states i = 0; 1 ≤ j < C:

Fig. 2 shows the transit rate diagram in the batch blocking b0,j π(0, j) = (j + 1)μw π(0, j + 1)
scheme with relaxed C, Nn and Nw . Let π(i, j) be the steady min (j,Nw )

state probability distribution for a valid state (i, j) ∈ Γ. In + xw,k λw π(0, j − k) + μn π(1, j) (9)
what follows, bi,j represents the total transit rate out of the k=1

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 3954 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 10, OCTOBER 2008

where
min (C−j,Nw ) (C−j,Nn ) 
 
b0,j = jμw + xw,k λw + xn,k λn (10) π (q)
(0, j) = ωb−1
0,j (j + 1)μw π (q−1) (0, j + 1)
k=1 k=1
min (j,Nw )

4) For the states 0 < i + j < C: + xw,k λw π (q−1) (0, j − k)
k=1
bi,j π(i, j) = (i + 1)μn π(i + 1, j) + (j + 1)μw π(i, j + 1) 
min (i,Nn )
 + μn π (q−1) (1, j) + (1 − ω)π (q−1) (0, j);
+ xn,k λn π(i − k, j)
k=1
1≤j<C
min (j,Nw )

+ xw,k λw π(i, j − k) (11) 
k=1
π (q)
(i, j) = ωb−1
i,j (i + 1)μn π (q−1) (i + 1, j)
where
+ (j + 1)μw π (q−1) (i, j + 1)
bi,j = iμn + jμw
min (i,Nn )
min (C−i−j,Nn ) min (C−i−j,Nw ) 
  + xn,k λn π (q−1) (i − k, j)
+ xn,k λn + xw,k λw (12)
k=1
k=1 k=1 
min (j,Nw )

5) For the states i + j = C: + xw,k λw π (q−1) (i, j − k)
min (i,Nn ) k=1

bi,j π(i, j) = xn,k λn π(i − k, j) + (1 − ω)π (q−1) (i, j); 0<i+j <C
k=1
min (j,Nw ) where bi,j are given by (12).

+ xw,k λw π(i, j − k) (13)
π (q) (i, j) = ω(iμn + jμw )−1
k=1 ⎡
min (i,Nn )
where bi,j = iμn + jμw . 
× ⎣ xn,k λn π (q−1) (i − k, j)
In addition, the summation of all steady state probabilities
k=1
satisfies the normalization constraint, such that ⎤
 min (j,Nw )

π(i, j) = 1 (14) + xw,k λw π (q−1) (i, j − k)⎦
(i,j)∈Γ k=1

Combining the equations (6)-(14), we can solve the set of + (1 − ω)π (q−1) (i, j); i+j =C
the linear equations and consequently the steady state prob-
ability distribution. In particular, the set of linear equations 5) If we have
can be solved by using an iterative method called successive  
 
over-relaxation (SOR) [18]. We devise the following iterative  π (q) (i, j) − π (q−1) (i, j)
algorithm to calculate:   < , (16)
 (q) 
(i,j)∈Γ π (i, j) + π (q−1) (i, j)
1) Compute the number of total valid states in the set Γ;
2) For ∀(i, j) ∈ Γ, let the initial probability be then exit to step 6); otherwise let q = q + 1 and re-
(0) 1 execute steps 4) and 5).
π (i, j) = ; (15) 6) Output the steady state probability and calculate the
Number of total states in Γ
3) Let the convergence criteria be , the relaxation factor performance metrics.
be ω (1 ≤ ω < 2), and q = 1;
4) Calculate the following equations: Let Pn and Pw denote the narrow-band and wide-band
  call blocking probabilities, respectively. At the time when a
π (q) (0, 0) = ωb−1
0,0 μn π (q−1)
(1, 0) + μw π (q−1)
(0, 1) particular narrow-band (or wide-band) call with batch size k
arrives, if the number of available subcarriers is less than k,
+ (1 − ω)π (q−1) (0, 0) the call is blocked. As a result, the narrow-band call blocking
probability and wide-band call blocking probability can be
 expressed respectively as
π (q)
(i, 0) = ωb−1
i,0 (i + 1)μn π (q−1) (i + 1, 0) ⎧
 Nn
⎪
⎪
 ⎪
⎪ P = π(i, j) xn,k
min (i,Nn )
 ⎪
⎨ n
(q−1) (q−1) (i,j)∈Γ k=C−i−j+1
+ xn,k λn π (i − k, 0) + μw π (i, 1) (17)
⎪  Nw
k=1 ⎪
⎪
⎪
⎪ P = π(i, j) xw,k
+ (1 − ω)π (q−1) (i, 0); 1≤i<C ⎩ w
(i,j)∈Γ k=C−i−j+1

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 ZHANG et al.: QUEUEING ANALYSIS FOR OFDM SUBCARRIER ALLOCATION IN BROADBAND WIRELESS MULTISERVICE NETWORKS 3955

Let Mn and Mw denote the average number of subcarriers 3) For the states i = 0, 1 ≤ j < C:
used by a narrow-band call and a wide-band call, respectively, b0,j π(0, j) = (j + 1)μw π(0, j + 1) + μn π(1, j)
we always have
min (j,Nw )
⎧  
⎪ iπ(i, j) + xw,k λw π(0, j − k) (23)
⎪
⎪
⎪
⎪ k=1
⎪
⎪ Mn =
(i,j)∈Γ
⎪
⎪ λn where b0,j = jμw + λn + λw .
⎪
⎪ (1 − Pn )
⎪
⎨ μn 4) For the states 0 < i + j < C:
 (18) bi,j π(i, j) = (i + 1)μn π(i + 1, j) + (j + 1)μw π(i, j + 1)
⎪
⎪
⎪
⎪ jπ(i, j) min (i,Nn )
⎪
⎪ 
⎪
⎪ (i,j)∈Γ + xn,k λn π(i − k, j)
⎪ Mw =
⎪
⎪
⎪ λw
⎩ (1 − Pw ) k=1
μw min (j,Nw )

The bandwidth utilization is defined as the ratio of the average + xw,k λw π(i, j − k) (24)
k=1
number of busy subcarriers to the total number of subcarriers
C, or where bi,j = iμn + jμw + λn + λw .
 5) For the states i + j = C:
(i + j)π(i, j) 
min (i,Nn ) Nn

 
(i,j)∈Γ bi,j π(i, j) = xn,l λn π(i − k, j)
γ= (19)
C k=1

l=k

min (j,Nw ) Nw
 
+ xw,l λw π(i, j − k) (25)
B. Partial Blocking Scheme
k=1 l=k
In the partial blocking scheme, when a narrow-band call where bi,j = iμn + jμw .
arrives, if the number of available subcarriers is less than the In addition, the summation of all steady state probabilities
required number, the call is not blocked but is accepted with satisfies the normalization constraint, such that
degraded QoS. In other words, if an arriving narrow-band call 
π(i, j) = 1 (26)
requests k (1 ≤ k ≤ Nn ) subcarriers, while the number of
(i,j)∈Γ
free subcarriers CF is smaller than the batch size k, then the
narrow-band call will be accepted with a provided service of Combining the equations (21)-(26), we can solve the set
CF subcarriers, i.e., a data rate of CF Rb . This policy also can of the linear equations and consequently the steady state
be applied to wide-band calls. probability distribution. Following the similar SOR iterative
On the basis of the above reasoning, we develop the queue- algorithm in the Section II-A, we are able to calculate the
ing model for a cell, as shown in Fig. 1. Define the system state stationary state probability distribution.
as (i, j) with i representing the number of subcarriers used Assume that the system is in state (i, j) when a narrow-band
by narrow-band calls and j the number of subcarriers used call arrives. Then, the number of free subcarriers is C − i − j.
by wide-band calls. Let π(i, j) be the steady state probability If the batch size of the narrow-band call k is greater than
distribution. The two dimensional state space Γ is defined by the number of idle channels, the C − i − j subcarrier will be
  allocated to the narrow-band call. If there is no free subcarrier,
 i.e., i + j = C, the narrow-band call is blocked. Hence, the

Γ = (i, j)  0 ≤ i ≤ C, 0 ≤ j ≤ C, 0 ≤ i + j ≤ C (20) narrow-band call blocking probability is expressed as

Following the similar reasoning as discussed in the scheme Pn = π(i, j) (27)
(i,j)∈Γ and (i+j)=C
BBS, we are able to analyze the state transition entering or
exiting any state. In addition, we are able to plot the transition The wide-band call blocking probability is equal to the
rate diagram. Consequently, we express the global balance narrow-band call blocking probability.
equations as follows. Pw = Pn (28)
1) For the state (i, j) = (0, 0):
The average number of subcarriers used by a narrow-band call
b0,0 π(0, 0) = μn π(1, 0) + μw π(0, 1) (21) and a wide-band call is given by
⎧ 
⎪
⎪ iπ(i, j)
where b0,0 = λn + λw . ⎪
⎪
⎪
⎪ (i,j)∈Γ
2) For the states 1 ≤ i < C; j = 0: ⎪ Mn =
⎪
⎪
⎪ λn
⎪
⎪ (1 − Pn )
bi,0 π(i, 0) = (i + 1)μn π(i + 1, 0) + μw π(i, 1) ⎨ μn
 (29)
min (i,Nn )
 ⎪
⎪
⎪
⎪ jπ(i, j)
+ xn,k λn π(i − k, 0) (22) ⎪
⎪
⎪
⎪ (i,j)∈Γ
k=1 ⎪
⎪ Mw =
⎪
⎪ λw
⎩ (1 − Pw )
where bi,0 = iμn + λn + λw . μw

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 3956 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 10, OCTOBER 2008

The bandwidth utilization is expressed as 1) For the state y = 0:

 S S
(i + j)π(i, j)  
λs π(0) = μs π(ds )1ds (35)
(i,j)∈Γ
γ= (30) s=1 s=1
C
2) For the states y ∈ ΓS satisfying y = 0 and |y| < C:
III. G ENERALIZED M ULTISERVICE WITH P RIORITY ⎛ ⎞
  min (Cs −ys ,C−|y|,Ns )
D IFFERENTIATION ⎝ ys μs 1y−ds + xs,k λs ⎠ π(y)
In a multiservice wireless networks, we assume that there s∈S s∈S k=1
are S classes services. For each class-s (where s ∈ S ≡ 
= (ys + 1)μs π(y + ds )1y+ds +
{1, 2, · · · , S}), the batch arrival rate is denoted by λs . The
s∈S
maximum size of a batch arrival is denoted by Ns (s ∈ S) ⎡ ⎤
min (ys ,Ns )
with Ns being no larger than C. Then, the highest data rate  
⎣ xs,k λs ⎦ π(y − kds )1y−kds (36)
of class-s is Ns Rb . The batch size of class-s, xs , follows the
s∈S k=1
pmf xs,k (where s ∈ S; k = 1, 2, · · · Ns ), or
Ns
3) For the states y ∈ ΓS satisfying |y| = C:
  
xs,k = 1; 1 ≤ k ≤ Ns ≤ C; s ∈ S (31) 
k=1 ys μs 1y−ds π(y)
s∈S
The average value of the group size in a class-s call is given ⎡ ⎤
min (ys ,Ns )
by  
= ⎣ xs,k λs ⎦ π(y − kds )1y−kds (37)
Ns
 s∈S k=1
xs = kxs,k ; s∈S (32)
k=1
In addition, the summation of all steady state probabilities
satisfies the normalization constraint, or
Hence, xs Rb represents the average data rate that a class-s 
call requests. In addition, to provide service priority among π(y) = 1 (38)
various classes, we limit the maximum number of subcarriers y∈ΓS
employed by class-s to Cs (s ∈ S) with Cs ≤ C. Therefore, the transit rate matrix, Y, which governs the
evolution of the multidimensional Markov chain, can be
A. Batch Blocking Scheme obtained. Let Π be the stationary state probability vector of
Define the system state as y = (y1 , y2 , · · · , yS ) with ys the system and be in lexicographic order based on the state.
representing the  number of subcarriers used by the class-s Then, Π is the solution of linear equation Π = ΠY and
calls. Let |y| = s∈S ys . As described, when a class-s call Πe = 1 with e representing the column vector with all ones
arrives in the batch blocking scheme, the call is blocked if the [17]. Analogously, the set of linear equations can be solved
number of requested subcarriers is larger than 1) the number of by using an iterative method SOR [18]. We present the SOR
total free subcarriers (C − |y|); or 2) the additional maximum algorithm to calculate the steady state probability as follows:
allowed class-s calls, which is equal to the difference between 1) Construct the matrix Y;
Cs and the number of subcarriers currently used by class-s 2) Let the convergence criteria be , the relaxation factor
calls (Cs − ys ). The state space is consequently expressed by be ω(1 ≤ ω < 2);
   3) Let initial state probability be Π(0) ;
 4) Repeat and calculate
ΓS = y  0 ≤ ys ≤ Cs (s ∈ S), 0 ≤ |y| ≤ C (33)
Π(n) = ωΠ(n−1) Y + (1 − ω)Π(n−1) (39)
Let the indicator function 1E be equal to one if the event
E is true, and zero otherwise. For presentation simplicity, let until the following inequality is held, or
 
1y be equal to one if the state y is a valid state in the state  (n) (n−1) 
 in ΓS Πk − Πk
Total states 
space ΓS , and zero otherwise. Let 0 denote the S-dimension   < . (40)
vector with all its elements equal to zero. Let ds represent the  (n) (n−1) 
k=1 Πk + Πk 
S-dimension vector with its sth element equal to one while
all other items are zero. That is 5) Output the steady state probability Π.
⎧ Let Pclass−s denote the class-s blocking probability. The
⎪ 0 = (0, 0, · · · , 0)
⎪
⎨    class-s (s ∈ S) call is blocked when either: 1) the summation
S-dimension vector
(34) of the requested batch size and the number of currently
⎪
⎪ ds = (0, · · · , 1, · · · , 0)
⎩    allocated subcarriers ys is greater than the maximum allowed
the sth element is 1; all others are zero number of subcarriers for class-s Cs ; or 2) the number of total
Then, y ± kds represents the new vector with its sth element free subcarriers is less than the requested subcarriers.
equal to ys ± k, while all other elements remain unchanged Hence, the class-s blocking probability is given by
as the elements in y. Let π(y) be the steady state probability  Ns

distribution for a state y ∈ ΓS . We develop the set of global Pclass−s = π(y) xs,k (41)
balance equations as follows. y∈ΓS k=min (Cs −ys ,C−|y|)+1

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 ZHANG et al.: QUEUEING ANALYSIS FOR OFDM SUBCARRIER ALLOCATION IN BROADBAND WIRELESS MULTISERVICE NETWORKS 3957

TABLE I
Let Mclass−s denote the average number of subcarriers used D EFAULT VALUES FOR THE INPUT PARAMETERS
by a class-s call. Then, we obtain
 Parameter Value
ys π(y) Bandwidth 5 MHz
y∈ΓS C 32
Mclass−s = (42)
(λs /μs )(1 − Pclass−s ) FFT/IFFT block size 256
The bandwidth utilization is given by Channel perfect
  AMC modulation 16-QAM
( ys )π(y) Turbo coding rate 2/3
y∈ΓS s∈S Spatial multiplexing mode 20.6 Mb/s
γ= (43) Rb 659.2 kb/s
C
Nn 4
B. Partial Blocking Scheme Nw 8
1/μn = 1/μw 180.0 sec
When a class-s call arrives in the partial blocking scheme, λn = λw 0.01 −→ 0.1
if the number of available subcarriers is less than the required
number of subcarriers, the call is not blocked but accepted with
the available subcarriers. Here, the number of available subcar- is less than Cs but all of the subcarriers are busy. Hence, the
riers is the minimum value of the maximally allowed class- class-s (s ∈ S) call blocking probability is given by
s calls and the number of free subcarriers. The maximally 
allowed class-s calls is equal to the difference between Cs Pclass−s = π(y)
and the number of subcarriers currently used by class-s calls, y∈ΓS and ys =Cs

i.e., Cs − ys . The number of free subcarriers is the difference + π(y) (46)
between C and the number of all occupied subcarries, i.e., y∈ΓS ,ys <Cs and |y|=C
C − |y|. The state space is identical to ΓS . We develop the
set of global balance equations as follows. The average number of subcarriers used by a class-s call is
1) For the state y = 0: given by

S
 S
 ys π(y)
λs π(0) = μs π(ds )1ds (44) y∈ΓS
Mclass−s = (47)
s=1 s=1 (λs /μs )(1 − Pclass−s )
2) For the states y ∈ ΓS satisfying y = 0 and |y| < C: The bandwidth utilization is given by
   
 
ys μs 1y−ds + λs 1y+ds π(y) ( ys )π(y)
s∈S s∈S y∈ΓS s∈S
 γ= (48)
C
= (ys + 1)μs π(y + ds )1y+ds
s∈S IV. N UMERICAL R ESULTS
⎧⎡ ⎤ ⎫
 ⎨ min (ys ,Ns ) ⎬
We now validate the analytical model via the simulation re-
+ ⎣ xs,k λs ⎦ π(y − kds )1y−kds 1ys <Cs
⎩ ⎭ sults. Illustrative numerical examples are presented to demon-
s∈S k=1
⎧   ⎫ strate the performance tradeoff in the proposed call admission
 ⎨min (ys ,Ns ) Ns
 ⎬ control algorithms. The results will also show the interactions
+ xs,l λs π(y − kds )1y−kds 1ys =Cs between the performance metrics and key parameters. We
⎩ ⎭
s∈S k=1 l=k follow the setting similar to that used in [2], and the downlink
3) For the states y ∈ ΓS satisfying |y| = C: is based on a MIMO-OFDM air interface. The average coding
  rate is 2/3 with 16-QAM and spatial multiplexing mode.
 There are two transmit antennae and up to four receiver
ys μs 1y−ds π(y)
antennae. The system has a bandwidth of 5 MHz and C = 32
s∈S
  subcarriers. Based on the setting, the peak throughput is 20.6
(ys ,Ns ) Ns
 min  Mbps [2]. As a consequence, each subcarrier has an average
= xs,l λs π(y − kds )1y−kds
data rate of Rb = 20.6 Mbps/32 = 659.2 kbps. If not
s∈S k=1 l=k
specified, the default values are given in Table I. To validate
The summation of all steady state probabilities satisfies the the analytical model, we developed a discrete event simulation
normalization constraint as program in C++.

π(y) = 1 (45)
y∈ΓS A. Comparison of call admission control algorithms
We assume that the system state is y when a class-s (s ∈ S) Fig. 3 shows the performance comparison between the batch
call arrives. The call is blocked when either: 1) the number blocking scheme (BBS) and the partial blocking scheme (PBS)
of currently used subcarriers of class-s ys is equal to Cs ; or in terms of narrow-band call traffic intensity. The narrow-band
2) the number of currently utilized subcarriers of class-s ys traffic intensity is defined as ρn = xn λn /(Cμn ). The solid

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 3958 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 10, OCTOBER 2008

1 1
BBS: solid line
0.9 PBS: dashed line 0.9

γ
0.8 0.8

Blocking Probability/Bandwidth Utilization

Blocking Probability/Bandwidth Utiliztion

0.7 0.7

0.6 0.6
P
w
0.5 0.5

0.4 P 0.4 P
w n
Pn
0.3 0.3

0.2 0.2 circle: EXP

square: HyperErlang
0.1 0.1 cross: Erlang

0 0
0 0.5 1 1.5 0 0.5 1 1.5
ρ ρ
n n

(a) Pn , Pw and γ (a) Pn , Pw and γ

4.5 4.5

Mw BBS: solid line circle: EXP

4 PBS: dashed line square: HyperErlang
4 cross: Erlang
Average number of used subcarriers

Average number of used subcarriers

3.5

3.5

3
Mn M
2.5 w

2 2.5
M
n

1.5
0 0.5 1 1.5 2
ρ 0 0.5 1 1.5
n
ρn

(b) Mn and Mw
(b) Mn and Mw
Fig. 3. Performance comparison between BBS and PBS. The analytical
results for BBS and PBS are represented by solid line and dashed line, Fig. 4. Performance metrics in the batch blocking scheme with different
respectively. The simulation results for BBS and PBS are represented by the service time distributions. The simulation results for BBS and PBS are
circle symbol ’o’ and square symbol ’2’, respectively. represented by the circle symbol ’o’ and square symbol ’2’, respectively.

curve and the dashed curve indicate the performance metrics probabilities Pn and Pw in PBS are lower than the prob-
for BBS and PBS, respectively. The simulation results for BBS abilities given in BBS. This can be explained as follows.
and PBS are respectively indicated by the circle symbol ’o’ In PBS, a narrow-band call or wide-band call is accepted
and the square symbol ’2’ for the purpose of validation. It can even if there are fewer free subcarriers than the number of
be seen that the simulation and the analysis agree very well. requested subcarriers. In comparison, in BBS, a narrow-band
The blocking probabilities and bandwidth utilization increase call or wide-band call is blocked when there are fewer free
as traffic intensity ρn increases. The average number of subcarriers than requested subcarriers. The enhancement in
subcarriers used by narrow-band or wide-band calls decreases reducing the call blocking probability in PBS is clear with
as ρn increases. We take BBS as the example to explain. respect to Pw , while the reduction is insignificant for Pn .
With a higher ρn , more narrow-band and wide-band calls This is due to the much smaller batch size of narrow-band
with large bandwidth requirement will be blocked. Hence, calls than that of wide-band calls. With respect to the average
the accepted narrow-band or wide-band calls have the trend number of used subcarriers, narrow-band and wide-band calls
of lower subcarrier demand. This will additionally lead to may use fewer subcarriers than those requested in PBS. In
decreasing number of used subcarriers per accepted call. comparison, the calls always utilize the requested subcarriers
For either narrow-band or wide-band calls, the call blocking in BBS. Hence, Mn or Mw in PBS is smaller than that given

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 ZHANG et al.: QUEUEING ANALYSIS FOR OFDM SUBCARRIER ALLOCATION IN BROADBAND WIRELESS MULTISERVICE NETWORKS 3959

0.8 3 1

Average used subcarriers

Mclass−1
Blocking Probabilities

0.6 2.5 Mclass−2 0.8

Mclass−3
0.4 2 0.6

γ
Pclass1

0.2 Pclass2 1.5 0.4

Pclass3
0 1 0.2
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
ρ1 ρ1 ρ1

Fig. 5. Performance metrics in the batch blocking scheme with three service classes in terms of class-1 traffic intensity ρ1 .

0.8 3 1

Average used subcarriers

Mclass−1
Blocking Probabilities

0.6 2.5 Mclass−2 0.8

Mclass−3
0.4 2 0.6

γ
P
class−1
0.2 Pclass−2 1.5 0.4
P
class−3
0 1 0.2
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
ρ1 ρ1 ρ1

Fig. 6. Performance metrics in the partial blocking scheme with three service classes in terms of class-1 traffic intensity ρ1 .

in BBS. This accords with the observations in this numerical The comparison indicates that the discrepancy is very minor
example. In addition, the comparison shows that the bandwidth when employing different service time distributions to evaluate
utilization γ in PBS is higher than that in BBS. The reason is network performance. Hence, employing the exponential dis-
also due to the different policies used in BBS and PBS. The tribution for the service times in the batch blocking scheme is
results suggest that BBS may be adopted for the case in which able to provide a sufficient accuracy. We have also calculated
a shorter service time is preferable for accepted calls; PBS the performance metrics in the partial blocking scheme with
may be employed for a situation in which the call blocking different service time distributions. Similar results can be
probability is sensitive and can tolerate a longer service time. observed, i.e., the discrepancy is insignificant when employing
We have also investigated other scenarios, e.g., longer wide- different service time distributions for the partial blocking
band service time, and similar phenomena can be observed. scheme. The figure and more explanations for the partial
Due to the limited space we will not discuss them in detail blocking scheme is ignored due to the limited space.
here.
C. Performance in multiple classes
B. Effect of service time distributions In the previous examples, we focused on the situation
Fig. 4 shows the performance metrics in the batch blocking with two service classes. We now present the analysis and
scheme with different service time distribution functions. In simulation comparison for three classes. The parameters are
each figure, the exponential distribution, hyper-Erlang dis- set as follows: N1 = 1, N2 = 2, N2 = 4; C1 = 12, C2 = 24,
tribution and Erlang distribution for the narrow-band and C3 = 32; λ1 = λ2 = λ3 ; 1/μ1 = 1/μ2 = 1/μ3 = 180.0 sec.
wide-band call service times are employed for comparison. Fig. 5 shows the performance metrics in the batch blocking
The hyper-Erlang distribution has been proven to be able scheme in terms of class-1 traffic intensity ρ1 . Simulation
to approximate arbitrarily closely to the distribution of any results are presented as symbols in the figure to validate
positive random variable, as well as to measured data [19]. the analysis. It is clear that the simulation and the analysis
This general approximation property has been applied to a agree with each other very well, which validates the analytical
variety of scenarios, e.g., call holding time modeling [13], model and also the simulation program. The average number
cell residence time modeling [12] and the serving GPRS of used subcarriers by class-1 is a constant 1.0 due to the fixed
support node (SGSN) residence time [14]. For such universal number of requested subcarriers 1.0. The average number of
approximation property and wide applications, it is necessary subcarriers used by class-2 and class-3 decreases as traffic
to investigate the performance under hyper-Erlang assumption. intensity increases, due to the increasing blocking probability.
The hyper-Erlang distributed service times have a pdf defined Fig. 6 shows the performance metrics in the partial blocking
as follows: scheme in terms of class-1 traffic intensity ρ1 . Similarly, the
average number of subcarriers used by class-1 is 1.0, due to
0.4 × 0.8μk e−0.8µk t + 0.6 × (2 × 1.2μk )2 te−2×1.2µk t , (49)
the constant number 1.0 of requested subcarriers. The average
where k ∈ {n, w} and μk is the arrival rate. number of subcarriers used by class-2 and class-3 decreases

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 3960 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 10, OCTOBER 2008

as traffic intensity increases. The decreasing gradient of the [8] D. D. Yao, “Some results for the queues MX /M/c and GIX /G/c,”
average number of subcarriers used in the partial blocking Operations Research Lett., vol. 4, pp. 79-83, Feb. 1985.
[9] M. L. Chaudhry, U. C. Gupta, and V. Goswami, “Modeling and analysis
scheme is steeper than that in the batch blocking scheme. The of discrete-time multiserver queues with batch arrivals: GIX /Geom/m,”
reason is given as follows. In the partial blocking scheme, any J. Computing, vol. 13, pp. 172-180, Mar. 2001.
class call will be accepted if there is a subcarrier available. By [10] D. R. Manfield and P. TranGia, “Analysis of a finite storage system
with batch input arising out of message packetization,” IEEE Trans.
contrast, in the batch blocking scheme, a call will be accepted Commun., vol. 30, pp. 456-463, Mar. 1982.
only if there are sufficient subcarriers. Hence, comparatively, [11] I. Kabak, “Blocking and delays in M(n) /M/c bulk queueing systems,”
there are fewer subcarriers used by class-2 and class-3 in the Oper. Res., vol. 16, pp. 830-840, 1968.
[12] Y. G. Fang, “Hyper-Erlang distribution model and its application in
partial blocking scheme than in the batch blocking scheme. wireless mobile networks,” ACM Wireless Networks, vol. 7, no. 3, pp.
211-219, 2001.
[13] K. Yeo and C.-H. Jun, “Modeling and analysis of hierarchical cellular
V. C ONCLUSIONS networks with general distributions of call and cell residence times,”
We proposed a multiclass multiserver batch arrival queue- IEEE Trans. Veh. Technol., vol. 51, no. 6, pp. 1361-1374, Nov.2002.
[14] Y. B. Lin and Y. K. Chen, “Reducing authentication signaling traffic in
ing system for the performance analysis in the context of third-generation mobile network,” IEEE Trans. Wireless Commun., vol.
OFDM subcarrier allocation in future generation OFDM-based 2, no. 3, pp. 493-501, May 2003.
wireless multiservice networks. Two call admission control [15] Y. Zhang and B. Soong, “Performance evaluation of GSM/GPRS net-
works with channel re-allocation scheme,” IEEE Commun. Lett., vol. 8,
algorithms, the batch blocking scheme and partial blocking no. 5, pp. 280-282, May 2004.
scheme, were studied. We first analyzed the cases of two [16] D. Gross and C. M. Harris, Fundamentals of Queueing Theory. New
service classes. We then extended the queueing model to York: Wiley, 1974.
[17] L. Kleinrock, Queueing Systems: vol.1, Theory. New York: Wiley, 1976.
generalized multiple classes with service differentiation. We [18] R. Cooper, Introduction to Queueing Theory. New York: North-Holland,
presented simulation results to validate the analytical model 1981.
and the two match with each other very well. The analyt- [19] F. P. Kelly, Reversibility and Stochastic Networks. New York: Wiley,
1979.
ical model provides an efficient tool for evaluating future-
generation broadband wireless access networks. A future Yan Zhang Yan Zhang received Ph.D. degree in
topic can be efficient call admission control schemes in the School of Electrical & Electronics Engineering,
context of OFDM subcarrier allocation. For instance, hybrid Nanyang Technological University, Singapore. From
August 2006, he works with Simula Research Lab-
mechanism may be employed, where BBS (or PBS) is used oratory, Norway. He is a regional editor, asso-
for narrow-band calls and PBS (or BBS) for wide-band calls. ciate editor or on the editorial board of S ECU -
The methodology and the results can be easily extended to this RITY AND C OMMUNICATION N ETWORKS (Wiley);
I NTERNATIONAL J OURNAL OF N ETWORK S ECU -
scenario to investigate the system performance. In addition, RITY ; I NTERNATIONAL J OURNAL OF U BIQUITOUS
channel re-allocation [15] or buffer mechanism can also be C OMPUTING; T RANSACTIONS ON I NTERNET AND
taken into account for complex traffic types in our future work. I NFORMATION S YSTEMS (TIIS); I NTERNATIONAL
J OURNAL OF AUTONOMOUS AND A DAPTIVE C OMMUNICATIONS S YSTEMS
(IJAACS);, I NTERNATIONAL J OURNAL OF U LTRA W IDEBAND C OMMUNI -
ACKNOWLEDGMENT CATIONS AND S YSTEMS (IJUWBCS) and I NTERNATIONAL J OURNAL OF
S MART H OME (IJSH). He is currently serving as the Book Series Editor for
The authors would like to thank the three anonymous the book series on Wireless Networks and Mobile Communications (Auerbach
reviewers for their careful reading and constructive comments Publications, CRC Press, Taylor & Francis Group). He serves as guest co-
that have significantly improved the quality of this paper. This editor for W ILEY S ECURITY AND C OMMUNICATION N ETWORKS special
issue on Secure Multimedia Communication; guest co-editor for S PRINGER
work was supported in parts by the Norwegian Ministry of W IRELESS P ERSONAL C OMMUNICATIONS special issue on selected papers
Transport and Communications, and National Science Council from ISWCS 2007; guest co-editor for E LSEVIER C OMPUTER C OMMU -
NICATIONS special issue on Adaptive Multicarrier Communications and
of Taiwan (NSC96-2221-E-110-050 and NSC96-2221-E-110-
Networks; guest co-editor for I NDERSCIENCE I NTERNATIONAL J OURNAL
035). OF AUTONOMOUS AND A DAPTIVE C OMMUNICATIONS S YSTEMS (IJAACS)
special issue on Cognitive Radio Systems; guest co-editor for T HE J OURNAL
OF U NIVERSAL C OMPUTER S CIENCE (JUCS) special issue on Multimedia
R EFERENCES Security in Communication; guest co-editor for S PRINGER J OURNAL OF
[1] A. Jamalipour, T. Wada, and T. Yamazato, “A tutorial on multiple access C LUSTER C OMPUTING special Issue on Algorithm and Distributed Comput-
technologies for beyond 3G mobile networks,” IEEE Commun. Mag., ing in Wireless Sensor Networks; guest co-editor for EURASIP J OURNAL ON
vol. 43, pp. 110-117, Feb. 2005. W IRELESS C OMMUNICATIONS AND N ETWORKING (JWCN) special issue
[2] C. Dubuc, D. Starks, T. Creasy, and Y. Hou, “A MIMO-OFDM prototype on OFDMA Architectures, Protocols, and Applications; guest co-editor for
for next-generation wireless WANs,” IEEE Commun. Mag., vol. 42, pp. S PRINGER J OURNAL OF W IRELESS P ERSONAL C OMMUNICATIONS special
82-87, Dec. 2004. issue on Security and Multimodality in Pervasive Environments. He is
[3] I. Koffman and V. Roman, “Broadband wireless access solutions based serving as co-editor for several books. He serves as organizing committee
on OFDM access in IEEE 802.16,” IEEE Commun. Mag., pp. 96-103, chairs for many international conferences, including Track Co-Chair for
Apr. 2002. ITNG 2009, Workshop Co-Chair for IEEE APSCC 2008, Symposium Co-
[4] J.-C. Chen and W.-S. E Chen, “Call blocking probability and bandwidth Chair for ChinaCom 2008, Industrial Co-Chair for ACM MobiHoc 2008,
utilization of OFDM subcarrier allocation in next-generation wireless Program Co-Chair for UIC-08, Program vice Co-Chair for IEEE ISM 2007,
networks,” IEEE Commun. Lett., vol. 10, no. 2, Feb. 2006. Publication Chair for IEEE ISWCS 2007, and Program Co-Chair for IEEE
[5] C. Y. Wong, R. S. Cheng, K. B. Letaief, and R. D. Murch, “Multiuser PCAC 2007. He serves as a member of Technical Program Committee for
OFDM with adaptive subcarrier, bit and power allocation,” IEEE J. numerous international conferences, including ICC, GLOBECOM, PIMRC,
Select. Areas Commun., vol. 17, pp. 1747-1757, Oct. 1999. VTC, CCNC, AINA, ISWCS, HPCC etc. He received the Best Paper Award
[6] D. Kivanc and H. Lui, “Subcarrier allocation and power control for and Outstanding Service Award in the IEEE 21st International Conference on
OFDMA,” in Proc. 34th Asilomar Conference on Signals, Systems and Advanced Information Networking and Applications (AINA-07). His research
Computers, vol. 1, pp. 147-151, 2000. interests include resource, mobility, spectrum, data, energy and security
[7] M. V. Cromie, M. L. Chaudhry, and W. K. Grassmann, “Further results management in wireless networks and mobile computing. He is a member
for the queueing system MX /M/c,” J. Operational Research Society, of IEEE and IEEE ComSoc.
vol. 30, pp. 755-763, Aug. 1979.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.
 ZHANG et al.: QUEUEING ANALYSIS FOR OFDM SUBCARRIER ALLOCATION IN BROADBAND WIRELESS MULTISERVICE NETWORKS 3961

Yang Xiao (SM’04) worked in industry as a MAC chapters related to these research areas. Dr. Xiao’s research has been supported
(Medium Access Control) architect involving the by the US National Science Foundation (NSF), U.S. Army Research, Fleet
IEEE 802.11 standard enhancement work before & Industrial Supply Center San Diego (FISCSD), and The University of
he joined Department of Computer Science at The Alabama’s Research Grants Committee. He is the Principle Investigator (PI)
University of Memphis in 2002. He is currently with for several federal research grants over 1.2 million dollars.
Department of Computer Science at The University
of Alabama. He was a voting member of IEEE H siao-Hwa Chen (S’89-M’91-SM’00) is currently
802.11 Working Group from 2001 to 2004. He is an a full Professor in the Department of Engineering
IEEE Senior Member. He is a member of American Science, National Cheng Kung University, Taiwan.
Telemedicine Association. He currently serves as He obtained his BSc and MSc degrees from Zhe-
Editor-in-Chief for I NTERNATIONAL J OURNAL OF jiang University, China, and a PhD degree from the
S ECURITY AND N ETWORKS (IJSN), I NTERNATIONAL J OURNAL OF S ENSOR University of Oulu, Finland, in 1982, 1985 and 1991,
N ETWORKS (IJSNet), and I NTERNATIONAL J OURNAL OF T ELEMEDICINE respectively. He has authored or co-authored over
AND A PPLICATIONS (IJTA). He serves as a referee/reviewer for many funding 250 technical papers in major international journals
agencies, as well as a panelist for NSF and a member of Canada Foundation and conferences, five books and three book chapters
for Innovation (CFI)’s Telecommunications expert committee. He serves on in the areas of communications. He served as general
TPC for more than 100 conferences such as INFOCOM, ICDCS, MOBIHOC, chair, TPC chair and symposium chair for many
ICC, GLOBECOM, WCNC, etc. He serves as an associate editor for several international conferences. He served or is serving as an Editor or/and Guest
journals, e.g., IEEE T RANSACTIONS ON V EHICULAR T ECHNOLOGY. His Editor for numerous technical journals. He is the founding Editor-in-Chief of
research areas are security, telemedicine, sensor networks, and wireless net- Wiley’s S ECURITY AND C OMMUNICATION N ETWORKS J OURNAL. He is the
works. He has published more than 280 papers in major journals (more than 60 recipient of the best paper award in IEEE WCNC 2008.
in various IEEE journals/magazines), refereed conference proceedings, book

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 08:59 from IEEE Xplore. Restrictions apply.