Performance analysis of dynamic packet assignment

in cellular systems with OFDMA

R. Jayaparvathy, S. Anand and S. Srikanth

Abstract: The authors present a performance analysis of orthogonal frequency division multiple
access (OFDMA) based cellular systems with dynamic packet assignment (DPA). Most
approaches in the current literature for analysing wireless data networks do not take into account
the interference conditions on the radio channel. An analytical model based on a two-dimensional
continuous-time Markov chain (CTMC), that takes into account the interference conditions on the
OFDM subcarriers in addition to the traffic conditions, is proposed. The authors derive
expressions for the mean delay and the average system throughput on the uplink and the downlink
in OFDMA based cellular systems with data traffic.

1 Introduction a discrete-time priority queueing approach in a system with

High spectral efficiency and flexible data-rate access are the
main focus of future wireless systems. Third generation (3G)
wireless systems such as wideband code division multiple
access (WCDMA) use interference averaging techniques and
provide bit rates of 50–384 kbit/s in macrocellular systems
and 2 Mbit/s in microcellular systems [1]. Interference
avoidance techniques provide better spectrum efficiency
compared with interference averaging techniques [2]. In [3],
Chuang and Sollenberger showed that fourth-generation
(4G) wireless systems can provide data transmission rates
of 2–5 Mbit/s in macrocellular environments, and up to
10 Mbit/s in microcellular environments by deploying
dynamic channel assignment (DCA) based on interference
avoidance, combined with orthogonal frequency-division
multiplexing (OFDM). In [4], Erikkson proposed and
evaluated dynamic radio resource management schemes for
nonreal-time packet-mode communication over an OFDM
based downlink in terms of spectrum efficiency, fairness and
computational efficiency. In [5], Cimini et al. proposed the
concept of packetised DCA or dynamic packet assignment
(DPA), which involves fast measurements on all subcarriers
in parallel. In [6], Chuang and Sollenberger proposed a
system with DPA, that makes fast channel measurements
using the multicarrier nature of OFDM. A variation of
OFDM called orthogonal frequency-division multiple access
(OFDMA) has been proposed for the IEEE 802.16a wireless
metropolitan area networks (WMAN) in the 2–11 GHz
band [7, 8]. It is possible to obtain physical-layer rates of up
to 75 Mbit/s in the IEEE 802.16 WMAN for fixed broad-
band wireless access [9] using OFDM or OFDMA.
Analytical models have been developed in [10–13] to
study the performance of multiple-access systems. In [10],
Rubin and Tsai obtained message-delay distribution, using
r IEE, 2005
IEE Proceedings online no. 20041020
doi:10.1049/ip-com:20041020
Paper first received 14th November 2003 and in revised form 28th June 2004
two classes of traffic. In [11], Khan and Peyravi compared
the effects of five bursty distributions on buffer size and
end-to-end delay in cellular data networks. In [12], Prakash
and Veeravalli present analytical techniques for cellular
wireless packet data systems with incremental redundancy.
The authors presented a time-scale separation approach to
evaluate the mean delay and per-user throughput. The
models in [10–12] do not take the interference conditions
into account. In [13], Anand et al., presented an analysis to
evaluate the blocking probability in channelised cellular
systems with DCA; the analysis took into account voice-
only traffic. However, next-generation cellular systems will
typically use both voice and data traffic [14].
In this paper, we present a performance analysis of DPA
in OFDMA-based cellular systems with data traffic. In
particular, our analysis is applicable for fixed broadband
wireless access (FBWA) systems like the IEEE 802.16d [9],
i.e. the analysis can be used to evaluate the capacity of IEEE
802.16 WMAN. We consider a cellular OFDMA system
where each user requires a block of OFDM subcarriers for
transmission. We derive expressions for the mean data traffic
delay and the average system throughput on the uplink
(mobile-to-base station link) [15] and the downlink (base-
station-to-mobile link). Our analysis takes into account the
interference conditions on the OFDM subcarriers. We
model each cell as a buffer of infinite size. The state of the
buffer as seen by newly arriving data traffic in a cell is
modelled by a two tuple of non-negative integers ðm; kÞ,
where m represents the sum of the number of data bursts
currently being transmitted and the number of buffered data
bursts, and k represents the number of subcarriers that
cannot be used by newly arriving data traffic due to
violation of interference constraints. We model the state
space of the two tuples ðm; kÞ as a continuous-time Markov
chain (CTMC), and solve the CTMC to obtain the mean
delay and the average system throughput. We illustrate the
accuracy of our analysis by comparison with simulations.
2 System model

R. Jayaparvathy was with the AU-KBC Research Centre and is now with the Consider a cellular system with 61 circular cells as shown in
Coimbatore Institute of Technology, Coimbatore, India Fig. 1. Although a hexagonal model for cells is commonly
S. Anand is with Samsung India Software Operations, Bangalore, India used, we model the cells to be circles for analytical
S. Srikanth is with the AU-KBC Research Centre, Chennai, India simplicity. It is of interest to develop an analytical model

IEE Proc.-Commun., Vol. 152, No. 1, February 2005 45

 38
OFDM system since it is possible to measure the
61 39 interference on all subcarriers in parallel using the multi-
60 20 40 carrier nature of OFDM [16]. Hence, though the analysis is
59 37 21 41 valid for any FDMA system in general, it is particularly
58 36 8 22 42
35 19 9 23 applicable to OFDMA-based cellular systems.
57 18 2 10 43 The subcarrier allocation strategy to the data bursts is as
34 7 3 24 follows.
56 17 1 11 44
33 6 4 25
On the uplink, when a data burst arrives at a user in cell i,
55 16 5 12 45
32 15 13 26 the base station of cell i measures the interference on all the
54 31 14 27 46 subcarriers. A subcarrier c is called a ‘feasible subcarrier’ for
53 30 28 47 the data burst in cell i if the interference measured on
52 29 48
51 49 subcarrier c by the base station of cell i is below a specified
50 threshold E. The data burst is transmitted on a feasible
subcarrier, if available. If there are no feasible subcarriers
Fig. 1 Cellular system with 61 circular cells available, the data burst is buffered. The buffered data
Cells 2–7: first tier of interferers to cell 1 bursts are transmitted on a first-come first-serve (FCFS)
Cells 8–19: second tier of interferers to cell 1 basis if a subcarrier becomes feasible following the
Cells 20–37: third tier of interferers to cell 1 departure of a data burst from the system.
Cells 38–61: fourth tier of interferers to cell 1

On the downlink, when a data burst arrives for a user in
cell i, the user in cell i measures the interference on all the
active subcarriers. A subcarrier c is called a feasible subcarrier for
data burst
the data burst if the interference measured on subcarrier c
idle idle by the user in cell i is below a specified threshold, E. The
period period base station transmits the data burst to the user on a
T1 T2 T3 T4 T5 T6 feasible subcarrier, if available. If there are no feasible
time, t subcarriers available, the data burst is buffered. The
buffered data bursts are transmitted to users on an FCFS
Fig. 2 Data traffic with active bursts and idle periods basis if a subcarrier becomes feasible following a departure
from the system.
to compute the mean delay and the average system
throughput on the uplink and the downlink. We consider This subcarrier allocation strategy is a decentralised or
data-only traffic in our analysis. Data traffic consists of distributed mechanism, whereas the system for DPA
active data bursts and idle periods as shown in Fig. 2. In proposed in [3] considered a partially centralised algorithm
practice a data burst is a data packet of variable length, for for subcarrier allocation where the interference conditions
example an IP packet with zero idle time between a finite set were measured not only for the newly arriving user but also
of consecutive packets. Each active data burst requires a for estimation of the interference caused to the existing users
block of OFDM subcarriers for transmission. Allocation of in the system if the newly arriving user is admitted. We
a block of subcarriers (typically five to ten subcarriers) to consider the distributed system as it results in lower
each data burst enables large data-rates in 802.16 systems [9] communication complexity. Therefore in the system
and in 4G systems [3]. The blocks and block sizes are proposed in [3] data bursts do not undergo retransmission,
chosen such that the physical-layer propagation character- whereas in the system we consider, both on the uplink as
istics of each block are independent and identically well as on the downlink it is possible that a data burst
distributed (IID). Analysis of a system with varying block currently under transmission can be corrupted due to
sizes and varying requirements of number of blocks per violation of interference constraints during the period of
active burst is complex. Therefore to simplify the analysis transmission due to admitting a newly arriving data burst in
we assume that all blocks consist of equal number of another cell. This is further explained as follows.
subcarriers and it is necessary to allocate a block of
subcarriers to each active data burst. This is valid because
On the uplink, consider a data burst being transmitted on
the service stations or the user nodes in the IEEE 802.16d subcarrier c in the ith cell. Let there be a newly arriving data
system are usually of the grant per service station (GPSS) burst in cell j ðj ¼6 iÞ, which is also transmitted on subcarrier
type [8], for which it is reasonable to assume that all users c. If the additional interference caused by the newly arriving
require equal resources on an average [9]. data burst in cell j at the base station of cell i is such that it
The analysis we present is valid irrespective of the size of results in the interference measured on subcarrier c by the
the block. Therefore we present the analysis for a block size base station of cell i going above the threshold, E, it results in
of one subcarrier per block. We consider systems like the the data burst on subcarrier c in cell i getting corrupted.
IEEE 802.16d WMAN in which the traffic on the uplink
On the downlink, consider a data burst being transmitted
and on the downlink is expected to be symmetric [8]. on subcarrier c to a user in the ith cell. Let there be a newly
Therefore the block sizes are considered to be the same both arriving data burst in cell j ðj ¼6 iÞ, which is also transmitted
on the uplink and on the downlink. Our analysis is on subcarrier c. If the additional interference caused by the
independent of the block size. Henceforth, throughout the newly arriving data burst in cell j at the user in cell i is such
paper, we use the term ‘subcarrier’ to represent a block of that it results in the interference measured on subcarrier c by
subcarriers. the user in cell i going above the threshold E, it results in the
Subcarriers are allocated to the active data bursts and data burst on subcarrier c in cell i getting corrupted.
released during the idle periods. We consider OFDMA
because it is suitable for high data-rate transmissions and
provides flexibility in multiple-access and channel quality Both on the uplink and the downlink the corrupted
measurements [3]. Also, DPA is specifically designed for an data bursts are retransmitted after a random time. The
46 IEE Proc.-Commun., Vol. 152, No. 1, February 2005
retransmitted data bursts are treated as newly arriving data
bursts. Data bursts that are buffered due to nonavailability
of subcarriers suffer a delay. The delay increases due to
retransmissions. The mean delay is defined as the mean time
spent by a data burst in the buffer, which includes time
spent in the buffer by all the transmissions (i.e. first-time
transmission and all retransmissions) of the data burst.
The assignment of a subcarrier to a data burst among the
feasible subcarriers can be based on strategies like the best
subcarrier allocation, i.e. the feasible subcarrier with the
least interference, or random subcarrier allocation, i.e. any
of the feasible subcarriers at random. Simulation studies in
[17] show that the mean data burst delay and the average
system throughput performance of cellular systems with
DPA are similar for both the best and random subcarrier
allocation schemes. We consider the random subcarrier
allocation scheme for analytical simplicity. We make the
following assumptions to carry out the performance
analysis.

There are N ¼ 61 cells. All cells are of equal radius R,
with the base stations situated at the centres of each cell.

There are n subcarriers available for allocation in the
entire system.

The data burst arrival process in each cell is a Poisson
process with mean arrival rate l. The data burst holding
times are exponentially distributed with mean 1/m seconds.

The positions of users in any cell are uniformly
distributed over the area of the cell. The positions of
different users are independent of each other.

No two data bursts in the same cell are allocated the
same subcarrier, i.e. in any cell there is atmost one data
burst transmitted on every subcarrier.

The signal undergoes Rayleigh fading, log-normal
shadowing and attenuation due to the distance between
the users and the base stations. The path loss exponent is
taken to be four.

The signal propagation characteristics are independent
and identically distributed for all the subcarriers.

The users are assumed to have no mobility.
These assumptions are made on both the uplink and the
downlink.
3 Performance analysis
Consider the cellular system shown in Fig. 1. To obtain the
mean data burst delay and the average system throughput
for the system it is necessary to model the interference
conditions on the uplink and the downlink. It is also
necessary to take into account the arrival rates and holding
times of the data bursts.
3.1 Uplink
Let a user in cell i transmit a data burst on subcarrier c. Let
Si be the set of all cells other than cell i that have a data
burst transmitted on the same subcarrier c and hence cause
interference to cell i. Let Di ¼ jSi j be the number of
interferers to cell i on subcarrier c. To simplify the analysis
we assume that a base station experiences significant
interference only from the users in the immediate neigh-
bouring cells (i.e. the first tier of neighbouring cells).
Therefore in Fig. 1 S1  f2; 3; 4; 5; 6; 7g, i.e. 0  Di  6.
Let x be the probability that a subcarrier c is not feasible
for a data burst in cell i. A subcarrier c is not feasible in cell i
due to one of the following two reasons. There is no other
IEE Proc.-Commun., Vol. 152, No. 1, February 2005
data burst being transmitted on subcarrier c in cell i but the
interference on subcarrier c measured by the base station of
cell i is above the threshold E, or the data burst in cell i is
transmitted on subcarrier c and gets corrupted and hence
retransmitted. Let Pth(Di) be the probability that a
subcarrier is not feasible for a data burst in cell i due to
Di interferers causing the interference measured at the base
station of cell i to go above E, and let Prt(Di) be the
probability of data burst retransmission due to Di
interferers. Then x is given by
X
6 X
6
x ¼ ð1  pÞ Pth ðDi ÞbDi þ p Prt ðDi ÞbDi ð1Þ
Di ¼1 Di ¼1
where p is the probability of there being a data burst in cell i
transmitted on subcarrier c, and bDi is the probability mass
function of Di which is given by
(

6 Di ð6Di Þ
0  Di  6
bDi ¼ Di p ð1  pÞ ð2Þ
0 otherwise
The value of x obtained from (1) is used in Section 3.4 in
(4)–(12) to evaluate the mean delay and the average system
throughput.
3.2 Downlink
Let a user in cell i receive a data burst on subcarrier c. The
user in cell i receives interference from the base stations of
cells that have a data burst also being transmitted in
subcarrier c. As on the uplink, we make the assumption that
a user receives significant interference only from the base
stations of the first tier of neighbouring cells. Defining Si
and Di as previously (i.e. Section 3.1), the expression for x
given in (1) is also valid on the downlink. However, on the
downlink, Pth(Di) is redefined as the probability of the
interference measured by a user in cell i from Di
neighbouring base stations being above the threshold E.
3.3 Markov chain model
Both on the uplink as well as on the downlink each cell is
modelled as a buffer of infinite size. A newly arriving data
burst sees m1 subcarriers being used by other data bursts in
cell i, m2 data bursts buffered in cell i, and k subcarriers not
being feasible due to interference constraints. We model the
state seen by a newly arriving data burst in cell i as a two-
tuple of nonnegative integers ðm; kÞ, where m ¼ m1 þ m2 : It
is observed that m þ kon for a subcarrier to be available
for allocation to a newly arriving data burst. The state space
of two-tuples ðm; kÞ is modelled as a two-dimensional
continuous-time Markov chain (CTMC), with transition
rates as shown in Fig. 3. Transitions in the CTMC also
occur from the states (m, k) to ðm  1; k þ 1Þ due to
retransmissions of data bursts. We neglect these transitions
to simplify the analysis. However, we account for data burst
retransmissions by modifying the arrival rates to be l0 ¼
l=ð1  PR Þ; where PR is the data burst retransmission
probability, given by
X
6
PR ¼ Prt ðDi ÞbDi ð3Þ
Di ¼1
in which Prt ðDi Þ is as defined in Section 3.1 and bDi is as
given in (2). The CTMC in Fig. 3 has transition rates of the
form m0 a and k0 b, where m0 ; k 0 2 f1; 2;    ; ng: The
analysis to obtain the values of a and b is complex. It is
shown in (4)–(12) that the mean delay and the system
throughput depend on the ratio b/a, and not on the actual
values of a and b. Hence to compute b/a we model the state
of a subcarrier seen by a newly arriving data burst as a two-
47

′
′ .....................................................................................................................
0,n 1,n

b na b na

′
′ ......................................................................................................................
0,n−1 1,n−1 .......................................................................................................................
 .
.

′
′ .
′
′
′ .........
0,2 1,2 2,2 ..............n−2,2 n−1,2 n,2 n+1,2 .........
 2 (n−2) (n−2) (n−2)

(n−1)b 2a (n−1)b 2a (n−1)b 2a (n−1)b 2a(n−1)b 2a(n−1)b 2a (n−1)b 2a

′
′
′
′
′
.........
0,1 1,1 2,1 ................. n−2,1 n−1,1 n,1 n+1,1 .......
 2 (n−1)  (n−1)  (n−1) 
nb a nb a nb a nb a nb a nb a nb a

′
′
′
′
′
0,0 1,0 2,0 n−2,0 n−1,0 n ,0 n+1,0 ...........
............. ..........
 2 (n−1) n n

Fig. 3 Markov chain model for state seen by a newly arriving data burst

b The details are given in the Appendix. From the two-state

CTMC in Fig. 4, b/a can be obtained as
b x
¼ ð7Þ
subcarrier
subcarrier a 1x
not
feasible feasible The probability of a subcarrier being allocated to a data
burst p is given by
Xn Xnm
minðm; n  kÞ
a p¼ pðm; kÞ ð8Þ
m¼0 k¼0
n
Fig. 4 Two-state CTMC model to evaluate b/a
The value of p obtained from (8) is used in (1) and (2) to
state CTMC as shown in Fig. 4, in which the probability of compute b/a, which in turn is used in (4)–(6) to compute
a subcarrier being feasible is 1  x; and that of a subcarrier pðm; kÞ: From (1), (2), (4), (7) and (8) it is observed that p
not being feasible is x, where x is obtained from (1). and pðm; kÞ are to be computed iteratively. The mean
The structure of the CTMC shown in Fig. 3 is valid for number of buffered data bursts Nb is given by
X
both the uplink and the downlink and therefore the analysis Nb ¼ ðm þ k  nÞpðm; kÞ ð9Þ
we present in the following Section is valid for both. mþkn
However, we point out that the values of the transition rates
is obtained using Little’s theorem [18] as
The mean delay D
a and b are different for the uplink and the downlink.
Therefore the values of the mean delay and the average
¼ Nb
D ð10Þ
system throughput are expected to differ on the uplink and l0
downlink. The average system throughput Z is given by
X m
3.4 Mean delay and average system Z ¼ ð1  PR Þ ð1  Pb Þ pðm; kÞ ð11Þ
n
throughput mþkn
It can be shown that the CTMC in Fig. 3 is positive where PR is obtained from (3), and Pb is the probability of a
recurrent if and only if r^9l0 =mon: When the CTMC is data burst being buffered, which is given by
positive recurrent the steady-state probability for the X
occupancy of state (m, k), pðm; kÞ can be obtained as Pb ¼ pðm; kÞ ð12Þ
8      mþkn
>
>
> 1 r ^m n b k To evaluate (4)–(12) it is necessary to evaluate x, and hence
>
> m  n; k  n
< AB m! k a Pth(Di) and Prt(Di). We present the analysis to obtain Pth(Di)
pðm; kÞ ¼     k
1 ^m n
r b and Prt(Di) in Section 3.5.
>
> m4n; k  n
>
> n mn AB n! k a
>
: 3.5 Evaluation of Pth(Di) and Prt(Di)
0 otherwise
ð4Þ 3.5.1 Uplink: On any subcarrier c, the interference
received by the base station of cell i from the users in Di
where A and B are normalisation terms given by ðuÞ
neighbouring cells Ii (Di) is given by
^nþ1
r X n
^m
r X
A¼ þ ð5Þ ðuÞ
n!ðn  r^Þ m¼0 m! Ii ðDi Þ ¼ D4 ðBi ; Pj Þ10cij =10 Uij ð13Þ
j2si
jsi j¼Di
and
 n where Uij represents the Rayleigh fading loss from the user
b
B¼ 1þ ð6Þ in cell j to the base station of cell i, D(Bi,Pj) is the distance
a between the base station of cell i (located at Bi), and the user
48 IEE Proc.-Commun., Vol. 152, No. 1, February 2005
in cell j (located at Pj), and 10cij =10 represents the log- the number of interferers Di. To determine Prt (Di) in (15) it
normal shadow loss from the user in cell j to the base is necessary to evaluate the joint probability
n o
station of cell i. In (13), cijBNð0; s2 Þ. Pth (Di) and Prt (Di) ðuÞ ðuÞ
PJ ¼ Pr Ii ðDi Þ4E; Ii ðDi  1Þ  E ð23Þ
are given by
n o
ðuÞ
Pth ðDi Þ ¼ Pr Ii ðDi Þ4E ð14Þ and the marginal probability
n o
ðuÞ
PM ¼ Pr Ii ðDi  1Þ  E ð24Þ
and
n  o
ðuÞ  ðuÞ PM in (24) is evaluated using the method similar to the one
Prt ðDi Þ ¼ Pr Ii ðDi Þ4EIi ðDi  1Þ  E ð15Þ
used to evaluate Pth (Di). In [20], Anand et al., computed an
respectively. Therefore to evaluate Pth (Di) and Prt (Di) it is expression similar to PJ in (23) for CDMA systems by using
necessary to evaluate the cumulative distribution function Fenton’s approximation and ln ð1 þ xÞ
x forj xjo1: We
of I(u)
i (Di). Conditioned on the positions of the users and the
adopt the approach in [20] to evaluate PJ and hence Prt (Di).
Rayleigh fading loss, each term in the summation of (13) is The retransmission probability PR is obtained from (3). The
log-normal of the form 10Oij =10 , where Oij  Nðmij ; s2 Þ values of Prt (Di) obtained fro (15) and Pth (Di) obtained
and mij is given by from (22) are used to evaluate x in (1), which in turn is used
in (4)–(12) to evaluate the mean data burst delay and the
mij ¼ 40 log10 ½DðBi ; Pj Þ þ 10 log10 U ð16Þ average system throughput on the uplink.
In (16), U is an exponentially distributed random variable
with unit mean. Using the approach described by Fenton in
3.5.2 Downlink: On any subcarrier c the interference
ðuÞ received by a user in cell i from the base stations of Di
[19], Ii ðDi Þ can be approximated to be a log-normally ðdÞ
neighbouring cells Ii (Di) is given by
distributed random variable of the form 10ODi =10 , where X
ðdÞ
ODi  NðmDi ; s2Di Þ and s2Di and mDi can be obtained as Ii ðDi Þ ¼ D4 ðPi ; Bj Þ10cji =10 Uji ð25Þ
2
2 2  3 j2si
w s 2wmij
1 e  1  j2s i
e jsi j¼Di
s2Di ¼ 2 ln41 þ  2 5 ð17Þ where Uji represents the Rayleigh fading loss from the base
w j2s ewmij i
station of cell j to the user in cell i, D (Pi,Bj) is the distance
and between the user in cell i located at Pi, and the base station
" # of cell j located at Bj, and 10cji =10 represents the log-
wðs2Di  s2 Þ 1 X
mDi ¼  ln ewmij ð18Þ normal shadow loss from the base station of cell j to the
2 w j2si user in cell i. In (13) cji  Nð0; s2 Þ:
Pth (Di) and Prt (Di) are given by
In (17) and (18), w9ð1n 10Þ=10, and mij is obtained from n o
ðdÞ
(16). Averaging over the Rayleigh fading loss and the Pth ðDi Þ ¼ Pr Ii ðDi Þ4E ð26Þ
position of the user, Pth ðDi Þ is given by
  Z Z   and
1 Di EdB þ mDi n  o
Pth ðDi Þ ¼    Q ðdÞ  ðdÞ
pR2 sDi ð19Þ Prt ðDi Þ ¼ Pr Ii ðDi Þ4EIi ðDi  1Þ  E ð27Þ
eU ti01    ti0D dri01 dyi01    dri0D dyi0D dU respectively. Equation (26) is evaluated as we evaluated (14)
i i i
in Section 3.5.1. Thus on the downlink Pth (Di) is obtained
where EdB ¼ 10 log10 E; and
Z N 2
from (22) by replacing s2Di and mDi by
1 y =2 2
2 2  3
QðxÞ ¼ pffiffiffiffiffiffi dy e w s
 1 j2si e2wmji
x 2p 2 1 4 5
sDi ¼ 2 ln 1 þ  2 ð28Þ
From (19) it is noted that the computation of Pth (Di) w j2s ewmji
i
involves 2Di+2 integrations, which results in 14 integrations
when Di ¼ 6. To simplify the numerical computations we and
" #
assume that the interferers are IID. This assumption is valid wðs2Di  s2 Þ 1 X
wmji
due to the symmetry of the system. The expressions for s2Di mDi ¼  ln e ð29Þ
2 w j2si
and mDi are then given by
2
2 2 3 In (28) and (29) mji is given by
w s
1 e  1  

s2Di ¼ 2 ln 41 þ 5 ð20Þ mji ¼ 40log10 D Pi; Bj þ 10log10 U ð30Þ

w Di
The expression for Prt ðDi Þ in (27) is evaluated by once again
and adopting the approach in [20]. The values of Pth ðDi Þ and

 Prt ðDi Þ thus obtained are used to evaluate x in (1), which in
s2Di  s2 1 turn is used in (4)–(12) to compute the mean data burst
mDi ¼ mij þ w  ln ½Di  ð21Þ delay and the average system throughput on the downlink.
2 w
Pth (Di) is then given by
  Z Z Z   4 Results and discussion
1 Di EdB þ mDi U
Pth ðDi Þ ¼ Q e rdrdydU We use the following values for the computations: N ¼ 61
pR2 sDi
cells, n ¼ 150 subcarriers, 1/m ¼ 125 ms, E ¼ 13 dB,
ð22Þ s ¼ 8 dB, and R ¼ 1. The data burst arrival rate l is varied
From (22) it is observed that only four integrations have to to obtain a load r9l=m in the range 1 to 40 Erlangs per
be performed to evaluate Pth (Di), irrespective of the value of cell. The IEEE 802.16d standard for WMAN specifies a

IEE Proc.-Commun., Vol. 152, No. 1, February 2005 49

maximum of 2048-point FFT (i.e. 2048 subcarriers) for the
OFDMA-based system [8]. However, typical OFDMA
based WMAN systems deploy a 1024-point FFT [7] of
which about 768 subcarriers are used for data traffic and the
remaining for pilot signals. Hence we consider about 150
blocks with about 5 to 6 subcarriers per block. As
mentioned in Section 2, our analysis is independent of
block size and hence we perform the analysis for 150
subcarriers.
The performance of dynamic channel allocation (DCA)
is also provided for comparison. DCA refers to allocation
of subcarriers to data traffic both to the active data bursts as
well as during the idle periods. The DCA traffic model is a
circuit-switched version of DPA. Although this model is not
used in wireless data networks we provide the comparison
to emphasise the advantage of DPA and that of OFDMA
in cellular systems as DPA is feasible only over systems with
OFDM or OFDMA which is deployed only in cellular
systems like the IEEE 802.16 or in 4G cellular systems. The
arrival rate and holding time for the DCA computations are
obtained as follows. Consider the representation of data
traffic with active data bursts and idle periods as shown in
Fig. 2. Usually active bursts form 25% of the data traffic i.e.
in Fig. 2 (T2T1+T4T3+T6T5)/(T6T1)E0.25. For
data traffic comprising of an average of 100 active data
bursts, the mean call holding time to perform the DCA
computations 1/mdca is given by 1/mdca ¼ 100/(m * 0.25) ¼ 50 s
if m ¼ 125 ms. The arrival rate for DCA computations ldca is
given by ldca ¼ l=100:
Figures 5 and 6 present the mean delay performance as a
function of the data traffic per cell on the uplink and the
downlink, respectively. The analytical results match closely
with the simulations, thus validating our analytical
approach. The Erlang capacity of the system is defined as
the maximum carried traffic (in Erlangs) to obtain a
specified mean delay performance. From Fig. 5, the Erlang
capacity to obtain a mean delay of 1 ms is about 32.5
Erlangs with DPA and 7 Erlangs with DCA. This results in
an improvement of the Erlang capacity on the uplink by
about a factor of 4.5 with DPA, as compared with DCA.
This improvement in the Erlang capacity with DPA is
because in DPA, the subcarrier is released during the idle
periods of data traffic, and hence can be allocated to the
active data burst corresponding to the data traffic for
another user in the cell, thereby reducing the delay, whereas
in DCA the subcarrier is held both by the active data bursts
as well as during the idle periods. In other words, the
102 DPA− analysis
DPA− simulation
DCA
100
mean delay, s
10 −2
10 −4
10 −6
5 10 15 20 25
data traffic per cell, Erlangs
Fig. 5 Mean data burst delay: uplink, N ¼ 61 cells, n ¼ 150
subcarriers
50
100
DPA −analysis
DPA −simulation
10−1 DCA
10−2
mean delay, s
10−3
10−4
10−5
10−6
10 15 20 25 30 35 40
data traffic per cell, Erlangs
Fig. 6 Mean data burst delay: downlink, N ¼ 61 cells, n ¼ 150
subcarriers
resources are better utilised when subcarriers are allocated
to the active data bursts alone and released during the idle
periods.
This can be quantitatively explained as follows. Consider
an arrival rate of 1 call per second in DCA. As mentioned
earlier, m1
dca ¼ 50 s thus leading to a load of r ¼ 50 Erlangs.
For the same data traffic, if DPA is used, this corresponds
to an arrival rate of 100 bursts per second and for
m1
dpa ¼ 125 ms it results in a load of r ¼ 12.5 Erlangs. Thus
for the considered traffic statistics, DPA loads the system
less than DCA by a factor of about 1/4. Hence a gain of
about four times can be obtained in the system capacity by
using DPA as against DCA. However, the actual value of
the gain in capacity need not be exactly four since the active
bursts are randomly distributed within the data session.
Similarly, from Fig. 6 it is observed that the Erlang
capacity to obtain a mean delay of 1 ms is about 34 Erlangs
with DPA, and about 12 Erlangs with DCA. This results in
an improvement of the Erlang capacity on the downlink by
about a factor of three with DPA, as compared with DCA.
The absolute value of the mean data burst delay is smaller
on the downlink than on the uplink (e.g. a mean delay of
1ms occurs at r ¼ 34 Erlangs with DPA and r ¼ 12 Erlangs
with DCA on the downlink, and at r ¼ 32.5 Erlangs with
DPA and r ¼ 7 Erlangs with DCA on the uplink). The
reason for the reduction in the mean data burst delay on the
downlink is as follows. On the downlink, interference is
caused to a user from six neighbouring base stations. A user
is always nearer to three of the six neighbouring base
stations and far away from the other three base stations.
Therefore only three of the six neighbouring base stations
contribute significantly to the interference measured by a
user. However, on the uplink the interference caused to a
base station is due to users present in the six neighbouring
cells and the average interference caused is equivalent to the
interference caused when all the six users are positioned at
the centres of their respective cells. Thus significant
interference is caused by users present in all the six
neighbouring cells. Therefore the average interference at a
user on the downlink is lower than the average interference
at a base station on the uplink. Hence, the buffering and
retransmission probabilities on the downlink are lower than
those on the uplink. This leads to lower mean delay on the
30 35 downlink. Since the absolute values of mean delays on the
uplink and the downlink are different with both DPA and
DCA, the factor of improvement in the Erlang capacities is
also different.
IEE Proc.-Commun., Vol. 152, No. 1, February 2005
 100
DPA −analysis
DPA −simulation
DCA
average system throughput
10−1
10−2
10−3
0 5 10 15
data traffic per cell, Erlangs
Fig. 7 Average system throughput: uplink, N ¼ 61 cells, n ¼ 150
subcarriers
100
DPA − analysis
DPA − simulation
DCA
average system throughput
10−1
10−2
10−3
0 5 10 15
data traffic per cell, Erlangs
Fig. 8 Average system throughput: downlink, N ¼ 61 cells,
n ¼ 150 subcarriers
Figures 7 and 8 present the average system throughput
performance for varying data traffic on the uplink and the
downlink, respectively. The throughput remains the same
with both DCA as well as DPA. This is because throughput
is the effective utilisation of the subcarrier, and DPA
represents the effective utilisation periods (i.e. active bursts)
of DCA.
5 Conclusions
We have presented a performance analysis of OFDMA-
based cellular systems with DPA. We derived expressions
for the mean delay and the average system throughput on
the uplink and the downlink. Our analysis took into
account the interference conditions on the OFDM sub-
carriers in addition to the traffic load conditions, and was
shown to be accurate. We also compared the performance
of DPA with that of DCA and showed that higher Erlang
capacities are possible with DPA as compared with DCA.
This result is useful because it was already shown in [2] that
interference avoidance techniques of DCA could result in
capacities higher than that of CDMA, which in turn
provided larger capacities compared with circuit-switched
systems like GSM and GPRS. Hence the advantage in
using DPA makes DPA a very attractive mechanism to
provide high data-rates over wireless channels.
IEE Proc.-Commun., Vol. 152, No. 1, February 2005
An extension of our approach to incorporate varying
block sizes and varying requirements per active data burst is
a topic for further investigation. Our analytical approach
can also be extended to study the performance of OFDM-
based wireless adhoc networks.
6 References
1 Ojanpera, T., and Prasad, R.: ‘An overview of air interface multiple
access for IMT-2000/UMTS’, IEEE Commun. Mag., 1998, 36, (9), pp.
82–95
2 Pottie, G.J.: ‘System design choices in personal communications’,
IEEE Pers. Commun., 1995, 2, (5), pp. 50–67
3 Chuang, J., and Sollenberger, N.: ‘Beyond 3G: Wideband wireless
data access based on OFDM and dynamic packet assignment’, IEEE
Commun. Mag., 2000, pp. 78–87
4 Erikkson, M.: ‘Dynamic single-frequency networks’, IEEE J. Sel.
20 25 30 35 Areas Commun., 2001, 19, (10), pp. 1905–1914
5 Cimini, L.J. Jr., Chuang, J.C.I., and Sollenberger, N.R.: ‘Advanced
cellular internet service’, IEEE Commun. Mag., 1998, pp. 150–159
6 Chuang, J.C.I., and Sollenberger, N.R.: ‘Spectrum resource allocation
for wireless packet access with application to advanced cellular
internet service’, IEEE J. Sel. Areas Commun., 1998, 16, (6), pp. 820–
829
7 IEEE P802.16a/D3-2001: ‘Draft amendment to IEEE standard for
local and metropolitan area networks – Part 16: Air interface for fixed
wireless access systems – Medium access control modifications and
additional physical-layer specifications for 2–11 GHz’ March 2002
8 Eklund, C., Marks, R.B., Stanwood, K.L., and Wang, S.: ‘IEEE
standard 802.16: A technical overview of wireless MAN air interface
for broadband wireless access’, IEEE Commun. Mag., 2002, pp. 98–
107
9 IEEE P802.16-Revd/D3-2004: ‘Draft amendment to IEEE standard
for local and metropolitan area networks – Part 16: Air interface for
fixed wireless access systems – Medium access control modifications
and additional physical-layer specifications for 2–11 GHz’ January
2004.
10 Rubin, I., and Tsai, Z.H.: ‘Message delay analysis of multiclass
priority TDMA, FDMA and discrete time queueing systems’, IEEE
Trans. Inf. Theory, 1989, 35, (3), pp. 637–647
11 Khan, K., and Peyravi, H.: ‘Delay and queue size analysis of TDMA
systems with general traffic’. Presented at 6th Int. Symp. on Modeling,
Analysis and Simulation of Computer and Telecommunication
Systems, July 1998
12 Prakash, R., and Veeravalli, V.V.: ‘Wireless packet data systems with
incremental redundancy – Uplink analysis’. Presented at Conf. on
Information Science and Systems, Princeton University, March 2002
20 25 30 35 13 Anand, S., Sridharan, A., and Sivarajan, K.N.: ‘Performance analysis
of channelised cellular systems with dynamic channel allocation’,
IEEE Trans. Veh. Technol, 2003, 52, (4), pp. 847–859
14 Gummalla, A.V., and Limb, J.O.: ‘Wireless medium access control
protocols’, IEEE Commun. Surveys and Tutorials, Second Quarter
2000
15 Jayaparvathy, R., Anand, S., and Srikanth, S.: ‘Dynamic packet
assignment in OFDM-based cellular systems – performance analysis’.
Presented at IEEE Conf. APCC 2003, Sep 2003
16 Qiu, X., Chawla, K., Chuang, J., and Sollenberger, N.R.: ‘Network-
assisted resource management for wireless data networks’, IEEE J.
Sel. Areas Commun., 2001, 19, (7), pp. 1222–1234
17 Sridhar, M., and Namrata, D.: ‘Performance of uplink MAC layer in
4 G cellular networks’. B.E. Project Report, AU-KBC Research
Centre, Madras, India
18 Bertsekas, D., and Gallager, R.: ‘Data networks’ (Prentice Hall of
India, 2000)
19 Fenton, L.F.: ‘The sum of log-normal probability distributions in
scatter transmission systems’, IRE Trans. Commun. Syst., 1960, CS-8
20 Anand, S., Chockalingam, A., and Sivarajan, K.N.: ‘Outage and
capacity analysis of cellular CDMA with admission control’.
Presented at IEEE Conf. WCNC’ 2002, March 2002
7 Appendix: Derivation of pðm; kÞ in (4)
The differential–difference equations of the CTMC in Fig. 3
is given by
p0 ð0; 0Þ ðtÞ ¼a pð0; 1Þ ðtÞ þ mpð1; 0Þ ðtÞ
 ðnb þ l0 Þpð0; 0Þ ðtÞ ð31Þ
p0 ð0; nÞ ðtÞ ¼b pð0; n  1Þ ðtÞ þ mpð1; nÞ ðtÞ
 ðna þ l0 Þpð0; nÞ ðtÞ ð32Þ
51
 p0 ðm; 0Þ ðtÞ From (40) and (41) we obtain
¼ apðm; 1Þ ðtÞ þ ðm þ 1Þmpðm þ 1; 0Þ ðtÞ
þl0 pðm  1; 0Þ ðtÞ pðm; kÞ
8    k
ðnb þ l0 þ mmÞpðm; 0Þ ðtÞ 1  mon >
> ^m n
r b
>
> pð0; 0Þ 0  m  n; 0  k  n
ð33Þ >
> m! k a
<
¼  mn    k
p0 ð0; kÞ ðtÞ > r
^ n b
>
> pðn; 0Þ n nomo1; 0  k  n
¼ ðk þ 1Þapð0; k þ 1Þ ðtÞ þ mpð1; kÞ ðtÞ >
> k a
>
:
þðn  k þ 1Þ bpð0; k  1Þ ðtÞ 0 otherwise
½ðn  kÞb þ l0 þ kapð0; kÞ ðtÞ 1  kon ð34Þ
ð42Þ
0
p ðm; kÞ ðtÞ i.e.
¼ l0 pðm  1; kÞ ðtÞ þ ðk þ 1Þapðm; k þ 1Þ ðtÞ
þðm þ 1Þ mpðm þ 1; kÞ ðtÞ 1  mon pðm; kÞ
8    k
þðn  k þ 1Þ bp ðm; k  1Þ ðtÞ >
> ^m
r n b
>
> pð0; 0Þ m! 0  m  n; 0  k  n
½ðn  k Þb þ l0 þ kaj mmp ðm; kÞ ðtÞ 1  kon >
> k a
<
ð35Þ ¼ m
   k
> r
^ n b
>
> pð0; 0Þ mn nomo1; 0  k  n
p0 ðm; nÞ ðtÞ >
> n!n k a
>
:
¼ l0 pðm  1; nÞ ðtÞ þ ðm þ 1Þmp ðm þ 1; nÞ ðtÞ 0 otherwise
þbpðm; n  1Þ ðtÞ  ðl0 þ na þ mmÞpðm; nÞ ðtÞ 1  mon ð43Þ
ð36Þ P1 Pn
Since m¼0 k¼0 pðm; kÞ ¼ 1
0
p ðm; 0Þ ðtÞ X    k "X #
n
n b n
^m 1 X
r 1
^m
r
¼ apðm; 1Þ ðtÞ þ nmpðm þ 1; 0Þ ðtÞ pð0; 0Þ þ
k a m! n! m¼nþ1 nmn
þl0 pðm  1; 0Þ ðtÞ k¼0 m¼0
ðnb þ l0 þ nmÞ p  ðm; 0Þ ðtÞ 1  mon ð37Þ ¼1 ð44Þ
p0 ðm; kÞ ðtÞ i.e.
¼ l0 pðm  1; kÞ ðtÞ þ ðk þ 1Þapðm; k þ 1Þ ðtÞ   " n m 1  mn1
#
þnmpðm þ 1Þ nmpðm þ 1; kÞ ðtÞ n  mo1 b n X r
^ ^nþ1 X
r r
^
pð0; 0Þ 1 þ þ
þðn  k þ 1Þ bp ðm; k  1Þ ðtÞ a m¼0
m! n! m¼nþ1 n
½ðn  k Þb þ l0 þ ka þ nmp ðm; kÞ ðtÞ 1  kon ð38Þ
¼1 ð45Þ
and
The summation in the LHS of (45) exists if and only if
p0 ðm; nÞ ðtÞ ^on; which is also a necessary and sufficient condition for
r
¼ l0 pðm  1; nÞ ðtÞ þ nmpðm þ 1; nÞ ðtÞ ^on; pð0; 0Þ is
the stability of the system. Therefore when r
þbpðm; n  1Þ ðtÞ obtained as
ðl0 þ na þ nmÞpðm; nÞ ðtÞ n  mo1 ð39Þ
1
To prove (4) we proceed as follows. Let the steady-state p ð0; 0Þ ¼ ð46Þ
AB
probability of the system being in state ðm; kÞ be p ðm; kÞ.
At steady state, i.e. as t-N, pðm; kÞðtÞ ! pðm; kÞ, and where
p0 ðm; kÞðtÞ ! 0. Therefore at steady state the probability ^nþ1
r X
n
^m
r
pðm; kÞ can be obtained from (31)–(36) as A¼ þ ð47Þ
n!ðn  r^Þ m¼0 m!
pðm; kÞ
8   and
>
> pðm; k  1Þ ðn  k þ 1Þ b
< 0  m  n; 0  k  n  n
k a b
¼ B¼ 1þ ð48Þ
>
> r
^ a
: pðm  1; kÞ 0  m  n; 0  k  n
m From (43) and (46)
ð40Þ
pðm; kÞ
where r^9l0 =m. Similarly from (37)–(39), p(m,k) can be 8      k
written as >
> 1 r^m n b
>
> 0  m  n; 0  k  n
>
> AB m! k a
piðm; kÞ <
8   ¼   m
   k
pðm; k  1Þ ðn  k þ 1Þ b > 1 r
^ n b
>
> nomo1; 0  k  n >
> nmn AB n! k nomo1; 0  k  n
< >
> a
k a >
:
¼ 0 otherwise
>
> r
^
: pðm  1; kÞ nomo1; 0  k  n
n ð49Þ
ð41Þ

52 IEE Proc.-Commun., Vol. 152, No. 1, February 2005