See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228650814

Opportunistic fair scheduling for the downlink of IEEE 802.16 wireless

metropolitan area networks

Article · January 2006

DOI: 10.1145/1185373.1185441

CITATIONS READS
40 83

4 authors, including:

Mehri Mehrjoo Xuemin Sherman Shen

Institute of Electrical and Electronics Engineers University of Waterloo
21 PUBLICATIONS   267 CITATIONS    1,239 PUBLICATIONS   36,302 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Data Routing in VANETs View project

Mobile social network and vehicular network View project

All content following this page was uploaded by Mehri Mehrjoo on 14 July 2014.

The user has requested enhancement of the downloaded file.

 Opportunistic Fair Scheduling for the Downlink of IEEE 802.16 Wireless
Metropolitan Area Networks
Mehri Mehrjoo, Mehrdad Dianati, Xuemin (Sherman) Shen, Kshirasagar Naik
Department of Electrical and Computer Engineering
University of Waterloo
Waterloo, Ontario, Canada N2L 3G1
mmehrjoo, mdianati, xshen, snaik@uwaterloo.ca
Abstract
In this paper, we propose a novel scheduling scheme for
the downlink of IEEE 802.16 networks. A scheduler at the
Base Station (BS) decides the order of downlink bursts to
be transmitted. The decision is made based on the quality
of the channel and the history of transmissions of each Sub-
scriber Station (SS). The scheduler takes advantage of tem-
poral channel fluctuations to increase the BS’s throughput
and maintain fairness by balancing the long term average
throughput of SSs. Simulation results are given to demon-
strate the performance of the proposed scheduling scheme.
1 Introduction
Wireless Metropolitan Area Networks (WMANs) have
gained considerable attractions because they provide wire-
less broadband services over a large coverage area. High
data rates as well as cost effective installation of WMANs
have motivated tremendous effort to introduce industry
standards for the deployment of WMANs. The initial ver-
sion of the IEEE 802.16 standard [1] defines specifica-
tions for the air interface of WMANs to operate at 10 to
66 GHz. The IEEE 802.16a [2] includes another opera-
tion frequency at 2-11 GHz which is favorable for Non-
Line-Of-Sight (NLOS) transmission at urban areas. The
IEEE 802.16d standard [3], the most recent version, ex-
tends the frequency band by combining the IEEE802.16 and
IEEE 802.16a standards.The IEEE 802.16a/d standards de-
fine two infrastructures for the air interface: Point to Multi-
point (PMP) and Mesh. In PMP, Subscriber Stations (SSs)
messages are transmitted via the Base Station (BS). By con-
trast, in Mesh, SSs can communicate either directly or an SS
relay an ongoing communication between two other SSs. In
PMP configuration, a single BS controls the communica-
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.
QShine’06 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
tions with multiple SSs in both downlink (DL) and uplink
(UL) directions. The IEEE 802.16 standard does not specify
transmission scheduling schemes. Therefore, vendors have
to define proper scheduling schemes in order to utilize the
network resources efficiently.
In this paper, we propose an effective scheduling scheme
for an IEEE 802.16 WMAN. We consider a BS in PMP in-
frastructure that forwards both real-time and non-real-time
traffic to the SSs. In PMP infrastructure, the scheduling is
a centralized one. The BS schedules packets in UL and DL
subframes, respectively. Therefore, improving the perfor-
mance of the DL is crucial in PMP infrastructure, to pre-
vent the BS from becoming a bottleneck. This is an im-
portant issue as the traffic load is expected to be higher in
the DL than the UL. Furthermore, the efficient utilization
of the DL bandwidth for non-real-time traffic allows for
more bandwidth for real-time traffic such as VoIP and video
MPEG. Real-time traffic has tight transmission time con-
straints. On the other hand, non-real-time and best-effort
traffic types are delay tolerant. Therefore, an opportunistic
scheduling scheme can take advantage of channel fluctua-
tions to improve network throughput by scheduling delay
tolerant packets in proper time subframes.
The proposed scheduling scheme deploys a cross-layer
design concept. That is, the scheduler in the BS schedules
packets of each SS based on the maximum capacity strat-
egy. The better the channel quality that an SS has in the
current frame, the higher the priority that an SS can trans-
mit in the following frame. In other words, transmission to
the SSs with bad channel condition can be postponed until
their channel recover from bad condition. As this scheme
may not be fair to the SSs that have a bad channel qual-
ity most of the time, a balancing mechanism is required to
maintain fairness [4][5]. We adopt a utility-based [6][7] no-
tion of fairness. The fairness mechanism takes into consid-
eration the amount of previously transmitted data to each SS
[8][9]. The proposed scheduling is practical and conforms
with the IEEE 802.16 adaptive transmission rate and signal-
ing mechanisms. Also, our simulation results demonstrate
that the proposed scheme significantly improves the perfor-
mance efficiency metrics such as throughput and fairness.
The rest of this paper is organized as follows. In sec-
tion 2, the IEEE 802.16 broadband wireless access standard
is introduced. The opportunistic fair scheduling scheme is
proposed in section 3. Section 4 presents the simulated sce-
nario and results. The concluding remarks and future work
are given in section 5.
2 IEEE 802.16a/d Specification
This section provides an overview of the IEEE 802.16a/d
specifications relevant to the scheduling at the MAC layer.
The MAC layer of IEEE 802.16a can interact with both sin-
gle carrier and multicarrier modulation air interfaces. The
multicarrier modulation is based on orthogonal frequency-
division multiplexing (OFDM). OFDM based transmission
simplifies the equalization process of a receiver and re-
duces the interference of NLOS transmissions. Moreover,
resource management techniques can take advantage of Dy-
namic Subcarrier Assignment (DSA). In other words, in
each assignment interval, a BS assigns a set of subcarri-
ers to an SS that are not in deep fading in that interval. In
this paper, we focus on a WirelessMAN-OFDM air inter-
face and assume the required channel information can be
obtained by a pilot based channel estimation method [10].
The channel information are then deployed in the adaptive
modulation and coding technique of the IEEE 802.16 pro-
tocol to achieve robust transmissions in case of interference
and poor channel conditions. Upon receiving feedback in-
formation about a channel status, a transmitter switches to a
proper modulation and coding. In a bad channel condition,
the transmission rate is decreased to achieve an acceptable
level of Bit Error Rate (BER) and vice versa.
In a WMAN with PMP topology, the UL channel is
shared by all SSs. SSs gain access to the channel based
on the Time Division Multiple Access (TDMA) technique.
On the other hand, DL path is a broadcast channel. The BS
broadcasts information based on the Time Division Mul-
tiplexing (TDM) into the DL subframe. UL and DL du-
plexing methods are Time Division Duplexing (TDD), Fre-
quency Division Duplexing (FDD), and Half-duplex FDD
(HFDD). In TDD, the UL and DL transmissions may share
the same frequency band but take place alternatively at time
intervals. On the contrary, in FDD, the UL and DL trans-
missions may occur at the same time interval, but the UL
and DL transmission frequencies are different. HFDD is
the same as FDD except that the UL and DL transmission
time intervals are concatenated.
In the context of the IEEE 802.16 standard, transmis-
sions to or from an SS are in the form of bursts. Each burst
in a DL channel may contain more than one Medium Access
Protocol Data Units (MPDUs) destined for different SSs.
However, in an UL channel each burst carries MPDUs of an
individual SS. An MPDU is a variable size packet formed
by aggregation or defragmentation of upcoming packets
from the higher layer to the MAC layer. An MPDU consists
of a MAC header, subheaders, a PDU payload, and a 32-bit
Cyclic Redundancy Check (CRC-32) as shown in Fig. 1. A
fixed-length MAC header is mandatory in each MPDU, but
the other parts of an MPDU can be optional. A MAC header
indicates the connection that the MPDU belongs to as well
as the length and the type of the payload and subheaders.
If an MPDU consists of only a MAC header, it is consid-
ered as bandwidth request message. Subheaders, when pre-
sented, demonstrate some information regarding the current
MPDU content. For instant, subheaders may determine the
presence of a fragmentation or a packing in the MPDU pay-
load, or they may convey bandwidth management messages
such as a piggyback request message. A PDU payload is
of variable size. This specification enables an MPDU to
encapsulate the upper layer packets without any concern of
their content. The payload may be data or a MAC man-
agement message. A large number of MAC management
messages can be exchanged in the PDU payload.
Generic MAC Header Payload (optional) CRC(optional)
Figure 1. A MAC PDU of 802.16
Provided that the duplexing method is TDD, an UL and
a DL subframe are concatenated in a MAC frame, as shown
in Fig. 2. Each frame consists of some control information,
a number of DL and UL bursts, contention intervals and
guard bands. The DL subframe header synchronizes the SSs
with the BS and carries the burst profiles, e.g. the code rate
or the length of each burst. The first DL burst after frame
header may carry a DL-MAP and an UL-MAP that deter-
mine the bandwidth allocation maps of SSs. The DL-MAP
and UL-MAP specify the receiver and transmitter SSs re-
spectively in the DL and UL sub-frames. The rest of the DL
sub-frame is filled with DL bursts containing one or more
MPDUs for individual SSs. The UL sub-frame starts with
contention time slots for initial ranging and bandwidth re-
quest. The initial ranging interval has been provided for new
SSs joining the network and preparing to associate with the
BS. The bandwidth request time slots have been dedicated
to the SSs’ contention for bandwidth allocation request. UL
bursts follow the contention slots in an order defined by the
UL-MAP.
A connection between a BS and an SS is set up as fol-
 Frame n-1 Frame n Frame n+1
DL subframe
DL subframe
Frame Headers DL burst #1 DL burst #2
UL subframe
Contention slots for Contention slots for UL PHY PDU
initial ranging bandwidth request from SS #1
Figure 2. A TDD frame of the 802.16
lows. First, a new SS should find an opportunity in the
ranging period to associate with the BS. Then the SS and
BS negotiate connection parameters. Finally, the connec-
tion is registered by the BS. Each SS is granted a portion
of bandwidth by the BS periodically or upon bandwidth re-
quest. Accordingly, the BS specifies the UL bandwidth as-
signed to each SS in the UL-MAP of the next frame. Also,
the schedule of DL bursts in the DL channel, DL-MAP, is
broadcast by the BS. The mechanism of bandwidth assign-
ment to each SS depends on the required service flow of
the SS. Four service flows have been defined by the IEEE
802.16 standard. The service flows and their corresponding
bandwidth request mechanisms are described as follows:
• Unsolicited Grant Service (UGS): this service supports
constant bit rate traffic. Bandwidth is granted to this
service flow periodically or in case of traffic presence
by the BS.
• Real-time Polling Service (rtPS): this service is pe-
riodically provided for real-time service flows with
variable-size data packets. rtPS flows are polled by
the unicast polling mechanism of the BS to send their
bandwidth request to the BS.
• Non-real-time Polling Service (nrt-PS): this service is
for non-real-time traffic with variable-size data pack-
ets. nrtPS can gain access to the channel using mono-
cast or multicast polling mechanisms. Upon receiving
a multicast polling, the nrtPS service can take part in a
contention in the bandwidth contention range.
• Best Effort (BE) service: this service provides the
minimum required QoS for non-real-time traffic. The
channel access mechanism of this service is based on
contention.
Frame n+2 Frame n+3 Despite the fact that the IEEE 802.16 standard has exten-
sively specified the service flows and bandwidth allocation
mechanisms, the details of the scheduling and admission
UL subframe
control are undefined for IEEE 802.16 WMAN.
3 Opportunistic Fair Scheduling Scheme
DL burst #N
This section describes the proposed opportunistic fair
scheduling scheme for DL scheduling of nrtPS and BE ser-
vice flows of the IEEE 802.16. The scheduling scheme
deploys an opportunistic service discipline with a fairness
UL PHY PDU
mechanism. Opportunistic scheduling takes advantage of
From SS #N
random channel fluctuations. The variations of the chan-
nel of each SS are mutually independent from the others
in a multiuser environment [11]. In a scheduling interval,
some SSs may experience deep fading while the others may
not. Therefore, if there is no fairness constraint, an optimal
scheduling policy to maximize the total DL throughput is
to transmit to a group of SSs with the best channel quality
at the maximum achievable rates in each scheduling inter-
val. However, this policy can cause sever unfairness when
the temporal and probabilistic characteristic of channels are
different. SSs with a relatively better average channel qual-
ity enjoy higher average transmission rates than those with a
relatively poor channel quality. To mitigate this problem to
some extent, a fairness mechanism is deployed to assign the
resource, i.e. bandwidth, to each SS according to its history
of transmission and average channel quality.
The measure of the history of transmission, denoted by
R̃, is an estimated average rate of transmission to each SS
over an appropriate interval of time. In this paper, a low-
pass filtering technique from [8][9] is adopted to estimate
the average transmission rate of each SS among N other
SSs. The estimation formula is given by
1 1
R̃j (n) = (1 − )R̃j (n − 1) + ( )rj (n) (1)
Tc Tc
j = 1...N
where R̃j (n) is the estimated average rate in the current
subframe, R̃j (n − 1) is the estimated average rate up to the
current subframe, and rj (n) is the allocated transmission
rate of SSj in the current subframe. Tc is the time constant
of the low pass filter. The value of R̃ is updated after each
DL subframe transmission. SSs with the lower value of R̃
have the higher probability to transmit in the next DL sub-
frame and vice versa. This mechanism avoids starving SSs
with the low average channel quality as it may happen in a
pure opportunistic scheme.
The measure of the channel quality is the average gain
of the channel of each SS. Proportions of bandwidth are as-
signed to SSs according to their average gain of channels.
The higher is the average gain of channel, the higher is the
proportions assigned to the SS. Fair assignment of propor-
tions to SSs is challenging. Consider an assignment mech-
anism that grants even proportions to the SSs with different
average gain of channel. This assignment may satisfy ser-
vice requirement of SSs (service fairness) but wastes the
resources of the network and causes effort unfairness. In
other words, a significant amount of network effort may be
spent to satisfy the SSs with poor channel condition. This
fact reveals the need for an elaborated fairness mechanism
for proportions assignment to balance between service and
effort fairness criteria.
Recently, variations of utility fairness have been de-
ployed in wireless resource allocation schemes [11]. A util-
ity function, a concept borrowed from economy, is a notion
of network performance that represents an SS’s satisfaction
of provided services. A proper choice of a utility function
for a service depends on its required QoS. For instance, the
utility function of real-time traffic may be represented by
a delay-based function and the utility function of non-real-
time traffic may be represented by a concave and increasing
function of the rate. The fairness mechanism of the net-
work tries to maximize the sum of the utility functions of
SSs. That is, a fair rate allocation vector is the solution of a
concave maximization problem as follows.

N
maxR j=1 Uj (Rj )
(2)
F (R) = 0
where R = (R1 , . . . , RN ) is the rate allocation vector, Rj
is the allocated rate to SSj , Uj (·) is the utility function of
SSj , N is the number of SSs, and F (R) = 0 specifies a set
of constraints. A fair rate allocation vector depends on the
definition of utility functions. In this paper, a set of utility
functions suitable for wireless resource allocation schemes
defined in [12] has been adopted. Accordingly, the utility
function of an SS is defined as
⎧ real-time traffic
⎨ ln(1
 + x) x < xmin
U (x) = ln(1 + x) + b x min ≤ x ≤ xmax (3)
⎩ 4
ln(1 + x) + c x > xmax
where b and c are embedded to force continuity of the
utility function, and are given by

b = ln(1 + xmin ) − ln(1 + xmin )
c = ln(1 + xmax ) + ln(1 + xmin )
 
− ln(1 + xmin ) − 4 ln(1 + xmax ) (4)
In the utility function, x = R · α, xmin = Rmin · α,
and xmax = Rmax · α are the weighted average rate, the
weighted minimum average rate, and the weighted maxi-
mum average rate of an SS, respectively. α is a scaling
factor that determines the proximity of the fair allocation
vector to an even rate allocation vector. The minimum and
maximum average rates define a desired operating region
for the SS to make the utility function sensitive to the Qual-
ity of Service (QoS) in terms of the allocated rate. Allo-
cating a rate below Rmin is considered as a significant per-
formance degradation. Allocating a rate beyond Rmax does
not increase the actual utility of an SS. The proposed notion
of fairness is sensitive to QoS, effort, and service unfair-
ness. The parameter α can be used to adjust the sensitivity
of total network utility to effort and service unfairness. For
larger values of α the solution of (2) is close to an even allo-
cation vector, i.e. even rate distributions. On the other hand,
smaller values of α result in more even effort distribution.
Fig. 3 shows the shapes of the utility function for different
values of α. To form the optimization problem and derive
Utility Function
1
0.99
0.98
0.97
0.96
0.95
0.94
0.93
0.92
alpha=100
alpha=100.05
0.91 alpha=100.08
0.9
0 0.2 0.4 0.6 0.8 1
Allocated Rate
Figure 3. A concave utility function for non-
the allocation rates, we assume that there is a perfect cod-
ing that can achieve a transmission rate of Shannon’s upper
bound. Therefore, the normalized average transmission rate
of SSj is approximated by
Rj = log2 (1 + aj pj ) bits/sec/Hz (5)
where aj is the average power gain of the channel of SSj
and pj is the allocated power to SSj . If the total power
budget (constraint) of the BS is P, and the utility function
of SSj is given by (3), the optimization problem (2) can be
simplified as follows

⎧ N
⎪
⎪ max(p1 ,...,pN ) H(x1 , . . . , xN ) = j=1 Uj (xj )
⎨ • IEEE 802.16 implements an estimation channel gain
xj = α · Rj
(6)
⎪
⎪ R j = log2 (1 + aj pj ) bits/sec/Hz
⎩ N
j=1 pj = P
The nonlinear optimization problem (6) can be solved by
nonlinear programming techniques.
To obtain the fair proportions, we define fair share
weights, denoted by (w1 , . . . , wN ) as follows:
wj = log2 (1 + aj pj ) (7)
The fair share weights are used as proportions to assign the
fair share of bandwidth to each SS.
We consider an allocation vector (R1 , . . . , RN ) as an ap-
proximately utility fair allocation vector if
R1 RN
= ... = (8)
w1 wN
With the utility fair rate allocation vector, (R1 , . . . , RN ),
the fairness of a scheduling scheme can be dynamically
monitored. If the proportions among the actual average al-
located rates for different SSs divert from the fair propor-
tions, the scheduling scheme can adapt itself to improve
fairness. However, an efficient fairness mechanism should
also consider the deviations among the average quality of
the channels for different SSs. When the deviations are
large, a high degree of unfairness is expected; therefore,
strong fairness provision is required. The maximum devia-
tion among the average quality of channels is given by
σ = max{aj |j = 1...N } − min{aj |j = 1...N } (9)
To factor σ into the scheduling scheme, we define a normal-
ized achievable rate of SSj in time slot n as
rj (n)
rj
(n) = (10)
Yj (n)(1+σ)
where
Rj (n)
wj
Yj (n) = Rj (n)
(11)
N
j=1 wj
N burst is 0.6 msec long. Once all of the SSs put up their time
and rj (n) is the achievable rate of SSj in time slot n. Yj (n)
represents the ratio between the normalized average trans-
mission rate of SSj and the mean normalized average trans-
mission rate of all SSs. Yj (n) > 1, means that SSj has
received less than its fair share in the past Tc time slots, and
vice versa [12].
In summary, the opportunistic fair scheduling scheme
can be implemented in the following steps:
• According to the knowledge of average gain of chan-
nels the fair share weights are calculated.
routine for its adaptive modulation that can be used in
the scheduling algorithm. The estimated channel gain
values by the SSs are reported back to the BS before
every DL subframe begins.
• The associated data rate of SSj , rj , is calculated by
the adaptive modulation process in the BS.
• The scheduler sorts the SSs based on the descending
order of the ratio of rj
. Once the order and the rate
of transmission of SSs determined, the DL subframe is
filled by the bursts in an order defined.
• The BS updates R of all of the SSs after each DL sub-
frame transmission.
4 Simulation results
In this section, we present the simulation results to
demonstrate the performance of the proposed scheduling
scheme. The simulated model consists of a number of
equally spaced SSs around a BS operating in PMP mode.
An IEEE 802.16 WirelessMAN-OFDM operating at unli-
censed band (5 GHz) has been considered. The maximum
power budget of the BS is 60 Watt. The fading channel
of non-line-of-sight transmission paths are characterized by
a Rayleigh distribution. The average gain of channel span
is between 5 to 25 dB. There are equal number of chan-
nels with average gain of channels, 5, 10, 15, 20, and 25.
Once values of channels gain at each subframe are gener-
ated, with an exponential distribution, the optimum mod-
ulation and coding rate for the corresponding channels are
chosen from Table 1. Except the preamble and FCH that are
sent by BPSK, the rest of each burst is sent with the opti-
mum rate. The adopted network parameters for simulation
are listed in Table 2.
The frame length is fixed-size and equally divided be-
tween UL and DL. The available length of the DL subframe
for DL-burst transmissions is calculated by deduction of the
frame header and the guard band between the UL and DL
subframes. It is assumed that the DL and UL-MAP have
the same size in each subframe. Bursts are fixed size; each
slot into the subframe and there is still some bandwidth left,
the remaining subframe is filled with the burst of SSs in the
same order defined for that subframe.
The BS forwards both real-time and non-real-time traf-
fic. The real-time traffic is supposed to be an aggregation
of VoIP packets. All real-time traffic arrived at the BS up
to the current DL transmission are forwarded in the ongo-
ing DL subframe on the First-Come First-Served (FCFS)
 maximization are the main performance metrics. Through-
Table 1. Modulation and coding schemes of put value in this paper shows the layer 2 throughput, ex-
the IEEE 802.16 [13] cluding the MAC overheads. To evaluate the fairness of
the schedulers, we need to use a measure of inequality of
Modulation Coding Channel Data Rate resource sharing. Several measure of inequality have been
rate gain (dB) (MBPS) introduced in economics that are recently being used in fair-
BPSK 1/2 6.4 6.91 ness analysis of computer network protocols [6][15][16].
QPSK 1/2 9.4 13.82 Gini index is a simple and mature measure of imbalance
QPSK 3/4 11.2 20.74 and inequality that have been deployed in several fairness
16QAM 1/2 16.4 27.65 studies of resource allocation schemes for wireless network
16QAM 3/4 18.2 41.47 [17][18]. We adopt a modified Gini index, I, to quantify the
64QAM 2/3 22.7 55.30 schedulers fairness.
64QAM 3/4 24.4 62.21 N N
1
I= |uk − ul |, (12)
2N 2 ū
k=1 l=1

basis. The length of a DL subframe that is occupied by the

real-time traffic has a normal distribution with an average
of 2.5 msec and a variance of 0.5 msec. The performance
of the scheduling schemes will be evaluated for two cases.
First, it is assumed that the non-real-time traffic has already
been admitted to the BS, and the buffer capacity for non-
real-time traffic is infinite. Next, the network is considered
under light load of non-real-time traffic. The interarrival
time and the length of the non-real-time traffic, arrived at
the BS, have exponential distribution.
Table 2. Frame Specification Parameters [14]
Parameter Value
frame size 10msec
preamble 2 OFDM symbol
FCH 1 OFDM symbol
TTG 2 OFDM symbol
MPDU header 6 byte
MPDU CRC 4 byte
DL-MAP 9 + 4 ∗ n byte
n number of transmitted
bursts in each DL subframe
OFDM 13.891µ sec
symbol duration
To demonstrate the efficiency of opportunistic fair
scheduling scheme, we compare its performance with that
of a pure opportunistic and a round robin scheduling
scheme. A pure opportunistic scheduling scheme does not
implement a fairness enforcement mechanism. On the other
hand, a round robin scheduling scheme allocates the equal
amount of resources to each SS, irrespective of the chan-
nel status of that SS. SSs have the same utility function as
defined in (3). The time constant of the low-pass filter in
equation (1), Tc, is equal to 1000. Fairness and throughput
where uk = Rk /wk , u = (u1 , . . . , uN ), and ū =
N
( k=1 uk )/N . Rk and wk are the long term average trans-
mission rate and the fair share weight of SSk , respectively.
The Gini fairness index, I, varies between 0 and 1. I=0
means the rate allocation vector is perfectly fair. The higher
value of I, near 1, indicates that there is higher unfairness
among the proportion assignment of rates and fair share
weight values.
For the first part of the simulation, a network with a
heavy load of non-real-time traffic is considered. The
throughput performance is shown in Fig. 4 versus the num-
ber of SSs in the network. It can be seen that if the number
of SSs increases, the throughput of the round robin sched-
uler remains constant while the throughput of the oppor-
tunistic and opportunistic fair schedulers increase. Once the
number of SSs increases, the percentage of the SSs in good
channel conditions will also increase. As a limited number
of SSs can receive service in each subframe, the SSs in good
channel conditions have a higher chance to be serviced. It
can be seen that the opportunistic scheme outperforms the
opportunistic fair scheme in terms of throughput. However,
superiority of the opportunistic scheme in terms of through-
put is achieved by a more unfair rate allocation among the
SSs as shown in Fig. 5. It can be seen that opportunistic fair
scheduling even performs better than round robin schedul-
ing in terms of fairness index. This improvement is due
to the fact that the opportunistic fair scheduling assigns re-
sources based on fair shares, but the round robin scheduling
assigns the resources evenly which is not necessarily fair.
An alternative comparison between the schedulers can be
observed in Fig. 6. The figure shows the throughput of each
SS in the network with 45 SSs. Each group of five SSs have
the average gain of channels equal to 5, 10, 15, 20, and 25.
As an adaptive modulation and coding scheme is deployed,
SSs with higher gain of channels have larger throughput val-
ues. However, the pure opportunistic scheduling allocates
significantly different amount of services to the SSs in each
 Overall Throughput per user throughput
26 3
Round Robbin
24 Opportunistic
2.5
Round Robbin Opp. fair
22
Opportunistic
Opp. fair 2
20
Mbps

Mbps
18 1.5

16
1
14
0.5
12

10 0
5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
Number of stations station number

Figure 4. Network Throughput Figure 6. Subscriber Stations Throughput

Gini Fairness Index

0.8
SS’s history of receptions on the scheduler decision making.
0.7 Round Robbin Consequently the SS has more chance to compensate for
Opportunistic the previous intervals which has been in a deep fading. The
Opp. fair
0.6 network throughput decreases with Tc decrement. For the
second part of the simulation, the network under light load
0.5
of non-real-time traffic is considered. As a consequence, in
0.4 some of DL subframes, there is no traffic in BS waiting to
be transmitted to some of the SSs. Therefore, in each DL
0.3 scheduling, opportunistic and round robin schemes are ap-
plied to the channels that have some traffic to transmit to
0.2
the SSs. Under these circumstances, the opportunistic fair
0.1
scheme should be applied with some modification. In the
5 10 15 20 25 30 35 40 45 opportunistic fair scheme, the history of traffic transmission
Number of stations
to each SS, R̃, is updated in each scheduling interval accord-
ing to the equation (1). In the current simulation, we update
Figure 5. Gini Fairness Index
R̃ of each SS as long as there is some traffic in the BS to be
transmitted to that SS. Otherwise, we keep the value of R̃
unchanged until the next time that some traffic is available
group of five SSs. The difference of throughput among SSs for the transmission to the SS.
in each group of five SSs is less for the opportunistic and The simulated network consists of 25 SSs. The aver-
round robin schemes than the pure opportunistic scheme. age of the destined traffic toward each SS, from the BS, in-
This fact can also be seen in Fig. 7 that shows the first 50 creases from 1 Kbps to 100 Mbps. The variation of through-
seconds of the transmission history of each SS in a group of put and Gini fairness index of the network versus the traf-
five SSs. After a few seconds, the average transmission rate fic load are shown respectively in Fig. 9 and Fig. 10. In
of each SS reaches to a steady value. A comparison shows the lightly loaded network, the throughput increases for
that the opportunistic fair scheduling maintains a more rea- all the scheduling schemes as long as the traffic load in-
sonable level of fairness among the SSs than the pure op- creases. However, the throughput increase is not unlimited.
portunistic scheme. It can also be seen that the throughput As the traffic load is intensified, the throughput of the net-
variation of SSs in the round robin scheduling has the least work reaches to an approximately constant value for each
amount with respect to the two other scheduling schemes. scheduling scheme. Except for the very light traffic load, 1
The effect of Tc, the time constant of the low pass filter, on to 2 Kbps in this simulation, the overall network throughput
the network throughput is shown in Fig. 8. Tc indicates the with pure opportunistic scheduling scheme is higher than
amount of time that the scheduler keeps track of any SS’s that of the opportunistic fair and round robin. In the very
receptions. The longer is Tc, the more is the effect of the light traffic load, opportunistic scheduling schemes choose
 4
x 10 Opportunistic Fair Overall Throughput
6 18

Tc1=1000
4 Tc2=600
bps 17.5 Tc3=200
2 Tc4=50

17
0
0 5 10 15 20 25 30 35 40 45 50
4
x 10 Opportunistic
6 16.5

Mbps
bps

16
2

15.5
0
0 5 10 15 20 25 30 35 40 45 50
4
x 10 Round Robin
6 15

4
bps

14.5
2

14
0 15 20 25 30 35 40 45
0 5 10 15 20 25 30 35 40 45 50 Number of stations
sec.

Figure 7. Average Rate of Transmission of 5 Figure 8. The Effect of Tc on Network

SSs Throughput

the channels with high average gains, irrespective of the
available traffic for each channel. When these channels
have little amount of traffic to transmit, some part of the
bursts will remain idle which causes the throughput drop.
Fig. 10 depicts the Gini fairness index of the three schedul-
ing schemes. The fairness index of the opportunistic fair
scheduling has been drawn for two values of Tc. The allo-
cation of resources by opportunistic and round robin in the
light traffic load is more evenly than that in the high traffic
load situations. Therefore, the Gini fairness index of these
two scheduling schemes increases with the traffic load in
a lightly loaded network. Although the opportunistic fair
scheduling is fairer than the other two scheduling schemes
in the heavily loaded network, its performance in the lightly
loaded network degrades. It the later situation, the fairness
of the opportunistic fair scheme is critically depends on the
value of Tc. The lower is Tc, the fairer is the scheme. This
is reasonable that the short term fairness, Tc=100, performs
better than long term fairness in the light traffic load due to
the short duration of traffic.
5 Conclusions
An opportunistic fair scheduling scheme for the down-
link of the IEEE 802.16 PMP WirelessMAN OFDM has
been proposed. The scheme implements an opportunistic
service discipline as the core. A fairness enforcement mech-
anism has also been integrated in order to maintain long
term fairness and smooth service delivery. In other words,
the opportunistic fair scheduling takes advantage of the tem-
porary variations of a fading channel to improve the overall
throughput, at the same time, avoids starving the SSs suffer-
ing of a long term bad channel conditions. Simulation re-
sults demonstrated that the opportunistic fair scheduling can
significantly improve the fairness among SSs and maintain
a good level of throughput for the network. In our further
research, we will consider the scheduling of real-time traf-
fic. The effects of the mobility of SSs on the scheduling
process will be involved.
References
[1] IEEE, “IEEE std 802.16-2001.” IEEE Standard for
Local and Metropolitan Area Networks, Part 16: Air
Interface for Fixed Broadband Wireless Access Sys-
tems, 2002.
[2] IEEE, “IEEE std 802.16a-2003.” IEEE Standard for
Local and Metropolitan Area Networks, Part 16:
Air Interface for Fixed Broadband Wireless Access
Systems-Amendment 2: Medium Access Control
Modifications and Additional Physical Layer Specifi-
cations for 2-11 GHz, 2003.
[3] IEEE, “IEEE std 802.16-2004.” IEEE Standard for
Local and Metropolitan Area Networks, Part 16: Air
Interface for Fixed Broadband Wireless Access Sys-
tems, 2004.
[4] F. Kelly, “Charging and rate control for elastic traf-
fic,” European Transactions on Telecommunications,
vol. 8, pp. 33–37, 1997.
 Overall Throughput Proc. IEEE VTC’2000, 2000.
25

[9] H. Bang, T. Ekman, and D. Gesbert, “A channel

20 predictive proportional fair scheduling algorithm,” in
IEEE 6th Workshop on Signal Processing Advances in
Wireless Communications, 5-8 June 2005.
15
Mbps

[10] S. Coleri, M. Ergen, A. Puri, and A. Bahai, “Channel

10 estimation techniques based on pilot arrangement in
OFDM systems,” IEEE Transaction on Broadcasting,
Round Robbin
Opportunistic vol. 48, no. 3, pp. 223–229, 2002.
5
Opp. fair, Tc=1000
Opp. fair, Tc=100 [11] G. Song and Y. Li, “Utility-based resource allocation
0 0 and scheduling in OFDM-based wireless broadband
1 2 3 4 5
10 10 10 10 10 10 networks,” IEEE Communications Magazine, vol. 43,
Load (Kbps)
no. 12, pp. 127–134, 2005.
Figure 9. Network Throughput [12] M. Dianati, X. Shen, and K. Naik, “Coperative fair
scheduling for the downlink of CDMA cellular net-
works,” IEEE Transactions on VT, to appear.
Gini Fairness Index
0.8
[13] A. Ghosh and D. R. Wolter, “Broadband wireless
0.7 access with WiMAX/802.16: Current performance
benchmarks and future potential,” IEEE Communica-
0.6 tions Magazine, vol. 43, pp. 129–136, 2005.
Round Robbin
Opportunistic
0.5
Opp. fair, Tc=1000
[14] C. Hoymann, “Analysis and performance evaluation
Opp. fair, Tc=100 of the OFDM-based metropolitan area network IEEE
0.4
802.16,” Computer Networks, vol. 49, no. 3, pp. 341–
0.3 363, 2005.

0.2
[15] A. Kumar and J. Kleinberg, “Fairness measures for re-
source allocation,” IEEE 41st Annual Symposium on
0.1 0 1 2 3 4 5
Foundation of Computer Science, November 2000.
10 10 10 10 10 10
Load (Kbps)
[16] R. Jain, D. Chiu, and W. Hawe, “A quantitative mea-
sure of fairness and discrimination for resource alloca-
Figure 10. Gini Fairness Index
tion in shared computer system,” DEC Technical Re-
port 301, 1984.

[5] P. Viswanath, D. Tse, and R. Laroia, “Opportunistic [17] M. Garetto, T. Salonidis, and E. Knightly, “Modeling
beamforming using dumb antennas,” IEEE Transac- per-flow throughput and capturing starvation in csma
tions on Information Theory, vol. 48, pp. 1277–1294, multi-hop wireless networks,” Proceedings of IEEE
June 2002. INFOCOM 2006, Barcelona, Spain, April 2006.

[18] H. Shi and H. Sethu, “An evaluation of timestamp-

[6] M. Dianati, X. Shen, and K. Naik, “A new fair-
based packet schedulers using a novel measure of in-
ness index for radio resource allocation in wireless
stantaneous fairness,” Proceedings of the IEEE Inter-
networks,” in Proc. of the IEEE WCNC05, vol. 2,
national Performance, Computing, and Communica-
pp. 712–717, 2005.
tions Conference, Phoenix, Arizona, USA, April 911,
[7] A. Maulloo, F. Kelly, and D. Tan, “Rate control in 2003.
communication networks: Shadow prices, propor-
tional fairness and stability,” Journal of Operation So-
ciety, vol. 2, no. 49, pp. 237–252, 1998.

[8] A. Jalali, R. Padovani, and R. Pankaj, “Data through-

put of CDMA-HDR a high efficiency data rate,” in

View publication stats