AbstractRadio Resource Management (RRM) techniques used

in the Worldwide Interoperability for Microwave Access

(WiMAX) can provide for many of the services and features
promised by 4G wireless networks, such as supporting
multimedia services with high data rates, and wide coverage
area, as well as all-IP with security and Quality of Service (QoS)
support. The IEEE 802.16 standard, associated with the WiMAX,
leaves the details of RRM components open for the vendors to
explore. In this paper we propose a novel scheme for the QoS
support in WiMAX including packet scheduling and bandwidth
allocation strategies. The proposed solution efficiently enhances
performance and utilizes resources, while being fair, practical
and in compliance with the IEEE 802.16 standard specifications.
Our solution provides QoS support to all traffic classes defined
by the standard, and it dynamically changes the bandwidth
allocation based on the traffic characteristics and service
demands. Simulation results show that the proposed solution can
deliver QoS support and be fair to all classes of service in a
WiMAX network.
Keywords- Packet Scheduling; QoS support; Inter-Class and
Intra-Class QoS support
I. INTRODUCTION
WiMAX is designed to provide high speed wireless access in
metropolitan area networks. RRM techniques and QoS support
are among the most important features of this technology.
WiMAX is associated with the IEEE 802.16 standard, which
defines five classes of traffic flows representing different
types of services in the following order: Unsolicited Grant
Service (UGS), Extended Real Time Polling Service (ertPS),
Real Time Polling Service (rtPS), Non-Real Time Polling
Service (nrtPS), and Best Effort Service (BE). The standard
defines a connection-oriented MAC protocol with a
mechanism for the QoS support. However, RRM techniques
such as Packet Scheduling (PS) and Bandwidth Allocation
(BA) schemes are left open for development by the vendors.
In this study, we propose a new scheme for the QoS support
that includes PS and BA called 2Tier-QoS Framework (2T-
QoSFW). The proposed solution supports all types of service
flows, and makes bandwidth allocation technique that is
dynamic and fair, and utilizes the resources in an efficient
way. The motivation for this study is to develop RRM
schemes that consider fairness and utilization, and to establish
parameters to quantify their values.
The proposed solution separates the traffic and provides
QoS support in two levels: Inter-Class and Intra-Class. The
components of the solution (PS and BA) are implemented in
two main building blocks: Tier-1 and Tier-2, and they work
based on two fundamental ideas: (i) Dynamic bandwidth
allocation based on traffic arrival rate, (ii) Intra-class versus
inter-class QoS support.
The QoS support is incorporated into the solution using
fairness and utilization. In WiMAX applications, fairness is
related to bandwidth allocation among different service
classes. There are several definitions and methods for
evaluating fairness such as Jains fairness, Gini coefficient,
Max-Min fairness [6]. Utilization in WiMAX networks
depends on application demands, user density, and available
network resources. In this paper we define fairness and
utilization as the ratios of allocated bandwidth (bw
alloc
) by the
requested bandwidth (bw
req
), and throughput achieved (Th) by
bw
alloc
respectively.
The rest of this paper is organized in the following order.
Section II reviews some of the works done by the research
community in PS and BA for WiMAX. Section III describes
our proposed solution including system architecture and
modeling features. Section IV includes preliminary simulation
results. Finally, we make a conclusion and list future plans for
this study in Section V.
II. LITERATURE REVIEW
RRM for IEEE 802.16 and WiMAX has been studied by
many groups in the wireless research community. Several
schemes are proposed for PS and BA. RRM schemes are
categorized based on centralized/decentralized, Real-Time
(RT) versus Non-Real-Time (NRT), Uplink (UL) and
Downlink (DL) or basic versus complex schemes.
In the basic schemes, often a simple traditional queuing
discipline is considered for PS in all classes [1]. Such systems
use one or two simple queue disciplines to separate the traffic.
Giving different weights to various classes to highlight their
priorities is also used as a common strategy. Proposals based
on complex schemes for PS and BA combine several queuing
disciplines with some parameters related to traffic and
medium characteristics. They often develop schemes in more
than one stage. Complex Hierarchical schemes are proposed in
[2, 3]. In hierarchical schemes, the PS and BA are designed in
multiple levels. Often the traffic from different service classes
is separated in the first level, and then scheduled within each
class in the second level. Different levels of RRM could all be
performed in the BS or the tasks could be divided between BS
and SSs. In [5], the authors propose an Ad-hoc scheme for
scheduling, where a separate queuing discipline is used for
each service flow.
From a different perspective, the RRM could be classified
based on QoS support strategies. Most of the proposed
Amir Esmailpour
Department of Electrical and Computer Engineering
University of Toronto
Toronto, Canada
Amir.Esmailpour@utoronto.ca
Nidal Nasser
Department of Computing and Information Science
University of Guelph
Guelph, Canada N1G 2W1
nnasser@uoguelph.ca
A Novel Scheme for Packet Scheduling and
Bandwidth Allocation in WiMAX Networks
978-1-61284-231-8/11/$26.00 2011 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

solutions that we have seen in this area are divided into two
major groups: fairness or priority based models. An example
of fairness-based proposal is Modified Deficit Round Robin
(MDRR) discipline, and for priority-based solution, we can
refer to Modified Priority Queue (MPQ) [4, 5]. We compare
the results from these models with those of our proposed
solution.
J. Liang et al proposed a novel scheduling scheme that
covers both RT and NRT applications that performs resource
allocation based on a polling interval adjustment [7]. They
show by simulation that their scheme could improve the
performance of nrtPS, while keeping that of rtPS in a steady
state.
S. Pizzi used a DRR-based schedulers, for the MAC layer
of a Point-to-Multipoint WiMAX network and compared two
methods of compensation-based versus and greedy algorithms.
Their simulation results show that the compensation-based
approach is more capable of service differentiation when it
comes to heterogeneous channel conditions[8].
III. PROPOSED SCHEME SYSTEM DESIGN
In this paper, we propose a solution for the two components of
RRM in WiMAX that provides QoS support for different
types of applications. The proposed scheme makes the
bandwidth allocation for the TDD mode of operation of the
IEEE 802.16 standard, and supports all types of service flows
as indicated by the standard. It is fair, efficient, and it follows
the IEEE 802.16 variable transmission rate and signaling
mechanism rules and regulations through dynamic
adjustments of the bandwidth allocation.
We have implemented the solution in two building blocks
to perform PS and BA for WiMAX classes of service: Tier-
and Tier-2. Each block has several modules performing the
tasks for scheduling and bandwidth allocation. Detail of each
block and its modules are explained in the following sections.








A. Packet Scheduling
In Tier-1 four different queuing disciplines are employed
for five classes of service to perform Intra-Class scheduling as
illustrated in Figure 1. For all connections holding UGS
traffic, the scheduler uses Earliest Deadline First (EDF),
where packets with the earliest deadline will be scheduled
first. The scheduler determines the packets deadline based on
its arrival time and maximum latency. EDF uses a dynamic
priority based scheduling algorithm that provides delay
guaranteed service to the UGS service classes.
For ertPS and rtPS connections, Weighted Fair Queuing
(WFQ) is employed, which is similar to the generic fair
queuing algorithm. The advantage of WFQ is that it is based
on the resource reservation, i.e. instead of giving an equal
share of resources to all users; it allocates a specific
amount of resource to each service. However, it does not
impose a strict time limit on each packet. ertPS and rtPS
packets are scheduled based on their weights, calculated
using the ratio between a connection reserved traffic rate
and the total sum of all reserved traffic rates of all
connections.
For nrtPS connections the scheduler uses RR queuing
discipline, where each service gets a fair share of the allocated
bandwidth in a RR fashion. RR assigns time slots to each
service flow in equal portions and in order, handling all
requests without priority. RR is both simple and easy to
implement, and it does not cause any starvation problem for
the lower priority services. BE connections will be treated in a
FIFO fashion. FIFO is the most basic discipline, yet suitable
for BE traffic with no particular QoS requests.
The scheduler goes through all queues in the order of their
priorities from highest priority access to the lowest, and
checks for any activities. If there is a packet waiting in a
queue, the queue will be added to an Active Queue List
(AQL). If a queue belongs to AQL, then the scheduler carries

































ertPS e e e e
rtPS r r r r
nrtPS n n n n
BE b b b b
UGS u u u u
Tier 1: Intra-class Traffic Separation Classifier Tier-2: Inter-class Traffic Separation
.
.
EDF
.
.
WFQ1
.
.
WFQ2
.
.
RR
.
.
FIFO BE-2
BE-1
BE-n
.
nrtPS-1
nrtPS-2
nrtPS-1
.
rtPS-1
rtPS-2
rtPS-n
.
ertPS-1
ertPS-2
ertPS-n
.
UGS-1
UGS-n
UGS-2
.
un ... u2 u1
en ... e2 e1
rn ... r2 r1
nn ... n2 n1
bn ... b2 b1
DAPQ

DAWFQ
Figure 1: 2-Tier Scheduling Scheme System Design
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

on to the final stage of scheduling, and forwards packets to
Tier-2 or the DAQ stage to perform Inter-Class scheduling.
DAQ will dynamically assign a portion of the data-burst for
each type of traffic. If a queue has no packets, it will be
removed from the AQL, and no bandwidth is allocated to it.
The process goes through several iterations. In each
iteration, it updates the AQL, the available frame portion, and
the number of packets in each queue. Then it goes through
several steps for each service round, processing higher
priorities first, and then the lower priorities. In each service
round, the highest priority queue will be served first until the
available bandwidth is lower than the requested bandwidth. If
the available bandwidth is lower than the bandwidth requested
by the higher priority queue, then the lower priority queues
will be served in order, which is a fair solution for the lower
priority queues.
DAQ has two types of implantation: Dynamically
Allocating Priority Queue (DAPQ) and Dynamically
Allocating Weighted Fair Queue (DAWFQ). In case of DAPQ
traffic flows are separated based on their priorities, whereas in
DAWFQ, they are set apart on a fairness basis. However, in
both cases, QoS is differentiated using relative dynamic Cap
(dCap
rel
, defined in the next section) values either as priority
factor for the DAPQ or as weights of DAWFQ respectively.
DAPQ is a modified version of the Priority Queue (PQ).
The problem with PQ is that the highest priority will get
allocated bandwidth, and if the flow continues for a long time,
the lower priorities such as BE could go through bandwidth
starvation. We propose DAPQ as a solution for this problem
and a fairness strategy based on a reservation technique.
DAWFQ is based on modified version of WFQ to allocate
resources to different classes of service. DAWFQ schedules
the packets based on the weight of each service class, which is
calculated using the corresponding fairness values for each
service class. In this case we use inter-class fairness. We
dynamically calculate the weights based on fairness to avoid
oversubscription of higher priority classes. In each round of
scheduling, class-level fairness is considered, as well as the
total fairness.
B. Bandwidth Allocation
Dynamic Bandwidth Allocation (DBA) adjusts resource
allocation based on traffic and network characteristics. Traffic
behavior is characterized by traffic arrival rate, whereas the
network conditions are incorporated using two performance
metrics: Fairness (Fr) and Utilization (Ut).
Traffic arrival rate is used to calculate a new parameter
called dynamic Cap (dCap). dCap is an important parameter
used for PS, CAC, and DBA by integrating traffic and
network characteristics, and allows the system to improve
fairness and utilization.
dCap is a delicately calculated parameter that takes into
account the traffic behaviors such as traffic arrival rate, as
well as network conditions such as fairness and utilization.
dCap value is calculated in Equation 1.
Where Ieff is the normalized effective traffic rate for the
connection j of service class n, and Fr and Ut are the
corresponding fairness and utilization factors. dCap is used for
both DAPQ and DAWFQ implementation of Tier-2
component of the proposed solution, and it allows us to
dynamically control the amount of resources allocated based
on the most recent network and traffic condition.
dCap =
`
1
1
1
1
I
e
n,j
, | (Fr
n,j
Fr
re|
n,j
) / (Ut
n,j
Ut
re|
n,j
)
I
e
n,j
+ Fr
re|
n,j
, | (Fr
n,j
< Fr
re|
n,j
)
I
e
n,j
+ Ut
re|
n,j
, | (Ut
n,j
< Ut
re|
n,j
)
I
e
n,j
+ Fr
re|
n,j
+Ut
re|
n,j
, | (Fr
n,j
< Fr
re|
n,j
) / (Ut
n,j
< Ut
re|
n,j
)

In this paper, we define fairness as the ratio of allocated
bandwidth divided by requested bandwidth for each service
flow, and by summing up this parameter over a period of time
for all flows within a service class we get total fairness of the
system towards a service class n, according to Equation 2.
Fr
1
n,]
= _
bw
uIIoc,
n,]-1
bw
cq,
n,]-1
_
I
=1


(2)
Where: i is the connection number arriving in the j
th
round
of scheduling for service class n.
Utilization, on the other hand, is the ratio of throughput
achieved divided by the allocated bandwidth, and it is
calculated for all service flows in class n, based on Equation 3.
ut
1
n,]
= _
Ib

n,]-1
bw
uIIoc,
n,]-1
_
I
=1


(3)
Upon calculation of dCap and corresponding Fr and Ut
values in Tier-1, then the bandwidth allocation for the next
round of scheduling is performed in Tier-2 block according to
Equations 4-6.
bw
uuI
1
= bw
cs,
n,]
I
=1
neN
+ bw
cm
1


(4)

bw
uIIoc
1
= _
JCop
n
JCop
1
] bw
uuI
1



(5)

bw
cs,
n,]
= bw
uIIoc,
n,]-1
- bw
cq,
n,]



(6)
IV. SIMULATION AND RESULTS
We simulated a simple WiMAX network in OPNET modeler
version 14.5-PL3. The network consists of one cellular
structure containing one BS and five SSs as illustrated in
Figure 2.
The results show that the solution satisfies the QoS
requirement of RT classes, while the NRT classes get their
share of bandwidth. The delay results in Figure 4 show value
for UGS is less than 10 ms which falls into acceptable delay
tolerance for UGS class. For other RT applications the delays
have reached levels above 30 and 50 ms for ertPS and rtPS
flows respectively for the duration of the simulations.
(1)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings


The source of traffic is an application server that could
provide five types of application, one for each type of traffic.
The assumption is that each SS carries the traffic from only
one user, and each user is using only one type of application
at-a-time. Applications, the type of traffic they represent, and
Subscribers that request them are listed in Table 1.

Table 1: Applications and type of traffic
Application voice video SSH FTP browsing
Traffic type UGS ertPS rtPS nrtPS BE
Requester SS_1 SS_2 SS_3 SS_4 SS_5

The simulation starts by downloading files from the
application server, one by each SS with the new code in place.
In the first case scenario, we look at the traffic separation and
resource allocation for all five types of traffic using the
proposed solution. Results of throughput achieved by each
type of service are presented in Figure 3. Results in this figure
show that the solution has separated the traffic from all
sources and delivered QoS requirements of all service
requests. Shortly after the beginning of the simulation, UGS
throughput reach its fixed bandwidth allocated at around
slightly below 5 Mbps, and subsequently other RT classes
reach their respective bandwidth allocations, with ertPS at
around 2.5 Mbps, and rtPS at 1.1 Mbps. As for NRT traffic,
nrtPS has achieved almost 64 Kbps and BE over 30 Kbps.
These results indicate that has been fair to NRT service flows
while maintaining the QoS requirements of RT applications in
terms of bandwidth requirements.
These results indicate that although the low priority flows
are not completely starved, however, the higher priority RT
applications are consuming a considerable portion of the
bandwidth. In this scenario, a link with high data rate is
employed, resembling a typical WiMAX over-provisioning
case, in which the total bandwidth is not completely utilized.


Figure 3: Throughput achieved by the
proposed scheduler for all applications
In the second scenario, we repeat the same experiments
with Modified Priority Queue (MPQ). The results are
presented in Figures 5 and 6.

Figure 4: Delay achieved by the
proposed scheduler for all applications
The results for RT application are slightly above the
expected values for delays generated by the NRT application
demands. Delay results for NRT applications are also above
200 and 500 ms for nrtPS and BE traffic, however for NRT
applications delay is not as important as bandwidth allocation,
which is quite satisfactory. These results indicate that although
the low priority flows are not completely starved, however, the
higher priority RT applications are consuming a considerable
portion of the bandwidth. In this scenario, a link with high
data rate is employed, resembling a typical WiMAX over-
provisioning case, in which the total bandwidth is not
completely utilized.

Figure 5 shows that MPQ provides throughput for UGS and
RT applications at levels close to bandwidth requirements and
at a faster paste. UGS reaches the 5 Mbps quickly after the
start of simulation period. Other RT applications follow the
same trend at 3 and 1.5 Mbps for ertPS and rtPS.

To investigate further, we look at the throughput results for
the NRT traffic flows using MPQ in Figure 8 and compare
them with those of the solution. Figure 8 shows throughput
values for NRT service flows in comparison with the lowest
priority RT application (i.e. rtPS). As it is observed in this
figure, RT application throughput is in Mbps ranges, close to
1.5 Mbps, while the NRT applications are in the lower ranges
of Kbps.

Figure 2: OPNET simulation of a small WiMAX
network including one cell, one BS and 5 SSs
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

Figure 5: Throughput achieved by MPQ
When comparing these results with those of the solution,
it is observed that in terms of RT applications both models
show high values in Mbps ranges, however, in terms of NRT
applications, the solution shows significant improvements in
Figure 7. Values for nrtPS and BE using the solution in
Figure 7 are at 63 Kbps and 30 Kbps, comparing to those of
MPQ in lower ranges of 17 and 10 Kbps respectively in Figure
6.

Figure 6: Throughput for NRT classes using MPQ
According to the results from Figures 3-7, RT traffic flows
reach their respective bandwidth allocations in MPQ faster
than the proposed the solution scheme. The bandwidth
allocations and throughputs are in a comparable range in both
cases. However, in terms of NRT applications, the proposed
solution proves superior in comparison with MPQ, while
keeping the RT QoS requirements in a satisfactory level.
IV. CONCLUSION AND FUTURE WORK
We proposed a solution for the RRM techniques in WiMAX,
which supports all types of service flows, and makes
bandwidth allocation that is dynamic, fair, and efficiently
utilized. We show by simulation results that the solution could
deliver QoS support while being fair to all classes of service
defined by the standard. We also introduced two new metrics
to evaluate the QoS support in WiMAX: fairness and
utilization. We used the new metrics to perform PS and BA
using traffic behavior and network conditions.

Figure 7: Throughput for NRT classes using proposed solution

In future, we plan to further develop this scheme by
incorporating the new metrics into the core of both intra-class
and inter-class QoS support. We also like to further investigate
the performance of various parts of the scheme using QoS
support, and perform a comparative analysis with other
proposed solutions.
REFERENCES
[1] K. Wongthavarawat and A. Ganz, Packet Scheduling for QoS Support
in IEEE 802.16 Broadband Wireless Access Systems, J. Comm.
Systems, 2003, vol. 16, pp. 81-96.
[2] J. Kim and K. Kim, Apparatus and Method for download Packet
Scheduling in Base Station of a Portable Internet System, United State
Patent Application Publication Number: US 2007/0121636 A1, May 31
2007.
[3] M. Mehrjoo, M. Dianati, X.Shen, K. Naik, Opportunistic Fair
Scheduling for the Downlink of IEEE 802.16 Wireless Metropolitan
Area Networks. Proceedings of the 3rd int. conference on QoS in
heterogeneous networks, New York, 2006, Vol. 191, Article No. 52.
[4] D. Tarchi, R. Fantacci, and M. Bardazzi, Quality of Service
Management in IEEE 802.16 Wireless Metropolitan Area Networks,
IEEE International Conference on Communications (ICC), Istanbul,
Turkey, 2006, Volume: 4, pp.1789-1794.
[5] K. Vinay1, N. Sreenivasulul, D. Jayaraml, and D. Das, Performance
Evaluation of End-to-end Delay by Hybrid Scheduling Algorithm for
QoS in IEEE 802.16 Network, International Conference on Wireless
and Optical Communications Networks (IFIP), Bangalore, India, 2006.
pp. 5-10.
[6] Mehrdad Dianati, Xeumin (Sherman) Shen, and Sagar Naik. A New
fairness Index for Radio Resource Allocation in Wireless Networks.
IEEE Communications Society / WCNC 2005, IEEE Xplore. pp. 712-
717.
[7] Jin Liang; Xuekang Sun; Ning Kang; , "Performance evaluation of an
integrated and efficient uplink scheduler for WiMAX network,"
Communications Technology and Applications, 2009. ICCTA '09. IEEE
International Conference on , vol., no., pp.326-330, 16-18 Oct. 2009
[8] Pizzi, S.; Molinaro, A.; Iera, A.; , " On the perfromance of
compensation based and greedy scheduling for the IEEE 802.16
standard. Proced. of the IEEE Int. Conference on Communication 2009
pp.978-984, 14-18 June 2009, Dresden, Germany.


.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings