3.rapport Activite Pub

Papiers publis dans des
JOURNAUX ISI
( Institute for Scientific Information)

1. Ridha Ouni, Jamila Bhar and Kholdoun Torki, A new scheduling protocol design based on deficit
weighted round robin for QoS support in IP networks, Journal of Circuits, Systems, and Computers,
Vol. 22, No. 3 (2013) (21 pages).
2. Ridha Ouni, Dynamic slot assignment protocol for QoS support on TDMA-based mobile networks,
Computer Standard and Interfaces, Vol.34, No.1, pp.146-155, 2012.
3. M. Z. Hourani, Ridha Ouni, Efficient data harvesting for inelastic traffic in vehicular sensor networks,
Science international, 24(1), pp13-19, 2012.
4. Monji Zaidi, Ridha Ouni, Rached Tourki, Wireless propagation channel modeling for optimized
handoff algorithms in wireless LANs, Computer and Electrical Engineering, Vol 37, (2011), pp 941-
957.
Sous reserve de correction
5. Ridha Ouni, Rafik Louati, Enhanced AODV routing protocol for energy-efficiency in wireless sensor
networks, under revision for the Journal of Circuits, Systems, and Computers (JCSC), 2014.

A NEW SCHEDULING PROTOCOL DESIGN BASED ON
DEFICITWEIGHTEDROUNDROBINFORQoS SUPPORTIN
IP NETWORKS
RIDHA OUNI
,
, JAMILA BHAR
and KHOLDOUN TORKI
College of Computer and Information Sciences,

King Saud University, Riyadh 11543, Kingdom of Saudi Arabia
Faculty of Sciences of Monastir, Tunisia
CMP, INPG, Grenoble, France
rouni@ksu.edu.sa
Received 23 March 2012
Accepted 4 October 2012
Published 21 February 2013
We present a study of the eects of active queue management (AQM) on the average queue size
in routers. In this work, three prominent AQM schemes are considered: packet classication,
checking service level agreements (SLA) and queue scheduling. This paper presents several
adaptive resource sharing models that use a revenue criterion to allocate bandwidth in an
optimal way. The models ensure QoS requirements of data ows and, at the same time, max-
imize the total revenue by adjusting parameters of the underlying schedulers. Decit round
robin (DRR) and decit weighted round robin (DWRR) scheduling techniques have shown their
ability in providing fair and weighted sharing of network resources for network devices. How-
ever, they are unable to use the total allocated network bandwidth even in burst trac. In this
paper, we propose a negative-decit weighted round robin (N-DWRR) technique as a new
packet scheduling discipline to improve the bandwidth utilization rate without increasing the
total latency. A fully hardware packet scheduler has been implemented and veried as part of an
intellectual property core. This is motivated by the fact that the design and analysis of hard-
ware/software architectures for such techniques requires new models and methods, which do not
fall under the domain of traditional embedded-systems design.
Keywords: Scheduling; DWRR; QoS; distributed queue; dierentiated service; AQM; ASIC.
1. Introduction
Routers, even with their basic \store-and-forward" functionality can be considered
as \packet processors", and this is still their default behavior in IP-based networks.
However, with networks extensively growing in size, and the Internet shifting from
a research network into the one being used for commerce, communication,
*This paper was recommended by Regional Editor Eby G. Friedman.
Journal of Circuits, Systems, and Computers
Vol. 22, No. 3 (2013) 1350012 (21 pages)
#.
c World Scientic Publishing Company
DOI: 10.1142/S0218126613500126
1350012-1
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
entertainment and information dissemination, routers became more and more
complex and incorporated new packet-processing functionality. The basic packet-
processing tasks at a router include:
. header parsing,
. packet classication to assign the packet a quality-of-service (QoS)-class,
. determination of the outgoing network interface (i.e., forwarding),
. checking service level agreements (SLA) (i.e., policing),
. queuing and link scheduling,
. means for implementing QoS guarantees to dierent packet ows.
Delivering QoS means guaranteeing given parameters within certain bounds for
connections made over a network.
1
QoS can be applied dierently to connections or
users, as well as to dierent types of trac and data ows. The parameters involved
in QoS can be classied as bandwidth, delay, jitter and packet loss. The routers must
use trac scheduling algorithms to serve packets carrying high-priority trac in
the network. Such trac scheduling algorithms should have low implementation
complexity and simple connection admission control to be able to operate at a high
speed. The latter increases complexity and limits the scalability of switching
systems. Thus, providing end-to-end QoS guarantees for high-priority trac in
a scalable and low-complexity fashion is an important issue in high-speed
communication networks.
The system presented in this paper provides guaranteed levels of QoS using
packet scheduling. The term \scheduling" encompasses a number of policies on
which decisions are made when processing packets arrive and depart from a router.
2
A number of dierent scheduling techniques exist for QoS and trac management.
Their main objective is to treat dierent trac classes or ows of packets with a
variable degree of priority in order to provide performance guarantees for a range of
dierent trac types and proles.
2
Link scheduling in packet networks is an im-
portant mechanism to achieve QoS as it directly controls packet delays.
3
Existing
QoS architectures like integrated services (IntServ)
4
and dierentiated services
(DiServ)
5
rely on link scheduling to provide the dierentiated bandwidth fairness
and delay among queues on each router. There are several tasks that any queue-
scheduling discipline should accomplish:
. Support the fair distribution of bandwidth to each of the dierent service classes
competing for bandwidth on the output port. If certain service classes are required
to receive a larger share of bandwidth than other service classes, fairness can be
supported by assigning weights to each of the dierent service classes.
. Furnish protection (rewalls) between the dierent service classes on an output
port, so that a poorly behaved service class in one queue cannot impact the
R. Ouni, J. Bhar & K. Torki
1350012-2
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
bandwidth and delay delivered to other service classes assigned to other queues on
the same output port.
. Allow other service classes to access bandwidth that is assigned to a given service
class if the given service class is not using all of its allocated bandwidth.
. Provide an algorithm that can be implemented in hardware, so that it can arbi-
trate access to bandwidth on the highest-speed router interfaces without nega-
tively impacting system forwarding performance. If the queue-scheduling
discipline cannot be implemented in hardware, then it can be used only on the
lowest-speed router interfaces, where the reduced trac volume does not place
undue stress on the software implementation.
Many schedulers have been proposed to address these issues. These algorithms in-
clude weighted fair queuing (WFQ),
2
weighted decit earliest departure rst
scheduling (WDEDF),
3
frame-counter scheduling,
6
resource allocation in an
IntServ/DiServ integrated EPON system
7
and user-oriented hierarchical band-
width scheduling.
8
End-to-end congestion control is widely used in the current internet to prevent
congestion collapse. However, because data trac is inherently bursty, routers are
provisioned with fairly large buers to absorb this burstiness and maintain high-link
utilization. The random early detection (RED) technique keeps the average queue
size low while allowing occasional bursts of packets in the queue.
9
It is designed to
accompany a transport-layer protocol such as TCP that avoids the global syn-
chronization of many connections while decreasing their window at the same time. In
this work, weighted RED (WRED) has been adopted and implemented because it
drops packets selectively based on IP precedence. Packets with a higher IP prece-
dence are less likely to be dropped than packets with a lower precedence. Thus,
higher priority trac is delivered with a higher probability than lower priority
trac.
10
This paper focuses on three main issues pertaining to the classication, active
queue management (AQM) and scheduling for such packet processors and toward
this proposes appropriate models and algorithms. It introduces the negative
weighted decit round robin (N-DWRR) scheduler, which aims to maintain the
weighted share of bandwidth among queues while reducing the queuing delay of
packets. The eectiveness and eciency of this technique, based on a performance
evaluation process, allows therefore addressing its hardware implementation. The
remainder of the paper is structured as follows. Section 2 analyzes the perceived QoS
as well as the classication disciplines in IP networks. It describes also related works
of recent scheduling algorithms. Section 3 discusses the basic operation details and
algorithm of the N-DWRR scheduler. Sections 4 and 5 present the performance
evaluation of the N-DWRR scheduler and its hardware implementation, respec-
tively. Finally, Sec. 6 concludes the paper.
A New Scheduling Protocol Design Based on DWRR
1350012-3
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
2. Background and Related Work
In this section, we describe four categories of service that eectively improve QoS in
networks. We present features of dierentiated service approaches and tasks ac-
complished by scheduling disciplines. We are interested in highlighting the impact of
these techniques on the QoS in large IP networks.
2.1. Impact of statistical multiplexing on perceived QoS
QoS is the ability of a network to dierentiate between dierent types of trac and
prioritize accordingly. Voice and video are very delay-dependent and have very
predictable patterns, whereas data is very bursty and is less delay sensitive. If all
three types of trac occur on a network, the data trac usually interferes with voice
and video and causes it to be unintelligible.
QoS can eectively improve the usage of existing bandwidth.
11,12
It deals with
four dierent categories of services such as bandwidth, latency, jitter and loss. The
rst service category, bandwidth, concerns itself with how the network manages the
entire stream of data packets owing through it, particularly in times of network
congestion. The second service category is latency, the end-to-end delay of a ow.
Numerous applications, including voice and video, have a specic end-to-end delay
budget. If a packet is delayed beyond the allocated budget, the data becomes stale or
is no longer relevant. The third category addresses the need to control jitter, the
variations in latency between packets. The nal category deals with the need to
manage packet loss. As a consequence of congestion, packet loss has two purposes.
First, reducing the number of packets competing for an output link can relieve the
level of congestion. Second, when sending hosts notice that some packets are being
discarded, they usually reduce the volume of trac they are injecting into the
network.
13
2.2. Dierentiated service approaches in large IP networks
In the internet, three service models are studied and developed: Best-eort (BE)
service, Integrated service (IntServ) and Dierentiated Service (DiServ).
4,5
In BE
service model,
14
the application program can be sent any amount of packets at any
time, and is not required to be approved in advance or notify the network. The
network provides no guarantee on packet transmission performance of reliability or
delay. The Intserv is a per-ow oriented QoS architecture that uses the resource
reservation protocol (RSVP) for dynamical resource allocation. It provides the
guaranteed service and controlled-load service. The guaranteed service allows
guaranteeing bandwidth and delay application requirements. The controlled-load
service guarantees, when the network congestion occurs, the low delay and high pass
rate for packets. However, the DiServ classies packets into a small number of
aggregated ows or \classes" that provide dierent levels of service for dierent
1350012-4
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
classes.
7
The DiServ model includes two services: expedited forwarding (EF) and
Assured Forwarding (AF) services. The network achieves packet classication,
trac shaping, trac policing and queuing (see Fig. 1). DiServ needs queuing
technology like WRED, priority queuing (PQ) and WFQ, which buers and dis-
patches congested packets to accomplish queue management.
In case the IP border network adopts IntServ system, the problem of intercom-
munication between IntServ and DiServ must be solved, including the processing
mode of RSVP in DiServ domains, and the mapping between services supported by
IntServ and PHB (Per-Hop Behavior) supported by DiServ.
2.3. Queue-scheduling disciplines related work
There are many dierent queue-scheduling disciplines, each attempting to nd the
correct balance between complexity, control and fairness
13,15
that describe a number
of popular queue-scheduling disciplines: rst-in-rst-out queuing (FIFO), priority
queuing (PQ), fair queuing (FQ), WFQ, weighted round-robin queuing (WRR) and
decit weighted round-robin queuing (DWRR).
The WRR scheduler is a pioneer in this area yielding dierentiated fairness among
queues. Packets, from dierent ows, are queued in separate queues and the
scheduler polls each queue in a cyclic manner in proportion to a weight pre-assigned
to each queue (see Fig. 1). WRR performs well when all packets have the same size.
16
The decit round-robin (DRR)
17
scheduler is a modication of the WRR scheduler
that handles variable packet sizes without knowing the mean packet size of each ow
in advance. The DRR scheduler provides near-perfect throughput fairness and ow
isolation at low implementation cost. The DWRR scheduler is a variation of the
DRR scheduler and provides unfair bandwidth allocation to dierent queues in the
scheduler. Dierent queues are allocated a dierent quantum value using a propor-
tionally weighted function.
RED
dropper
Shaper
Shaper
Meter
RED
dropper
Marker
S
c
h
e
d
u
l
e
r

Data
Data
EF Queue
AF Queue
BF Queue
C
l
a
s
s
i
f
i
e
r

D
S
C
P

p
r
e
c
e
d
e
n
c
e

Classification AQM Scheduling
Fig. 1. Architecture of the weighted round-robin schedulers based on the IntServ/DiServ integration.
1350012-5
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Recently, Ref. 7 proposed to apply IntServ model in DiServ-based EPON
(Ethernet passive optical network), which uses per-ow processing to guarantee QoS.
A combined DiServ and IntServ model is employed in an EPON system, with a
dynamic bandwidth allocation algorithm to provide more exible user-oriented
service quality. Later, Ref. 8 proposed new User-oriented Hierarchical bandwidth
Scheduling Algorithms (UHSAs) that support DiServ and guaranteed fairness
among end users.
8
Includes inter and intra-optical network unit (ONU) scheduling
processes. The inter-ONU scheduling adopts an improved hybrid cycle approach that
separates a frame into a static part for high priority trac and an adaptive dynamic
part for low priority trac. The intra-ONU scheduling proposes credit-based
scheduling approach to guarantee fairness among end users.
For implementation benets and due to the exible and scalable modular circuit
design approach, certain circuit architecture can be targeted for a full ASIC imple-
mentation. Thus, Ref. 2 proposed a full hardware implementation of a WFQ packet
scheduler in order to deliver 50 Gb/s throughput. The circuit comprises three main
components; a WFQ algorithm computation circuit, a tag/time-stamp sort and re-
trieval circuit and a shared buer. However, the overall performance of the WFQ
circuit is limited by the technology available on the development board, particularly
the memory bus between the FPGA and the RLDRAM II.
This paper proposes a new scheduling technique, called Negative-DWRR, to meet
the DWRR limitations and improve the bandwidth utilization rate without in-
creasing the total latency. Then, a fully hardware packet scheduler is implemented
and veried as part of an intellectual property core.
3. Negative Decit Weighted Round-Robin Scheduler
In this section, we propose a new approach for a queue-scheduling discipline. This
approach, called Negative-decit weighted round robin, is an extended technique
from the DWRR and WRR models.
3.1. WRR and DWRR limitations
WRR and DWRR models support ows with signicantly dierent bandwidth
requirements. They ensure that lower-priority queues are not denied access to buer
space and output port bandwidth.
13
However, WRR's inability to support the pre-
cise allocation of bandwidth when scheduling variable-length packets is a critical
limitation that needs to be addressed. In DWRR, several packets at the head of a
visited queue are not serviced and they would wait the next round robin only because
their sizes are slightly higher than the permitted byte number. Consequently, these
packets may be delayed before they can be serviced. The amount of delay, introduced
during a round needed for scheduling the other queues, causes QoS degradation. It
may cause the dropping of packets placed at the tail of the queue. Moreover, this
1350012-6
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
problem reduces packet throughput, increases end-to-end delay, causes jitter, and
can lead to packet loss if there is insucient buer memory to store all of the packets
that are waiting to be transmitted.
3.2. Negative DWRR parameters
The N-DWRR queue-scheduling discipline is proposed to address the limitations of
WRR and DWRR models and improve especially the bandwidth utilization rate
without increasing the total latency. It denes new parameters that rst allow in
supporting the weighted fair distribution of bandwidth when servicing queues that
contain variable-length packets. Secondly, they reduce the waiting delay for a packet
in a queue even if its size is large. This gives a precise allocation of bandwidth. In
N-DWRR queuing, each queue is congured with the following parameters:
. A weight reects the importance of the service class routed over the queue. It
denes the percentage of the output port bandwidth allocated to the queue. When
the weight of a queue is high, the number of bytes permitted for transmission is
high.
. A quantum of service is proportional to the weight of the queue and is expressed
in terms of bytes. Each round, the quantum is added to the number of bytes that a
queue can transmit.
. A credit is the number of bytes permitted for transmission but they are not yet
transmitted in the previous round. i.e., a queue that was not permitted to transmit
in the previous round, because the packet at the head of the queue was larger than
the value of the permitted bytes, could save transmission \credits" and use them
during the next service round.
. A negative credit is the number of bytes transmitted in the previous round over
the number of permitted bytes. The negative credit for a queue should not exceed
the sum of credits of all the queues.
. A decit-counter species the total number of bytes that the queue is permitted
to transmit each time that it is visited by the scheduler. The Decit-Counter for a
queue is incremented by the quantum each time that the queue is visited by the
scheduler. It depends also on the values of the credit and the negative credit
parameters.
3.3. Negative DWRR algorithm
In the N-DWRR algorithm, the scheduler visits each nonempty queue and deter-
mines the number of bytes in the packet at the head of the queue. The decit-counter
is variable and it takes, each round, a specic value according to the quantum, the
credit and the negative-credit values. The decit-counter is incremented by the value
1350012-7
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
quantum. Two scenarios exist as a function of the size of the packet:
(1) If the size of the packet at the head of the queue is less than or equal to the
variable decit-counter, the packet is transmitted on the output port. Then, the
variable decit-counter is reduced by the number of bytes in the packet. The rest
of the subtraction is called credit. The scheduler continues to dequeue packets
and decrements the variable decit-counter by the size of the transmitted packet
until either the size of the packet at the head of the queue is greater than the
variable decit-counter or the queue is empty.
(2) If the size of the packet at the head of the queue is greater than the variable
decit-counter, then one of the following two situations take place:
(a) If the size of the packet at the head of the queue is less than or equal to the
variable decit-counter incremented by the available credits of all the queues,
the packet is transmitted on the output port. The negative-credit takes then
the value of the out of range transmitted bytes (of this packet) among the
available credit. The latter depends on credits and negative-credits of all
queues as explained in Fig. 2. The variable decit-counter as well as the credit
are reset to zero. The scheduler moves on to service the next nonempty queue.
This situation can be explained by the fact that the current queue has taken
a part of the bandwidth unused by the other queues. In general, this part of
Queue (i)
Wi <= Weight queue(i)
Qi <= Quantum queue(i)
DCi <= Deficit counter queue(i)
Pckt_size
<
DCi
Transmit pckt DCi <= DCi pckt_size
Pckt_size
<
DCx
Transmit pckt
N_credit = pckt_size - DCi
Credit <= reset (0)
DCi <= reset (0)
Queue (i+1)
Credit <= DCi
N_credit <= reset (0)
Yes
No
Yes
No
Wi traffic type
Qi depends on Wi
DCi = f
ct
(Qi, credit(i), n_credit(i))
Av_credit = DCi * Beta-i (Available credits)
Adding Av_credit without exceeding total bandwidth
DCx <= DCi + Av_credit
next packet
Fig. 2. The N-DWRR algorithm state diagram.
1350012-8
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
bandwidth depends on (i) the credit and the negative credit of all queues and
(ii) the size of the packet at head of this queue. This portion of bandwidth
will be subtracted, at the next round, from the bandwidth allocated to this
queue. This approach allows reducing packet latency in queues, without
exceeding the total bandwidth.
(b) If the size of the packet at the head of the queue is greater than the variable
decit-counter incremented by the sum of the credit of all the queues, the
scheduler moves on to service the next nonempty queue. This queue will be
characterized by two values (credit and negative-credit), which will be used
within the next visit (round). The credit takes the last value of the decit-
counter. However, the negative-credit is reset to zero.
(3) When the queue is empty, the scheduler sets to zero the decit-counter, the credit
and the negative-credit values and moves on to service the next nonempty queue.
Figure 2 shows the NDWRR algorithm. It presents two test levels before deciding to
transmit a packet or to move on to service the next nonempty queue. These two test
levels give more probability to transmit a packet and decrease the average queue size,
which avoids congestion of packets in the tail of each queue.
4. N-DWRR Performance Evaluation
A queue-scheduling discipline allows managing the access to a xed amount of
output port bandwidth by selecting the next packet that is transmitted on a port.
The packet transmission order depends on two parameters: the bandwidth allocated
to each queue (weight Wi) and Beta. Beta is a new parameter introduced by
N-DWRR to improve the total bandwidth utilization. It represents an additional
amount of bandwidth to oer for each queue from the unused band. Our approach
provides a limited band without exceeding the total bandwidth. N-DWRR assumes
that the total bandwidth is not in general fully occupied. Therefore, the free band can
be distributed according to the needs of all queues.
4.1. Simulation environment and parameters
In this section, using Opnet modeler, we evaluate the performance of the most popular
scheduling techniques, at dierent scenarios, based on many service classes: VoIP,
FTP, HTTP and Email. We consider a topology/network architecture including
mainly two parts: network core and network edge. In total, this environment includes
112 hardware devices (routers, switches, workstations, servers and VoIP telephones),
115 physical link (serial, Ethernet) and 3 conguration utilities. The simulation
scenario is based on many communication features such as trac type, number of
sources, trac starting time, trac data rate, etc. Each simulation scenario is done
for 10 min including dierent trac conditions (light, burst and heavy loads).
Table 1 outlines the QoS requirements expected for the dierent types of trac.
1350012-9
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
4.2. Simulation results
In order to evaluate the performance of the proposed technique, two simulation levels
are suggested. First, the simulation of the most popular scheduling techniques
deployed for DiServ in IP networks is done. This allows selecting the most eective
technique in terms of bandwidth utilization and latency. Second, using computer
simulation, we evaluate the performance of the proposed N-DWRR algorithm
compared to the selected technique.
4.2.1. Popular scheduling algorithms evaluation
CQ, DWRR, FIFO, PQ and WFQ are the most popular scheduling techniques used
in the IP networks that we propose to evaluate their performances under dierent
scenarios and trac types. Figure 3 illustrates the main results of our simulation.
As mentioned in Table 1, all queuing strategies provide less than the acceptable
range. But, DWRR and WFQ queues oer the lower end-to-end delay and jitter due
to the weight concept. They also oer a reasonable response time for FTP and HTTP
applications. For these experiments, PQ and FIFO provide the worst performances
which cannot guarantee timing requirements in heavy or burst load.
Both WFQ and DWRR perform very well against the other technique, especially
in terms of end-to-end delay, jitter, throughput, download response time and
queuing delay. They also guarantee the lower dropping rate of IP packets. Finally,
Table 1. QoS requirements for the supported types of trac.
Trac class Delay Jitter Response time Loss Bandwidth
VoIP <150 ms <30 ms <1% High
FTP Med None 25 sec Zero High
HTTP <400 ms None 25 sec Zero <30.5 Kbps
Email Low None 25 sec Zero <10 Kbps
(a) VoIP End-To-End Delay (color online). (b) VoIP Packet Delay Variation (color online).
Fig. 3.
1350012-10
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
we demonstrate that the DWRR technique oers quite higher performances com-
pared to WFQ in all results (see Fig. 3). Moreover, DWRR is less computationally
complex and consequently easier to be implemented even in hardware. That is why
we suggest selecting DWRR for performance evaluation of our technique.
(c) FTP Download Response Time (color online).
(d) HTTP Download
Response Time (color online).
(e) Throughput (color online). (f) Utilization rate (color online).
(g) Queuing Delay (color online). (h) Dropped IP trac (color online).
Fig. 3. (Continued)
1350012-11
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
4.2.2. Bandwidth utilization
In this section, we evaluate the performance of the proposed N-DWRR technique
through extensive simulation experiments maintained within a single IP router based
on a scalable architecture of queues. Here, four queues having dierent weights are
deployed to store IP packets which are delivered randomly with variable size and
support dierent types of trac. For the purpose of comparison and to show the
provided advantages, the performance evaluation consists mainly of measuring two
QoS metrics: the bandwidth utilization rate and latency. The rst metric measures
the total bandwidth used by all queues during each round. The latency measures the
total queuing delay of all packets of the same weight.
Figure 4 shows the evolution of the total bandwidth used for each duty cycle of all
queues. These queues are characterized by dierent weights given by 0.4, 0.3, 0.2 and
0.1. Notably, this gure shows that the N-DWRR technique absorbs more band-
width than DWRR. The gap in bandwidth carried by N-DWRR reaches up to more
than 12% from the total bandwidth. In addition, N-DWRR aects the bandwidth
allocation speed which rapidly increases since the rst round. This is due to two
reasons. First, empty queues are skipped to serve next queue which improve band-
width allocation. Second, we start with an intensive trac input to check the e-
ciency of our algorithm. Thus, it can react appropriately within bursty trac
situations. This approach is also more suitable to support steady trac ow or
constant bit rate, as indicated by the last part of the curve (see Fig. 4).
All these interpretations have been conrmed by varying the weight of these
queues. Figures 5(a) and 5(b) justify these results and show the bandwidth opti-
mization rate by using the N-DWRR technique. Table 2 shows the dierent weights
of queues employed to extract results given in Figs. 4 and 5. This table also provides
the maximum bandwidth allocation gap between the two techniques. The gap
measures the dierence between the allocated bandwidth using both scheduling
70
80
90
100
0 10 20 30 40 50
U
s
e
d

b
a
n
d
w
i
d
t
h

(
%
)
Round robin number
DWRR
NDWRR
Fig. 4. Bandwidth utilization rate for (W
1
0:4; W
2
0:3; W
3
0:2; W
4
0:1) weighted queues.
1350012-12
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
techniques. This parameter also reects the increasing rate in terms of bandwidth
improved by our technique against DWRR. This gap reaches upto 25% which
represents an eective argument indicating the eciency of the N-DWRR scheduling
technique.
4.2.3. Latency evaluation
The advantage of the N-DWRR technique is to maximize the bandwidth utilization
rate without introducing an additional queuing delay. This technique solves
successfully the compromise between bandwidth allocation and queuing delay.
Figures 6(a) and 6(b) evaluate the total latency of packets according to the weight of
their queues. First, these gures conrm that the latency is inversely proportional to
the weight of queues. Second, they show that the latency maintained by N-DWRR is
slightly lower than DWRR.
4.2.4. Impact of Beta
N-DWRR specication (see Sec. 3.3) employs a generic parameter \Beta" which acts
on the performance of this technique. Beta represents the negative credit that a
queue is permitted to reach. In other terms, Beta is an additional amount of band-
width, computed as a fraction of the Quantum that N-DWRR oers to a queue. In
Table 2. Maximum gap of allocated bandwidth between N-DWRR and DWRR.
Weight Q
1
Weight Q
2
Weight Q
3
Weight Q
4
Max gap of BW
Figure 4 0.4 0.3 0.2 0.1 12%
Figure 5(a) 0.25 0.25 0.25 0.25 25%
Figure 5(b) 0.42 0.08 0.42 0.08 15%
50
60
70
80
90
100
DWRR
NDWRR
(a) W
1
W
2
W
3
W
4
0:25
60
70
80
90
100
DWRR
NDWRR
(b) W
1
0:42; W
2
0:08; W
3
0:42; W
4
0:08
Fig. 5. Bandwidth utilization rate for dierent weighted queues.
1350012-13
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
this section, we are interested in measuring our QoS metric evolution function in
Beta. Thus, the objective is to determine the value of Beta, which serves both to
maximize bandwidth and minimize latency. We kept the same simulation environ-
ment (scenario, queue weight) for various alternatives of Beta.
First, the aim of this simulation is to determine a precise value (or value range) for
beta allowing access to the maximum amount of output port bandwidth. In this
context, we simulated the bandwidth evolution used by all queues according to Beta
0
0.2
0.4
0.6
0.8
1
1.2
5 15 25 35 45 55
T
o
t
a
l
l
a
t
e
n
c
y
(
m
s
)
Queue weight (%)
Total latency (DWRR)
Total latency (NDWRR)
(a) W
1
0:05; W
2
0:15; W
3
0:2 and W
4
0:6
0
0.2
0.4
0.6
0.8
1
1.2
5 15 25 35 45 55
T
o
t
a
l
l
a
t
e
n
c
y
(
m
s
)
Queue weight (%)
Total latency (DWRR)
Total latency (NDWRR)
(b) W
1
0:1; W
2
0:2; W
3
0:3 and W
4
0:4
Fig. 6. Total queuing delay of packets related to their queue weights.
90
91
92
93
94
95
5 15 25 35 45 55
A
v
e
r
a
g
e

o
f

u
s
e
d

b
a
n
d
w
i
d
t
h

(
a
l
l

q
u
e
u
e
s
)

(
%
)
Beta (%)
Fig. 7. Bandwidth utilization average as a function of Beta.
1350012-14
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
variation. In Fig. 7, the result shows that the used bandwidth increases with Beta to
reach a peak near the range of 3545%. The bandwidth decreases slightly beyond the
peak value and then keeps almost constant. However, the queuing delay becomes
particularly high (see Fig. 8). Therefore, the choice of Beta should satisfy the com-
promise between bandwidth allocation and queuing delay.
Second, Fig. 8 shows the latency evolution in queues with dierent weights. The
evolution often includes a Beta Band where latency is minimal. In the beta range of
25% to 35%, we note that latency meets a minimum of about 0.989 ms, 0.654 ms and
0.335 ms for the light weighted queues (a, b, c) and a maximum of about of 0.168 ms
only for the high-weighted queue (d). For the latter, the latency increases from
0.16 ms (beta 0) to 0.168 ms (beta 35%) which results in an increase of 8.1 s.
However, for all the light weighted queues, the latency decreases by 54 s.
0.988
0.989
0.99
0.991
0.992
0.993
0.994
0.995
(a)
0.652
0.654
0.656
0.658
0.66
0.662
0.664
0.666
(b)
0.33
0.335
0.34
0.345
0.35
0.355
0.36
(c)
0.152
0.154
0.156
0.158
0.16
0.162
0.164
0.166
0.168
0.17
(d)
Fig. 8. Total queuing delay for dierent weighted queues.
1350012-15
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
In conclusion, the common Beta range providing optimized values is limited be-
tween 25% and 35%. We choose Beta equal to 30% in order to obtain optimal results
which are already outlined in Secs. 4.2.2 and 4.2.3.
5. RTL Design of Embedded Packet Processors
Until recently, such a router used to be implemented entirely in software, running on
a general-purpose processor in a host computer. However, such implementations are
increasingly becoming infeasible because of two reasons: performance requirements
and complexity of the packet-processing functions. During the last few years, the
available network bandwidth has been on the rise, and with the advent of optical
bers being deployed for networking, network bandwidth has increased exponen-
tially. This has led to very stringent performance requirements from routers since
they have to process packets at line speed. Further, any new service to be supported
by a network is implemented by extending or modifying the routers. This has led to
very complex functionality being built in routers, with the additional requirement of
being reasonably exible.
The real-time packet-processing constraints imposed on routers to support high-
line speeds motivate hardware-based solutions, where the router functionality is
implemented on application specic integrated circuits (ASICs). The requirements
for exibility and the complex nature of many of the processing functions, on the
other hand, favor software-based implementations on general-purpose processors. To
address these two conicting issues, recently a new class of devices called \network
processors" has emerged.
18
These are high performance programmable devices with
special architectural features that are optimized for packet processing. It is moti-
vated by the fact that the design and analysis of hardwaresoftware architectures
for such processors requires new models and methods, which do not fall under the
domain of traditional embedded systems design.
19,20
5.1. RTL design: Issues and trends
Most of today's embedded systems are implemented as a system-on-chip (SoC). In
the context of network packet processors, the architecture of such systems consists of
a heterogeneous combination of dierent hardware and software components. The
hardware components consist of dedicated hardware block cores, dierent kinds of
memory modules and caches, various interconnections and inputoutput inter-
faces.
21
All of these are integrated on a single chip and run specialized software to
perform packet processing. The process of determining the optimal hardware and
software architecture for such processors includes issues involving resource alloca-
tion, partitioning and design space exploration. To tackle the complexity of such
designs, and also to meet demands for short time-to-market and low cost, several new
design paradigms like platform-based design
19
have evolved. These are based on the
1350012-16
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
idea of concurrent design of hardware and software, and the design ow starts with
an abstract specication of the application.
According to a protocol stack architecture, routers are built especially over two
layers, MAC in layer-2 and IP in layer-3. Forwarding frames are based on layer-2
MAC address information. Because of the standardization of the IP as the layer-3
protocol in local- and wide-area networks, layer-3 switching made it possible to do
classication and forwarding of packets based on their layer-3 DSCP eld, without
resorting to complex routing algorithms. Since IP packets make up the major portion
of trac in any switch or router, we oer here a solution where IP classication,
discarding, queuing, scheduling and forwarding were hardwired into ASICs.
5.2. Logic synthesis
Logic synthesis is one of the most important phases of the design ow in state-of-the-
art circuits. It aims at transforming the HDL (usually Verilog HDL or VHDL)
description of the circuit into a technology-dependent, gate-level netlist. Through
this process, the hardware designer denes the environmental conditions, con-
straints, compile methodology, design rules and target libraries, in order to achieve
certain design goals set by the initial specications. The tool we use for the logic
synthesis of the circuit is Synopsys DC, the most widely used synthesis tool. DC
optimizes logic designs for speed, area and wire routability. From the dened goals,
DC synthesizes the circuit and tailors it to a target technology. The gate-level re-
presentation of the circuit is the input le to the Place and Route tool.
The synthesis process is completed relatively easily and timing constraints are
met, while circuit area is kept to a minimum. Timing constraints are of the greatest
importance, as we opted for a clock frequency of 300 MHz (3.33 ns clock cycle). We
were constrained to 300 MHz due to the fact that it was the maximum operating
frequency of the FIFO we had. The circuit integrates 12 FIFOs as queues needed for
multiple port reception, dierentiated service classication and scheduling. Hence,
we achieve 76.8 Gbit/s incoming router throughput (8 inputs of 9.6 Gbit/s link
throughput). Synthesis results are given in Table 3.
5.3. Post-synthesis verication
Post-synthesis verication is possibly the most important phase of the synthesis ow.
It aims at testing whether the initial RTL design has the same behavior as the gate-
level netlist produced by the synthesis tool. In most cases, the initial results are not
the same, and the designer has to carefully investigate the reasons for the erroneous
behavior of the gate-level netlist. Usual mistakes happen when the circuit does not
reset correctly, a mistake that can pass unseen from the HDL compiler and simu-
lator, but, of course, the actual circuit will not work correctly.
Our synthesized gate-level netlist is imported back to the gate-level DC and is
tested within the same environment as the initial RTL design. The nal netlist
1350012-17
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
proved to behave correctly. The netlist is now ready to be imported to the Place &
Route tool for the nal phase of the design process.
5.4. Place and route ow
We place and route (P&R) the router chip core with Cadence Encounter.
22
It is
comprised of various stages, some of them being optional, although important for
new design cases. At rst, the design is imported to the tool. The les needed are
technology-dependent and are either given by the technology vendor (in our case
CMOS 0.35 m) or are produced by the synthesis tool (in our case Synopsys Design
Compiler). The vendor provides the P&R tool with (a) technology library (.lib les),
which contains the exact electrical characteristics of the library standard cells, (b)
layout (.lef les), which includes standard cell actual layout, pin placement and
metal layer usage and (c) verilog (.v les), containing the interface of every standard
cell. The same vendor also provides the memory models specied by their timing
library (.tlf les). From the synthesis tool point of view, the only information needed
to be passed on to the P&R tool is the Synopsys technology mapped (.v) gate-level
representation of the design.
After importing the design, the designer has to oorplan the various HDL mod-
ules and/or black boxes in the actual chip. Power rings are created and block rings
are added for power/ground termination purposes. Input/Output pins are also
placed in this stage. The most important decisions are, of course, exact chip layout,
utilization percentage and I/O placement.
When trial routing is complete, we move to the Clock Tree Placement phase. This
is achieved with buer insertions and thorough computations of the tree node
weights, so as to minimize clock skew produced by unbalanced clock tree placement.
In the nal route phase, setup and hold times of registers are taken into account and,
having in mind the operating frequency constraint, the tools try to buer wires and
resize standard cells in order to minimize clock skew.
5.5. P&R results
The classier, AQM and scheduler chip we place and route has a total area of
88.72 mm
2
, with a square shape of (9:4 9:4 mm), as presented in Fig. 9. Area
results, as well as gate number can be seen in Table 4.
Table 3. Synthesis results: Memory area and cells are not included in this table.
Module Nb. I/O ports Nb cells (10
3
) Area (mm
2
)
Classier 458 183.6 10.2
AQM & NDWRR 192 226.8 12.6
Total 344 410.4 22.8
1350012-18
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Note the area dierences compared to the corresponding synthesis results (shown
in Table 3): area is larger, while core gate count has increased by 18.8%. These
discrepancies are due to: (a) wiring; (b) hierarchy overhead; (c) clock tree and op-
timization buer insertion; (d) gate features in CMOS 0.35 m technology and (e)
area reserved for FIFOs. The total area, occupied by these FIFOs in the circuit, is
17.7 mm
2
. Keep in mind that each module has to be fully optimized before importing
it to the top-level core. As a result, in most cases, the tool added buers in all
hierarchy levels in order to meet timing constraints. These buers, although usually
small, are numerous.
In this work, we presented a novel IP router organization and proved its feasibility
by designing classier, AQM and scheduler modules. They aggregate incoming
throughput of 76.8 Gbit/s. Final chip core area was 88.72 mm
2
in a 0.35 m CMOS
technology, while its power consumption dropped just below 6 W. Taking into ac-
count that a similar router would be optimized (area, consumption) in the emerging
0.09 m technologies, we can state that the adopted organization can become a
switching block for future applications.
6. Conclusion
In this paper, we have proposed a novel QoS scheduler called negative decit
weighted round robin scheduler. The motivation of the NDWRR scheduler is to
Table 4. Place and route results: Total numbers shown do not include FIFOs.
Module
Nb. cells
(10
3
)
Nb. gates
(10
6
)
Nb.
transistors
(10
6
)
Area
0.35 m
(mm
2
)
Nb. I/O
ports
Nb. pins
(10
3
)
All modules (without FIFOs) 487 1.46 5.84 71.02 344 575
FIFOs used as
queues
Fig. 9. Final IP router layout: Square core with dimensions of 9:41 9:41 mm (88.72 mm
2
).
1350012-19
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
provide end-to-end bandwidth guarantees for dierentiated service classes in large IP
networks. It can maintain its weighted share of bandwidth and provide delay dif-
ferentiation among queues. However, this is done at the cost of a higher algorithmic
complexity due to the rearrangement of service order according to the weight of the
packets at the head of each queue.
The N-DWRR scheduler provides two advantages. First, it works with variable
size packets. Second, the N-DWRR scheduler avoids having a packet waiting in
queue only because its size is slightly higher than the variable decit-counter. This
problem occurs in the DWRR scheduler. However, in the N-DWRR scheduler the
packet in the head of the queue can be transmitted, even there is no sucient
bandwidth, without exceeding the total bandwidth of the network. In conclusion, the
N-DWRR scheduler provides its queues with delay dierentiation in terms of average
and worst-case packet delay that is statistically less than the DWRR scheduler; and
with throughput fairness that is higher than the DWRR and WFQ schedulers.
Further, this work includes the design evaluation for an optimal architecture of
network processors. It introduces a new service scheme motivated by the require-
ments of multi-service access networks. Based on the synthesis and P&R steps, we
then evaluated dierent combinations of algorithms (for policing, queuing and link
scheduling) along with dierent hardware building blocks and memory architectures,
for the design of a packet processor to support the proposed service scheme.
Acknowledgments
This work is supported by the Research Center of the College of Computer and
Information Sciences King Saud University.
References
1. P. Giacomazzi, L. Musumeci, G. Saddemi and G. Verticale, Two dierent approaches for
providing QoS in the Internet backbone, Comput. Commun. 29 (2006) 39573969.
2. K. McLaughlin, D. Burns, C. Toal, C. McKillen and S. Sezer, Fully hardware based WFQ
architecture for high-speed QoS packet scheduling, Integration VLSI J. (2011),
doi:10.1016/j.vlsi.2011.01.001.
3. T. M. Lim, B. S. Lee and C. K. Yeo, Weighted decit earliest departure rst scheduling,
Comput. Commun. 28 (2005) 17111720.
4. R. Braden, D. Clark and S. Shenker, Integrated services in the internet architecture: An
overview, RFC (1633).
5. S. Blakem, D. Black, M. Carlson, E. Davies, Z. Wang and W. Weiss, An architecture for
dierentiated service, RFC (2475).
6. S. Ece (Guran) Schmidt and H. S. Kim, Frame-counter scheduler: A novel QoS scheduler
for real-time trac, Comput. Commun. 29 (2006) 21812200.
7. L. Qiong, L. Hui, J. Yue-Feng and Q. Yao-Jun, Resources allocation in an Intserv/
Diserv integrated EPON system, J. China Univ. Posts Telecommunications 16 (2009)
108113.
1350012-20
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
8. Y. Yin and G. Poo, User-oriented hierarchical bandwidth scheduling for Ethernet passive
optical networks, Comput. Commun. 33 (2010) 965975.
9. S. Floyd and V. Jacobson, Random early detection gateways for congestion avoidance,
IEEE/ACM Trans. Netw. 1 (1993) 397413.
10. M. Wurtzler, Analysis and simulation of weighted random early detection (WRED)
queues, EECS 891 project, 2002.
11. S. Mishima, L. Moy-Yee, G. Yee-Madera and E. Youse, Broadband packet switch
processor, Space Commun. 18 (2002) 9195.
12. Y. Mo, J. M. Nho, Yang, N. P. Mahalik, K. Kim and B. H. Ahn, A trac-class burst-
polling based delta DBA scheme for QoS in distributed EPONs, Comput. Stand. Interfac.
28 (2006) 721736.
13. Supporting dierentiated service classes: Queue scheduling disciplines, white paper chuck
semeria, marketing engineer. Part number 200020-001, December 2001.
14. J. Postel, Internet protocol, DARPA Internet program protocol specication RFC791,
September 1981.
15. L. Le, J. Aikat, K. Jeay and F. D. Smith, The eects of active queue management on web
performance, in SIGCOMM'03, Germany, 2529 August 2003, pp. 265276.
16. M. Y. L. Wong and C. K. Li, Low computational complexity weighted queuing using
weighted Decit Round Robin, Proc. IASTED Application Information, Innsbruck,
Austria (2002).
17. M. Shreedhar and G. Varghese, Ecient fair queuing using Decit Round Robin, IEEE/
ACM Trans. Netw. 4 (1996) 375385.
18. S. Chakraborty, System-level timing analysis and scheduling for embedded packet pro-
cessors, Thesis of the Swiss Federal Institute of Technology, Zurich, April 2003.
19. K. Keutzer, S. Malik, R. Newton, J. M. Rabaey and A. Sangiovanni-Vincentelli, System
level design: Orthogonolization of concerns and platform-based design, IEEE Trans.
Comput.-Aided Des. 19 (2000).
20. R. A. Bergamaschi, S. Bhattacharya, R. Wagner, C. Fellenz, M. Muhlada, W. R. Lee,
F. White and J.-M. Daveau, Automating the design of SoCs using cores, IEEE Des. Test
Comput. 18 (2001) 3245.
21. D. G. Simos, Design of a 32 32 variable-packet-size buered crossbar switch chip,
Thesis of institute of computer science FORTH-ICS/TR-339, July 2004.
22. Cadence Corporation; Available at http://www.cadence.com.
1350012-21
J

C
I
R
C
U
I
T

S
Y
S
T

C
O
M
P

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
w
o
r
l
d
s
c
i
e
n
t
i
f
i
c
.
c
o
m
b
y

4
6
.
2
3
0
.
8
2
.
6
8

o
n

0
2
/
2
8
/
1
3
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Dynamic slot assignment protocol for QoS support on TDMA-based mobile networks
Ridha Ouni
College of Computer and Information Sciences, King Saud University, Kingdom of Saudi Arabia
Faculty of Sciences of Monastir, EE, Tunisia
a b s t r a c t a r t i c l e i n f o
Article history:
Received 2 November 2009
Received in revised form 21 May 2011
Accepted 15 June 2011
Available online 2 July 2011
Keywords:
TDMA/FDD
WLAN
Bandwidth allocation
QoS
MAC protocol
An efcient bandwidth allocation scheme in wireless networks should not only guarantee successful data
transmission without collisions but also enhance the channel spatial reuse to maximize the system
throughput. The design of high-performance wireless Local Area Network (LAN) technologies making use of
TDMA/FDD MAC (Time Division Multiple Access/Frequency Division Duplex Medium Access Control) is a
very active area of research and development. Several protocols have been proposed in the literature as
TDMA-based bandwidth allocation schemes. However, they do not have a convenient generic parameters or
suitable frame repartition for dynamic adjustment. In this work, we undertake the design and performance
evaluation of a QoS (Quality of Service)-aware scheme built on top of the underlying signaling and bandwidth
allocation mechanisms provided by most wireless LANs standards. The main contribution of this study is the
new guarantee-based dynamic adjustment algorithm used in MAC level to provide the required QoS for all
trafc types in wireless medium especially Wireless ATM (Asynchronous Transfer Mode). Performance
evaluation of this approach consists of improving the bandwidth utilization, supporting different QoS
requirements and reducing call reject probability and packet latency.
2011 Elsevier B.V. All rights reserved.
1. Introduction
Several wireless networking solutions have been developed to
provide different types of services for various end user applications.
Wireless mobile data communication has been enlarged with the
developments of high performance wireless computers and other
mobile devices [1]. Bandwidth is a scarce resource that can be shared
either dynamically according to the amount of data required to be
transferred to or from each node or deterministically by assigning a
xed number of slots to each cell, as in a cellular network [2]. With the
deterministic assignment, a xed number of slots, as portion of the
bandwidth, is assigned to certain nodes (or groups of nodes) so that
they have exclusive access to the assigned bandwidth. The dynamic
assignment shares the bandwidth into the required number of slots
which can be changed according to the occurred events in the
network. Thus, a QoS (bandwidth, delay, jitter) guarantee can be
provided. However, traditional bandwidth allocation is pre-planned,
hence, it is less adaptive to trafc load variations and network
topology changes [2].
Recently, several high-performance wireless LAN technologies
making use of TDMA/TDD (Time Division Duplex) MAC have been
designed [35]. In this type of networks, a central unit allocates the
bandwidth among all active mobile terminals. These lasts use a set of
signaling primitives in order to place their resource requests. The
central unit then allocates the channel bandwidth among all the
competing mobiles [3].
TDMA/FDD (Frequency Division Duplex) MAC protocol is used to
separate uplink and downlink channels which are divided into a
contiguous series of xed-size TDMA frames. Each frame is further
subdivided onto a xed number of slots to be allocated for different
service classes. TDMA wireless networks, when operating under the
infrastructure mode, distinguish between two types of devices: the
Access Point (AP) and the Mobile Terminal (MT) [3]. The AP is the
entity responsible of providing connectivity with the core network as
well as for adapting the users requirements by taking into account the
characteristics of the core network and the services offered by the
wireless LAN. Furthermore, the AP distributes resources and main-
tains coordination between all the MTs located within the cell [3].
Traditional wireless networks deployed universally cannot pro-
vide the necessary QoS guarantees for bursty trafc such as real-time
multimedia applications [1]. Based on the MAC protocol, it is possible
to build up QoS mechanisms capable of providing the guarantees
needed by various applications. These mechanisms need to specify the
format and sequence of the control messages between MTs and AP.
However, they dont dene the specics regarding the timing and
numerical values of the system parameters, such as the bandwidth to
be reserved for each type of connection [3]. In other hand, these
mechanisms may consider distributed dynamic slot and power
allocation for a TD/SDMA broadband wireless packet network with
multiple access ports and adaptive antennas [6].
Computer Standards & Interfaces 34 (2012) 146155
CCIS, King Saud University, P.O. Box 51178, Riyadh 11543, Kingdomof Saudi Arabia.
Fax: +966 1 4675630.
E-mail address: rouni@ksu.edu.sa.
0920-5489/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.csi.2011.06.003
Contents lists available at ScienceDirect
Computer Standards & Interfaces
j our nal homepage: www. el sevi er. com/ l ocat e/ csi
This work considers the problem of providing a QoS guarantee to
users and simultaneously maintaining the most efcient use of scarce
bandwidth resources. It employs a certain bandwidth allocation
scheme which needs to be dynamic according to the trafc demand
variations at each node. An efcient bandwidth allocation scheme
should take into account generic parameters describing different
requirements of MT trafcs. This scheme should then guarantee
successful data transmissions without collisions, share bandwidth
dynamically and enhance channel spatial reuse to maximize system
throughput. In this way, at setup connection time, the applications
may specify their QoS requirements. The admission control mecha-
nisms may then accept or reject newcalls based on the level of activity
of the on-going connections. Due to the high degree of burstiness of
todays applications, it is important that the AP be regularly informed
of the status of the active connections [3]. This paper makes use of a
TDMA/FDD based dynamic channel assignment to improve the
networks ability to meet the QoS requirements of various types of
applications. This work proposes a dynamic TDMA bandwidth
allocation approach referred to as Dynamic Slot Assignment Protocol
(DSAP). This approach efciently utilizes the channel bandwidth by
assigning unused slots to new MTs and compensating additional time
slots from lower QoS classes when the number of slots in the frame is
insufcient to support newconnections. This protocol also gives more
priority to serve handoff calls.
The rest of this paper is organized as follows. Section 2 provides a
short overview of the frame structure and access mechanisms used in
TDMA/FDD networks. In Section 3, we describe our dynamic
bandwidth allocation proposal. A performance evaluation of this
proposal is given in Section 4. Finally, Section 5 concludes the paper.
2. MAC protocol: principles and description
2.1. Time division multiple access
Time division multiple access (TDMA) shares the available
bandwidth in the time domain. Each frequency band is divided into
several time slots. A set of such periodically repeating time slots is
known as the TDMA frame. Each node is assigned one or more time
slots in each frame, and the node transmits only in those slots [3]. For
two-way communication, TDMA systems employ the uplink and the
downlink time slots for transmitting and receiving data respectively.
The uplink and downlink communications can be achieved in the
same frequency band, known as time division duplex (TDMA/TDD), or
on different frequency band, called frequency division duplex (TDMA-
FDD) [2].
2.2. Wireless network environment
TDMA wireless networks are designed through a cellular structure
in which a set of mobile terminals (nodes) maintain their connectiv-
ities with the core network over an AP. the AP allocates the time slots
in response to the MT requests. With this objective, each MT has to
request the required resources to the AP by issuing a Resource
Request (RR) message, while the AP informs the MT of the positive
outcome by using a Resource Grant (RG) message.
In WATM architecture, Wireless Terminals (WTs) and APs (or BSs:
Base Stations) can be xed or mobile. Therefore, some network
components constituting a WATM can be different likewise. A WATM
system with mobile users and xed APs consists of three major
network components [1]. These are a mobile terminal with radio and
mobility enhanced software and mobility enhanced ATM switch. The
WATM protocol stack integrates additional modules to the standard
ATM which include radio channels among MTs and APs, a DLC (Data
Link Control) layer, a MAC layer and a wireless access control prole,
which supports such functions as radio resource management at
physical, MAC and DLC layers, as well as mobility management.
2.3. TDMA-FDD frame description
The downlink channel time is divided into time frames of equal
length, which are, in turn, divided into a number of time slots [1,7].
The channel frame duration accommodates 111 slots shared among
dynamic and xed applications-based allocation. Fixed allocation
requires a certain guaranteed timeslots to be assigned while the
connection is established. However, dynamic allocation ensures two
types of timeslots: guaranteed and non-guaranteed. Each time slot
includes a single 56 bytes wireless ATM packet as adopted by many
previous works [1,810] while it improves efciently the bandwidth
utilization rate. A slot allocation table (SAT) is used to manage and
update the time slot allocation. In our study, we divide the downlink
frame in two intervals:
(a). The control interval, which is placed at the beginning of each
frame, and
(b). The information (or data) interval which may include idle slots.
The control interval comprises two elds: the frame header and
the signaling eld. The frame header is the rst downlink control slot
used by the AP for signaling the beginning of a frame transmission and
denoting the current frame structure. The signaling elds divided into
a number of time slots in the downlink, corresponding to an equal
number of request slots in the uplink, plus a fewmore slots needed for
reasons which are explained in Section 3.2. Each signaling slot is
divided into two mini-slots of equal duration.
Uplink request slots are used by terminals to access the channel
and make a slot reservation, and are placed at the beginning of each
time similarly to the signaling slots in the downlink. Request slots are
also used by mobile terminals to acquire the additional information
slots when their bit rate increases.
The information interval consists of xed allocated slots and
dynamic allocated slots according the supported service classes. We
consider four types of trafc in our study: Constant Bit Rate (CBR),
Variable Bit Rate (VBR), Available Bit Rate (ABR) and Unspecied Bit
Rate (UBR). Specic trafc and QoS parameters have been dened to
characterize each service class. Based on the previous concepts and
references, Fig. 1 presents our DSAP frame structure proposal to
dynamically allocate slots for different service classes.
For Constant Bit Rate (CBR) trafc, slot allocation is performed
once during call establishment. A xed allocation of slots is assigned
according to user requests. When CBR slots are no longer available,
arriving CBR calls are blocked. Variable Bit Rate (VBR) slots are
assigned based on a statistical multiplexing algorithm. Unused CBR
and VBR slots are shared with other trafc classes. Arriving VBR calls
are also blocked when VBR slots are not available. Finally, for Available
Bit Rate/Unspecied Bit Rate (ABR/UBR) trafcs, slot allocation is
performed on a burst-by-burst basis via dynamic reservation of ABR/
UBR slots and unused CBR and VBR slots. Since slots are apportioned
for CBR, VBR, and ABR/UBR categories, the channel is not dominated
by the most demanding user. In order to minimize the waste of
bandwidth due to collisions, reservation slots are divided in mini-slots
[11].
2.4. Related works: bandwidth allocation strategies
For efcient utilization of the radio spectrum, a frequency reuse
scheme that is consistent with the objectives of increasing capacity
and minimizing interference is required. A variety of channel
assignment strategies have been developed to achieve these objec-
tives. Channel assignment strategies can be classied as either xed or
dynamic [12]. In a Fixed Channel Assignment (FCA) strategy, each cell
is allocated a predetermined set of voice channels. If all the channels
in that cell are occupied, the call is blocked and the subscriber does not
receive service.
147 R. Ouni / Computer Standards & Interfaces 34 (2012) 146155
In a Dynamic Channel Assignment (DCA) strategy, voice channels
are not allocated to different cells permanently. Given the central role
played by the MAC algorithms, there have been a large number of
studies focusing on the design and evaluation of QoS-aware MAC
protocols for TDMA networks [3].
One of the most well-known protocols of this category is the
Dynamic TDMA with TDD (DTDMA/TDD) protocol, which was
designed especially for the wireless ATM network (WATMnet) [11].
It is a microcellular network capable of providing integrated
multimedia communication service to remote terminals [13]. In
[13], users send transmission requests to the access point in the
dedicated reservation slots using slotted Aloha random access. The
requests are then processed, resulting in a schedule table based on the
QoS parameters of user trafc. The access point proceeds to broadcast
slot allocations and acknowledge successful reservations. The basic
disadvantages of this protocol are: (a) its use of the slotted Aloha
access algorithm, which leads to unstable systembehavior (unless the
channel utilization is low) and provides low throughput, and (b) the
fact that the scheme is based on the concept of not allowing the most
demanding user to dominate the channel seems a logical choice in
terms of fairness to lower priority users [11].
In [8], the authors describe a resource reservation protocol. In this
protocol, the MTs issue the rst request by using a contention-based
protocol, similar to S-Aloha. Once having been allocated a number of
channels, the data packets convey, via piggybacking, the following
MTs resource requests. Whenever, an MT does not have any more
packets to transmit, the resources (bandwidth) are freed. Later, when
the MT once again becomes active, it has to start the reservation
process by issuing a rst request via the contention process.
The MASCARA algorithm [9] makes use of a resource request
mechanism which integrates a scheduler based on the "token bucket"
scheme allowing the resource distribution among the active connec-
tions. The performance of these mechanisms severely degrades as the
number of active connections increases. This may make the allocation
mechanism prone to delay and losses; an undesirable condition when
developing QoS mechanisms [3].
Chang and Kim[14] proposed two efcient heuristic algorithms for
the channel allocation problem that minimize the average call
blocking probability of the whole network subject to the co-channel,
adjacent-site and co-site interference constraints, given the number of
available channels. Sung and Wong [15] developed a sequential
packing algorithm that provides optimal solutions for a special class
of network topologies they also derived a lower bound for the
minimum number of channels required to satisfy a given call trafc
demand under co-channel and adjacent-channel interference
constraints.
With connection oriented services, adaptive channel reallocation
and distributed power control signicantly improve system capacity.
In this context, [16] proposed an efcient resource (slot and power)
allocation scheme with a technique named power shaping, in a
multicell packet environment with sectorized base station antennas.
Unlike in traditional approaches, where time slots are allocated rst,
and then power control is performed, power shaping rst assigns in
advance (static allocation) a maximum power level to each channel
and then dynamically performs channel allocation and ne power
control. Later, the same authors proposed, in [6], an allocation
algorithm which works in a distributed SDMA environment and is
able to support power shaping. Power shaping imposes a constraint
on each slot of the frame about the maximum transmit power. The
allocation algorithm tries to ll each slot with a set of packets,
depending on users spatial separability, channel quality, interference
estimation and available power. In addition, [6] also described two
other allocation algorithms: centralized Max-min Fit (C-MMF) and
random allocation. The centralized Max-Min Fit algorithm ideally
knows the current status of the whole network where Slots are visited
and lled one by one starting from slot 0, with a centralized control,
i.e., each AP is aware of what the other APs are doing. With the
random allocation algorithm, each AP picks up each of its assigned
users in a random order and allocates it to a random slot.
3. QoS-aware model for dynamic bandwidth allocation in mobile
networks
In this section, we briey explain the DSAP model for a dynamic
bandwidth allocation in mobile networks. The TDMA format in this
model has N sessions with M slots each one in the frame. The AP
maintains its frame, and each frame includes a control slot table and
data slot table. M and N are variable values and are adjusted when the
frame does not have enough slots to support new connections. This
protocol controls the expansion and recovery of unassigned time slots
by dynamically changing the set of M values according to the trafc
load, the service classes and the number of mobile terminals in the
contention area.
3.1. Control channel descriptions
The control channel is formed by means of the rst slot in each
frame. The control phases of slot 0 are dedicated to the reservation of
the corresponding data slot. If a MT wants to reserve a data slot, it
contends for the corresponding control slot. The time slots order is
generated in the control slot, and is used in the corresponding data
slot of all the subsequent frames. In general terms, to make a
reservation, the MT rst sends out a request and second a feedback is
provided from the AP of the contention area. This request cannot
successful when it collides with other requests. A four-phase dialogue
is carried out between MTs when establishing connections and AP
Frame (i - 1) Frame ( i ) Frame (i+1)
Guard time
Time

Frame
header
Signaling
mini-slots
Dynamic allocated
ABR, VBR and UBR
slots
Fixed allocated CBR
slots
Idle slots

Fig. 1. The DSAP frame structure.
ensuring adequate resource allocation, by using four types of control
packets.
(1). Request phase: this is the initial phase in which a MT sends a
request packet (REQP) to obtain information on the assigned
slots from the AP. The request packet includes the connection
requirements expressed by specic number of time slots.
(2). Reservation phase: In this phase, the AP receiving the packet
request ensures managing the available resources to satisfy the
requirements. Other alternatives could be followed to allow
establishing the connection when there are not enough
resources (Section 3.2.2).
(3). Response phase: the AP reports the receipt of the request
packet using a response packet (RESP) to announce the number
and the order of assigned slots.
(4). Conrm phase: the MT sends back a conrmation packet
(CFMP) indicating that the suggestion is accepted.
Consider Fig. 2. Here MTi has packets ready to be transmitted via
AP1. During the request phase, node i transmits the request packet
(REQP). If the neighbors of node i send other REQPs ensuring collision,
the request cannot successful. Otherwise, the AP gets the REQP,
allocates the required time-slots and announces the assigned slots
using RESP. MTi collects the information of the assigned slots by
listening to each slot and sends conrmation packet (CFMP) notifying
that the previous allocation is accepted. Finally, a certain number of
time slots are assigned to MTi. The four-phase protocol with the
reservation procedure is thus completed. A similar approach has been
developed in [2].
3.2. Dynamic slot assignment protocol (DSAP)
The considered MAC protocol is centralized in nature with more
management and control functions at the AP. The uplink (MTs to AP)
and downlink (AP to MTs) communication is assumed to be time-
slotted and physically separate (i.e. different frequency channels).
Statistical multiplexing on a TDMA channel is used with a slot length
able to carry one wireless ATM packet. The slots are classied as
available and reserved slots. Available slots are used to send dynamic
reservation requests (signaling channel) and reserved slots are used
to send data packets. To coordinate the multiple accesses in the
uplink, the AP transmits slot-by-slot commands, in order to specify
whether a slot of the uplink channel is available or reserved, and
identify the MT enabled to transmit.
Based on time slot architecture, our proposed protocol controls the
number of unassigned time slots by dynamically changing the
allocated time slots according to the trafc load, the service class
and the number of MT in the contention area.
When a MT needs to communicate with any other, initially it asks
for a transmission channel from the AP. According to the QoS
requirements of this connection request, the AP assigns adequate
number of time slots for this connection using a dynamic Slot
Allocation Table (SAT). The SAT conguration dynamically changes
P
a
c
k
e
t

a
r
r
i
v
a
l
DP
HRP
DRP
Y
Y
Y
N
N
N
Send the packet to the MT.
Higher weight (W
h
) affected,
Compute the number of slots
that satisfy required QoS.
Release the connection slots.
NCRP
Y
N
Normal weight (W
n
) affected,
Compute the number of slots
that satisfy required QoS.
Discard packet
DP: data packet ,
HRP: handoff request packet ,
NCRP: new connection request packet ,
DRP: disconnection request packe t,
W: weight of the connection request.
Reservation procedure
Slots reservation & SAT update
Recovery procedure
Slots release & SAT update
Fig. 3. Signaling procedure of the DSAP protocol.
New
calls
MT0 MT1 MT2 Idle Idle
0 1 2 3 n
MTi
REQP
AP
RESP
MTi
CMFP
New
calls
MT0 MT1 MT2 MTi Idle
Current state of frame allocation
Next state of frame allocation
Phase 1
Reservation + Phase3
Phase4
Fig. 2. Four-phase protocol with reservation procedure in DSAP.
with an algorithm based on the QoS parameters of the service classes.
The proposed DSAP protocol can be mainly divided into three
complementary procedures operating at the AP for signaling,
reservation and recovery.
3.2.1. Signaling procedure
The signaling procedure includes three tasks according the type of
packet that the AP receives. These are namely, assigning adequate
number of slots for a connection considering the QoS requirements of
the requesting MT, forwarding any arrived data packets to their
destinations and terminating any active connection. The core function
of the signaling procedure manages efciently packets arriving at the
AP MAC level. This procedure is highly coupled with the reservation
procedure as described in Fig. 3.
The signaling procedure is used to compute the available
bandwidth for MT applications and allow data packet transfer to the
destined MTs. The trafc delivered to the AP MAC level is identied
according to the packet formats. Four types of packet may be
identied (NCRP, HRP, DP and DRP) and each one needs specic
processing. The receipt of a NCRP allows getting the number of slots
that can provide the required QoS and reserves them from the SAT.
Handoff connection requests are assigned higher priority to be
accepted. When there are not enough slots, a solution is given
through a resource compensation step integrated in the reservation
procedure and will be described below in Section 3.2.2. DRP and DP
handle connection termination requests and data packets deliveries,
respectively. In most cases, packet processing leads to change the SAT
content.
3.2.2. Reservation procedure
The reservation procedure and its resource compensation algo-
rithm in the AP are the most vital part of our DSAP protocol. This
procedure supports QoS guarantee for different and bursty trafcs
such as multimedia applications based on an effective and efcient
management of the SAT. The SAT content depends on the allocated
and available slot number, the MT number in the contention area and
the service classes.
As outlined in ATM standard, the QoS parameters and trafc
descriptors are used to determine the required number of slots. The
PCR (Peak Cell Rate), SCR (Sustainable Cell Rate) and MCR (Minimum
Cell Rate) are the basis parameters which allow determining required
slot number for CBR, VBR and ABR trafcs respectively. With these
parameters (high level allocation category), the slots guarantee eld
are restricted, in the SAT, for these trafcs. With other QoS
parameters, the guarantee eld for VBR, ABR and UBR trafcs may
be reassigned for a new connection based high level allocation.
However, the guarantee eld of UBR slots can be used by any
guaranteed CBR, VBR or ABR connection since UBR service does not
provide any QoS guarantees. These allocation categories are used by
the compensation algorithm.
A CBR connection request is characterized by the PCR and Cell
Transfer Delay (CTD) parameters which their values determine the
number of required slots. Moreover, the CDV parameter allows
distributing these slots through the TDMA frame (as well as through
the SAT) in order to guarantee the Cell Delay Variation (CDV).
A VBR connection request is characterized by the SCR, CTD and MBS
(MaximumBurst Size) parameters which their values determine the
Y
Y
Y
N
N
N
Packet arrival
C
B
R
V
B
R
A
B
R
U
B
R
If unused slots satisfy connection request
requirements for different services classes.
Actualizing / Terminating connection
DRP Timer
Assign additional non -guaranteed slots from
UBR, ABR and VBR connections successively to
satisfy connection requirements (equation1),
Identify MTs contributing for compensation
process and save their number of offered slots.
Get the packet, its requirements & Weight
Requirement negotiation: offering lower QoS
than those required.
Reject connection & Notification with release
information.
Connection parameters buffering,
Send the changes in the SAT to all affected active
MTs.
USATP: phase 1
AP releases slots, and sends update packets
(rewards non-guaranteed slots).
CSATP: phase2
MTs report the USAT packets and send CSAT
packets,
AP updates and reschedules its SAT.
Recovery procedure
Reservation procedure
CRP
Unused slot within an
amount of time
CRP: connection request packet,
DRP: disconnection request packet,
SAT: slot allocation table,
USATP: update SAT phase,
CSATP: confirmation SAT phase,
W: weight of the connection request.
Fig. 4. Reservation and recovery procedures in DSAP.
number of guaranteed slots. In addition, the PCR value determines
the number of the non-guaranteed slots. Similarly, CDV parameter
serves to distribute slots through the SAT.
For an ABR connection request, the MCR and PCR values determine
the numbers of guaranteed slots and non-guaranteed slots
respectively. Finally, a UBR connection request doesnt require QoS
guarantees. It uses any amount of remaining bandwidth from the
other service classes without exceeding the part provided by the
PCR.
In these last sections, we assumed that the available empty slots
are sufcient to satisfy the request requirements. In this case, the
connection is established and the requesting MT is informed for its
reserved slots. Otherwise, when there are not enough slots for the
request, two solutions can be considered (Fig. 4):
(1). The rst solution tries to assign additional slots from the non-
guaranteed slots pre-allocated to UBR, ABR and VBR connec-
tions in the given order. Based on a compensation algorithm,
the additional slots are selected, from specic service classes
with a computed number, in order to avoid QoS degradation in
the active connections. Eq. (1) computes the number of
additional slots using one or more terms (a, b, c) that satisfy
connection requirements. The compensation algorithm affects
the service classes with different weights used to calculate the
number of slots that could be liberated for the new connection
request. This algorithm takes into account the weight of the
service classes (active and requesting), the number of active
connections (i.e. MTs in the contention area) and the trafc
load. The weight depends on the service class and the type of
the connection request (new request or handoff request).
Handoff requests are affected by higher weights to reduce the
call reject probability, minimize the transition delay between
APs and guarantee the same QoS level.
Nb:additional:slots =
UBRclass
UBR:slots +
ABRclass
non:grnteed:slots +
VBRclass
non:grnteed:slots
0ACRMCR
a
MCRACRPCR
b
SCRACRPCR
c
1
(2). The second solution is initiated when the rst one does not
satisfy the QoS requirements of the connection request. This
solution is based on negotiation step which consists of offering,
to the MT, lower QoS than those required. The calculation
achieved in the rst solution can be used to generate the QoS
parameters and the trafc descriptors that the AP can offer. This
last then sends a signaling packet to the MT which can accept or
refuse these parameters. Finally, if the second solution does not
provide acceptable QoS level, the connection is rejected and a
release notication is sent to the MT (Fig. 4).
3.2.3. Recovery procedure
Our frame recovery procedure in the DSAP protocol improves the
efciency of the frame. This procedure starts when a slot in the frame
is released. This occurs when (1) a disconnection request is received
or (2) a slot remains idle for a certain amount of time. Then, the AP
checks its Slot Allocation Table (SAT) to identify the slots in the frame
which are unreserved and not agged.
The AP releases its unused slots in the data session. It updates its
SAT and then noties its MTs to update their SATs by following the
two phases described below using two types of control packets
(Fig. 4).
(1). Update SAT Phase (USATP): This is the rst phase in which an
AP releases the slots and sends the update SAT packets to notify
its MTs to update theirs SATs. These MTs are those providing
additional slots from their non-guaranteed slots during the
compensation phase (Section 3.2.2).
(2). Conrmation SAT Phase (CSATP): this is the second phase in
which the MT reports receiving the USATP packet sent in
phase 1. The AP receiving the CSATP packet must update and
reschedule its SAT.
4. Performance evaluation
In this section, we are interesting to improve DSAP performances
in order to justify its capability to manage the MAC level in wireless
ATM environment. We propose to study and analyze the evolution of
certain metrics which allow evaluating DSAP behavior. Different
scenarios are established to simulate the DSAP protocol using several
network events specied by the number of new and handoff calls, the
supported service classes and the trafc load. According the wireless
ATM service class, trafc parameters are xed as outlines Table 1.
4.1. Metrics
In our study, we have been interested in assessing the performance
in terms of the following metrics: total normalized throughput,
connection reject probability (new call, handoff call), resource
utilization average and overhead. In the following, we provide the
denitions of all metrics being considered. The term upstream trafc
refers to the trafc being sent from the MTs to the AP.
Total normalized throughput: is the ratio between the upstream
trafc having been effectively sent through the channel over all
trafc having been submitted by all types of sources [3]. This metric
can be simply dened as follows:
Throughput =

service classes
i

connections
j
traf
out i;j
service classes
i

connections
j
traf
in i;j
2
Call blocking probability: is the ratio between the rejected connec-
tions over the number of all connection requests per unit time. It
depends on two criteria: the DSAP behavior and the trafc load.
Resource utilization average (Rua): is the average channel utilization
of the assigned time slots at a particular [2]. Let:
CH i; j =
n
1 slot i is usedfor MTj
0 Otherwise
FL = theframelength
Then Rua =
MTs
j =1
slots
i =1
CH i; j
FL
3
4.2. Modeling and simulation environment
We assume the use of a wireless LAN consisting of several mobile
terminals and an access point connected to a wired node that serves as
sink for the ows from the wireless domain. The background ow
generated by the MTs is a combination of different service classes of
trafc with which many scenarios (light, heavy and burst trafcs) are
deployed as simulation inputs. The simulation parameters, listed in
Table 1
Simulation parameters.
Downlink frame Duration=2 ms, xed number =111 slots
CBR parameters PCR=100 kbps10 Mbps, CDV1 ms
VBR parameters SCR=100 kbps8 Mbps, PCR=300 kbps10 Mbps, CDV=1 ms.
ABR parameters MCR=100 kbps4 Mbps, PCR=300 kbps10 Mbps
UBR parameters PCR=100 kbps5 Mbps
Table 1, have been respected and MTs are modeled by components
responsible of generating various trafcs at up to 10 Mbps. In this
work, modeling the AP behavior constitutes the principal contribution
to dynamically allocating resources for optimal requirements satis-
faction. Therefore, the AP is modeled by several concurrent and
sequential processes allowing transmit, receive, signaling, reserva-
tion, recovery and scheduling packets.
In order to evaluate the DSAP protocol compared to TDMA, we
developed a simulator generating trace-les describing the evolution
of different metrics according the input trafcs and their features. This
simulator has been developed using the frame structure described in
Fig. 1 with a time duration of 2 ms. Our simulator combines three
sequential processes: (a) insertion of metrics, (b) trace-le loading
and (c) result visualization (Fig. 5).
(a). First, several QoS metric-related instructions are inserted at
specic processing levels in the algorithm pseudo-code. These
instructions allow measuring the evolution of the indicated
metrics associated to different times and inputs.
(b). Second, simulating the algorithm using different events and
network workloads leads to generate a trace le(s) loading
metric values evolution which reects the algorithm behavior.
(c). Third, the trace le is organized into tables to correlate between
metric evolutions and allows results visualization. Now,
performance evaluation of the proposed algorithms can be
done. But, the insufcient results leads to reformulate the
pseudo-code and go back to the process (b).
As simulation environment input, the algorithm pseudo-code
should be described in programming tool allowing concurrent
processes and time sensitive instructions which are a primary
attribute of hardware as offered by Hardware Description Language
(HDL). These features introduce exibility to describe variable and
real scenarios as well as initiating specic process when detecting
certain event.
The real-time packet-processing constraints imposed on APs to
support high line speeds motivate hardware based solutions, where
the AP functionality is implemented on application specic integrated
circuits (ASICs). The requirements for exibility and the complex
nature of many of the processing functions, on the other hand, favor
software based implementations on general-purpose processors. To
address these two conicting issues, recently a new class of devices
called network processors has emerged [17]. These are high
performance, programmable devices with special architectural fea-
tures that are optimized for packet processing. It is motivated by the
fact that the design and analysis of hardware-software architectures
for such processors requires new models and methods, which do not
fall under the domain of traditional embedded-systems design. But,
The co-design ow is addressed only when performance evaluation
proves the effectiveness and efciency of the proposed protocol.
- - - - - - -
- - - - - - -
store (ti, throughput, ) in trace file
- - - - - - -
- - - - - - -
store (tj, BW, ) in trace file
Process 1
Process 2
a) insertion of metric for measurement
t
0
=0s, throughput= x,
t
1
=100s, used BW=y,
t
2
=300s, blocked call Nb=z,
t
i
=10ms, throughput=m,
t
n
=5s,
b) measure and load metric in trace file
Algorithm pseudo-code
Trace file structure
Time 0 100 300
throu
Time
BW
Input% 10 20
throu
c) result organization & visualization
Call blocking
prob
Fig. 5. Simulation environment processes.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.2 0.4 0.6 0.8 1
random access
power shaping
C-MMF
DSAP
Pure-TDMA
T
h
r
o
u
g
h
p
u
t
Load
Fig. 6. Throughput vs. Load: comparison between the proposed algorithm, the pure-
TDMA, the power shaping, the centralized and the random allocation algorithms.
4.3. DSAP evaluation
The most works for resource allocation focus on one point of view
by evaluating the impact of their approaches on a single QoS metric.
However, this work includes several factors and processes allowing
simultaneously improving resource utilization rate, throughput, call
blocking probability (new call, handoff) and packet delay variation.
As described in Section 2.4, [6] presented three allocation
algorithms with power shaping, CMMF and random allocation. As
rst performance evaluation result, Fig. 6 compares, in terms of
network throughput, the different algorithms for increasing loads. The
highest curve refers to the Dynamic Slot Allocation (DSAP) algorithm,
whereas the bottom one refers to the random algorithm. As the load
increases when the MTs have to go through a contention mechanism
to place their requests, the basic allocation algorithms cannot serve
these requests with the desired requirements especially when there
are not enough timeslots. This increases dramatically the collision and
call blocking probabilities as the number of MTs increases, which in
turn leads to decrease the throughput. In the same situation, the
compensation mechanism in DSAP consists of liberating new time
slots to serve additional requests and meet these problems.
Using the same simulation scenario, DSAP performs also to increase
the resource utilization average as shown in Fig. 7. The total resources
utilization is more important for DSAP compared to TDMA for two
raisons. First, the compensation algorithm employs non-used and non-
guaranteed slots which may decrease the call blocking probability and
then increases the number of allocated slots (i.e. Rua). Second,
disconnections are followed by rewarding released slots for MT
transmitting with lower QoS than those requested in signaling phase.
In summary, DSAP includes compensation and rewarding mechanisms
to improve the assigned slots average when a connection request and
connection release events occur, respectively. Consequently, DSAP uses
more than 25% of resources compared to TDMA as shown in Fig. 7.
Furthermore, the DSAP affects the call blocking probability since it
includes algorithm capable to serve new connection requests even in
overloaded state. For the same trafc scenario, Fig. 8 compares the
reject probability evolution between classical TDMA and DSAP.
Within the rst period (until 3 second), the DSAP reject probability
is very lower than TDMA. This is due to the availability of resources
that can be compensated between requests. Moreover, the rest of the
simulation scenario shows that the average reject probability is still
lower for DSAP than TDMA.
DSAP is also designed to improve handoff performances allowing
to maintain a transparent transition between APs. DSAP gives more
priority for handoff requests processing since it is usually preferred
maintaining an established communication than accepting new call
[10]. As a result, the handoff reject probability is lower for handoff
calls while the scenario includes the same number of handoff and new
calls (Fig. 9).
DSAP protocol is designed to process CBR applications with more
priority during resource allocation or compensation. This leads to
reduce the blocking probability for new CBR calls especially when the
trafc load is low. As result, the CBR call blocking probability is under
0.2 when the supported CBR trafc is lower than 50% because there
are usually enough resources to be allocated or compensated for
new CBR calls. However, the CBR call blocking probability brutally
increases when the CBR trafc load exceeds 50%. In this case, there
arent enough non-guaranteed time slots that can be used as resources
for the new CBR calls (Fig. 10).
The cell delay variation (CDV) is a sensitive QoS parameter which
depends on the type of service class. CDV should take on almost
constant value especially for CBR applications. DSAP protocol provides
0
10
20
30
40
50
60
70
80
90
1 2 3 4 5 6 7 8
R
u
a

(
%
)
time (s)
TDMA
DSAP
Fig. 7. DSAP resource utilization average compared to TDMA behavior.
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0 1 2 3 4 5 6 7 8 9
C
a
l
l

b
l
o
c
k
i
n
g

p
r
o
b
a
b
i
l
i
t
y
time (s)
TDMA
DSAP
Fig. 8. Call blocking probability evolution vs. a typical scenario for TDMA and DSAP.
0
0.1
0.06 1.31 2.58 3.83 5.08 6.32 7.58
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
H
a
n
d
o
f
f

c
a
l
l

b
l
o
c
k
i
n
g

p
r
o
b
a
b
i
l
i
t
y
time (s)
total connections
Handoff connections
Fig. 9. Handoff and new calls blocking probabilities under light the same scenario.
a reduced CDV for CBR connections compared to classical TDMA
technique since it includes a suitable model for time slot repartition
and update based on the slot allocation table (SAT). Fig. 11 outlines
the CDV evolution for three CBR connections maintained with
different PCRs ((a)- 1.5 Mbps, (b)- 1 Mbps and (c)- 0.5 Mbps). The
CDV uctuates around a constant value according to the PCR and equal
to (0.51 s, 0.76 s and 1.54 s) given by Fig. 11-(a), (b) and (c)
respectively. DSAP protocol minimizes CDV uctuations which im-
proves QoS to be offered for mobile terminals and meets application
delayconstraints. It gives a sufcient resources distributionstrategy that
arranges fairly all the connections.
5. Conclusions
In this work, we have proposed a QoS-aware mechanism and
evaluated its performance in terms of various metrics of interest. This
mechanism combines signaling, reservation and recovery processes
aiming to provide the QoS guarantees required by applications when
coexisting with other services in TDMA/FDD mobile networks. This
mechanismutilizes the bandwidth resources in an efcient way in the
presence of four different types of services. Based on the simulation
results, we show the benets of designing a simple and effective
bandwidth allocation mechanism.
Compared to other TDM mechanisms, the proposed DSAP
improves channel throughput, achieves lower call blocking probabil-
ity and stabilizes cell delay variation. In particular, DSAP improves
resource utilization by more than 25%.
Acknowledgements
This work is supported by the Research Center of the College of
Computer and Information Sciences - King Saud University. The
author thanks the reviewers for their helpful comments. Their
remarks greatly improved the presentation of the paper.
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
10 20 30 40 50 60 70 80 90
C
B
R

c
a
l
l

b
l
o
c
k
i
n
g

p
r
o
b
a
b
i
l
i
t
y
CBR load
Fig. 10. CBR call blocking probability function in CBR load.
0.44
0.46
0.48
0.5
0.52
0.54
0.56
0
0
.
1
0
.
2
0
.
2
9
0
.
3
9
0
.
4
9
0
.
5
8
0
.
6
8
0
.
7
8
0
.
8
7
0
.
9
7
1
.
0
7
1
.
1
7
1
.
2
6
1
.
3
6
1
.
4
6
C
D
V

(
s
)
time (ms)
CBR(1,5Mbps)
0.69
0.71
0.73
0.75
0.77
0.79
0.81
1
.
3
6
1
.
7
2
2
.
0
9
2
.
4
5
2
.
8
1
3
.
1
8
3
.
5
4
3
.
9
4
.
2
7
4
.
6
3
C
D
V

(
s
)
time (ms)
CBR (1 Mbps)
1.47
1.49
1.51
1.53
1.55
1.57
1.59
1.61
0
0
.
4
7
0
.
9
3
1
.
3
9
1
.
8
5
2
.
3
2
2
.
7
8
3
.
2
4
3
.
7
4
.
1
7
C
D
V

(
s
)
time (ms)
CBR(500kbps)
a
b c
Fig. 11. Cell delay variation for different CBR trafcs.
References
[1] C. Ceken, I. Erturk, C. Bayilmis, A new MAC protocol design based on TDMA/FDD
for QoS support in WATM networks, Computer Standards & Interfaces 28 (2006)
451466.
[2] Chien-Min Wu., Dynamic frame length channel assignment in wireless multihop
ad hoc networks, Computer Communications 30 (2007) 38323840.
[3] Francisco M. Delicado, Pedro Cuenca, Luis Orozco-Barbosa, QoS mechanisms for
multimedia communications over TDMA/TDD WLANs, Computer Communica-
tions 29 (2006) 27212735.
[4] Broadband Radio Access Networks (BRAN), High Performance Radio Local Area
Networks (HIPERLAN) Type 2; System Overview, ETSI Std. TS 101 683, , 2000.
[5] IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface
for Fixed Broadband Wireless Access Systems Amendment 1: Detail System
Proles for 1066 GHz, IEEE Std. 802.16c-2002, December 2002.
[6] R. Veronesi, V. Tralli, J. Zander, M. Zorzi, Distributeddynamic resource allocationwith
power shaping for multicell SDMA packet access networks, the wireless commu-
nications and networking congference (WCNC), March 2004, Atlanta, 2004.
[7] P. Koutsakis, M. Va.adis, H. Papadakis, Prediction-based resource allocation for
multimedia trafc over high-speed wireless networks, AEU International Journal
of Electronics and Communications 60 (2006) 681689.
[8] M. Karol, Z. Liu, K. Eng, An efcient demand-assignment multiple access protocol
for wireless packet (ATM) networks, Wireless Networks 1 (1995) 267279.
[9] N. Passas, S. Paskalis, D. Vali, L. Merakos, Quality-of-service oriented medium
access control for wireless ATM networks, IEEE Communications Magazine 35
(11) (1997) 4250.
[10] Bih-Hwang Lee, Hsin-Pei Chen, Su-Shun Huang, Dynamic Resource Allocation for
Handoff in WATM Networks, 11th International Conference on Parallel and
Distributed Systems (ICPADS05) IEEE, 2005.
[11] Polychronis Koutsakis, On improving channel throughput by restricting data
trafc access in multimedia integration over wireless channels, AEU International
Journal of Electronics and Communications 62 (2008) 110.
[12] Mohan R. Akella, Rajan Batta, Moises Sudit, Peter Rogerson, Alan Blatt, Cellular
network conguration with co-channel and adjacent-channel interference
constraints, Computers and Operations Research 35 (2008) 37383757.
[13] D. Raychaudhuri, L.J. French, R.J. Siracusa, S.K. Biswas, Y. Ruixi, P. Narasimhan, C.A.
Johnston, WATMnet: a prototype wireless ATM system for multimedia personal
communication, IEEE Journal on Selected Areas in Communications 15 (1) (1997)
8395.
[14] K.N. Chang, S. Kim, Channel allocation in cellular radio networks, Computers and
Operations Research 24 (1997) 849860.
[15] C.W. Sung, W.S. Wong, Sequential packing algorithm for channel assignment
under co-channel and adjacent channel interference constraint, IEEE Transactions
on Vehicular Technology 46 (1997) 676685.
[16] R. Veronesi, V. Tralli, DCA with power-shaping (PS-DCA) in TDMA and TD/CDMA
cellular systems with centralized and distributed control, The 14th IEEE
International Symposium on Personal, Indoor and Mobile Radio Communications
(PIMRC), 710 Sept, 2003.
[17] S. Chakraborty, System-Level Timing Analysis and Scheduling for Embedded Packet
Processors, thesis of the Swiss federal institute of technology, Zurich, April 2003.
Ridha Ouni. is Assistant professor in the college of Computer and Information Sciences
of the King Saud University, KSA, since Feb 2009. He was Assistant Professor in the
Institut Prparatoire aux tudes d'Ingnieurs de Monastir (IPEIM), Tunisia, since 1999.
Currently, he prepares his HDr in the wireless networking eld. His research interests
include wireless communication, mobility, QoS management and wireless sensor
networks.Previously, he Received his MSc in Physic Micro-electronic, his DEA degree in
Matriaux et Dispositif pour l'lectronique and his PhD from the Sciences Faculty of
Monastir, Tunisia in 1995, 1997 and 2002, respectively.
Sci.Int.(Lahore),24 (1),13-19,2012 ISSN 1013-5316; CODEN: SINTE 8

13
EFFICIENT DATA HARVESTING FOR INELASTIC TRAFFIC IN
VEHICULAR SENSOR NETWORKS
Ridha Ouni, Mohammad Zuheir Hourani
Department of Computer Engineering, CCIS, King Saud University, P.O.Box 51178, Riyadh 11543,
Saudi Arabia
ABSTRACTThe basic idea behind intelligent transportation system (ITS) is how to deploy vehicular sensor
network that have many characteristic such as high computation power ,enough storage space and mobile sensor
node in order to design an effective and efficient architecture for data collection and data exchange. In this paper
we will introduce an intelligent transportation system with new network paradigm to collect important information
from the road environment based on the vehicular sensor network (VSN). Data aggregate provides the end users
with valuable information in order to make the road safer and less congested. Our system framework consists of
active vehicular sensor node, passive vehicular sensor node and sink node distributed according to the road
segmentation for collecting data from active vehicular sensor node passing by, while active vehicular sensor node
collects data from passive vehicular sensor node in their segment using multihop data harvesting. .Finally, the
simulation shows the effectiveness of the proposed schema.

Keyword- ITS, VSN, IVS, data harvesting, hybrid architecture, data aggregation.

1. INTRODUCTION
Significant advances in manufacturing technology
equipment and the advent of Micro-Electro-Mechanical
Switches (MEMS) has opened the way for the construction
of intelligent sensor nodes which are able to perform three
major functions: sensing, processing and wireless
communication. These wireless sensor nodes are
characterized by their intelligence, their small size, low cost,
battery powered, and easy to install and repair. These
features open doors to deploy WSNs in the future for a wide
range of applications because it greatly expands our ability
to monitor and control the physical environment from
remote locations [1].
An interesting field where the use of WSNs proves
effectiveness is the field of Intelligent Vehicular Systems.
An Intelligent Vehicular System (IVS) uses technological
advances in computers and information technology to
improve the efficiency of both new and existing vehicular
systems.
Vehicular sensor networks (VSNs) is a technology where
sensors are deployed in the road side and in the vehicles to
sense various urban phenomenons and transmit information
for vehicular traffic control and monitoring. VSNs have
different characteristic from traditional sensor network
(static network), interns of mobility, computational, power
supply, memory storage and reliability. Moreover vehicular
sensor network VSN has a much more dynamic topology as
compared to the static WSN. It is often assumed that VSN
will move continuously in a random fashion, thus making
the whole network a very dynamic topology. This dynamic
nature of VSN is reflected in the choice of other
characteristic properties, such as routing, MAC level
protocols and physical hardware, beside this, dynamic
topology of vehicular sensor network VSN, communication
links can often become unreliable[2].The previous
characteristics allow deploying vehicular wireless sensor
network to design intelligent transportation system.
Compared to traditional static sensor network vehicular
sensor network may generate a sheer amount of data due to
the mobility of the vehicles, its impossible to deliver all the
data collected to the sink. First because just too much is
detected by such powerful sensor platform, second because
the network capacity is too thin and limited time it takes for
vehicle to drive through the coverage area of the sink
because of the mobility. In order to solve most of these
problems we introduce active and passive node solution
beside simple data representation that will reduce the
amount of data transferred between nodes.
In this paper we are interested in designing optimal system
architecture for such vehicular sensor network for vehicular
traffic control and monitoring. Several assumptions have
been made. First, we assume that vehicles communicate
through a wireless interface, implementing a CSMA/CA
MAC layer protocol that provides a RTS/CTS/DATA/ACK
handshake sequence for each transmission. Vehicular sensor
network adopt IEEE 802.11 as a cost efficient and widely
deployed solution for network communication. IEEE
802.11b is a draft amendment to IEEE 802.11 standard to
add wireless access in vehicular environment. It supports
data exchange for vehicle to vehicle (V2V) and vehicle to
infrastructure (V2I) in the licensed ITS band. Performance
evaluation of IEEE 802.11b is done in [3], [4], [5].
The number of sink nodes that are distributed beside the
road is very small in proportion to the number of vehicular
sensor nodes. So, we assume that sink node has a relatively
fast processor and a large storage device and has enough
energy resources. In addition, it has very large data base to
store information from the vehicles. However, vehicular
sensor node has lower storage space and low processor
capability. It is assumed that each vehicle has unique ID to
identify the vehicular node and included in each message
sent.
Finally, we assume devices participating in vehicular
networks are highly mobile with a speed up to 180km/h.But,
their mobility patterns are predictable due to the constrained
movement imposed by the road system and constrained
speed imposed by speed limits, traffic conditions and
signals. In fact, the mobility of vehicular sensors poses
challenges to the communication system. Mobility
undermines the reliability of communication and also causes
the topology to continuously change.
ISSN 1013-5316; CODEN: SINTE 8 Sci.Int.(Lahore),24(1),13-19,2012

14
This paper focuses on how to deploy WSN as an intelligent
transportation system over effective and efficient
architecture for data collection and data exchange allowing
vehicle traffic monitoring and control. The rest of the paper
is organized as follow. Section II provides a back ground
and related work in vehicular sensor networks. Section V
introduces the proposed scheme for data harvesting and
dissemination. Section IV evaluates the simulation results
conducted with C++. The final section concludes the paper.
RELATED WORK
Recently, there is a strong interest from researchers in
deploying WSNs in VSNs in many applications that
involve constraints related to the traffic conditions such as
traffic monitoring and control, traffic estimation and
monitoring parking. Some research focus in moving
vehicles to enable wireless sensor communication between
roadside and vehicle or between vehicles. These
applications aim to make roads safer and less congested in
order to save the time for people. Its important to note that
these applications encounter three types of communications
[6]:
Vehicle-to-Vehicle (V2V) communication: vehicles
are equipped with sensors in order to exchange
information that is crucial to avoid severe situations
like traffic jam avoidance.
Vehicle-to-Infrastructure (V2I) communications:
information flow from vehicles to sensors installed on
roadway infrastructure
Hybrid communication: uses both V2V and V2I
architecture
In [7] the author proposed a scheme based on the hybrid
communication. Vehicles will send all their sensed data to
infostations, where the data will be forwarded to
corresponding infostation based on the infostations
management area. Later, any vehicle requesting sensed data
can request these infostation, which is more of an indirect
form of vehicle to vehicle communication using relay nodes
forming another type of data harvesting protocol. However,
this technique requires installation of an infostation
infrastructure, which can be very costly and complex. An
effective traffic monitoring system is studied in [8] which is
capable of detecting, classifying and determining the
direction of travel of vehicles on a two-lane road.
One of the biggest issues in realizing VSN is concerned
with data harvesting which is a technique where sensors
create data that summarize the characteristic of the data and
send it to the target. In [9], the author proposed a novel
multi-hop data harvesting (MDH) method for the V2I
architecture. MDH have two scheme proposed for VSN.
The MDH scheme using replicas (MDH-R) is proposed for
requesting data from single sensor node, while data
aggregation scheme is designed (MDH-RA) for cases when
the request was made to a geo cast region. Many
applications in VSN may require multi-hop data
transmission to meet real-time constraints. The author see
multi-hop data dissemination capabilities may become ideal
for future researches in this area.
In some situations it becomes obvious that flooding a
message into a whole network is not appropriate and will
cause a high level of congestion. At the same time
request/reply schemes might not be appropriate, e.g. since
the information is needed by many nodes at the same time. It
has therefore been proposed to replace the unstructured
flooding of packets by some sort of hierarchical distribution
using clustering.
Clustering come out as new brand of vehicular networks,
whose propose is the real time gathering and diffusion of
information. In [6] the author used a Clustering Gathering
Protocol (CGP) that is across layered protocol based on
hierarchal and geographical data gathering, aggregation and
dissemination. The goal of CGP is to gather from all node in
the vehicular ad hoc network in order to offer different kind
ITS services, it allows telecommunication/service providers
to get valuable information about the road environment in a
specific geographical area, using V2V network to minimize
the high cost links usability and base station to gather
information from the vehicles
Another cluster-based approach is presented in [10]. Here,
cars within a small region autonomously form a cluster.
Owing to the close vicinity, direct communication between
those cars is possible. However, communication with other
vehicles or clusters has to be performed by relaying vehicles
that are part of several clusters.
2. OVERVIEW
Our system framework is consisting of static road side node
(sink node), and mobile vehicular sensors. Road side nodes
are distributed according to the road segmentation for
collecting data from mobile vehicular sensors passing by
and to exchange data about traffic condition. While mobile
sensors on vehicles monitor the road condition and send
this information to active mobile neighbors when they are
close enough then to the road side sink node (see Figure 1).
We focus on vehicular mobility, collaboration between
mobile and static nodes, and information exchange among
mobile vehicles. Mobile vehicles can gather latest
information spreading on the map out of the reach of static
node, whereas static node can gather information from
more active vehicles coming across, where the connectivity
between static and mobile nodes and also between mobile
and mobile nodes are most likely meaningful and useful.

Figure 1: Vehicular Sensor Architecture
3. ROAD SEGMENTATION
The roads are divided into small segments. On each road
segment, there are two road side node (sink node) located at
the both ends of the segment, as shown in Figure 2.
Usually, the road side nodes are placed on the roadside with
different distances based on the road environment
to collect data from active vehicles passing by. So, drivers

15
can get the road condition before entering this segment;
while vehicular mobile sensors, assisted by themobility of
the vehicles, can know the road information along their
own path.
The road is divided into virtual segments with the different
length (Figure 2). In each segment an active node is selected
to gather data from all nodes in its segment, aggregate them,
and send the result to the next sink node.
4. PROPOSED SCHEME
The proposed scheme consists of providing a feasible,
efficient and robust vehicular sensor network framework to
monitor road traffic and provide desired and reliable
information for users, particularly for drivers in
automobiles. In the context we decided to use active node
based solution for the V2V dissemination. The scheme will
be divided into three parts: Active vehicular sensor node
selection phase, data harvesting/dissemination phase, and
the data sharing phase.

Figure 2. Road segmentation

Figure 3. Active node selection flowchart
ACTIVE VEHICULAR SENSOR NODE SELECTION
PHASE
Every road segment has two such sink near the ends.Every
vehicle enter the segment will send hello message to the sink
node at the beginning of each segment containing the vehicle
ID and its speed.Then, the sink node will store this
information in the data base.Using this information, the sink
node will create an active node based on two parameters as
threshold; the maximum number of vehicles (passive nodes)
detected in the road, and the elapsed time.
First, the sink node stores the data about each vehicle
entering the segment until reaching the maximum number of
vehicles. Second, the sink node will choose one of these cars
randomly to be an active node by broadcasting control
information as a request including the ID of the vehicle and
its velocity.When collecting this request, the other vehicle
nodes (passive nodes) identify the target node dedicated for
forwarding their information. All other nodes must know the
active node in their segment. To do so, the AVS will include
its ID in the packet as new destination and then diffuse reply
to the sink node which will be also received and processed
by its neighbored vehicles.The mobility of vehicular sensor
network can affect the topology of the network.Therefore,
we also use the elapsed delay to control when the sink sends
request to create an active node exactly before the group of
vehicle leave the wireless range of the sink.This time will be
calculatedusing theequation below.

Where is the distance that our wireless communication can
support (IEEE 802.11), and reflects the mobility of the
vehicles which is the velocity of the vehicular sensor
node.As a result, we need two counters in the sink node one
for time and the other for the number of vehicles. So, in this
way we guarantee that we create an active node for each
group of vehicles, figure 3 and the following pseudo-
codeprovides more details in explaining the algorithm.
1. while (EOS)
2. ENSURE VEE
3. if (N = TL
V
) then
4. CREATE an active node
5. elseif (T = TL
T
) then
6. CREATE an active node
7. else goto 5
8. end if
9. end while
Pseudo code notations:
EOS: End of simulation.
VEE: Vehicle wants to enter/exit the current base stations
wireless range.
TL
V
: Threshold level to create a new active node.
N: Admitted no. of Vehicles for the current base station.
TL
T
: Threshold level in terms of average time of the vehicles
to leave the range of the current base station.
T: Current Time.


16

Figure 4. Active Node Selection
Segment-based active node solutions provide less
propagation delay and high delivery ratio with also
bandwidth fairness. In [6] the authors use a distributed
clustering algorithm to create a virtual backbone that allows
only some nodes to broadcast messages and thus, to reduce
significantly broadcast storms.
When the active node receives the data from the passive
node originally holding the data, it will process it in store &
forward fashion instead of sending directly to the sink node
when the next sink node is far. Similarly, the passive node
keeps the data in its memory during a parametric time, and
waits for active node to be closest enough from it. An
example is given in figure 5, where the node (F) cannot
reach the active node. In this case, it will store its data till it
reaches the active node or wait another active node.
All nodes in the segment, transmits in unicast their sensed
data to the active node using a mechanism similar to DCF
(Distributed Coordination Function). Themechanism
consists of:
First, each node wait a random bounded back-off time,
Then at the end of the back-off time, the node sends a
Request To Send (RTS) to the active node,
Next, the active node acknowledges the reception by
sending a Clear To Send (CTS) message,
Finally, the node sends its data to the active node.
The road architecture and main concept of this mechanism
are shown in figure 5.

Figure 5. Data Harvesting Model.
Passive vehicular sensor nodes will monitor the traffic
condition by measuring the speed of the vehicle to send it to
the active node in its segment. The active vehicular sensor
node will store the data message and count the number of
vehicles in its segment, supporting limited number of
vehicles. Related to the event representation the sink node
( ) compare the number of vehicles and their average
velocity that reside in its segment, in order to take decision
about congestion. This data will be forwarded to sink node
( ) that is located at the end of this segment. In turn, the
sink node will forward this data to the preceding sink
node to update its data about traffic condition in this
segment and floods it to the new coming vehicles that wish
to enter this segment.
When the vehicle is going to enter a new road segment, the
sink node at the near end will communicate with this
vehicle. So, the vehicle can know the road condition of this
segment in advance. There is no need to place more sink
node in the middle of one segment, because even if the
vehicle get information at the middle of a segment, drivers
still cannot change their direction or change the route trip.
DATA SHARING PHASE
In some areas, data traffic may increase dramatically due to
many vehicles requesting for data at the same time. In this
case, there is a high probability that more than one vehicle is
requested to be an active vehicular node from the sink node
( ). When the active node reach its limit from the passive
node due to the congestion that involve many vehicle in the
segment, the sink node sends request to create new active
node to provides fairness. This is very important in a sensor
network where every node has to send its data. It also
reduces significantly broadcast storms and thus avoids
collisions. Moreover, case that we can have new active node
when there are two separate groups of cars where one of
them reaches the end of wireless communication range of
sink( ) but the maximum number of passive node still not
complete. In this case, the sink node will request new active
node from the coming group based on time factor in order to
ensure that each group have active node to send information
to it. It can be seen from figure 6, when there are many
vehicles having data to be sent about the traffic condition
which means congestion occurred on a specific segment. In
this case, the active node sends a message to the sink node
( ) in order to create new active node.

Figure 6. Sharing Phase.
5. SIMULATION
The simulation was carried out in C++ for 1000 seconds.
Depending on the number of base stations and their relative
ranges the road site is segmented. IEEE 802.11b is
considered to operate with CSMA-CA with minimum back-

17
off exponent set to 3 and having 4 as maximum number of
back-off with channel sensed every 0.1 seconds and
operating in 2.4GHz frequency, ACK waiting duration is set
to 0.05 seconds for the participating active and passive
nodes, and having a channel bandwidth of 20MHz. The
maximum data rate is 11 Mbit/s and the modulation is
achieved using DSSS with only 1 MIMO stream into
consideration. The threshold level for the admitted vehicles
that will create an active node is varied with 5, 10 and 15
numbers to study the effect of congestion based on the
number of vehicles which is random. The road site can have
a maximum capacity of 100 or 200 vehicles for the above
mentioned threshold levels and the speed of the vehicles
have a range from 0-180km/hr.
First of all, the number of active agents is of utmost
importance to study the performance of the system. In figure
7, there is a clear difference between the total number of
vehicle that enter the segment and the subsequent active
agents. This reflects that the number of messages between
the vehicles and infrastructure would be reduced to 90%.
The total number of active agents/nodes during a given time
interval is calculated as number of total vehicle in that
interval over threshold plus the number of active nodes that
left the boundary of each segment and wait for other
segments to determine their active and passive nature.
Figure 8 illustrates the number of messages transferred from
passive to active vehicles on the whole system when the
average speed is varying for 100 vehicles. It is observed that
when all the vehicles are present within the system, the
number of messages is quite stable. It can be concluded that
active nodes solution is convenient for different kind of
mobility model such as high speed (low density) or low
speed (high density) and also that nodes velocity doesnt
affects the active nodes performance. Indeed, the number of
messages doesnt increase significantly with rising speed.

0 100 200 300 400 500 600 700 800 900 1000
0
20
40
60
80
100
120
140
160
Time(sec)
N
u
m
b
e
r

o
f

V
e
h
i
c
l
e
s
Number of vehicles vs Time

Total Number of Vehicles
Number of Active Nodes for N
th
=10
th
=5
th
=4

Figure 7. Number of Vehicles vs Time

0 20 40 60 80 100 120 140 160 180
4000
4500
5000
5500
6000
6500
7000
7500
8000
Velocity (km/h)
N
u
m
b
e
r

o
f

P
a
c
k
e
t
s

f
r
o
m

P
a
s
s
i
v
e

t
o

A
c
t
i
v
e
Number of Packets vs Velocity for 100 vehicle

Figure 8. Number of packets from passive to active nodes
In figure 9, when the system is procured with 10
communication channels and the velocity of the vehicles are
kept in this range (20160km/hr), the number of lost packets
will abruptly increase from 18 62 ( approx. 3.5 times).
This increase is due to the high mobility of the vehicles that
affect the access channel. But when the communication
channel is increased to 15, the vehicles have high probability
to access the channel and the number of lost packets
decreases to 40 which is approx. 1.5 times reduced as
compared to 62.
Comparison of figure 9 and figure 10 reveals the fact that
there is an influence of mobility on packet loss. As
highlighted above in figure 9, packet loss occurs from even
at low velocity due to the mobility difference between the
active nodes and the base station. However, usually there
will be a small difference of velocity between active and
passive nodes. As a result, the packet loss is negligible.
Figure 11 show how the blocking probability evolves
between active and sink station for a rising velocity of
vehicles. When the vehicles begin to enter a segment, the
average system velocity is low and the system is in a
congested state. In this scenario, more nodes will try to
communicate with the base station resulting in a higher
blocking probability.
As the average system velocity increases the blocking
probability decreases due to the fact that the vehicles are far
apart. Finally, the blocking probability starts rising again
after 55km/hr which is the optimal speed to access the
channel. For an increasing velocity the system goes through
the back-off due to packet loss and resulting in a rising
blocking probability.
When the distance margin to access the base station for any
active node increases the blocking probability will be
affected as illustrated in figure 12. Because with the
increased distance the node will take much time for data
transmission and even for back-off. The channel
communication factor also contributes towards this effect on
the blocking probability. Higher communication channel
leads to a lower the blocking probability.


18
0 20 40 60 80 100 120 140 160 180
0
20
40
60
80
100
120
140
160
180
200
Velocity (km/h)
N
u
m
b
e
r

o
f

L
o
s
t

P
a
c
k
e
t
s

f
r
o
m

A
c
t
i
v
e

t
o

B
a
s
e
s
t
a
t
i
o
n
Number of Lost Packets vs Velocity

N
ch
= 15
N
ch
= 5
N
ch
= 10

Figure 9. Number of Lost Packets from Active to Base Station
0 20 40 60 80 100 120 140 160 180
0
200
400
600
800
1000
1200
1400
Velocity (km/h)
N
u
m
b
e
r

o
f

L
o
s
t

P
a
c
k
e
t
s

f
r
o
m

P
a
s
s
i
v
e

t
o

A
c
t
i
v
e
Number of Lost Packets vs Velocity

N
ch
= 5
N
ch
= 10
N
ch
= 20

Figure 10. Number of Packets Lost from Passive to Active
CONCLUSION
In this paper, a scheme for data harvesting and data
exchange based on active vehicular sensor nodes for real
time traffic flow is proposed. We provide an effective and
collaborative hybrid method to deliver important
information to particular end users. We use road side sink
and vehicular sensor nodes to restore and exchange data, and
then we study our novel scheme to prove that the mobility of
the nodes affects the data transmission and also the blocking
probability to access the channel. An optimal system
velocity has been defined in order to have a stable system.
Finally, the active node solution is used to reduce the
broadcast storm and also the number of vehicles that
communicate with the base station in the form of active
nodes.
0 20 40 60 80 100 120 140 160 180
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
velocity (km/h)
B
l
o
c
k
i
n
g

P
r
o
b
a
b
i
l
i
t
y
Blocking Probability vs Velocity (A->BS)

N
ch
= 7
N
ch
= 11
N
ch
= 15

Figure 11. Blocking Probability
10 20 30 40 50 60 70
0
0.01
0.02
0.03
0.04
0.05
0.06
Margin
B
l
o
c
k
i
n
g

P
r
o
b
a
b
i
l
i
t
y
margin choosing effect

N
ch
=11
N
ch
=14
N
ch
=17
N
ch
=20

Figure 12. Blocking Probability in terms of margin
ACKNOWLEDGMENT
The authors would like to thank the Research Center of the
College of Computer and Information Sciences - King Saud
University, for its support in funding this work.
REFERENCES
[1]. J. Yick, B. Mukherjee, D. Ghosal, Wireless sensor network
survey, Computer Networks, Science Direct, April 2008.
[2]. Munir, S.A.; Biao Ren; Weiwei Jiao; Bin Wang;
DongliangXie; Man Ma , Mobile Wireless Sensor Network:
Architecture and Enabling Technologies for Ubiquitous
Computing, Advanced, pp:113 120, August 2007.
[3]. Vladimir Bychkovsky, Bret Hull, Allen Miu, Hari
Balakrishnan, and Samuel Madden. A measurement study of
vehicular internet access using in situ Wi-Fi networks. In
MobiCom 06:Proceedings of the 12
th
annual international
conference on Mobile computing and networking, pages 50
61, New York, NY, USA, 2006.
[4]. David Hadaller, Srinivasan Keshav, Tim Brecht, and
Shubham Agarwal. Vehicular opportunistic communication
under the microscope. In MobiSys 07: Proceedings of the

19
5th international conference on Mobile systems, applications
and services, pages 206219, New York, NY, USA, 2007.
[5]. J. Ott and D. Kutscher. Drive-thru internet: IEEE 802.11b
for automobile users. In 23rd Annual IEEE Conference on
Computer Communications (INFOCOM), 2004.
[6]. I. Salhi, M. O. Cherif, S. M. Senouci, A New Architecture
for Data Collection in Vehicular Networks, IEEE
International Conference onCommunications ICC'09, pp.16,
August 2009.
[7]. U. Lee, E. Magistretti, B. Zhou, M. Gerla, P. Bellavista, A.
Corradi. Efficient data harvesting in mobile sensor
platforms, Proceedings of the 4th annual IEEE International
Conference on Pervasive Computing and Communications
Workshops PERCOMW06, pp. 352, 2006.
[8]. H. Ng, S.L. Tan, J. G. Guzman, Road traffic monitoring
using a wireless vehicle sensor network, International
Symposium on Intelligent Signal Processing and
Communications Systems, ISPACS08. pp. 14, Feb. 2009.
[9]. K.W. Lim Y.-B. Ko, Multi-hop data harvesting in vehicular
sensor networks, Communications, IET, Vol.4, no7,
pp.768775, April 2010.
[10]. Chennikara-Varghese J, Chen W, Hikita T and Onishi
R,Local peer groups andvehicle-to-infrastructure
communications. GLOBECOM07: Proceedings of the
IEEE GlobalTelecommunications Conference Workshops,
pp. 16, 2007.
Wireless propagation channel modeling for optimized Handoff
algorithms in wireless LANs
q
Monji Zaidi
a,
, Ridha Ouni
b
, Rached Tourki
a
a
Electronic and Micro-Electronic Laboratory (ElE, IT-06) FSM, University of Monastir, Monastir 5000, Tunisia
b
College of Computer and Information Sciences, King Saud University, Saudi Arabia
a r t i c l e i n f o
Article history:
Received 9 November 2009
Received in revised form 5 September 2011
Accepted 12 September 2011
Available online 24 October 2011
a b s t r a c t
In this paper, we present a time-series analysis technique which covers the basic concepts
and mechanisms driving the wireless propagation channel. We also use a generated series
for simulation study of Handoff performance showing the impact of multipath phenomena.
Moreover, the extraction of the average signal has been used to reduce signicantly the
number of unnecessary Handoffs.
The wireless propagation channel modeling is based on the linear model concept of the
received power from the access point (AP). This concept has a crucial role in modeling new
decentralized Handoff based on the ratio of expected and current signal slopes already pro-
vided by the linear model. Hence, a fuzzy-based solution is developed and a comparison
with the analytical solution results is established. Until recently, Handoff mechanisms
are implemented entirely in software, which increasingly becoming infeasible. Therefore,
this work attempt to follow the top-down co-design approach providing hardware proto-
type which leads to reduce the power consumption and support high processing speed.
2011 Elsevier Ltd. All rights reserved.
1. Introduction
The man-made structures [1] such as buildings or small houses in suburban areas, with sizes ranging froma few meters to
tens of meters, dramaticallyinuencethe wireless propagationchannel. Inurbanareas, the size of structures canbe evenlarger.
Likewise, in rural and suburban environments may reach similar dimensions. These features are similar or greater in size than
the transmitted wavelength (metric, decimetric, centimetric waves) and may both block and scatter the radio signal causing
specular and/or diffuse reections. These contributions may reachthe mobile station(MS) by wayof multiple paths, inaddition
to that of the direct signal. In many cases, these echoes make it possible that a sufcient amount of energy reaches the receiver,
so that the communication link is feasible. This happens especially when the direct signal is blocked. Hence, in addition to the
expected distance power decay, two main effects are signaled in mobile propagation: shadowing and multipath [2].
Wireless links have intrinsic characteristics that affect the performance of Handoff protocols. In this paper, we review
many simulation models for cellular and WLAN links used in the design of Handoff protocols. We also consider the interplay
between wireless links and Handoff. We argue that the design and evaluation of Handoff protocols can be improved by pro-
viding available models of wireless links that strike a balance between realism, generality, and detail. We consider how the
appropriate models for wireless links can help in Handoff optimization and evaluation.
We can identify three levels in the variation rate of the received signal as a function of the distance between the access
points (AP) and MS, namely, very slow variations due to the range, slow or long-term variations due to shadowing and fast or
0045-7906/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compeleceng.2011.09.003
q
Reviews processed and approved for publication to Carvalho.
Corresponding author. Tel.: +216 22691815; fax: +216 73501785.

E-mail addresses: Monji.Zaidi@fsm.rnu.tn (M. Zaidi), rouni@ksu.edu.sa (R. Ouni).
Computers and Electrical Engineering 37 (2011) 941957
Contents lists available at SciVerse ScienceDirect
Computers and Electrical Engineering
j our nal homepage: www. el sevi er. com/ l ocat e/ compel eceng
short-term variations due to multipath. Other operating scenarios, where both ends of the link are surrounded by obstacles,
are indoor communications where walls, ceiling, or various pieces of furniture will clearly determine the propagation con-
ditions. In an indoor environment, the corner effects are very seldom studied. It causes dropping off the received signal
strength (RSS) at the MS by 20dB or more in few meters when turning a corner [3]. Therefore, several characteristics of
the communication environment should be taken into account while designing Handoff algorithms. Moreover, many Hand-
off issues should be revived before its development in microcells.
The major issue deals with the implementation process which includes specic features to meet the Handoff timing con-
straints and QoS guarantee. Therefore, in the simulation environment, the Handoff algorithm pseudo-code could be
described in programming tool allowing concurrent processes and time sensitive instructions which are a primary attribute
of hardware as offered by the hardware description languages (HDL). These features introduce exibility to describe variable
and real scenarios as well as initiating specic process when detecting certain event [4]. In one hand, the real-time packet-
processing constraints imposed on the APs to support high line speeds motivate hardware based solutions, where the AP
functionality is implemented on application specic integrated circuits (ASICs). The requirements for exibility and the com-
plex nature of many of the processing functions, on the other hand, favor software based implementations. As result, the
design and analysis of hardware-software architectures for such algorithms requires new models and methods, which do
not fall under the domain of traditional embedded-systems design.
The rest of this paper is organized as follows. Section 2 provides an overview of related works. Section 3 briey introduces
the basic concepts and mechanisms driving the wireless propagation channel. Section 4 evaluates the importance of the
overall Signal extracted from the fast variation in the removal of unnecessary Handoffs based simulation. The Section 5
establishes a comparison of the proposed scheme with other methods existing in the literature in term of speed and power
consumption. Section 6 describes the linear regression method applied to a measurement series of the received power from
an AP. Section 7 develops a decentralized Handoff algorithm using fuzzy logic based approach on the slopes ratio between
current and expected signals provided by the linear regression method. Finally, Section 8 summarizes the paper.
2. Related work
Many metrics have been used to support Handoff decisions, including the received signal strength (RSS), signal to inter-
ference ratio (SIR), distance between the MS and AP, trafc load, and mobile velocity, where RSS is the most commonly used
one. The conventional Handoff decision compares the RSS from the serving AP with that from one of the target APs, using a
constant Handoff threshold value (Handoff margin). The selection of this margin is crucial to Handoff performance. If the
margin is too small, numerous unnecessary Handoffs may be processed. Conversely, the quality of service (QoS) could be
low and calls could be dropped if the margin is too large. The uctuations of signal strength associated with shadow fading
cause a call sometimes to be repeatedly handed over back and forth between neighboring APs, in what is called the ping-
pong effect Over recent years, many investigations have addressed Handoff algorithms for cellular communication
systems.
A novel terminal-controlled handover scheme in heterogeneous wireless networks was presented in [5] and [6]. A
network discovery algorithm with fuzzy logic and a Handoff decision algorithm using multi-criteria decision making
(MCDM) based on vague sets are derived, which are both user-centric. In [7], we have proposed the hardware transformation
and implementation of the Handoff protocol from its initial software description. We have implemented two models that
reduce the scan phase during the Handoff process. These models have been implemented on an application-specic inte-
grated circuit (ASIC). A local averaging technique, which moves fast fading component from the received signal strength,
was proposed in [8] to allow the conventional Handoff decision reacting more quickly to corner effects. A timer-based hard
Handoff algorithm was presented in [9] to prevent unnecessary Handoffs caused by uctuations due to shadowing, by which
the choice of timer interval introduces the tradeoff between Handoff number and Handoff delay. A dynamic handover mar-
gin decision based on a trafc balancing rule was proposed in [10] to resize the cells according to the spatial variability of
trafc. A speed-sensitive Handoff algorithm in a hierarchical cellular system was described in [11], in which microcells serve
the slowly moving mobiles and macro-cells serve fast-moving mobiles. In [12] and [13], RSS and the mobile location and
velocity were used as metrics for making Handoff decisions using fuzzy logic. A table lookup approach, proposed in [14],
determines Handoff margins based on the mobile location, the mean of the intensity of the signal and the standard deviation
thereof.
Distance hysteresis for mitigating the fading effect on Handoff performance was presented in [15]. Making Handoff
decisions in various Handoff scenarios was presented in [15] and [16], where a suitable Handoff decision is provided for a
predened scenario. In the literature, most Handoff algorithms, based on the mobile location, suffer from a lack of practica-
bility. The computational complexity of making a Handoff decision using fuzzy logic is excessive and establishing and updat-
ing a lookup table to support a Handoff margin decision is time-consuming. The selection of a Handoff algorithm based on
the Handoff scenario only succeeds in cases that the mobile environment is similar to one of the pre-classied environments,
and involves complicated processes to dene the Handoff scenarios. It also relies on an updated database when applied in a
new mobile user environment. Furthermore, most studies assume that location of the mobile can be perfectly determined
using the global positioning system (GPS), which is not available for most existing mobile telephones. In reality, the perfor-
mances of available location estimators are far from that obtainable using GPS technique.
942 M. Zaidi et al. / Computers and Electrical Engineering 37 (2011) 941957
A normal Handoff scheme for channel carrying in mobile cellular networks has been reported in [17]. It increases the
channel reuse factor by 1 and it becomes difcult to take advantage of channel carrying if trafc were to exhibit strong deter-
ministic correlations between neighboring cells. Handoff techniques in cellular networks are reported in [18,19].
Performance analysis with bandwidth efcient Handoff scheme is presented in [20]. Fuzzy logic based technique for
Handoff in cellular communication have been reported in [21,22]. The Handoff criteria use threshold values to membership
functions. The possible weakness in [21] and [23] is the jump values inherent in some fuzzy sets and it only takes a non line
of sight (NLOS) transmission into consideration.
This paper proposes an optimized Handoff (in terms of reduction of unnecessary Handoffs) algorithm based on local aver-
aging technique, which moves fast fading component from the received signal strength. (Not using additives equipments).
However, since the MS has limited energy source, in the form of the battery pack, energy consumption should be minimized.
An important factor in this works is to minimize and estimate the power consumption that our technique requires to per-
form optimized Handoff.
Then, based on the possibility concept, the linear model of the received power from the actual AP has been proposed and
its pivotal role in a new decentralized Handoff modeling has been established. The Handoff based on the ratio of the slopes of
normal signal loss to the actual signal loss, provided by the linear model, is presented and the fuzzy based solution is sup-
ported by comparing its results with the results obtained in analytical solution.
3. Basic concepts driving the WLAN propagation channel
3.1. Wave propagation model
Currently the 3 G wireless systems are being deployed in the 2 GHz band while wireless Local Areas Networks (WLANs)
are beginning to be deployed in the 5 GHz band while, still, the 2.4 GHz band is the most popular mode for this application.
The WLANs typically operates on unlicensed frequency bands, either the 2.4 GHz or the 5 GHz bands. They do not require
line-of-sight (LOS) conditions, which is a very desirable feature. The key element of a WLAN is a wireless AP. APs are con-
nected to an Ethernet hub or server. They transmit a radio frequency over an area of several hundred feet, which can pen-
etrate walls and other nonmetal barriers.
The signal from the AP can reach a target through its LOS path, or reection paths, or diffraction paths and their combi-
nation. Consider a scenario with maximum numbers of n access points, n
r
reection paths, n
d
diffraction paths, n
c
combined
path and one LOS path from the jth access point to a target, the received signal y(t) at the target is given by the equation 1.
Y
j
(t) = x
j
(t s
1;j
)L
j

X
nr 2
i=2
x
j
(t s
i;j
)R
i;j

X
nr n
d
3
i=nr 3
x
j
(t s
i;j
)R
i;j

X
nr n
d
nc 4
i=nr n
d
4
x
j
(t s
i;j
)C
i;j
e(t); \j 1; . . . ; n (1)
where i is the ith path to the target, t is the time, x(t) is the emitted signal from the access point station, e(t) is the back-
ground and system noise at the source, and L
j
, R
j
and C
i,j
are the weighting factors for LOS path, reection paths, diffraction
paths and combined paths respectively. The time delay s
i,j
or time of arrival (ToA), of the signal through the ith path with
distance d
i,j
is given by: s
i;j
=
d
i;j
V
air
.
Where, V
air
represents the signal speed in air. The magnitudes of the factors are inversely proportional to the signal
strength losses of their correspondent paths. Each weighting factor depends on its path distance and geometric spreading.
Additionally, the weighting factor R
i,j
is dependent on the surfaces and positions of obstacles, while the factor D
i,j
is
dependent on the frequency of the signal, the shape and positions of obstacles. The factor C
i,j
can be expressed as
C
i,j
= D
i,j
R
i,j
.
In a dynamic environment, all of the weighting factors may change with time depending on obstacles. Modeling RSS in
such an environment requires taking into account the all information of the movements, sizes, shapes and surface materials
of all obstacles. In NLOS conditions, where the LOS path is not available, the LOS path weighting factor L
j
is zero. The diffrac-
tion of an acoustic signal increases with the increase in the ratio between the wavelength and the obstacle size [24]. The
diffraction of a high-frequency signal, such as ultrasonic, with short wavelength is negligible. Therefore, for a high frequency
signal, D
i,j
and C
i,j
are approximately zero. This means that the signals passing through the diffraction and combined paths are
too weak to be detected.
Consider a small dynamic obstacle obstructing a signal path, the low-frequency acoustic signal will pass through the
obstacle if its wavelength is larger than the dimensions of the obstacle, even though the correspondent weighting factor de-
creases. As a result, small dynamic obstacles do not hinder any low-frequency signal.
3.2. Received signal characteristics
Two representative and extreme scenarios may be considered: (a) the case where a strong direct signal is available to-
gether with a number of weaker multipath echoes, i.e., LOS conditions; and (b) the case where, a number of weak multipath
echoes is received and no direct signal is available i.e., NLOS conditions.
M. Zaidi et al. / Computers and Electrical Engineering 37 (2011) 941957 943
Case (a) occurs in open areas or in very specic spots in city centers, in places such as crossroads or large squares with a
good visibility of AP. This situation may be modeled by a Rice distribution for the variations of the received RF signal enve-
lope. Under these conditions, the received signal will be strong and with moderate uctuations (Fig. 1).
Case (b) will typically be found in highly built-up indoor environments. This is a worst case scenario since the direct sig-
nal is completely blocked out and the overall received signal is only due to multipath, thus being weaker and subjected to
marked variations (Fig. 2). This kind of situation may also occur in indoor environments where the signal is obstructed by
dense masses of walls. The received signal amplitude variations in this situation are normally modeled with a Rayleigh
distribution.
The received eld strength or the received voltage may be represented in the time domain, r(t) or in the traveled distance
domain, r(x) . For carrying out the propagation channel measurements, the mobile speed, V, should remain constant. In such
cases, the traversed distance needs to be recorded too. In our simulations and in the series analyzed, we will assume a con-
stant MS speed. For a constant V, it is quite practical to make the conversion directly between the representation in the time,
r(t), and the traveled distance domains, r(x) = (t = x/V).
The variable x may either be expressed in meters or in wavelengths. Based on such signal recordings plotted in the dis-
tance domain, it is possible to separate and study individually the fast and slow variations, which are due respectively to
multipath and shadowing. Fig. 1 and Fig. 2 show slow and fast variations (i.e., long-term and short-term variations) of
the received power r(t), or r(x) from an AP respectively.
The received signal may, therefore, be described as the product of these two terms, when expressed in linear units
r(t) = s(t) f(t) or alternatively r(x) = s(x) f(x). In dB, the products become additions, i.e., R(t) = s(t) f(t) or R(x) = s(x) f(x).
0 50 100 150 200 250
5
10
15
20
25
30
35
40
45
50
55
R
e
c
e
i
v
e
d

v
o
l
t
a
g
e
:
s
l
o
w

v
a
r
i
a
t
i
o
n
s
(
d
B

V
)

Traveled distance (m)
Fig. 1. Received power: Slow variation.
0 50 100 150 200 250
-130
-120
-110
-100
-90
-80
-70
-60
-50
-40
R
e
c
e
i
v
e
d

p
o
w
e
r

:

F
a
s
t

v
a
r
i
a
t
i
o
n
s

(
d
B
m
)
Fig. 2. Received power: Fast variation.
4. Mean extraction and impact on the undesirable Handoffs
4.1. The average signal computing and construction
With the approach mentioned above, we are assuming that the fast variations are superposed on the slow variations.
Fig. 3 illustrates an overall time series where the slow variations are also plotted. The gure also shows the fast variations
after removing (ltering out) the slow variations. The slow variations can be extracted from the fast variations through low-
pass ltering by computing a running means. This is equivalent to calculating the signal average for the samples within a
route section of length 2L equal to a few tens of wavelengths (equation 2).
S(X
i
) =
1
2N 1
X
k=N
k=N
f
ik
for f
iN
. . . f
i
. . . f
iN
X=X
i
L < x < x
i
L (2)
Where S(x
i
) is the local mean value at x
i
, x
i
which is the route position dened by, x
i
=
2L
2
, with 2N being the number of
samples received from -L to +L. Here f
i
represents the i
th
fast or short-term variations and [-L, +L] is the 2L length of the route
considered for the computation of the local mean. It is used to separate out the fast from the slow variations.
Typically lengths, of 10k to 10k, are used [5]. For example, for the 2.4 GHz (k=12.5 cm) band used in 3G mobile commu-
nications, the average length could be 2L ~ 3. The average value, S(x
i
) computed for a given route position x
i
is usually called
the local mean at x
i
.
The length (2L) of route considered for the computation of the local mean, i.e., used to separate out the fast variations
from the slow ones, is usually called a small or local area. It is within a small area that the fast variations of the received
signal are studied since they can be described there with well-known Rayleigh distributions.
4.2. Handoff review
In this section, we will take into consideration a scenario of the WLAN Handoff process; this is meant to show the advan-
tage of performing Handoff based on the overall signal extracted from the fast variations. MS is assumed to travel from AP1
to AP2. Midway between the two APs, the averages of the received signals strength are similar. Therefore, a threshold is set to
initiate a simple Handoff algorithm. Fig. 4 illustrates the two received signals from, both, old and new APs as well as the xed
threshold. Based on the actual signal variation, the Handoff algorithm runs whenever the received signal is below the thresh-
old and the alternative signal is high.
Fig. 5 illustrates the numerous switches that take place. If the extracted mean from the former received power is strictly
decreasing and the extracted mean fromthe newreceived power is strictly increasing, subsequently undesirable Handoff can
be eliminated. Finally the number of AP changes is drastically reduced as shown in Fig. 6. Of course, real Handoff algorithms
are much more complex, and involve different criteria which are part of the network optimization process.
This method selects the AP delivering the higher power at the MS location during the Handoff processing duration. The
decision is based only on the RSS measurement which leads to many unnecessary Handoffs, even when the signal of the cur-
rent AP is still at an acceptable level.
0 50 100 150 200 250
-140
-130
-120
-110
-100
-90
-80
-70
-60
-50
-40
R
e
c
e
i
v
e
d

p
o
w
e
r

a
n
d

i
t
s

m
e
a
n

e
x
t
r
a
c
t
i
o
n

(
d
B
m
)

Fast variations
Overall signal
2L
xi
Fig. 3. Fast variation and its mean extraction.
Fig. 4. Threshold and received signals from old and new APs.
Fig. 5. Aps handling the communication during Handoff and using fast variation signals and threshold.
Fig. 6. APs handling the communication during Handoff and using the overall signal and hysteresis.
This scheme allows a MS to Handoff only if the overall signal of new AP is sufciently stronger (by a hysteresis margin, H)
than the current one. This technique prevents the so-called ping-pong effect, the repeated Handoff between two APs caused
by rapid uctuations in the received signal strengths from both APs. The rst Handoffs, however, may be unnecessary if the
serving AP is sufciently strong.
4.3. Handoff methods: description and comparison
In this simulation, there are four Handoff methods which performance will be compared.
v Normal hysteresis method
v Hysteresis plus dwelling timer method
v Trajectory-Aware vertical Handoff method and
v Proposed Local and Variable Average Extraction plus Hysteresis (LVAEH) method.
4.3.1. Normal hysteresis method
In this method, Handoff is performed whenever the RSS of new AP is higher than the RSS of old AP bay a predened value.
[25]. Simple pseudo codes describe the Handoff algorithm.
If (RSS
new
- RSS
old
) > Hys
Handoff;
End.
In this simulation, the Hysteresis value is assumed to be 10dB.
4.3.2. Hysteresis plus dwelling timer method
In this method, whenever the RSS of new AP is higher than the RSS of old AP bay a predened hysteresis value, a Timer is
triggered. When this Timer reaches certain specied value, Handoff is processed. The dwelling is reset when the RSS of new
AP is not higher than the RSS of old AP bay the hysteresis value or when Handoff is processed [26,27]. The following pseudo
codes may describe this algorithm.
If (RSS
new
- RSS
old
) > Hys
Timer ++ = Sampling time;
Else
Reset Timer;
End
If Timer > dwelling Delay
Handoff;
Reset Timer;
End;
In this simulation, the Hysteresis value is assumed to be 10dB and dwelling delay is assumed to be 3 sec.
4.3.3. Trajectory-Aware Vertical Handoff (TAVH) method
In this method, the position, velocity and RSS of MS are calculated and every data is taken in account in Handoff decision.
A Handoff is carried out whenever the position of MS has reached to a certain boundary, regardless of the RSS. This reduces
the rate of Handoff failures. Here, the boundary is a safety distance of MS from the AP to assure a successful Handoff and this
boundary is not xed and is varying according to the position and velocity of the MS. On the other hand, Handoff is also car-
ried out whenever the RSS of the MS has dropped belowpredened threshold value (-60 dB in this simulation). The following
pseudo code describes the dwelling TAVH method which is deployed in [28].
If (position_of_MT > Safe_Boundary) or (RSS_LSL < Thershold)
Handoff;
End;
4.3.4. Proposed Local and Variable Average Extraction plus Hysteresis (LVAEH) method
This method continuously computes the average of the received power from two APs (new and old). This computation
aims to combine each 2L portion of the traveled distance with an S(x
i
) value. The length (2L) is not xed but, it is variable
according to the velocity of MS and the frequency band used in 3G communication. For example, when the 2.4 GHz band is
deployed, the wavelength is k = 12.5 cm. So, when typically lengths of 10k to 10k are used [1], the average length would be
2L ~ 3-6 m. The local mean at x
i
is computed using the equation 2.
When a new AP provides higher S(x
i
) than the old AP bay a predened hysteresis value (-15 dB in this simulation), Hand-
off is processed. Our pseudo code includes equations computing power signal average plus other parameters taking decisions
to initiate Handoff.
Old communication established
If x>2L
S =
X
k=N
k=N
f
ik
for f
iN
. . . f
i
. . . f
iN
X=X
i
L < x < x
i
L;
S
X
i
=
X
k=N
k=N
f
ik
for f
iN
. . . f
i
. . . f
iN
X=X
i
L < x < x
i
L;
If (S
X
i
)
new
(S
X
i
)
old
> Hys
Handoff;
Else
Maintain old communication
Under ve different scenarios, these methods are simulated to evaluate the number of initiated Handoffs. Fig. 7 estab-
lishes a comparison between our proposed solution and all methods previously described in section 4.3. As result, our pro-
posed method guarantees a minimumnumber of Handoff processes avoiding ping-pong situations. Especially, it performs till
65% Handoff initializations lower than the TAVH method.
For performance evaluation, the handoff latency and the loss rate are considered as two QoS metrics for both standard
and LVAEH algorithms. The Handoffs latency is dened as the time interval from the time when the MS is disconnected from
its current AP till the time when the MS receives the re-association response for its new location. In order to evaluate the
Handoff latency, different cellular transition scenarios have been developed to simulate these algorithms. Fig. 8 shows
the impact of our proposed LVAEH scheme on the handoff latency compared to the standard algorithm. For 9 scenarios,
the Handoff latency tolerated in LVAEH scheme is reduced for 64 to 85.7 % compared to the standard protocol.
We assume that during the Handoff processing delay, all the data packets destined for the MS are lost which leads to
interrupt the communication. Consequently, the proposed LVAEH scheme also guarantees lower data loss rate which reects
the quality of the communication during the Handoff process.
5. Hardware implementation and power consumption
There are two alternatives to implement wireless algorithms in the Medium Access Control (MAC) layer. The rst
alternative is a Central Processor Unit (CPU)-based solution. It uses software for protocol analysis and CPU, such as
Digital Signal Processing (DSP), for process management. It is more exible in design stage and easy to modify and update
the protocol. However, the low processing speed and the higher cost present its major weakness. The second alternative
implements a hardware prototype following the top-down co-design approach. This alternative offers the circuit
reconguration and very high speed processing suitable for real time communications. But, it needs long development
time. This work adopts the second alternative to implement our proposed LVAEH method at the MAC abstraction
level.
Yet, the very fast growth of the modern VLSI technology offers a hardware realization of an ever-growing share of
mathematical means. New algorithms which satisfy VLSI-technology requirements are needed. These algorithms perform
the features of the hardware prototype in terms of processing delay, chip area, and power consumption.
Fig. 7. Number of unnecessary Handoffs for each method.
5.1. Mean extraction Process: Design and implementation
Fig. 9 shows the mean extraction process ow chart responsible to provide S(x
i
) values (equation 2) which make useful
decision for handoff initialization. It also allows detecting and removing unnecessary Handoff. Fig. 9 correlates between in
one side the nite state machine operating as detailed in section 4.3 and in other side the sub-modules deployed to achieve
arithmetic operations (additions, divisions) and process data and events (Fig. 9).
Fig. 8. LVAEH and QoS parameters. (a): Handoff latency, (b): lost data.
Fig. 9. Flow chart of the signal mean extraction process.
Five MAC components have been developed in order to implement the average extractor process. As illustrated in Fig. 10,
these components include a controller (or scheduler), receiver, Adder, Divider and a shared memory. The controller serves to
schedule tasks and synchronize module interactions for the whole process. The receiver is responsible to acquire the signal
and extract RSS. As given by the equation 2, the adder and divider components apply arithmetic operations at discrete posi-
tions providing one sample every 2L distance. The shared memory allows storing up to 256 words of 16 bits and serves as
mailbox between these components.
5.2. Building blocks specications
5.2.1. Receiver Module
The MS receives Message Protocol Data Units (MPDUs) from the Physical Layer Convergence Protocol (PLCP) level and
decodes packets. Table 1 lists the MAC receiver interface signals and describes their signications.
Fig. 10. Top level structure of the average extraction system.
Table 1
Receiver interface signals (I. Input, O. Output).
Name Type Description
clk I:bit Operation clock
reset I:bit Initialization signal
Recep_valiv I:bit Input from the controller module, it requests receiving data
Rts_recep I:bit Noties transmitter that receiver got Request to Send (RTS) frame
In_recep I[15:0] Data bus from physical layer: (Received power)
End_recep O:bit Noties Controller that reception is completed
Cts_recep O:bit Noties that transmitter sent Clear To Send (CTS) frame
Write O:bit Enable memory for writing data
Out_recep O:[15:0] Enable to save the received data in memory
Adr_recep O:[7:0] Address bus
Table 2
The adder in-out put ports.
Clk I:bit Operation clock
Reset I:bit Initialization signal
In_somme I[15:0] Power value
Lire_mem I[15:0] Input port from the controller
Adr_data O[7:0] Memory bus address
Read O:bit Enables memory for reading mode
Somme O[15:0]
Addition result, returns
P
k=N
k=N
f
ik
value
5.2.2. Adder component
For each distance x equal to 2L, 2n+1 values of the received power must be calculated. We developed a pseudo-code using
Hardware Description Language (HDL) for 2n+1 added samples. The addition function reads 2n+1 samples from the memory
(in_somme), adds these values and returns the Sum (Somme) as result. The function is generic able to support different sizes
of inputs. In the mean extraction process, the main function of the adder is to calculate
P
N
k=N
f
ik
value after each 2L dis-
tance. Table 2 outlines the adder interfaces and describes their signications.
5.2.3. Divider component
We developed a HDL pseudo-code for generic division. The function uses as inputs two unsigned numbers (num and den)
with the same size and delivers as division result the quotient with similar features as the input parameters. In the mean
extraction process, the divider consists of calculating S(X
i
) =
1
2N1
P
N
k=N
f
ik
values after every 2L propagated distance. The
main interface signals deployed for unsigned number division are described in Table 3.
5.2.4. Controller module
In this design, all the actions are controlled and organized by the control module (scheduler). The controller is mainly
responsible to handle information or data from the physical layer and to coordinate the interaction between all the other
modules especially arbitrating the trafc from/to the shared memory.
The controller schedules the tasks as follows:
v Activates, at the same time, the receiver and the memory to receive and store received data from the current AP.
v Compares the value of the traveled distance with 2L.
v When traveled distance is equal to 2L
Stops receiving data
Makes memory accessible in reading
v Enables Adder to calculate
P
N
k=N
f
ik
v Enables Divider to calculate S(X
i
) =
1
2N1
P
k=N
k=N
f
ik
v Compares S(x
i
)
new
-S(x
i
)
odd
with the Hysteresis value
5.2.5. SRAM Memory
It is an external memory generated by STMicroelectronics library. It is used as shared memory to store data delivered by
an AP and exchange them between the different components. This memory allows storing up to 256 words of 16 bits.
5.3. Synthesis results
In the RTL level, VHDL has been deployed as hardware description language allowing to translate the functionalities of the
different blocks realizing the mean extraction process. Recently, the FPGA circuit solution is considered as one of the most
used prototyping environments. In fact, with the increase of the available programmable cells in this circuit, it is possible to
implement a wide range of sophisticated algorithms as reusable IPs blocks. Xilinx ISE tool has been also used for synthesis
and design implementation on Xilinx Virtex V FPGA circuits. This environment allows implementing and reconguring hard-
ware and programmable systems. Table 4 shows synthesis results and power consumption estimations for all modules.
Table 3
The divider in-out put ports.
Clk I:bit Operation clock
Reset I:bit Initialization signal
num I[15:0] Dividend
den I[15:0] Divisor
quot O[15:0] Quotient (division result)
rest O[15:0] Division remainder
Table 4
FPGA Implementation results (VIRTEX V).
Nb. of Slice Registers Nb. used Flip Flops Nb. of Slice LUTs Nb. bonded IOBs Frequency (MHz) Power consumption (W)
static dynamic Total
Receiver 50 50 235 47 238 0.3 0.09 0.39
Adder 557 557 539 44 157 0.4 0.09 0.49
Divider 85 85 63 66 342 0.25 0.02 0.27
Controller 9 9 9 30 424 0.3 0.009 ~0.31
6. Linear interpolation for decentralized Handoff
6.1. Linear interpolation
Linear interpolation is a topic usually well-covered in statistic courses since it is very important to any engineer. Linear
regression is not a difcult task to carry out, but the process of understanding and deriving the equations used can be chal-
lenging as much as it is tempting. Indeed, the calculators use the so-called linear regression systematically, i.e., they apply
the least square method in cases of the afne function (y=ax+b).
We want to make linear interpolation automatically i.e., on each portion of the root the MS must interpolate the measures
that it has collected from the current AP by a linear form. This linear form allows it to effectively perform the decentralized
Handoff that we will specify in the next section.
The linear regression process consists of deploying the distance between the AP and the MS to generate the signal
strength that will be received by the MS. This is a complex task, because in an indoor environment, several parameters
are included in this process. Mainly, the linear regression represents, in a linear form, the received power by the MS as a
function of the traveled distance or the elapsed time (equation 3).
RSS(dB) = m + D(m) b or RSS(dB) = m + t(s) b (3)
Where m is the line slope and b is the y-intercept. Usually D
0
is chosen as the origin of movement. In this case, the y-inter-
cept should be near to the LOS power.
6.2. The received signal representation
The data reported below correspond to the collections of the signal level leaving an AP and arriving to the MS. Note that
the MS is moving in an Indoor environment.
Now, how does one nd a calibration line that accurately describes the data? One could simply draw a line through the
data and assume that it describes the data accurately. This does not seem harsh enough. So, we need to come up with a way
to measure the error of our line. One logical way to measure the error is to take the difference of each data point with the
curve t and add up all of these errors. These errors are shown on the Fig. 11. If the error is too large, the slope and intercept
could be changed. The equation 4 describes this context.
SSE =
X
i
(y
real
y
est
)
2
(4)
Where SSE is the sum of the squared errors and y
real
and y
est
represent the measured and estimated powers
respectively.
The question remains, is there any way to nd the best t line for the data? The answer requires high knowledge of
calculus, so the full derivation will be abridged. The squared error can be expressed in terms of the independent variable
x
i
as outlines equation 5.
X
i
(y
real
y
est
)
2
=
X
i
(y
real
m + x
i
b)
2
(5)
0 50 100 150 200 250
-130
-120
-110
-100
-90
-80
-70
-60
-50
-40
R
e
c
e
i
v
e
d

p
o
w
e
r

(
d
B
m
)
Elapsed time (s)
Fig. 11. Measured power as function of elapsed time or (traveled distance).
To nd the best t line, the sum of the squared errors should be minimized. In Calculus, the minimum or the maximum of
a function can be found by considering its derivative and setting it to zero. In other words, the slope of a line at a maximum
or minimum point is zero. This is dealt with thoroughly in the m-le program. For this problem, the derivative of SSE is taken
with respect to m then again with respect to b. This yields two equations with two unknown parameters m and b.
With proper rearrangement, the nal result outlines two equations (6 & 7) evaluating m and b which depend on the sums
of data.
M =
n
P
i
X
i
y
i
(
P
i
X
i
)(
P
i
y
i
)
n
P
i
X
2
i
(
P
i
X
i
)
2
(6)
b =
P
i
y
i
P
i
X
2
i

P
i
x
i
y
i
P
i
X
i
n
P
i
X
2
i
(
P
i
X
i
)
2
(7)
In these equations, n is the number of data points. The sum of x
2
i
is obtained by squaring all the elements of x and then
adding them all up. The sum of x
i
y
i
is computed by multiplying x and y elements and then summing them all up.
The slope and y-intercept are found for each interval as (-0.0019,-60.0517), (-0.3356,-22.5582) and (-0.8700, 85.1815)
respectively (Fig. 12). These lines can then be plotted over the data to trace the linear form(s) that can describe effectively
the whole repartition of points.
7. Decentralized Handoff in wireless communication based on linear interpolation
Handoff is dened as the process of changing the current radio channel to a new radio channel [29] which mainly takes
place because of the movement of MS and unfavorable radio conditions (deterioration of received signal quality) inside an
individual cell or between a numbers of adjacent cells.
7.1. Analytical solution for Indoor-Handoff
The higher value of hysteresis (Th
max
-Th
min
) effectively prevents unnecessary Handoffs but causes undesired cell drag-
ging. This undesired cell dragging causes interference or could lead to dropped calls in microcellular environment.
When turning in a corner, the signal strength may drop by more than 20 dB at the MS [3] (Fig. 13). Especially, the signal
from AP1, which is in line of sight, suddenly disappears when the MS turns the corner. However, the link with AP2, which
was blocked, it becomes in a line of sight. In addition, the signals arriving from AP3 and AP4 interfere and will be received by
the MS. In some cases this interference can be handled by the MS as a good signal level .The MS can perform the Handoff
based on this interference that ultimately leads to a communication breakdown. Thus, there are Handoff issues that must
be revived before the development in microcells.
In this section we present the Handoff on the basis of ratio of slopes of normal decaying signal to the actual signal, the
received signal strength and the hysteresis value (Th
max
-Th
min
). In this work the emphasis given to the Handoff is due to the
corner effect. The rapid decrease in signal strength is due to the corner effect or any other NLOS condition in a cellular sys-
tem. This may result in a dropped call if Handoff algorithm does not support. The decrease in the received signal strength is
always monitored by determining the ratio of the slopes of expected normal signal and the actual signal (Fig. 14).
The slope of the expected signal is obtained by the difference in the previous sample and the expected new sample deter-
mined by the regression method (section 5).
Fig. 12. The measured power and its linear regression as function of the elapsed time.
The nature of the decaying signal can be determined by the slope or tangent of angle subtended by the two corresponding
points on the signal line with respect to two axes. The Handoff process initialization depends on the nature of the decaying
signal, the received signal strength and the hysteresis. In other terms, certain condition on these parameters leads initiating
the Handoff process by the MS to maintain the communication over another AP. The Handoff process should be fast to reduce
the call dropping probability.
v The slope of expected signal is evaluated by the equation 8.
v The slope of actual signal after the corner effect at A is given by the equation 9.
v The slope ratio of normal and actual signals is computed by the equation 10.
Tgh
1
=
DRSS
DS
1
(8)
Tgh
2
=
DRSS
DS
2
(9)
R
P
=
Tgh
1
Tgh
2
=
DS
2
DS
1
(10)
Where, RSS is the received signal strength.
If h
1
=h
2
then DS
1
=DS
2
and R
p
=1
DS
1
and DS
2
are small changes in distances of MS from AP with respect to the normal and actual signals respectively and
R
p
is the slope ratio.
For the different analytical values of h
1
and h
2
the various values of R
p
are obtained. But for the Handoff to take place h
2
must always be greater than h
1
. Thus, we have only taken the values for h
2
> h
1
into consideration. The ratio from h
1
=5 to
Expected Actual
1
1 S
2 S
RSS(dB)
Traveled distance(m)
H
e
s
t
e
r
i
s
i
s
2
A
Fig. 14. Slopes of actual and expected signals.
AP3
AP1
AP2
AP4
Mobile route
Fig. 13. Mobile station turning a corner in microcell.
h
2
=70 has been considered with high probability (Fig. 15). To avoid the overcrowding of data, the difference of two degrees
in h
1
and h
2
is considered.
7.2. Fuzzy logic solution for decentralized Handoff
The fuzzy set theory allows a linguistic representation of the control and operational laws of a system. The main strength
of the fuzzy set theory is that it excels in dealing with imprecision. The fuzzy logic can be used to control a complex system
[30,31]. The block diagram of the fuzzy logic system considered for this work is shown in Fig. 16.
For Handoff initiation there are three membership functions are used: Received Signal Strength (RSS), Slope ratio (R
p
) and
Hysteresis value. The fuzzy sets contain elements that have a varying degree of membership in a set. The membership values
are obtained by mapping the values obtained for a particular parameter onto a membership function. This function is a curve
or line that denes howeach data or value is mapped onto a membership value. It is represented graphically in Fig. 17, where
the ve lines represent the range available for a very low, low, Average, High and very high value.
The fuzzied data is passed to the inference engine, where the fuzzied data is matched against a set of fuzzy rules using
the fuzzy techniques to produce output fuzzy sets. Fuzzy rules can be dened as a set of possible scenarios utilizing a series
of IF-THEN rules, which decides whether the Handoff is necessary. Following this, a set of different Handoff decisions can be
obtained. An example of IF-THEN rules is as follows:
IF signal strength is high and the slope ratio is average and the hysteresis is low, then execute the Handoff slightly.
The output fuzzy sets are then passed to the defuzzier which computes a crisp output value. The fuzzy IF-THEN rules
provide the knowledge basic to the system and results in proper Handoff. The simulation data obtained from Mumdani infer-
ences system is shown in Fig. 18. These results are in coincidence with the results obtained by the analytical solution.
7.3. Comparison of results
From Fig. 14, it is evident that with the increase in slope ratio values, the curves become smoother indicating that Hand-
offs decrease. The higher value of slope ratio means decreasing denominator value in
tgh
1
tgh
2
or in other words h
2
approaches h
2
thereby the actual and normal signals approach coincidence. From Fig. 18, the slope ratio and Handoff factor act inversely
proportional to each other. Thus higher value of slope ratio produces lower Handoffs and vice versa.
0 10 20 30 40 50 60 70
0
0.2
0.4
0.6
0.8
1
variable Teta2
s
l
o
p
e

r
a
t
i
o
Teta1=5
Teta1=8
Teta1=11
Teta1=14
Teta1=17
Teta1=20
Teta1=23
Teta1=27
Teta1=30
Teta1=32
Fig. 15. Slope ratios with varying angles.
Fig. 16. Block diagram of fuzzy logic system.
8. Conclusion
In this work, three main parts are included. First, we presented a time-series analysis techniques which covers the basics
concepts and mechanisms driving the wireless propagation channel. We proposed a method allowing to extract a signicant
signal from the fast variations caused by the combined effects of shadowing and multipath. This method is well affordable by
the mobile device resources like power consumption and memory space. This overall signal was used to reduce signicantly
the number of unnecessary Handoffs.
Second, we implemented our technique using the hardware co-design approach allowing a performance evaluation in
terms of frequency, area and power consumption. The Handoff latency is also evaluated compared to different proposals.
Third, we applied the linear regression method to the above time-series in order to better optimize the Handoff process.
In addition, we developed a new fuzzy based Handoff algorithm capable of responding to the fast changes that occur in a
microcellular environment especially in indoor conditions. The results obtained by both analytical and fuzzy methods are
similar and better than provided by conventional approaches. The Handoff is independent of the received signal strength
of neighboring access points, so it is fast and mobile controlled.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
RSS
Slope ratio
Hesterisis
Handoff action
Fig. 18. Handoff with varying RSS, Slope ratio and Hysteresis.
1
0
Received signal strength
Very low Average Low High Very high
1
0
Slope ratio
1
0
hysteresis
Fig. 17. Various membership functions.

References
[1] W.C.Y. Lee. John Wiley & Sons, Ltd, Chichester. Mobile Communications Design Fundamentals. Wiley Series in Telecommunications and Signal
Processing, UK, 1993.
[2] Michel C. Jeruchim, Philip Balaban, K. Sam Shanmugan. Simulation of Communication Systems: Modeling, Methodology, and Techniques. 2
nd
Edition
2000.
[3] G. M. Mir, N. A. Shah, and Moinuddin. Decentralized Handoff for Microcellular Mobile Communication System using Fuzzy Logic. Proceding of World
Academy of Science, Engineering and Technology, Vol 38 Feb 2009.
[4] Ridha Ouni. Dynamic slot assignment protocol for QoS support on TDMA-based mobile networks. Computer Standards & Interfaces 2011. doi:10.1016/
j.csi.2011.06.00.
[5] Haibo Xu, Hui Tian, Ping Zhang. A novel terminal-controlled handover scheme in heterogeneous wireless networks. Computers and Electrical
Engineering 2010;36:26979.
[6] Nguyen-Vuong, Agoulmine N, Ghamri-Doudane Y. Terminal-Controlled Mobility Management in Heterogeneous Wireless Networks. IEEE
Communications Society 2007;45(4):1229.
[7] Zaidi Monji, Ouni Ridha, Torki Kholdoun, Tourki Rached. Low Power ASIC designs for fast Handoff in IEEE802. 11. International Journal of Computers
Systems and Signals 2009;10(1).
[8] Leu E, Mark BL. Analysis of Fast Handoff Algorithms for Microcellular Networks. Proc. 10th IEEE Symp. Modeling, Analysis and Simulation of Computer
and Telecomm Systems 2003:3218.
[9] Leu AE, Mark BL. An Efcient Timer-Based Hard Handoff Algorithm for Cellular Networks. Proc. IEEE Wireless Comm. And Networking Conf.
2003;2:120712.
[10] Tang Shensheng, Li Wei. Modelling and analysis of hierarchical cellular networks with bidirectional overow and take-back strategies under generally
distributed cell residence times. Telecommun Syst 2006:7191.
[11] Xia Liu, Jiang Ling-ge, He Chen. A Novel Fuzzy Logic Vertical Handoff Algorithm with Aid of Differential Prediction Pre-Decision Method ICC07. IEEE
International Conference on Communications 2007:566570.
[12] Israt Presila, Chakma Namvi, Hashem MMA. A Fuzzy Logic-Based Adaptive Handoff Management Protocol for Next-Generation Wireless Systems.
Journal of networks 2009;4(10).
[13] Majlesi A, Khalaj BH. An Adaptive Fuzzy Logic Based Handoff Algorithm for Hybrid Networks. Proc. IEEE Conf. Signal Processing 2002;2:12238.
[14] Wang SS, Green M, Malkawi M. Adaptive Handoff Method Using Mobile Location Information. Proc. IEEE Emerging Technology Symp. Broadband
Comm. for the Internet Era Symp. 2001:97101.
[15] Teerapabkajorndet W, Krishnamurthy P. Comparison of Performance of Location-Aware and Traditional Handoff-Decision Algorithms in CDPD
Networks. Proc. IEEE Conf. Vehicular Technology 2001;1:2126.
[16] Markopoulos A, Pissaris P, Kyriazakos S, Dimitriadis C, Karetsos G, Sykas ED, et al. Increased Handover Performance in 2G and 3G Wireless Systems
Based on Combined Mobile- Location and Area. Proc. IEEE Intl Symp, Wireless Personal Multimedia Comm. 2002;1:4751.
[17] Li Junyi, Shroff Ness B, Chong Edvin KP. Channel Carrying: A Noval Handoff Scheme for Mobile Cellular Networks. IEEE/ACM Transactions on
Networking 1999;7(1):001.
[18] Ekiz Nasif, Salih Tara, Sibel, Fidanboylu Kermal. An Overview of Handoff Techniques in Cellular networks. International Journal of Information
Technology 2005;2(3):1326.
[19] Wie SH, Jang JS, Shin BC, Cho DH. Handoff Analysis of the Hierarchical Cellular System. IEEE Trans Vehicular Technology 2000;49:202736.
[20] Li B, Wu CK, Fukuda A. Performance Analysis of Flexible Hierarchical Cellular Systems with a Bandwidth Efcient Handoff Scheme. IEEE Trans.
Vehicular Technology 2001;50:97180.
[21] Kassar Meriem, Kervella Brigitte, Pujolle Guy. An overview of vertical handover decision strategies in heterogeneous wireless networks. Computer
Communications 2008:260720.
[22] Edwards George, Sankar Ravi. Microcellular Handoff using Fuzzy Techniques. Wireless Networks 1998;4:4019.
[23] Muaoz-Rodiguez D, Cattermole K. Handoff procedure for dened radio cells. Proc. IEEE Vehicular Technology 1987:3844.
[24] Kapralos Bill, Jenkin Michael RM, Milios Evangelos E. Acoustical diffraction modelling utilizing the Huygens-Fresnel principle. In IEEE International
Workshop on Haptic Audio Visual Environment and their applications 2005;11(4):41621.
[25] Zhang N, Holtzman JM. Analysis of handoff algorithms using both absolute and relative measurements. IEEE Transaction on Vehicular Technolgy
1996;45:1749.
[26] Pollini GP. Trends in handover design. Communications Magazine, IEEE Computer Society, Washington DC, USA 1996:8290.
[27] H. Park, S. Yoon, T. Kim, J. Park, M. Do, J. Lee. Vertical Hando Procedure and Algorithm Between IEEE802.11 WLAN and CDMA Cellular Network,
Springer, USA, 2003, 103-112.
[28] Ei Thazin, Furong Wang. Trajectory-Aware Vertical Handoff Protocol Between WiMAX and 3GPP Networks. Information Technology Journal
2010;9(2):20114.
[29] Arunesh Mishra, Minho Shin, William Arbaugh, An empirical analysis of the IEEE 802.11 MAC layer handoff. ACM SIGCOMM Computer
Communications Review (ACM CCR) 33(2) 93-102.
[30] Astrain JJ, Villadangos J, Castilo M, Garitagoitin JR, Farina F. Mobility Management in Cellular Communication Systems Using Fuzzy Systems.
International Federation for Information Processing 2004:7991.
[31] Habetha Jrg, Walke Bernhard. Fuzzy Rule-Based Mobility and Load Management for Self-Organizing Wireless Networks. International Journal of
Wireless Information Networks 2000;9(2).
Monji Zaidi received the Dipl.-Ing. in electrical engineering for automation and processes control in 2005 from the National engineers school of Sfax and the
Mastere degree in Nanostructures, devices and systems micro-electronics from the University of Monastir,, Tunisia 2007. He is currently working toward
the PhD degree in electronic and communication. His research interests include Management of the WLAN technologies.
Ridha Ouni is Assistant professor in the CCIS of the King Saud University, KSA, since Feb 2009. He was in the IPEIM, Tunisia, since 1999. His research
interests include wireless communication, mobility, QoS management and sensor networks. Previously, he Received his MSc in Physic Micro-electronic, his
DEA degree and his PhD from the Faculty of Sciences of Monastir (Tunisia) in 1995, 1997 and 2002, respectively.
Rached Tourki received his B.S. degree in Physics from Tunis University in 1970, the M.S. and the Doctorat in Electronics from the Institut dElectronique
dOrsay, Paris-South University in 1971 and 1973, respectively. He received his Doctorat dEtat in Physics from Nice University in 1979. Since this date, he
has been a Professor in Microelectronics and Microprocessors with the Facult des Sciences de Monastir. His current research interests include digital signal
processing and HardwareSoftware Codesign for rapid prototyping in telecommunications.
A new energy-efficient neighbor discovery and load balancing protocols for
mobile sensor networks

Ridha Ouni
1
, Rafik Louati
2

1
College of Computer and Information Sciences (CCIS), King Saud University, KSA
rouni@ksu.edu.sa
2
Ecole Nationale dIngnieurs de Monastir (ENIM), Tunisia
rafik.louati@topnet.tn
Abstract
The wireless sensor networks (WSNs) are formed by a large number of sensor nodes working together to provide a
specific duty. However, the low energy capacity assigned to each node prompts users to look at an important design
challenge which is lifetime maximization. Especially the availability of nodes, the sensor coverage, and the connectivity
have been included in discussions on network lifetime. Therefore, designing effective techniques that conserve scarce
energy resources is a critical issue in WSNs. In this regard, we are interested in developing various mechanisms to save
energy based on the constraints involved in the energy consumption in WSNs. Three mechanisms are proposed to
improve the management of control packets responsible for path discovery, self-organization, and balance the load
distribution in the network.

Keywords: Wireless Sensor Networks, lifetime, AODV Routing, Energy-aware routing

1. Introduction
With the proliferation of wireless sensor networks
(WSN), completely new application domains for
wireless ad hoc networks have emerged. These later
motivated an increasing demand for developing more
efficient sensor networks despite their limitations in the
available resources (energy, processing speed, storage).
Besides these restrictions, WSNs are also exposed to
various requirements dealing with dynamic topologies
and variable environmental conditions [1]. Moreover,
routing and data dissemination constraints [2], self-
organization issues [3] and the efficient deployment of
sensor nodes [4] have already been investigated. While
the study of network lifetime as a key characteristic of
WSNs is still in progress [5].
Energy is a critical factor in wireless sensor
networks (WSN). To increase the network lifetime, the
energy must be conserved in hardware as well as in
software that comprise the network architecture. The
data communication is the most responsible factor in the
consumption of energy budget in comparison with
sensing and computing [6]. Therefore, it is desirable to
use short-range nodes instead of long range due to the
power required in transmission. In most scenarios, the
events can be detected by several nodes close to the
source of the phenomenon. Then, using short-range
communication leads necessarily to a multi-hop data
packets transmission.
Consumption optimization is a crucial issue in WSNs
that requires optimizing radio communication
processing and traffic load. This includes digital
devices, a microcontroller, application software and the
various algorithms for energy management.
A great number of algorithms and methods were
proposed to increase the lifetime of a sensor network by
reducing energy consumption according to their
application types, constraints, and network architecture.
Some previous works discuss reliable multipath routing
with energy efficiency [7] and exploiting latency and
density [8]. Other works reduce radio activities as key
management protocols [9] and multiple base stations
[10].
In this paper, we are interested to propose an
Enhanced model of the Ad Hoc On Demand Distance
Vector (EAODV) protocol. EAODV is designed to
reduce the number of control packets, create an historic
routing data base and balance the network load which
result decreasing the energy consumption. Then, we
discuss a comparative study of capabilities offered by
both of AODV and EAODV protocol. Several metrics
are undertaken to evaluate performances of EAODV
protocol.
This paper is organized as follows. Section 2
summarizes the related works. Section 3 introduces the
concept of EAODV protocol and its mechanisms to
decrease energy consumption. Section 4 outlines the
simulation results evaluating the performances of
EAODV protocol compared to AODV. Finally,
section 5 concludes the paper.

2. Related work
Many techniques have been proposed for energy
aware routing. [11] proposed an energy-efficient
neighbor discovery protocol (ENDP) to reduce the need
of network scans by distributing synchronized
information of nodes in two-hop neighborhood. This
information is carried in the beacon payloads of
underlying MAC protocol and utilized for establishing
new communication links. In addition, ENDP
introduces an efficient network beacon signaling
scheme to make network scans more energy efficient.
ENDP can effectively minimize network energy
consumption in dynamic WSNs. The energy efficiency
and operation fidelity are verified by analytical
performance models and experimental measurements
using real WSN prototypes.
[12] proposed a Geographical and Energy Aware
Routing (GEAR) protocol based on energy aware
neighbor selection using localization system such as
GPS device. This protocol delivers packet towards the
target region and restricts flooding algorithm to
disseminate the packet only in the destination region. In
[13], the authors proposed controlling the sink
movements to improve network lifetime. The sink
movements depend on a Mixed Integer Linear
Programming (MILP); an analytical model whose
solution determines sinks position that maximize
network lifetime. The authors define the Greedy
Maximum Residual Energy (GMRE) technique that
heuristically moves the sink from its current location to
a new site where nodes have the highest residual energy.
They also introduce a simple distributed mobility
scheme (Random Movement or RM) which makes the
sink moves uncontrolled and randomly throughout the
network.
Xu and Hong proposed, in [14], a chessboard
clustering scheme to maximize network lifetime. To
achieve good scalability and performance, they
proposed to form a heterogeneous sensor network by
deploying a small number of powerful high-end sensors
beside a large number of lowed sensors.
In [15], a Probabilistic Forwarding Protocol (PFR)
is introduced to combine energy efficiency and fault-
tolerance. In PFR, next hop is always exactly selected;
therefore the number of transmitted packets is
optimized. PFR succeeds with smart dust architecture
since smart dust networks have a very high density and
robustness.
Authors in [16] proposed Variable Transmission
Range Protocol (VTRP) based on varying the range of
data transmissions. This protocol allows the
transmission range to increase in various ways. By
bypassing obstacles or faulty sensors, VTRP has a high
fault-tolerance capability and increases network
lifetime. Varying the range of data transmissions
presents better performance compared to typical fixed
transmission range. This occurs when using low density
network. In this case, a network using fixed range may
cause failure in delivering packets when a routing node
to the sink is not found. VTRP can avoid this situation
by increasing transmission range.
Many improvements for Ad-hoc On-Demand
Distance Vector (AODV) protocol are addressed. In
[17], the authors proposed an improved AODV routing
protocol based on node-grade to reduce the energy
consumption of nodes. G-AODV can reduce energy
consumption using two techniques; (1) avoids any
useless route request (RREQ) packet broadcasted; (2)
provides a path to the packet with fewer resources and
less number of hops. G-AODV assigns a grade to each
node according to the hop distance between the node
and the sink. Therefore, each node doesnt accept any
route request packet from another one having a stronger
grade than itself.
DEAR [18], Dynamic Energy Aware Routing, is an
energy efficient routing algorithm for data querying
sensor networks that propagates routing instructions and
builds data paths by considering both the hop count to
the sink node and the minimum residual energy of that
path. Then, a node selects a data path with low energy
consumption and high residual energy.
3. Enhanced-AODV protocol design
In this section, we present our theoretical
foundations and algorithms to enhance AODV protocol
and improve energy efficiency. Three mechanisms are
involved for this goal. First, a partial diffusion
mechanism divides each nodes neighborhood into
sections to avoid flooding when sending control
packets. The node chooses the most likely section in
which the target could be present according to statistical
and historical data.

Fig 1. Decreasing number of involved nodes.

The second mechanism minimize search packet
lifetime by avoiding random packets, hence a target is
reached with less number of search attempts. The third
mechanism monitors the load distribution in order to
ensure connectivity and concentrates on consuming
energy equitably.
Partial diffusion method
TTL optimization
Charge
distribu-
tion
Sink node
As a result, every mechanism will limit the geographical
area of control packet flooding. Hence, the number of
nodes involved in data dissemination will decrease
(Fig.1).
3.1. Partial diffusion mechanism
EAODV routing protocol is based on the same
principle as AODV. Responding to a request, the
network establishes an instantaneous path to send data.
To discover the path, a certain number of control
packets is required. EAODV proposes a post simulation
of the basic idea in order to reduce the number of
control packets.
When establishing a path, the node sends a packet
(RREQ) to discover the shortest way. The node asks
neighbors whether the destination address exists in their
current routing tables. In turn, these nodes return the
same packet to their neighbors when the address does
not exist in their tables. This packet contains the desired
address (destination), the source node address, the
lifetime of this packet and a sequence number that
serves as a unique identifier of the packet. By mobility
effect of the destination node, this packet may flood the
entire network. RREQ packet is continuously
retransmitted until requested destination is reached,
unless the declared life time of the packet (TTL: Time
To Live) has expired.
In Figure 2, the source tries to send a message to
the destination. It sends a route request (RREQ) to its
three neighbor nodes with time to live equal to 5.

Fig 2. Route request packet.

The cost of this request is 10 packets of RREQ, 4
packets of RREP (Route REPly) and 6 packets of
RERR (Request ERRor), as shown in Figure 3. It
results in 20 packets with 85% of the routing tables are
updated. This reduces the number of discovery packets
by 6% in the future path requests. Figure 3 illustrates
the responses received according to the previous path
request.
If we reduce the number of RREQ (Route
REQuest) packets by requesting only two neighbors 2
and 3, the cost of discovery will be 10 packets. This
reduces 50% of energy consumption for the request
operation but only 45% of routing tables are updated
with the destination path. It also reduces, by 3%, the
number of discovery packets in the future path
requests. The neighbors are not selected randomly but
according to the occurrence of the target node in the
routing tables of neighbors. The source node starts with
neighbors which has the highest number of occurrence.
This method highly reduces the number of search
packets for networks with low node mobility
constraint.

Fig 3. Route request response.

Checking the number of occurrence is not always
efficient because the routing tables may be blank {I
insist on blank not empty} at the launch of network,
or a part of it may be not discovered yet. In this case,
the source randomly selects the neighbors, and then
diffuses packets partially through it. Information
collected in the search task is stored then used in a
future search attempts. This information is about the
energy level of neighbors and the occurrence of
addresses that appear in each routing table.
3.2. RREQ life time: TTL (Time To Live)
Time To Live (TTL) is defined by the number of
hops followed by a control packet while searching the
destination. This number increases when the packet
does not reach the destination, and then the source
restarts the search operation with a new higher number
of hops. In this work, we propose to reduce the search
attempts of the destination by selecting the optimal
number of hops. As a result, the control packet reaches
the destination from the first cycles.
When sending data, EAODV protocol proceeds as
the figure 4 illustrates. TTL value is chosen according
to historical data stored in routing tables; when a node
lunches a search attempt, it selects the last successful
TTL value. If no RREP packet received, the node
selects a higher value of TLL which is the result of an
earlier successful attempt. But the packet can find its
path earlier (figure 5, right path); in this case EAODV
proceeds as explained in what follows.

RREP

RREP

RREP

RREP
RERR

RERR

RERR

RERR

RERR

RERR
RERR

1

2

3

RREQ
RREQ
RREQ
RREQ
RREQ
RREQ
RREQ
RREQ
RREQ
RREQ
RREQ

Fig 4. Discovering mechanism involving TTL.

The neighbor leading to the desired destination
stops routing packets even if the control packet did not
achieve the number of hops. But, this occurs only in the
exact path because the packet completes the number of
hops in case of wrong path as shown in figure 5. For
example, if a node knows that the destination is through
its neighbor A with X hops, it returns the RREP packet
and cuts the dissemination of RREQ packet to its
neighbors, while other RREQ packets complete their
courses in other directions according to its lifetime
value TTL (figure 5).

Fig 5. Request through two neighbors (TTL= 6).

We assign bits representing an integer in the RREP
packet in order to calculate the number of hops toward
the destination. This number is incremented by one hop
and saved at each node on its path back to the source.
Therefore, each traversed node will have an idea about
the number of jumps needed to reach the desired
destination. As a result, every node can assign the
optimal TTL value when sending a discovery request
packet. Then, the new format of RREP is:

Fig 6. RREP packet.

3.3. Load distribution
To ensure good connectivity in a sensor network, it
is essential to keep alive all the links between different
parts of the network. The elements responsible for the
connectivity of the network parts may be similar (flat
network) or nodes with highest hardware performance
(hierarchical network). In this section, we propose
optimizing the management of collaborative work
between the flat network nodes to ensure a distributed
consumption and to share delivering data load between
nodes. When more neighboring nodes are organized to
share the work, the more we avoid redundancy of
packets and losing connections. Using AODV protocol,
redundancy of packets can occur when many
neighboring nodes detect the same stimulus.

Fig 7. Event detection.

Figure 7 shows an event S1 in the monitoring field
of three nodes and S2 in the monitoring field of node N3.
The contribution needed here consists of optimizing the
number of generated and delivered packets as well as
distributing the load for more energy consumption
efficiency. The following sections clarify how nodes
react when detecting event and delivering data.

S2
S1
N1
N2 N3
ID Source Destination Number of hops
Right path
Bad paths
S
D
Node leading to desired destination
Translate data
TTL0<=TTL1;
TTL1<=TTL2
Data to send !
Path discovery attempt

History?
TTL = TTL0
RREP packet
received?

TTL0 = Nb of hops
b c
TTL0 TTL1 TTL2
AODV
proc
Sending data
EAODV
AODV
a
3.3.1. EAODV behavior for event detection

According to EAODV protocol, the nodes transmit
periodically their energy levels to their neighbors with
each Hello message. Each node can classify its capacity
in terms of energy and then can assign a priority order
for each neighbor to deliver the data to the sink node.
Therefore, the node which has the highest energy level
is allowed to deliver data.
We consider ti the time interval between detecting the
event and starting the transmission, and Cn is the class
in terms of energy assigned to the node n. N nodes are
considered as neighbors if each one can see the (N-1)
other nodes. We consider the radius RS of monitoring
field to be less than the range of transmission radius RP.

RS < RP
The process code for event detection is as follows:

1. I F (Cn=1) THEN
{
2. EAODV transmission attempt
3. Broadcast (Hello !)
}
4. ELSE
{
5. Wait (ti* (Cn-1)) AND Listen
6. I F (no transmissions)
{
7. EAODV transmission attempt
8. broadcast (Hello !)
}
9. ELSE
{
10. Wait for(Hello !)
11. Classify new energy data
} END I F
} END I F

Notes.
Cn=1 considers the highest power level (line 1).
Update energy level after transmission activity
(line 3).
Wait and listen if any of the neighbors initiates
transmission (line 5).
Only one node detected the event (line 7).

The node having full power, immediately launches the
transmission of the packet. Otherwise, the node n waits
a period (Cn-1)ti in order to allow another higher class
neighbor to deliver the data. If so, the node waits for the
Hello message broadcasted by the neighbor which
performed the work and announced its new energy level
and then updates its energy level classification.
However, the node is required to deliver the packet and
announces its new power level whether none of the
neighbors has detected the event.
Figure 8 illustrates the activities of four neighboring
nodes where N3 and N4 only detect the same event. E1
to E4 are respectively the energy levels of N1 to N4 in a
decreasing order.

Fig 8. Sequence of event detection process in
EAODV.

3.3.2. EAODV behavior for packet delivery

If routing data can be done through only one path, it
must be through the shortest while load distribution is
maintained to avoid losing connections. EAODV
provides a direct solution as follows. Each node can see
only its neighbors and does not care about what will
happen after providing the data through them. This
creates a problem if traffic has a high priority and it may
decrease the rate of successful delivered packets. But
this problem disappears when network is dense with
nodes.
After determining the possible paths by adopting
the partial diffusion method shown above, the node
provides the packets to the neighbor which has the
highest energy level and admits a link to the destination.
Information about the estimated charge levels are then
transmitted using Hello packets.
4. Simulation environment and performance
evaluation
Several scenarios including fixed destination nodes,
various speeds of mobile sensors and different covered
area are developed to simulate our approach for routing
and energy optimization.

Table 1. Simulation scenarios.

sink
nodes
sensor
nodes
target
nodes
Scenario 1 Fixed
8 moving at
low speed
Mobile
Scenario 2
Mobile
different
speeds
Fixed Mobile

Performance evaluation of AODV and EAODV
protocols are realized following four aspects: (1)
energy consumption estimator model (2) discovering
N1
N2
N3
N4
E1
Sensing and listening to channel

Event
Wait & listen
Sensing and listening to channel

Event

Detection
Wait & listen

Listen
Hello!
Data
receiving

Transmitting
Data
receiving

Data
receiving
E2
E3
E4
2 T
3T
attempts optimization (3) time to live effect
optimization and (4) network survivability.

4.1. Simulation metrics

Many metrics are deployed to evaluate and
compare performances of these protocols.
The number of discovering attempts is a statistical
metric affecting the number of control packets.
EAODV protocol improves decreasing control
packets depending on the speed of nodes and
represented by the improvement rate (Impr%)
(Equation 1). The purpose of this metric is to
specify to which limit of speed the improvements
of EAODV algorithm are significant.

% =
100 Equation 1

Network survivability is tested based on two
aspects. In one hand, this metric reflects the
network lifetime and how faster the nodes die in
full consumption regime. In the other hand, the
percentage of nodes remaining alive reflects the
fairness of energy consuming and the network
capability for maintaining connections. The
lifetime of a sensor node basically depends on
two factors: how much energy it consumes over
time, and how much energy is available for its
use.

4.2. Energy consumption estimator model

We have adopted a purely statistical energy
consumption estimator. At the beginning of simulation,
the power level in each node is initialized to the full
charge by using a real number. We assume that all
nodes have the same hardware architecture and
therefore behave similarly. But, considering the flat
architecture, nodes consumes different energy levels as
they operate in:

Transmission mode.
Receive mode.
Idle mode.
Sleep mode.

In order to update the power level in each node, we
developed a routine responsible to detect both the
starting and the ending times for each operating mode.
As result, the routine measures the period in the
appropriate operating mode which allows calculating
the consumed energy. Then, the remaining power in the
node is updated by subtracting the consumed energy
from the initial power level per cycle.
Ec = (t end t begin) * Pm
Er = Ei Ec
(a)
Equation 2
(b)

Where:
Ec: energy consumed during the current cycle.
Ei: remaining energy after the previous cycle.
Er: remaining energy after the current cycle.
Pm: power level dedicated to the appropriate operating
mode.

The state of the node is permanently updated
whether it is still alive (energy > 0) or exhausted
(energy = 0). In the last case, the node no longer appears
in the topology and the network must adapt to this new
change from the first search request sent by the
neighbors.

4.3. Optimization of the discovering attempts

In AODV protocol, the TTL value is randomly
initialized. Then, it increases if the search result is
negative. At the beginning of simulation, TTL is
dynamically changing due to the virginity of the routing
tables. Table 2 shows the number of attempts for
discovering path every 20 minutes when simulating
AODV protocol.

Table 2. Number of attempts during 2 hours: AODV

Delay Number of attempts
From 0 to 20 mn 724
From 20 to 40 mn 740

After two hours, the total number of attempts
becomes almost constant. With EAODV protocol, we
set up a history-based approach which stores
successively the state of node and topology. Figure 9
shows how the number of discovering attempts evolutes
in both protocols AODV and EAODV for the same
scenario during 120 minutes. AODV and EAODV
behave similarly until minute 70, due to the absence of
historical data in the nodes. However, EAODV allows
optimizing significantly the number of attempts in the
second part of the simulation (after 70 mn).
Consequently, this result leads to save energy in nodes.

Fig 9. Number of discovering attempts.
275
295
315
335
355
375
10 30 50 70 90 110
N
u
m
b
e
r
o
f

d
i
s
c
o
v
e
r
i
n
g

a
t
t
e
m
p
t
s
Simulation time (mn)
AODV
E-AODV
4.4. Time To Live optimization effects
Scenario 1 (see table 1) has the ability to create
dynamic topology needed to test TTL optimization.
Nodes start discovering paths in the network by using
queries across the entire field. A sensor network with
dynamic topology is very greedy in energy. This is due
to the huge number of control packets that will be
periodically transmitted between nodes to discover the
paths. Such mobility decreases EAODV protocol
performances as the node speed increases outlined in
figure 10.
Figure 10 shows the variation of the improvement
rate related to the number of RREQ packets depending
on the speed of sink nodes. A strong constraint of
mobility (see table 1, scenario 2) requires a very recent
history where the age has not exceeded the time set by
the mobile node in order to admit a new neighborhood.
Even if this is feasible, it does not bring significant
improvements in number of discovery packets from
3 m/s. These results depend on the bandwidth, traffic
rates, density of nodes in the field and the required
Quality of Service (QoS).

Fig 10. Rate of improvement in the number of RREQ
packets: EAODV (Mobility constraints).

4.5. Network survivability
In order to increase the rate of traffic and ensure a
full energy consumption regime, we placed targets
moving across the entire field with different speeds.
Section 4.1 has described how calculating the
consumed energy in each sensor. The remaining energy
is therefore deducted easily and the node is dead when
this energy is null. Table 3 specifies when the first and
the tenth nodes are dead and lost from the topology for
both AODV and EAODV protocols starting with
different initial levels of energy. The survivability of the
network with EAODV is significantly improved
especially when comparing the time of the tenth dead
nodes. These protocols behave similarly at the
beginning. Thats why, we obtain the same time for the
first dead node with both protocols. But, the historic
concept introduced by EAODV allows increasing the
lifetime of the network performed by extending the
lifetime of the 10
th
dead node (table 3).

Table 3: Time of 1
st
and 10
th
node dies for several
energy values (minute).

Energy/
node (J)
Protocol
1
st
node
dead in
10
th
node
dead in
25
AODV 23 56
EAODV 25 73
50
AODV 42 119
EAODV 44 148
100
AODV 91 249
EAODV 90 303

Figure 11 shows the percentage of nodes alive
according to a heavy traffic-based simulation scenario.
EAODV allows keeping the totality of the nodes in life
until 270 mn i.e. 75 % of the total network lifetime. But,
the curve shows an abrupt fall after 320 mn, that the
network disappears within a short time from the first
died node.

Fig 11. EAODV/AODV network lifetime.
0
5
10
15
20
25
30
35
0 0.5 1 3 10 20 30
R
a
t
e

o
f

i
m
p
r
o
v
e
m
e
n
t

%
Node speed (m/s)
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
10 60 110 160 210 260 310 360
R
a
t
e

o
f

a
l
i
v
e

n
o
d
e
s
Simulation time (mn)
E-AODV
AODV
(a) (b) (c)
Zone (a): shows similar results for both protocols
because EAODV behave yet as AODV in the majority
of nodes.
Zone (b): shows linear decreasing regime in terms of
alive nodes for the AODV protocol while EAODV
keeps alive the totality of nodes in the network. At the
end of this phase AODV loses 40 % of the total number
of nodes. Thus, the network connectivity may not be
possible and clusters may appear to interrupt the
network operation.
Zone (c): EAODV keeps alive the total number of
nodes until the minute 270 while AODV keeps only
37%. If we assume that the network lifetime
corresponds to 80% of nodes remaining alive, EAODV
network operates 2 times the AODV network lifetime.
However, there is a difference in existing
definitions of sensor network lifetime that are used in
relevant publications. Some of them focus on the
coverage or connectivity rate, while other combine the
coverage and connectivity to form a single requirement
called connected coverage. In many publications, a
specific rate of nodes remaining alive measures the
network lifetime. Figure 12 represents the improvement
realized by EAODV compared to AODV towards a
generic definition of sensor network lifetime based on
different rate of the remaining node alive (i.e.
threshold). As result, EAODV allows higher network
lifetime especially when keeping high the generic
threshold.

Fig 12. EAODV vs AODV network lifetime with
generic definition.
4.6. EAODV vs DEAR
Dynamic Energy Aware Routing (DEAR) is an
enhanced version of Energy Aware Routing (EAR)
protocol which propagates routing instructions and
builds data paths using a Routing Decision Function,
Path Building Process and Path Revising Process. Those
techniques depend on hops count and residual energy,
which make it closer to EAODV in terms of behavior
than any other protocols. DEAR protocol consists of a
single sink node and lots of sensor nodes; thats why we
adopted scenario 2 with single sink node for a
comparative simulation with EAODV. A high data
density is considered to be at the level of DEAR
efficiency.

Fig 13. Comparative test: EAODV vs. DEAR.

Figure 13 shows that the time of first node loss
decreases with nodes density. DEAR protocol is more
efficient in a topology with low density. However,
EAODV acquires better performance with high nodes
density which is the typical topology of WSN.
Moreover EAODV does not show a remarkable
variation with nodes density constraint comparing to
DEAR protocol.
5. Conclusion
In this paper, we presented an enhanced version of
AODV protocol in terms of energy management.
EAODV protocol introduces three new algorithms
allowing (1) minimizing control packets using partial
diffusion method, (2) decreasing RREQ packets life
time and (3) ensuring a distributed consumption in the
network.
According to several evaluation metrics, simulation
results show that EAODV is able to reduce control
packets, keep alive the totality of node during sufficient
period and improve the network lifetime compared to
AODV.
Acknowledgement
This work is supported by the research center of the
college of Computer and Information Sciences King
Saud University.
References
[1] Chong, C.-Y. Kumar, Sensor networks: Evolution,
opportunities, and challenges. Proc. IEEE Vol 91,
No8, 2003, pp.12471256.
[2] Akkarya, K. Younis, A survey of routing protocols
in wireless sensor networks. Elsevier Ad Hoc
Network. Vol3, No3, 2005, pp.325349.
[3] Dressler, A study of self-organization mechanisms
in ad hoc and sensor networks. Elsevier Computer.
Communication. Vol 31, No13, 2008, pp.3018
3029.
[4] Bai, X., Kumary, S., Xuany, D., Yunz, Z., Lai, T.
H., Deploying wireless sensors to achieve both
Coverage and connectivity. In Proceedings of the
7th ACM International Symposium on Mobile Ad
0
50
100
150
200
250
300
350
400
40 50 60 70 80 90
n
e
t
w
o
r
k

l
i
f
e
t
i
m
e

(
m
n
)
Generic threshold - rate of remaining node alive
EAODV
AODV
20
30
40
50
60
70
80
90
100
110
50 100 150 200 300
T
i
m
e

o
f

t
h
e

f
i
r
s
t

n
o
d
e

d
i
e
s
Node density
EAODV
DEAR
Hoc Networking and Computing (ACM Mobihoc),
2006, pp.131142.
[5] Dietrich, I. and Dressler, On the lifetime of wireless
sensor networks. ACM Transaction on Sensor
Networks. Vol 5, No.1, 2009, 39 pages.
[6] J. Hill, et al. System Architecture Directions for
Network Sensors, Proc. of the 9th International
Conference on Architectural Support for
Programming Languages and Operating Systems,
pp. 93104, Nov. 2000.
[7] U.B. Mahadevaswamy, M.N. Shanmukhaswamy,
An Energy Efficient Reliable Multipath Routing
Protocol for Data Gathering In Wireless Sensor
Networks, (IJCSIS) International Journal of
Computer Science and Information Security, Vol.
8, No. 2, May 2010.
[8] C. Schurgers, V. Tsiatsis, S. Ganeriwal, and M.
Srivastava. Topology management for sensor
networks: Exploiting latency and density. In
Proceedings of MOBICOM, 2002.
[9] Sungha Pete Kim Bo-Cheng Charles Lai, David D.
Hwang. Reducing radio energy consumption of
key management protocols for wireless sensor
networks. ACM 1-58113-929-2/04/0008, pp. 9-11,
August 2004.
[10] Rao, S., Gandham, S., Dawande, M., Prakash, R.,
and Venkatesan, S. (2003). Energy efficient
schemes for wireless sensor networks with
multiple mobile base stations. In Proc. IEEE
Globecom, volume 22, pp. 377381, San
Francisco, USA.
[11] Mikko Kohvakka , Jukka Suhonen, Mauri
Kuorilehto, Ville Kaseva, H. Marko, Timo D.
Energy-efficient neighbor discovery protocol for
mobile wireless sensor networks, ACM/Elsevier
Ad hoc Networks, 2007.
[12] R. G. Yan Yu and D. Estrin. Geographical and
energy aware routing: a recursive data
dissemination protocol for wireless sensor
networks, May 2002.
[13] S. Basagni, A. Carosi, E. Melachrinoudis, C.
Petrioli, and M. Z.Wang. Controlled sink mobility
for prolonging wireless sensor networks lifetime.
ACM/Elsevier Wireless Networks, 2008.
[14] X. Du, and Y. Xiao. Energy efficient chessboard
clustering and routing in heterogeneous sensor
networks. International Journal of Wireless and
Mobile Computing, 121130, 2006.
[15] I.Chatzigiannakis, T. Dimitriou, S. Nikoletseas,
and P. Spirakis. A probabilistic algorithm for
efficient and robust data propagation in smart dust
networks. In Proceedings of the 5th European
Wireless Conference on Mobile and Wireless
Systems Beyond 3G (EW 2004), 2004.
[16] T. Antoniou, A. Boukerche, I. Chatzigiannakis, G.
Mylonas, and S. Nikoletseas. A new energy
efficient and fault-tolerant protocol for data
propagation in smart dust networks using varying
transmission range. In Proceedings of the 37th
Annual ACM/IEEE Simulation Symposium
(ANSS 04), 2004.
[17] Fei Tong Wan Tang Li-Mei Peng Rong Xie
Won-Hyuk Yang Young-Chon Kim. A Node-
Grade Based AODV Routing Protocol for
Wireless Sensor Network. In Proceeding of the
Second IEEE International Conference on
Networks Security Wireless Communications and
Trusted Computing (NSWCTC), 2010, pp. 180
183, 2010.
[18] Li-Min Sun, Ting-Xin Yan, Yan-Zhong Bi, and
Hong-Song Zhu. A Self-adaptive Energy-Aware
Data Gathering Mechanism for Wireless Sensor
Networks. ICIC 2005, Part II, LNCS 3645,
pp. 588 597, Springer 2005.

JOURNAUX INTERNATIONAUX

1. Ridha OUNI, Graph splitting based self-organization approach for energy and routing optimization,
International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): pp 738-
744, 2012.
2. Jamila BHAR, Ridha OUNI, Monji ZAIDI, Salem NASRI, A new TDMA-CR model for dynamic
resource allocation in wireless networks, IJCSES International Journal of Computer Sciences and
Engineering Systems, Vol.4, No.1, January 2010.
3. Monji ZAIDI, Ridha OUNI, Jamila BHAR, Rached TOURKI, New approaches reducing handoff
latency in 802.11 wireless LANs, IJCSES International Journal of Computer Sciences and Engineering
Systems, Vol.3, No.3, July 2009.
4. Monji Zaidi, Ridha OUNI, Kholdoun Torki and Rached Tourki, Low power ASIC designs for fast
Handoff in IEEE802.11, International Journal of Computers, Systems and Signals, Vol. 10, No.1,
pp 27-39, 2009.
5. Jamila BHAR, Ridha OUNI, Abdelhamid HELALI, Salem NASRI, Evaluation of Handover protocols
in Wireless ATM networks, Information Technology Journal (Asian Network for Scientific
Information), Vol 6, N2, pp 275-282, 2007.
6. Jamila BHAR, Ridha OUNI, Kholdoun TORKI, Salem NASRI, Handovers strategies challenges in
wireless ATM networks, International Journal of Applied Mathematics and Computer Sciences, Vol 4,
N2, pp 636-641, April 2007.

International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): 738-744
The Society of Digital Information and Wireless Communications, 2012(ISSN 2225-658X)

738

Graph splitting based self-organization approach for energy and routing optimization
Ridha OUNI
College of Computer and Information Sciences, King Saud University, KSA.
rouni@ksu.edu.sa

ABSTRACT
Recent advances in wireless sensor networks
have led to many new protocols specifically
designed for sensor networks where energy
awareness is an essential consideration. Both
self-organization and routing protocols
significantly impact the energy consumption
and therefore the network life time. This
paper surveys recent self-organization and
routing protocols for sensor networks and
proposes enhanced approaches for low duty
cycle and more reliable environment. We
compare some of the relays-based protocols
in a random graph model. Simulation results
will show the impact of a graph splitting
approach on the active node set and the
topology connectivity. Moreover, we analyze
the performance of that approach in terms of
transmitted and received packet rate.

KEYWORDS

WSN, MPR, MPR-CDS, Graph split.

1. INTRODUCTION

Smart environments represent the next
evolutionary development step in building,
utilities, industrial, home and transportation
systems automation. The information needed
by smart environments is provided by
Distributed Wireless Sensor Networks
(WSN), which are responsible for sensing as
well as for the first stages of the processing
hierarchy. The individual devices in a WSN
are inherently resource constrained: They
have limited processing speed, storage
capacity, and communication bandwidth.
These devices have substantial processing
capability in the aggregate, but not
individually, so they must organize
themselves and provide a means of
programming and managing the network as
an ensemble, rather than administering
individual devices.
Energy conservation is one of the most
challenging problems because batteries have
very limited capacities. Two particular
important problems are activity scheduling
and broadcasting. In activity scheduling
problem, some nodes decide to sleep to
preserve the energy, but should have an active
neighbor to collect messages for them or take
over some sensing tasks. In broadcasting
problem, one host needs to send a particular
message to all the other ones in the network.
Broadcasting is applied for publication of
services, alarming and other operations. In a
straightforward solution to broadcasting,
hosts only need to blindly relay packets at
least once to their neighborhood. However,
this leads to the well-known broadcast storm
problem as redundancy and collisions [1,2].
Self-organization protocols are proposed in
literature trying to avoid these problems by
choosing active and connected nodes set
charged of broadcasting and routing [3,4,5,6].
The concept of multipoint relaying [7,8]
consists to reduce the number of duplicated
re-transmissions when forwarding a broadcast
packet. This technique restricts the number of
re-transmitters to a small set of neighbor
nodes, instead of all neighbors, like in pure
flooding. This set is kept small as much as
possible by efficiently selecting the neighbors
which covers (in terms of one-hop radio

739

range) the same network region as the
complete set of neighbors does. This small
subset of neighbors is called multipoint relays
(MPR) for a given network node. The
technique of multipoint relays (or MPRs)
provides an adequate solution to reduce
flooding of broadcast messages in the
network, while attaining the same goal of
transferring the message to every node in the
network with a high probability.
With so many potential applications,
researchers are interested in algorithms to
develop "backbones" within these networks
reliably and quickly. A backbone, more
technically, is a dominating, independent set
in the graph of nodes, meaning that the nodes,
themselves, are out of one another's range, but
would be able to relay information between
intermediate nodes. Several graph models
have been proposed to impact energy but are
connectivity constrained. In this paper, a
graph splitting model will be proposed and
evaluated.
In this work, we propose to split the
entire network into two sub-networks before
applying self-organization protocols to
minimize the dominating set which leads to
reduce the number of retransmission. The
paper is organized as follows. Section 2
presents and analyzes different protocols used
for self-organization especially, MPR and
MPR-CDS (Connected Dominating Set).
While, section 3, develops an optimized
MPR-CDS model based on a graph splitting
approach and evaluates its performances.
Finally, section 4 concludes the paper.

2. WSN SELF-ORGANIZATION
PROTOCOLS

Self-organization allows devices to
recognize their surroundings, cooperate to
form topologies, and monitor and adapt to
environmental changes, all without human
intervention. Subsequently, self-organization
in wireless sensor networks provides a variety
of functions: sharing processing and
communication capacity, forming and
maintaining structures, conserving power,
synchronizing time, configuring software
components, adapting behavior associated
with routing, with disseminating and querying
for information, and with allocating tasks, and
providing resilience by repairing faults and
resisting attacks.

2.1. Flooding

A network consists of many nodes, each
with multiple links connecting to other nodes.
Information moves hop by hop along a route
from the point of production to the point of
use. In WSNs, each node has a radio that
provides a set of communication links to
nearby nodes. By exchanging information,
nodes can discover their neighbors and
perform a distributed algorithm to determine
how to route data according to the
applications needs. A basic capability in such
networks involves disseminating information
over many nodes. This can be achieved by a
flooding protocol in which a root node
broadcasts a packet with some identifying
information. Receiving nodes retransmit the
packet so that more distant nodes can receive
it. However, a node can receive different
versions of the same message from several
neighboring nodes.

Fig.1. Primitive flooding WSN.
Primitive network is based on blind
flooding where the packet is retransmitted by
all the intermediate nodes (Fig.1). It is simple,
Source node
Transmitted packet


740

easy to implement, and gives a high
probability that each node, which is not
isolated from the network, will receive the
broadcasted message. However, it consumes a
large amount of bandwidth and energy due to
many redundant retransmissions.

2.2. Multipoint Relay (MPR)

Many techniques are described in the
literature to reduce traffic flooding in WSNs.
But, each technique is developed for a target
application and characterized by its own
advantages and weaknesses. Here, we will
discuss the multipoint relaying mechanisms
(MPR and MPR-CDS) as possible solutions.
These mechanisms are based on two-hop
neighbors knowledge using HELLO
messages defined by the Mobile Ad hoc
Network (MANET) [9].
HELLO messages are broadcasted to all
neighbors at regular intervals. They contain
information about the neighbors and the link
state. Fig. 2 describes the process for three
nodes A, B and C. Two rounds of HELLO
messages are needed to establish the whole
one-hop and two-hop neighborhood.

Fig.2. HELLO process.
Multipoint relay was presented as a
technique to reduce the number of redundant
re-transmission in the wireless sensors
networks by electing a special node set to
cover the entire network based on the 2-hop
neighbors knowledge. Several rules and
algorithms are proposed for this calculation.
In this paper, we resort to the greedy [7] MPR
set computation described in the algorithm
below.

MPR Algorithm [10]

1. Start with an empty multipoint relays
set.
2. Add nodes which are the only neighbor
of some nodes in the 2-hop neighbors.
3. If there still exist some two-hop nodes
which are not yet covered, compute the
one-hop nodes degrees and choose the
maximum one.
4. Repeat step 3 until all two-hop
neighborhoods are covered.
Fig. 3 shows an example where a
broadcast message is diffused in the network
using the multipoint relays where
7 retransmissions are needed to reach all
nodes.

Fig.3. WSN deploying MPR protocol.

Fig.4. WSN deploying MPR-DS protocol.
Dominant node Source node
Transmitted packet
MPR node Source node
Transmitted packet

A
1

B
3

C
2
Hello (Empty)
Hello (Empty)
Hello (A, C) Hello (A, C)
1-hop: A,C
2-hop:
1-hop: B
2-hop: A
1-hop: B
2-hop: C

741

2.3. Multipoint Relay Connected
Dominating Set (MPR-CDS)

Adjih and al. proposed a novel extension
of the MPR to construct a small CDS source
independent using two simples rules based on
the node ID and the greedy algorithm [5]. In
the present work, the node is selected into the
dominating set of the network if:
It has the largest degree than all its
neighbors.
It is a multipoint relay selected by its
neighbor with the largest degree.
The multipoint relays are selected using
the greedy algorithm. Nodes with highest
degrees calculate their relays and call them to
join the CDS at the third round of HELLO
messages. Fig. 4 shows the MPR-CDS
application for a wireless sensor network
composed by 12 nodes where five
retransmissions are required to reach the
entire network.

3. CDS OPTIMIZING

Bridging the gap between the hardware
technologys raw potential and the broad
range of applications presents a systems
challenge. The network must allocate limited
resources to multiple concurrent activities,
such as sensing, processing, network
supervision and data communication. The
potential interconnections between devices
must be discovered and information routed
effectively from where it is produced to
where it is used. In this section, a graph
splitting method is developed to meet these
requirements and constraints.

3.1. Graph creation

The graph creation algorithm is an iterative
process which creates n vertices with random
x and y coordinates between 0 and 1. When
working with a square, all coordinates within
[0,1] are acceptable. We compute the
Euclidian distance from the origin of the
square in order to know whether or not a
coordinate is within the acceptable bounds of
that area (square). An efficient method to
connect the nodes is to divide the graph into
smaller pieces and connect the nodes only
within these pieces or "cells". An intelligent
graph division can minimize the number of
connections which would overlap between
"cells" and save even computation.

3.2. Graph splitting

Let G = (V,E) be a graph with a set of
nodes V and a set of edges E. The number of
nodes in G is denoted by N. The degree d(u)
of a node u is the number of edges adjacent to
u (number of neighbors). L and l are the
length and the width of the random graph,
respectively. All nodes have the same
coverage radius R.
We define p as:

The setting of ensures that
generated graphs are connected with high
probability [10,11]. In these simulations, we
study the impact of p on the CDS size. For
N=200, l=200 and R=80 and a variable L.
The simulation results of the MPR-CDS are
shown in the figure 5.
0 2 4 6 8 10 12 14 16
x 10
4
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Graph surface
%

o
f

d
o
m
i
n
a
n
t

Fig 5. The CDS for the MPR-CDS in a variable graph
Length.


742

We observe that the percentage of dominants
decreases with the length because p increases
and therefore the probability of graph
connectivity increases too. We conclude that
by reducing the length of the graph, we
optimize the CDS set. Therefore, we will
focus, in the rest of this paper, on the impact
of the global graph splitting into many
elementary sub-graphs.

3.3. Rank split

We assume that during HELLO message
process, nodes are able to calculate their
ranks defined as follow:
Definition: The rank of node u, rank(u), is
the minimum number of hops between source
node and node u.
The source node (sink node) has the rank
zero. HELLO packet contains the number of
hops between the node and the sink. The
nodes will receive many distances and they
have to take the minimum. The MaxRnk is
the greatest rank in the network.
We split the global graph G into two sub-
graphs G1 and G2. For each node u:

Fig.6. Graph decomposition concept.
Applying the MPR-CDS for G1 (CDS1)
and G2 (CDS2). The CDSs are connected
while the global CDS for the graph G is not
connected. We propose to create a new sub-
graph Gi (CDSi) in the interface of G1 and
G2 to connect the CDSs. The global CDS
becomes:

3.4. Simulation environment

To simulate the graph splitting approach,
a simulator was developed in MATLAB. It
generates a random topology in a rectangular
area with L=1000, l=200 and R=80. The
global graph connectivity is checked by a
simulation of blind flooding. CDS
connectivity check uses the same process as
global graph where only the dominants have
the right to diffuse. The delivery rate (DR)
describes the number of nodes which receive
at least one copy of message used in the
connectivity checks. The simulation process
is described by the following flowchart:

Fig.6. Simulation flowchart.

3.5. Results analysis

Figure 7 shows the simulation results of
100 connected graphs which have been
decomposed with ranks. For low density
(N < 100), the probability to keep DR = 100%
is around 50%. While, for N 100 nodes, this
probability increases substantially reaching
Rank = 0
Rank = 1
Rank = m
Rank = m+1
Rank = n
G1
G2
Gi
Graph generation G
DR = 100% ?
CDS = MPR-CDS(G)
CDS
DR = 100% ?
CDS' = MPR-CDS(G1,G2,Gi)
Data Gathering
No
No
Yes
Yes

743

more than 95%. Also, we note that during the
splitting process, a risk to create isolated
clusters appears for low density graphs. This
risk decreases with the increasing of the
number of deployed nodes.
60 80 100 120 140 160 180 200 220 240
0
10
20
30
40
50
60
70
80
90
100
Number of nodes (N)

G
r
a
p
h

(
%
)
DR=100%
DR<100%
Isolated clustering

Fig.7. Connected Graphs face to decomposition.
Figure 8 shows the difference between
the MPR-CDS with and without graph
splitting. The CDS size is performed after the
graph splitting. This difference is due to the
reduced graph area and to the nodes which
dont consider many of one-hop and two-hop
neighbors in the CDS calculation.
200 205 210 215 220 225 230 235 240 245
60
70
80
90
100
110
120
130
140
150
160
Number of nodes (N)
N
u
m
b
e
r

o
f

d
o
m
i
n
a
n
t
s
MPR
MPR-CDS
Split-MPR-CDS

Fig.8. Dominants set for MPR, MPR-DS and sub
graphs with MPR-DS.
The CDS set is responsible to control the
network. It allows transmitting and
forwarding both traffic and control messages
[9] to reduce the number of passive listening
(while not concerned). Figure 9 shows the
number of received message in the entire
network per cycle assuming that only one
node sends a message every cycle.
Consequently, the received messages are
reduced by 40% when splitting the graph and
applying MPR-CDS compared to the classic
MPR-CDS.
200 205 210 215 220 225 230 235 240 245 250 255
500
1000
1500
2000
2500
3000
3500
Number of nodes (N)
N
u
m
b
e
r

o
f

r
e
c
c
e
i
v
e
d

m
e
s
s
a
g
e
s
MPR-CDS
Split-MPR-CDS

Fig.9. Received messages per cycle.

4. CONCLUSIONS

This paper introduced a new concept to
optimize the MPR-CDS by splitting the
network graph into many elementary sub-
graphs. It reduces the dominant set in order to
minimize the number of retransmitted packet
in the network. The idea has been simulated
using different scenarios and comparisons
with MPR and MPR-CDS are established.
The results confirm that an amount of
dominants may be removed with connectivity
guarantee. The new approach is more
efficient for dense networks that are
composed by hundreds or even thousands
nodes.

REFERENCES

1. Rahim Kacimi, Technique de conservation
dnergie pour les rseaux de capteurs sans fils,
PhD thesis, Polytechnic of toulouse, 2010.
2. Ozan K.Tonguz, N. Wisitpongphan, Jayendra S.
Parikh, F. Bai, P. Mudalige, and Varsha K.
Sadekar, On the broadcast storm problem in ad
hoc wireless networks, Broadband
Communications, Network and Systems, 2006.

744

3. Karel Heurtefeux and Fabrice Valois, self-
organization protocols: behavior during the sensor
networks life, The 18th Annual IEEE
International Symposium on Personal, Indoor and
Mobile Radio Communications, 2007.
4. Jie Wu, Wei Lou and Fei Dai, Extended
Multipoint relays to determine connected
dominating sets in MANETs, IEEE transaction
on computer, vol. 55, NO. 3, MARCH 2006.
5. Xiao chen and Jian Shen, Reducing connected
dominating set size with multipoint relays in ad
hoc wireless networks, Proceed. of the 7th
international Symposium on parallel architecture,
algorithms and networks, 2004.
6. Rajiv Misra, On self-stabilization of multipoint
relays for connected dominating set in ad hoc
networks, IEEE region 10 conference, TENCON
2009.
7. Yongsheng Fu, Xinyu Wang, Wei Shi and
Shanping Li, Connectivity based greedy
algorithm with multipoint relaying for mobile ad
hoc networks, The 4th international conference
on mobile ad-hoc and sensor network, 2008.
8. Cedric Adjih, philipe Jacquet and Laurent
Viennot, Computing connected dominated sets
with multipoint relays, Technical report, INRIA,
Oct.2002, www.inria.fr/rrrt/rr-4597.html.
9. T.Claussen, P.Jacquet, C. Adjih, A. Laouiti, P.
Minet,P. Muhlethaler,A. Quayyam and L. Viennot,
Optimized link state routing protocol (OLSR),
RFC 3626, Oct. 2003, Network Working Group.
10. C. Bettestetter, On the minimum node degree and
connectivity of a wireless multi-hop network, in
Proc. ACM MobiHoc, Lausanne, Switzerland, Jun.
2002.
11. S. Crisostomo, J.Barros and C. Bettstetter,
Flooding the network: Multipoint relays versus
network coding, Circuits and Systems for
Communications, 2008.

IJCSES International Journal of Computer Sciences and Engineering Systems, Vol.4, No.1, January 2010
CSES International 2010 ISSN 0973-4406
1

A NEW TDMA-CR model for dynamic resource allocation in
wireless networks
Jamila BHAR, Ridha OUNI, Monji ZAIDI and Salem NASRI

Electronic and Micro-Electronic laboratory (EE)
Faculty of Sciences of Monastir (FSM), Tunisia
E-mail: jamilabhar@yahoo.fr

Abstract
Wireless networks architecture contains mobile terminals
randomly placed in cells managed by different APs (Access
Points). Terminals are considered node generating different
traffic types according to specific service classes. Supporting
heterogeneous traffics over AP poses many questions about
medium access and resource allocation. In this way, the present
work study and analysis these constraints and proposes a novel
technique for resources distribution. This technique, called
TDMA-CR (Time Division Multiple Access-Compensation
Reward) is designed at the mac layer of the AP. TDMA-based
systems are considered the efficient access solution for resources
allocation in wireless networks. A proposed TDMA/FDD-based
mechanism is designed by a generic model in which time on the
uplink and the downlink channels are divided into adjacent series
of fixed-size TDMA frames. Each frame is further subdivided
into a fixed number of slots to be dynamically allocated for
different service classes (CBR, VBR). In this context, the
solution may provide the ability to support multiple traffic types
and to process them according to generic parameters. The basic
idea is to provide slots reassignment, and to dynamically adjust
connection parameters based on signalling information
processing approach. This approach, based on resources
compensation-reward, performs the WCAC (Wireless Call
Admission Control) and gives solution to ameliorate link rate and
traffic conditions. Simulation shows that it achieves optimal
resources allocation, low connection reject probability, especially
for CBR connections, and resources degradation avoidance for
VBR and ABR connections, in comparable with TDMA
technique.

Key words: Wireless networks, TDMA, Resources allocation,
performances evaluation.
1. Introduction
TDMA (Time Division Multiple Access) allows multiple
users to access on a single frequency channel without
interference by allocating time slots to each user. TDMA
shares the available bandwidth in the time domain. Each
frequency band is divided into several time slots
(channels). A set of such periodically repeating time slots
is known as the TDMA frame. Each terminal is assigned in
those slots. In addition to increasing the efficiency of
transmission, TDMA offers the advantage to be easily
adapted to the transmission of data as well as voice
communication. There are different MAC schemes
proposed in the literature to improve channel access based
on TDMA mechanism. [3,9,11,13,17] give more details
about different MAC frames structure and allocation
strategies.

In this paper, we are interested to evaluate a frame
reservation strategy allowing efficient transmission of
multi-service traffic over TDMA/FDD channels in wireless
networks. This scheme is based on a dynamic bandwidth
allocation model for connections carrying different types
of traffic. The model strategy is to reserve bandwidth
(which changes dynamically) for each type of traffic during
each frame-time. The distribution of bandwidth on the
corresponding VCs depends on parameters of each traffic
type. Compared to classic TDMA mechanism, the
simulation results using TDMA based on a compensation-
reward model (TDMA-CR) shows more efficient resources
allocation.

Performances of TDMA-CR model are evaluated in terms
of connection reject probability and traffic load, for
various data traffic scenarios. The proposed MAC protocol
uses fixed duration frames. It controls terminals according
to traffic service class. The MAC mechanism requires
negotiation of connections parameter. Unfortunately, the
efficiency of such scheme can significantly decrease when
the number of terminals to be served is large and/or their
rates are high. Therefore, our principal goal, in this paper,
is to increase the traffic load and ameliorate QoS. Even if
the MAC protocol has no resources to satisfy connection
request, it can compensate slots from others connections.
These last can re-establish data transmit with their pic cell
rate in next frames duration. This approach makes then
priorities for different connections corresponding to
service classes.


2

The paper is organised as follows. Section 2 presents the
protocol aspect of TDMA scheme and invokes the WATM
architecture. Section 3 gives details of the proposed
algorithm and the experimental design. Regarding the
impact of resource allocation procedure we present our
simulation model and the derived performance results.
Finally, we present some concluding remarks.
2. TDMA scheme Description
Typical dynamic TDMA protocol is always selected for
resources allocation in wireless networks. It is suitable to
provide QoS for real-time multimedia traffics [9]. With
TDMA, the bandwidth is distributed using time-slot
allocation according to the service classes and leads to link
scheduling.

time
Fixed frame duration (cycle)
. . . . . .

Frame
header
Signalling
mini-slots
Dynamic allocated
ABR, VBR and
UBR slots
Fixed
allocated
CBR slots
Idle slots

Fig. 1 Dynamic TDMAUPLINK access control frame format.
As shown in figure 1, the TDMA frame is fixed and
divided into three sessions. They consist of a frame header,
a signalling session, a data transmission session and an idle
session. The header transport information for dynamic
access control synchronisation. Data transmission session
is variable because source terminal doesnt always have
data to send. The number of data slots allocated for each
connexion depends on the characteristics of the service
class. Four service classes (CBR, VBR, ABR and UBR)
are defined according to specific traffic and QoS
parameters. In this paper, we assume that the slot size is
equal to a WATM cell. When a signalling session
achieved, the AP knows all the terminals that have data to
transmit and calculate the slots number to be assigned for
each connexion. Data transmission session is composed by
dynamic allocated VBR, ABR and UBR slots and fixed
allocated CBR slots. Finally, the idle session is proceeded
when there are no data to send after that a new TDMA
frame begins.
3. Target architecture
Asynchronous transfer mode (ATM) based technology can
provide high speed wireless multimedia communications.
In fact, the fine-grain multiplexing provided by ATM due
to the fixed small cell size is well suited to slow-speed
wireless links since it leads to lower delay jitter and
queuing delays [1]. The wireless ATM protocol
architecture incorporates wireless access and mobility
related functions into the standard ATM stack. A high
speed and low complexity wireless access technique is
crucial for providing bandwidth-on-demand multimedia
services to mobile terminals. Typical target bit rates for the
radio physical layer of wireless ATM are around 25 Mbps.
A modem must be able to support burst operation with
relatively short preambles as well as short control packets
and ATM cells [10].

For efficient sharing the available wireless bandwidth
between multiple wireless terminals, a radio MAC layer is
required. A novel TDMA approach is adopted for medium
access control where several virtual circuits are
multiplexed in a single radio channel. The TDMA frame
structure supports constant bit rate (CBR), available bit
rate (ABR), variable bit rate (VBR) and unspecified bit
rate (UBR) services within each access point transmission
cell area.

Wireless
Control
+
Signaling
W-Data Link
Control
W-Medium Access
Control
W-Physic
control

Standard ATM
Physical Layer

User service
ATM Network Layer
ATM Adaptation
Layer
User service
ATM Network Layer
ATM Adaptation Layer
W-Physic
control
W-MediumAccess
Control

Wireless
Control
+
Signaling
W-Data Link
Control
WATM
NNI
ATM-Switch
Wireless
Terminal
Access
Point

Fig. 2 WATM networks architecture.
4. Design model
4.1 Wireless Terminal MAC Model
The Wireless Terminals (WT) generates CBR, VBR, ABR
and UBR traffic models. The ATMFs Traffic

3

Management specification defines four cell-based traffic
parameters namely the Peak Cell Rate (PCR), Sustainable
Cell Rate (SCR), Maximum Burst Size (MBS) and
Minimum Cell Rate (MCR) [13]. The PCR is a maximum
rate at which the user may transmit cells. Its inverse, the
minimum cell inter arrival time (1/PCR), may be easier to
measure in practice and it is useful to evaluate network
performances. The SCR is a possible average rate for
an ATM connection. The average rate is the number of
transmitted cells divided by the connections duration.
For ideal Constant Bit Rate (CBR) traffic, the PCR equals
the SCR. For Variable Bite Rate (VBR) traffic, the SCR is
typically less than the PCR.

For CBR VCs, slots are allocated according to their
required bit rates. A CBR traffic convention includes the
PCR and the Cell Delay Variation Tolerance (CDVT)
factors. For VBR source model, we consider an on-off
that transmits a number of cells at its SCR. Then, slots
allocated to different sources depend on traffic parameters
(Table 1).

Table 1: Service classes parameters
Service
class
Traffic parameters QoS parameters
CBR PCR CLR+CDV+CDT
VBR PCR+SCR CLR+CDT
ABR PCR+MCR
UBR

4.2. Access Point MAC model
The crucial networking algorithm is placed at the AP. It
includes receiving (data/signalling) packet, FCFSs
queuing, resources managements, etc. Hence, signalling
and data WATM cells are multiplexed and processed
according to resources allocation scheme. The AP controls
the uplink bandwidth allocation for WATM cells from
each WT, taking into account the number and the type of
active connections and their bandwidth requirements.

The medium bandwidth of WATM networks is divided
into two separate channels: uplink and downlink. The
uplink channel transfers information from WT to the AP.
Each channel is further partitioned into several sub frames,
carrying different classes of traffic. A set of buffer per-VC
cell scheduling schemes are used as first-come first-served
(FCFS) (figure 3). The FCFS cell-scheduling algorithm
could be easily hardwired with low cost, however it is
efficient only for homogeneous traffics. Consequently, in
order to meet this weakness, several weighted FCFSs are
allocated for different service classes queuing.

The radio resource manager, located at the AP, takes part
in the connection admission control (CAC) process for a
WATM terminal originated or terminated connection. It
performs the wireless connection admission control
(WCAC) and bandwidth allocation for ATM connection
over the radio interface.
The scheduler meets problems related to the allocation of a
limited amount of shared resources (buffer memory and
output port bandwidth) to support all users, applications
and service classes. It allows managing access to a fixed
amount of output bandwidth by selecting the next cell
which will be transmitted on a port [15]. CDV parameter
can be controlled since queuing data cells is required at
AP.

ATM
Interface
Data Cells
Scheduling Module
Queues (FCFS)

(2)
(n)
(1)
Signalling cells
Reception
Emission

Control Unit

.
.
.
.
.
.
Signalling module
MAC level Physical level

Fig. 3 AP radio medium architecture in WATM.
5. TDMA-CR proposal for dynamic resource
allocation
In our proposal, radio spectrum is divided into time slots
which are assigned to different connections. User
applications can send data only in their dedicated slots.
Due to the FDD duplexing technique, integrating the MAC
protocol, two distinct carrier frequencies are used for the
uplink and downlink channels. Results given here are
performed with symmetric traffic in both directions
(between AP and WT). The AP scheduler allocates the
same number of slots for uplink and downlink channels.
Uplink includes signalling mini-slots, followed by
allocated CBR, ABR, VBR and UBR data slots. The
signalling session allows the wireless terminals sending
their bandwidth requirements to the access point.
Terminals keep their radios on, since every signalling
session of the TDMA frame.

In fact, a WT attempting to communicate with another, it
sends firstly a connection request message to the AP.
According to the required QoS, the AP assigns adequate

4

number of time slots for this connection. Network
parameters are manipulated with an algorithm based on
characteristics typed ATM service. If there are not enough
slots for the request, a connection can not be established
with the required QoS guarantees. However, AP tries to
adjust resources through a compensation-reward
mechanism applied to all active connections. This
mechanism, called TDMA-CR, attempt to avoid reject of
new connections, especially CBR service class.

The proposal mechanism consists of rewarding time slots
from active ABR and VBR connections, without degrading
their performances, in order to serve a new connection to
be established. This mechanism integrates computational
steps in order to offer the desired resources in an optimal
rewarding way distributed on all the active service classes.
In other terms, ABR, UBR and VBR connections can
reward some of theirs slots to satisfy requirements (or
some of them) of new connections. Thereafter, time slots
of liberated connections will be used to compensate time
slots offered by those connections. Then, time slots of
TDMA-CR frame are actualized whether a new event
occurs (connection or disconnection). However, when
requirements are not satisfied, connection request is
rejected. Due to their strict requirements, reject probability
for CBR class could be higher than the others classes.
Moreover, ABR, UBR and VBR connections can start with
a lower rate which may increase during next frames. Figure
4 shows a finite state machine of signalling process.

Yes Yes
Signaling
packet in
wait
Start
No
Disconnection
packet

Disconnection notification
and parameters actualisation
No
Yes
Connection
Packet

Extract and save parameters
of WT (slots number, service
class, MVCI, ..

Fig. 4 Signaling process steps.
TDMA-CR approach starts to identify the type of request
messages. Disconnection request allows actualizing
available resources. However, a connection request
requires activating the corresponding VCs, in the AP, and
storing the reserved parameters. Second, the AP estimates
the resource needed by WT in order to decide, over a
calculation step, whether there are sufficient resources to
establish this connection (Figure 5). Finally, The AP
returns, to all the terminals, signalling cells describing
them the time slots allocation (number and position in the
frame). WT sends data to the AP within the allocated time
slots. If it has no data to send, the terminal operates in idle
mode.

Yes
Yes
.
.
No
No
No
Yes
Yes
No
CBR
service
Sufficient
remaining
slots

No
Yes Sufficient
remaining
slots
Connection reject
Start
Compensate needed
slots from others
connections (UBR,
ABR and VBR)
Sufficient
compensated
slots
VBR
service
Compensate needed slots:
from others connections
(UBR, ABR and VBR)
with minimal resources

Accept a connection and
send a notification

Fig. 5 Compensation-Reward protocol proposal.
6. Simulation results and performance
analysis
Different extension schemes of TDMA are proposed in the
literature for a fair and efficient operation of the MAC
protocol. Research strategies focus on resolving difficulties
to distribute carefully AP resources between WTs.
Difficulties are related to various traffic conditions like
buffer occupancy, connection parameter requirements, etc..
. Various solutions are proposed in the literature
[6,9,11,13,17]. [11] proposes that the access point traits
the average queue size. Then, information about wireless
terminals queues must be sent in signalling packets over a
special short control slots. It will be useful to distribute the
suitable number of slots for each WT. We mention also
that a scheduling technique, used for multiplexing
terminals data, impacts the average buffer queue.


5

Another challenge studied in this context takes part to
improve handoff performances. Then, buffer management
and optimal reservation of radio resources allows
ameliorating handoff efficiency. Many admission control
strategies have been discussed in the literature to give
priorities to handoff requests compared to the new
connection requests as shown in [3,5]. When a mobile
terminal moves from one area to another, the new AP area
should provide sufficient resources to this handoff
connection. Because, the premature termination of
established connections, due to insufficient resources, is
usually more objectionable than reject new connection
request.

In this work, we consider frame duration of 2 ms with a
rate of 25 Mbps. We evaluate the proposed protocol to
ameliorate some traffic conditions. Figure 6 shows the
proportionality between output and input flows for
different channel utilization. Channel utilization is defined
in [1] as the ratio of the number of slots allocated for
WATM data cells to the total available slots. This figure
depicts the influence of the service class on the output data
rate. CBR traffic is particularly served with a higher flow
ratio than VBR traffic. This is explained by a priority
given by TDMA-CR to CBR sources. These lasts can
support applications with strict temporal constraints (Real
Time).
0
10
20
30
40
50
60
70
80
15 30 50 70 85
input (%)
o
u
t
p
u
t

(
%
)
CBR
VBR

Fig. 6 Traffic Variation in function of service class.
Resource allocation protocol is useful to efficiently
distribute available bandwidth to active connections.
TDMA-CR performs to satisfy QoS parameters for each
connection and to increase the percentage of resources
utilisation as shown in figure 6. The resources utilisation
improvement explains that compensation approach avoid
wasting resources which allow increasing the number of
served connections. Consequently, it decreases the reject
probability, using the remainder and the compensated
resources to be available for connecting new terminals.

TDMA-CR is, then, a resource allocation protocol witch
affect the connection reject probability. Figure 7 compares,
for the same considered traffic scenario, a reject
probability between classical TDMA and TDMA-CR
schemes. We mention that the large gap between curves
during the first period explains the compensation
efficiency introduced by TDMA-CR. This is due to the
resource availability which could be compensated between
connections. In the remained simulation time, the gap of
reject probabilities becomes small. This means that
resources become limited or a big number of terminals
need to be served. However, the average reject probability
is significantly lower for the TDMA-CR scheme compared
with TDMA scheme.
0
0,02
0,04
0,06
0,08
0,1
0,12
0,14
0,16
0,18
0,2
0 1 2 3 4 5 6 7 8 9
time (s)
R
e
j
e
c
t

p
r
o
b
a
b
i
l
i
t
y
TDMA
TDMA-CR

Fig. 7 Connection reject probability.
Presented results show that a compensation mechanism
does not decrease a reject connections number only, but it
also increase the percentage of resources utilisation as
shown in figure 8. This figure outlines a traffic variation of
22 terminals equally distributed between CBR and VBR
sources. Simulation results show that TDMA-CR schemes
gives more than 25% of band regarding to TDMA. The
improvement of resources utilisation explains that
compensation approach avoid resources wasting and
increase a number of served connections.
0
10
20
30
40
50
60
70
80
90
1 2 3 4 5 6 7 8
time (sec)
%
i
n
p
u
t
TDMA
TDMA-CR

Fig. 8 Resources utilization percentage.


6

In this work, we have considered a priority aspect between
HandOver connection and new connection and a priority
according to a service class category (CBR, VBR). Figure
9 shows that reject probability depends on a proportion of
CBR connections. CBR service class is a unique class that
doesnt accept compensation of its resources. In
consequence, connexion CBR requests must be served
according to their traffic parameters or it is rejected. As
shown in figure 9, CBR connections presents minimal
reject probability when they are all established. In this
case, it is possible to use resources compensation (minimal
CBR traffic percentage). Otherwise, reject probability
increase because when CBR traffic is dominant there will
be little resources reserved for others classes to make
compensation.
0
0,01
0,02
0,03
0,04
0,05
0,06
0,07
0,08
0,09
15 30 50 70 85
CBRtraffic (%)
R
e
j
e
c
t

p
r
o
b
a
b
i
l
i
t
y

Fig. 9 CBR connection reject probability.
TDMA-CR is also designed to ameliorate HandOver
performances maintained by a transparent transition
between APs. In this way, it gives priority for HandOver
requests compared to the new connection requests. As a
result, the HandOver reject probability is notably
decreased (figure 10). Moreover, disconnecting Handover
terminal in order to serve a new connection is usually not
preferred such as explained in [3,4,15].
0
0,02
0,04
0,06
0,08
0,1
0,12
0,14
0,16
0,18
0,2
0
,
0
6
0
,
5
5
1
,
0
5
1
,
5
6
2
,
0
8
2
,
5
8
3
,
0
8
3
,
5
8
4
,
0
8
4
,
5
8
5
,
0
8
5
,
5
7
6
,
0
7
6
,
5
8
7
,
0
7
7
,
5
8
time (sec)
r
e
j
e
c
t

p
r
o
b
a
b
i
l
i
t
y
Total connections
Handover connections

Fig. 10 Reject probability of HandOver connections.
The Cell Delay Variation (CDV) is a sensitive QoS
parameter which depends on the type of service class.
CDV should take on almost constant value especially for
CBR classes. Figure 11 presents the CDV of CBR
connections characterized respectively by PCR of 500
kbps and 1 Mbps. TDMA-CR minimizes CDV fluctuations
which allow ameliorating QoS to be offered to mobile
terminals and meets application delay constraints. It gives
a sufficient resources distribution strategy that arranges
fairly all the connections.
1,47
1,49
1,51
1,53
1,55
1,57
1,59
1,61
0
0
,
3
1
0
,
6
1
0
,
9
2
1
,
2
2
1
,
5
3
1
,
8
3
2
,
1
4
2
,
4
5
2
,
7
5
3
,
0
6
3
,
3
6
3
,
6
7
3
,
9
7
4
,
2
8
4
,
5
8
Time (ms)
C
D
V

(
s
)
CBR(500kbps)

0,69
0,71
0,73
0,75
0,77
0,79
0,81
1
,
3
6
1
,
6
1
,
8
3
2
,
0
7
2
,
3
1
2
,
5
4
2
,
7
8
3
,
0
1
3
,
2
5
3
,
4
9
3
,
7
2
3
,
9
6
4
,
2
4
,
4
3
4
,
6
7
4
,
9
Temps (ms)
C
D
V

(
s
)
CBR (1Mbps)

Fig. 11 CDV variation of CBR connections.

7

The overhead is defined by signalling messages
introduced by TDMA-CR to insure compensation and
negotiation of desired resources. Signalling messages are
automatically and proportionally increased according to
network traffic conditions and active terminals number
(figure 12).

0
2
4
6
8
10
12
14
16
18
20
15 30 50 70 90
traffic level (%)
O
v
e
r
h
e
a
d

(
%
)

Fig. 12 Overhead variation in function of traffic percentage.
7. Synthesis results
7.1 Layout
Resource allocation process necessitates functions to the
extraction and the manipulation of traffic parameters. It
needs also functions to compute desired resources
according to saved parameters. Complex models including
arithmetic operators as addition, division and
multiplication are, then, required to be employed in the
resource allocation process.
Physical conception and verification present significant
steps for integrated circuits conception procedure. Figure
13 shows a layout of the obtained resource allocation
circuit. It indicates the complexity level of this circuit.
Regarding to circuits complexity, some routine are applied
to give fan-out trees of interconnections and generate
clock trees for adequate distribution.

Fig. 13 Layout in silicon level.
7.2 Clock Distribution
The quality of the clock Distribution in a circuit plays a
significant role in the performances of a synchronous
circuit. The majority of numerical applications are
implemented in synchronous logic, because the current
tools for synthesis do not allow the automation for every
design. These tools allow the automation for design with
combinative or sequential descriptions which rest on one
or more clock. It is vital for a certain implementation that
clock must be known and fixed in all the physical circuit,
and its geometrical propagation does not imply distortion
and dephasing.

The solutions to guarantee a uniform clock distribution
without skew dephasing are multiple. The best solution
used today to reduce the clock dephasing is based on the
concept of clock tree, it inserts on each level of hierarchy,
buffer or reverser which rectifies the clock signal locally.
This solution has the advantage of producing a
geographically distributed and optimized consumption.
Figure 14 presents clock trees of resource allocation
circuits designed to be integrated in the APs. Clock trees
quality play an imported role to increase synchronous
circuit performances.


8

Fig. 14 Clock tree in silicon level.
8. Conclusions
The MAC level should provide QoS management based
bandwidth allocation to multiple traffic classes. In fact,
various applications such as voice, video and high quality
multimedia services need QoS guarantees in order to
ensure delay constraints, prescribed data rate and loss. In
this paper we have evaluated an algorithm to distribute the
resources provided by the access point (AP) at the MAC
layer in wireless networks. We consider a classical
TDMA/FDD mechanism. It is designed by a generic model
in which time on the uplink and downlink channels are
divided into a contiguous sequence of fixed-size TDMA
frames. Each frame is further subdivided into a fixed
number of slots to be dynamically allocated to different
ATM traffic classes: CBR, VBR, ABR, and UBR.

This approach uses a compensation-reward process called
TDMA-CR. It controls all signalling information of
connected terminals in order to compute the suitable
allocation of slots according to the service classes.
Simulation results show the efficiency of such
compensation-reward approach to improve traffic
conditions.

References
[1] Ekram Hossain and Vijay K. Bhargava; Link-Level Traffic
Scheduling for Providing Predictive QoS in Wireless
Multimedia Networks; IEEE transactions on multimedia,
Vol 6, N1, Feb 2004.
[2] Xiaohua Li; Contention Resolution in Random-Access
Wireless Networks Based on Orthogonal Complementary
Codes; IEEE transactions on communications, Vol 52, N1,
Jan 2004.
[3] P. Ramanathan, K. M. sivalingam, P. Agrawal and S.
Kishore; Dynamic resource allocation schemes during
handoff for mobile multimedia wireless networks; IEEE
journal on selected areas in communications, Vol 20, July
1999.
[4] Bih-Hwang Lee Hsin-Pei Chen Su-Shun Huang; Dynamic
Resource Allocation for Handoff in WATM Networks; 11th
International Conference on Parallel and Distributed
Systems (ICPADS'05) IEEE 2005.
[5] Maria C. Yuang, Po L. Tien, and Ching S. Chen; A
Contention Access Protocol with Dynamic Bandwidth
Allocation for Wireless ATM Networks; p149-153, IEEE
2000.
[6] P. Narasimhan, S.K. Biswas, C.A. Johnston, R.J. Siracusa &
H. Kim; Design and Performance of Radio Access Protocols
in WATMnet, a Prototype Wireless ATM Network; Book:
Advances in Wireless Communications; book series: The
International Series in Engineering and Computer Science,
Vol. p 435, 2002.
[7] Tarek Bejaoui1, Vronique Vque, Sami Tabbane; Combined
Fair Packet Scheduling Policy and Multi-Class Adaptive
CAC Scheme for QoS Provisioning in Multimedia Cellular
Networks; International Journal on Communication System,
2005.
[8] Jaime Snchez, Jorge Flores Troncoso, Jos R. Gallardo;
Retransmission Algorithm based on Power Priorities for
Wireless Networks; PIMRC; IEEE 2002.
[9] Celal Ceken, Ismail Erturk, Cuneyt Bayilmis; A new MAC
protocol design based on TDMA/FDD for QoS support in
WATM networks; Computer Standards and Interfaces
Vol 28, p 451-466, 2006.
[10] H. Kim, S.K. Biswas, P. Narasimhan, R. Siracusa and C.
Johnston; Design and Implementation of a QoS Oriented
Data-Link Control Protocol for CBR Traffic in Wireless
ATM Networks; Wireless Networks, Vol 7, p 531540,
2001.
[11] Rolf Sigle, Thomas Renger; Fair Queueing Wireless ATM
MAC Protocols; Computer Networks 31(9-10): p 985-997
(1999).
[12] Wojciech Burakowski, Halina Tarasiuk, Andrzej Beben and
Marek Dabrowski; EuQoS IST project: Overview of the
QoS framework for EuQoS; IST-1999
[13] Ahcene Bouzoualegh, Thierry Val, Eric Campo and Fabrice
Peyrard; Study and simulation of an efficient medium access
control protocol for local area underwater networks; first
International Workshop on Wireless Communication in
Underground and Confined Area; 2005 Val dOr, Canada.
[14] Michael Wolf, Rolf Sigle; Medium Access in Wireless ATM
Systems for Industrial Applications: Requirements and
Solutions, International Performance, Computing, and
Communications Conference, p 18-22, IEEE1998.
[15] Jongho Bang, Sirin Tekinay, Nirwan Ansari; A novel
capacity maximisation scheme for multimedia Wireless
ATM systems providing QoS guarantees for handoffs;
VTC2000.
[16] Tae-Hee Kim, Kwan-Woong Kim, Jae-Hoon Kim, Ho-Jin
LEE; Adaptative CDV compensation algorithm for satellite
networks; IEICE Trans. Commun, pp 3401-3407, Vol E88-
B, N8, 2005.
[17] Sami A. EL-Dolil, Mohammed Abd Elnaby; An intelligent
resource management strategy for the next generation
WATM personal communication networks; Proceedings of
the 46th IEEE International Midwest Symposium on

9

Circuits and Systems, Vol 1, p 133-136, IEEE 2004.
[18] Der-Rong DIN; Wireless ATM Backbone Network Design
Problem; IEICE trans. Fundamentals, special section on
Multi-dimensional Mobile Information Networks; Vol E88-
A, N7, July 2005.
[19] Sungjin Lee, Optimal Loading Control Based on Region-
Time Division for Uplink Broadband Cellular Networks;
IEICE trans. fundamentals, special section on Wide Band
System, Vol E89-A, N11, Nov 2006.

IJCSES International Journal of Computer Sciences and Engineering Systems, Vol.3, No.3, July 2009

1

New approaches reducing handoff latency in 802.11 wireless New approaches reducing handoff latency in 802.11 wireless New approaches reducing handoff latency in 802.11 wireless New approaches reducing handoff latency in 802.11 wireless
LANs LANs LANs LANs
Monji ZAIDI, Ridha OUNI, Jamila BHAR and Rached TOURKI

Electronic and Micro-Electronic Laboratory (EE, IT-06)
FSM, Monastir, Tunisia
ridha.ouni@ipeim.rnu.tn
Abstract
Wireless local area networks (WLAN) have seen rapid growth
and deployment in the recent years. It has become the state-of-the
art campus networking option in many academic and corporate
campuses. As Wi-Fi technology becomes ubiquitous, it is
important to understand trends in the usage of these networks. As
the mobile clients are moving from one access point to another,
the convectional layer-2 handoff consumes more time in the
channel-scanning process. The proposed handoff mechanism for
mobility management is designed to minimize the handoff
latency in IEEE 802.11 wireless local area network. It reduces
the discovery phase according to two models extended from the
basis model. There are two methods to implement Handoff
functions in the MAC layer. The first method is CPU-based
solution. It uses software for protocol analysis and CPU, such as
DSP, for process management. It is more flexible in design stage
and easy to modify, however the low processing speed and the
higher cost present its major weakness. The second method
means that all functions are processed by hardware circuits. The
advantage of this method is circuit reconfiguration and
processing speed very high, but it needs long developed time. We
propose the last method to implement handoff functions in the
MAC layer.

Key words: Handoff, Latency, MAC, 802.11, Mobility, FPGA.
1. Introduction
IEEE 802.11 based wireless LANs have seen a very fast
growth in the last few years [1]. Voice over IP (VoIP) is
one of the most promising services to be used in mobile
devices over wireless networks [2]. One of the main
problems in VoIP communication is the handoff latency
introduced when moving from one Access Point (AP) to
another [3]. Then, the amount of time needed for the
handoff in 802.11 environments is too large for seamless
VoIP communications [4]. The definition of new mobile
network architectures able to improve the Internet
experience to the mobile users is becoming the primary
objective of the wireless research community. The future
4G access technologies will inevitably have to cooperate
with the existing cellular environments (e.g. 3G) and
indoor environments (e.g. 802.11 WLAN). These
technologies have been developed for provisioning
different specific services and thus they widely vary in
terms of bandwidth, latencies, coverage capability, etc.
The complementariness of these radio access technologies
can be an advantage, in order to offer adaptive and flexible
services to mobile users. The incompatibilities between
heterogeneous systems will be overcome by using both
new hardware solutions (for example FPGAs solutions for
providing physical layer reconfigurability) and new
software solutions (algorithms and protocols for providing
mobility management across heterogeneous systems). We
were able to reduce the handoff latency using extended
handoff mechanisms, with modifications being limited to
mobile devices and compatible with standard 802.11
behaviors.
This paper is organized as follow. Section 2 presents a
wireless communication environment based on
IEEE 802.11 standard. In section 3, we describe the
existing layer-2 handoff mechanism. In section 4, we detail
handoff mechanisms proposed for campus wide networks.
The simulation, analysis and synthesis are dealt in Section
5. Finally, section 6 concludes the paper.
2. IEEE 802.11 Standards
There are currently three IEEE 802.11 standards [5]:
802.11 a, b and g. The 802.11a standard operates in the 5
GHz ISM band. It uses a total of 32 channels of which only
8 do not overlap. Both 802.11b and 802.11g standards
operate in the 2.4 GHz ISM band and use 11 among the 14
possible channels. While 802.11b can operate up to a
maximum rate of 11 Mbit/sec, the 802.11g and 802.11a
standards can operate up to a maximum rate of 54 Mbit/sec.
The 802.11g standard is backwards-compatible with the
802.11b standard while the 802.11a standard, because of
the different ISM band, is not compatible with the two
other.

2

2.1. Wireless LAN architecture
In this section, we briefly summarize the network functions
and services of the 802.11 standard, in order to better
clarify the Handoff Process and the innovative
contributions. We outline this architecture as described in
[6].
A traditional WLAN architecture can be considered as a
type of cellular architecture, where each cell is called Basic
Service Set (BSS) and is controlled by a base station called
Access Point (AP). Each network has a name called
Service Set Identifier (SSID) which is advertised by the AP
in special control messages called beacons. When two or
more APs using the same SSID are connected via a
broadcast layer 2 network, called Distribution System (DS),
an Extended Service Set (ESS) is created (figure 1).
The standard also defines a set of networking services,
which are categorized into station services and distribution
services. Specifically, station services are the
authentication, desauthentication, confidentiality, and
MAC Service Data Unit (MSDU) Delivery services, while
the distribution services include association to the access
point, disassociation, reassociation, distribution in the
whole ESS, and integration towards non-802.11 networks.
Additional MAC services are defined in the standard and
in some extensions (e.g. 802.1le, 802.11i, 802.1lf) for
optimizing and protecting the use of the wireless resources
through rate adaptation, quality of service differentiation,
encryption/decryption, and so on.
IEEE 802.11 specification focuses on the two lowest layers
of the OSI model while they incorporate both physical and
data link components. The data link layer is partitioned
into the logical link control (LLC) and the media access
control (MAC). All 802.11 networks have both a MAC
and a physical component. The PHY layer consists of the
radio and the radios shared channel. The MAC layer
maintains communications among 802.11 stations by
managing the operation of the PHY and by utilizing
protocols that support and enhance communications over
the radio medium.

ESS
BSS BSS
Distribution Network
AP AP
MT MT

Fig. 1 IEEE 802.11 architecture.
2.2. IEEE 802.11 Management Frames
The IEEE 802.11 management frames enable stations to
establish and maintain communications. The following are
common IEEE 802.11 management frame subtypes, with
the description quoted from [7].
Probe request: A mobile terminal (MT) sends a probe
request frame when it needs to obtain information from
another station. For example, a MT would send a probe
request to determine which access points (APs) are within
range.
Probe response: A MT will respond with a probe response
frame, containing capability information, supported data
rates, etc., after it receives a probe request frame.
Authentication: The 802.11 authentication is a process
whereby the AP either accepts or rejects the identity of a
MT. The MT begins the process by sending an
authentication frame containing its identity to the AP. With
open system authentication (the default), the MT sends
only one authentication frame, to the selected AP which
could accept or reject the connection through an
authentication frame.

MT Active AP Old AP
New AP
Probe request
Probe response
Probe response
Probe response
Probe request
Probe response
Probe response
Probe response
...
.
.
.
C
h
a
n
n
e
l

1
C
h
a
n
n
e
l

N
Authentication request
Authentication response
Association request
Association response
Exchange
Stop accepting Traffic
Start accepting Traffic
H
a
n
d
o
f
f

L
a
t
e
n
c
y
P
r
o
b
e

d
e
l
a
y
A
u
t
h
e
n
t
i
c
a
t
i
o
n

a
n
d
A
s
s
o
c
i
a
t
i
o
n

d
e
l
a
y

Fig. 2 Handoff latency in IEEE 802.11 Networks.
Reassociation request: If a MT roams away from the
currently associated AP and finds another AP having a
stronger SNR, it will send a reassociation frame to the new
AP. This last then coordinates the forwarding of data
frames that may still be in the buffer of the previous AP
waiting for transmission to the MT.

3

Reassociation response: An AP sends a reassociation
response frame containing an acceptance or rejection
notice to the MT requesting reassociation. Similar to the
association process, the frame includes information
regarding the association, such as association ID and
supported data rates.

AP AP AP
Ch_1 Ch_N Ch_2
P
r
o
b
e

R
e
q
u
e
s
t

(
b
r
o
a
d
c
a
s
t
)
P
r
o
b
e

R
e
s
p
o
n
s
e
P
r
o
b
e

R
e
s
p
o
n
s
e
P
r
o
b
e

R
e
q
u
e
s
t

(
b
r
o
a
d
c
a
s
t
)
P
r
o
b
e

R
e
q
u
e
s
t

(
b
r
o
a
d
c
a
s
t
)
P
r
o
b
e

R
e
q
u
e
s
t

(
b
r
o
a
d
c
a
s
t
)
P
r
o
b
e

R
e
s
p
o
n
s
e
Probing Latency
...
...

Min Channel Time
Max Channel Time
CST
CST
CST
Wasted Channel
Max Channel Time
Wasted Probe-Wait
MT
AP

Fig. 3 Probing process in IEEE 802.11.
3. Existing layer-2 handoff mechanisms
The conventional Handoff process is divided into three
phases namely (a) Scan (b) Authentication and (c)
association as illustrated in figure 2 [5]. Scanning phase is
the dominating factor in handoff latency, accounting for
more than 90% of the overall latency [4].
The probing process (or scanning process) finds a new
available AP with the best signal quality with respect to the
station. Figure 3 illustrates the probing procedure as
described in the IEEE Standard 802.11. In this figure, N
distinct channels are selected to probe. Once the channels
to be probed are determined, the MT switches to each
selected channel and broadcasts a probe request frame. We
call this latency the Channel Switch and Transmission
overhead (CS&T). In figure 3, the arrows toward APs
represent such probe request frames broadcast on a
channel (numbered in a circle). Upon receiving a probe
request, APs respond with probe response frames to the
MT (downward arrows). After the transmission of the
probe request, the station waits for a certain amount of
time (probe-wait time) before switching to the next channel.
After probing all selected channels, the next AP is
determined from the information received in the probe
responses and their associated Signal to Noise Ratio (SNR).
During the transition phase (authentication and
association), a MT identifies a suitable candidate AP,
breaks its association with the current AP and then
reassociates with the targeted AP.
Existing handoff mechanism is based on Scanning
channels, authentication and association or reassociation
phases. This model is described with a hardware
description language using a Finite State Machine (FSM).
In this model, handoff starts when the SNR drops than a
specific threshold. Figures 3 explain, with more details, the
basis handoff model and its steps latencies.
Prior research has focused on improving handoff
performance using a single radio interface. Shin et al. [14]
in the Neighbor Graphs work explore techniques to
improve handoffs by implementing a topology inferencing
technique in both clients and APs. Ramani et al. [15]
defined a technique called SyncScan that requires
appropriate time synchronization between APs and clients.
SyncScan also requires synchronization of Beacon
broadcast times for different APs and periodic channel
hopping of clients. Both schemes seek to reduce the time
spent in the channel scanning phase when a handoff occurs.
In [16] authors proposed a method to completely eliminate
handoff time using two radio interfaces in wireless devices.
This method requires a high cost, significant power
consumption and increases the interference rate, thus a
high congestion.
4. Handoff proposal: New approaches
4.1. Model with reduced Scan phase
In this work, the objective consists of reducing handoff
latency. As a first solution, we propose to alleviate the scan
delay since it takes the major part of the handoff latency.
In fact, this solution consists of transmitting Probe requests
which the scanning channels, stops once a Probe response
indication is received with an adequate SNR. An SNR
threshold level has been defined to select AP that provides

4

QoS guarantee. Figure 4 explains the first approach based
on reduced scan phase.
During Scan phase of basis handoff model, the MT must
sweep the total number (N) of channels. The time allocated
to scan each channel is called MaxChannelTime. Thus, the
time of Scan is given by the Eq. 1.
e ChannelTim * mber ChannelsNu ScanTime = (1)
MaxchannelTime is the time interval separating the Probe
Request and the last Probe Response on each channel. The
scan time can be reduced when minimizing the channel
number to scan. By being unaware of negligible times, the
idealized latency of this active Scan is given by the Eq. 2.
) 2 ( * ) ( * )) ( 1 (
1
=
=
+ =
NumChannel c
c
Max c p Min c p Scantim

Where P(c) is the probability of one or more APs operating
in the same channel (c) . Whether Min Channel Time and
Max Channel Time values are respectively 1 ms an 11 ms
ideal latency should extend from 11 ms to 110 ms. Based
on Eq. (1), it becomes unuseful to sweep all channels while
the most adequate AP belongs to a channel already
scanned. In others terms, scanning the rest of channels
doesnt serve to find useful AP, but it loses time which
causes to higher scan latency.
The implementation of this model takes into account the
specific threshold once reached, the MT stops the scan
process and follows the rest of the basis handoff phases.
Then, it is obviously that the reduced scan model,
described in figure 4, leads to a fast communication
establishment.

Scan Channel i
AP(Ch i) selection
SNR i >Thershold
The rest of basis
model phases
N
e
x
t

c
h
a
n
n
e
l

Transmit probe Request on channel i.
Receive probe Response from channel i.
Yes
NO

Fig. 4 Second model: reduced Scan phase.
4.2. Predictive model
The second model, based on predictive approach, can be
adopted in two hypotheses. The first one occurs when
supported applications meet temporal constraints or
significant QoS level. Then, the Handoff process should be
achieved without (a) generating communication rupture at
the applicative level and (b) degrading QoS guaranties.
The second situation occurs when mobile stations follow
predefined trajectories. The solution consists of selecting
the future AP before handoff setup. Then, the AP is
selected in an advanced step in order to allocate resources
needs useful for the next MT hop and to minimize handoff
latency. This solution is a probabilistic based approach
since we predict AP. However, it takes into account several
arguments and parameters to outlines decision for the
probable AP that can be chosen to maintain an established
connection.
The mobile station achieves its first attachments using the
basis model. For each transition from one cell to another,
the MT records AP addresses which can be used to
determinate its trajectory. The future AP can be predicted
using addresses of the old APs present in the direction of
the MT. As well as the number of AP meted by the MT
becomes higher, so its trajectory is strictly defined. We
propose here that three successive attachments or AP
addresses allow discovering the MT trajectory. Figure 6
presents a cellular network and predefined trajectories
followed by MT [13].
AP 4
AP 10
AP 5
AP 6
AP 2
AP 9
AP 1
AP 7
AP 8
AP 3
First attachment
(AP4, - , -)
Snd attachment
(AP4, AP10 , -)
3rd attachment
(AP4, AP10 , AP9)
MT 1
MT 2

Fig. 5 Target cellular topology and mobility trajectory for predictive
handoff model.
Figure 6 represents the predictive algorithm flowchart
which outlines two main parts: basis model part and
predictive part. The basis model part is similar than the
model described in section 3. The predictive part works
with two threshold levels: Thmin and Thmax. When the
connection is established, the prediction of the next AP
starts when the current AP SNR degrades to the Thmax
threshold. The prediction phase takes place in an advanced
time to allocate resources within a predicted AP.
Furthermore, once the current AP SNR drops and reaches
the Thmin threshold, the MT initializes the authentifica-
tion and the association phases with the predicted AP.
5. Implementation
Our contribution for reducing handoff latency reposes on
the lower layers of the 802.11 networks (MAC and PHY).
Physical layer is divided into two sub-layers: the Physical

5

Layer Convergence Procedure (PLCP) and the Physical
Medium Dependent (PMD) [11,12]. We assume that a
specific data was done by PLCP. The PLCP is the bridge
between MAC and radio transmission layers over which it
translates MPDU to PMD frames. The PMD is responsible
for transmitting any bits it receives from the PLCP to the
wireless medium by using antenna. The major efforts of
the handoff processing are almost in the MAC layer, which
we did consider new approaches to be integrated in this
level.

Connection established
(N < 3)
SNR drops
Th
min

Basis Handoff
(N >= 3)
SNR drops
Th
min

SNR drops
Th
max

Prediction
Authentification
Association
Yes
Yes
Yes

Fig. 6 Predictive model state diagram.
In this section, we outline the implementation of the MAC
layer in a mobile terminal. A modular architecture proposal
of the MAC layer, based on receiver and transmitter
component, is represented in figure 7. It details interaction
between MAC layer and both LLC and PLCP layers.
Various types of handshaking signals are integrated to
manage and control transmission in both directions.
Table 1 defines the interfacing signals and briefly
describes their operations.
The receiver component receives PMDUs from PLCP and
decodes them into various types of packets [8,9,10]. First,
it identifies the type of frames (signalization, data or
control) and next processes them accordingly. For the
handoff initialization and operation, the main task of the
MAC receiver component consists of processing probe
response frames and controls SNR level.
The transmitter component allows (a) events detecting (b)
parameters buffering and (c) message generating.
Furthermore, it maintains a permanent interaction with the
receiver component in order to manage events (SNR, New
connection) and satisfy requirements (service classes,
addresses). The transmitter is mainly responsible first,
for handoff initialization by generating probe requests over
different channels. Second it allows automatizing handoff
phases according to the specific approaches described in
section 4.
Table 1: Receiver/transmitter component interface signals.
Signals
(interf ports)
Description
Clk Operation clock
Canal_val Sets Channel value to physical layer
Order Used to schedule responses in a specific order
Phy_start Physical layer notifies starting to receive data
Sig_level Indicates the SNR of the AP
Prob_resp MT receives Probe Response frames from APs
Phys_data Data signals from physical layer
Prob_reqst
Notifies that the transmitter sends probe
request frame
Ack
Notifies that the transmitter sends an
acknowledge frame
Scan_fini Indicates that all channels have been scanned.

P
h
y
_
s
t
a
r
t
P
r
o
b
_
r
e
s
p
S
i
g
_
l
e
v
e
l
P
h
y
s
_
d
a
t
a
P
r
o
b
_
r
e
q
s
t
A
c
k
C
a
n
a
l
_
v
a
l
O
r
d
e
r
S
c
a
n
_
f
i
n
i
Physical Layer:PLCP and PMD
MAC
LLC: Logic Link Control
C
l
k
Transmitter Part Receiver Part
1
6
424
1
6

Fig. 7 Modular architecture proposal of the MAC layer.

6

Probe Response Frames Probe Request Frame(First channel)
MT Address AP Address SNR
Probe Request Frame(2nd channel)

Fig. 8 Simulation of the Scan Phase on the first channel (basis model).
Probe Request Frame(First channel) Probe Response Frames Authentication frame Association frame
Association response Adequate thershold on the first channel Authentication response
handoff finished

Fig. 9 Simulation results of the reduced scan model.
5.1 Simulation and results
5.1.1 Basis model
A mobile station broadcasts probe request over three
channels. On each channel, it expects to receive responses
from three access points. Response frames as well as their
SNR are buffered and then used to select the AP that
satisfies the mobile requirements. Figure 8 gives an
example of an active scan timing diagram. It shows also
probe response frames identification, addresses extraction
and SNR measurement. The rest of the handoff process
reposes on authentication and association phases while
each one a frame is sent to the selected AP and a response
is received.
5.1.2 Handoff with reduced Scan
Figure 9 shows the simulation of a fast handoff processing
model. Optimal handoff latency is improved by reducing
the scanning phase, which finishes once an adequate AP is
detected. A higher detected SNR avoids scanning the other
channels and allows joining the corresponding AP.
Authentication and association phases are similar than
those of the basis model. This approach remains
probabilistic opposite of the adequate AP order while
another AP may be detected within the rest of channels. In
worst case, this approach improves the basis model
performances.
5.1.3 Predictive model
The predictive handoff model is similar than the basis
model during the first three transitions. This period serves
as training phase in order to discover the MT trajectory

7

over an historical setting created with signaling messages
between the MT and the meted AP. This model could
predict the future AP since the fourth handoff execution.
The MT authenticates with this AP without scanning step.

SNR drops
5th handoff initialisation

Adresses of the three last APs which
accomodated the MT
Old connection New connection
Handoff finished
...

Fig. 10 Attachment within the predictive model.
Figure 10 shows the predictive model simulation at the 5
th

handoff initialization. At this level, the MT trajectory is
discovered and the future AP is predicted according to the
historical transitions. When the SNR drops, the MT
authenticates and joins with this AP without scanning. This
approach improves reducing handoff latency compared to
the two others models as outlined in figure 10. This study
enables to conclude that the handoff latency depends on
scan period or in other terms on the number of AP which
operates under several channels. In fact, minimizing one
among N channels to be scanned allows reducing almost
1/N of the total handoff latency.
The handoff latency measures the time between a probe
requests is sent until an association reply is received. The
basis model requires to seen all channels. It needs
184 clock cycles without counting the Channel Switch and
Transition overhead (CST). The second model provides, in
general three responses according to the probe response
order of the adequate AP.
Scenario 1: We suppose that the adequate signal is
received during Scan of the third channel (the last). In
this case, the latency of the handoff is equivalent than
the basis model.
Scenario 2: means that the adequate signal is received
on the second channel (Half-time). In this case, the scan
of the third channel is useless.
Scenario 3: when the MT detects an adequate signal, on
the first channel such as outlined in figure 9. Then, the
second and the third channels are useless. It is clear that
there is significant decrease of handoff latency by the
proposed models (figure 11).
In subsequent, we consider five scenarios in order to
evaluate the handoff latency. Each scenario presents six
successive handoff executions. In our results, the k+1
th

handoff latency (k = 1, 2, 3, 4 or 5.) is equal to the sum of
all latencies of the earlier handoff executions.

1
8
1
4
8
1
8
1
8
9
9
1
8
1
8
5
1
1
8
1
8
1
4
8
1
8
Scan Association Authentication
1
8
5
1
8
Basis Model Model With reduced Scan Predictive Model
H
a
n
d
o
f
f

L
a
t
e
n
c
y

(
C
l
o
c
k

C
y
c
l
e
)
0
200
Scenario 1
Scenario 2
Scenario 3

Fig. 11 Handoff latency for the three models.
For the basis model, the scan phases are similar for all
executions while the MT is required each time to scan all
available channels on the network.
For the predictive model, we note that during the first three
attachments, the handoff latency is similar to the basis
model. The predictive model starts during the third
attachment since the terminal has a history reflecting its
trajectory. From the third attachment, the MT may provide
the next AP without conducting the scan phase.
Consequently, the handoff execution becomes transparent
due to its short duration. Then, the handoff latency of the

8

predictive model becomes lower than 25 % of the basis
model. This is more suitable for real time applications.
The reduced scan model depends on the detection of the
adequate SNR that can be received over the first, the
second or the third channel. In fact, the reduced scan
model remains a probabilistic model. The following
figures present the handoff latencies for the three models
using five scenarios.
1 2 3 4 5 6
0
200
400
600
800
1000
1200
Basis model
Model withreducedscan
Predictive model
T
o
ta
l la
te
n
c
y
o
f
h
a
n
d
o
f
f
s
(
c
lo
c
k
c
y
c
le
)
Initializationnumbersof handoff
1 2 3 4 5 6
0
200
400
600
800
1000
1200
Basis model
Model withreducedscan: Proposed
Predictive model: Proposed
T
o
ta
l la
te
n
c
y
o
f
h
a
n
d
o
f
f
s
(
c
lo
c
k
c
y
c
le
)
Initializationnumbers of handoff

Scenario 1 Scenario 2
1 2 3 4 5 6
0
200
400
600
800
1000
1200
Basis model
Predictive model
T
o
ta
l la
te
n
c
y
o
f
h
a
n
d
o
f
f
s
(
c
lo
c
k
c
y
c
le
)
1 2 3 4 5 6
200
400
600
800
1000
1200
Basis model
Predictive model
T
o
ta
l la
te
n
c
y
o
f
h
a
n
d
o
f
f
s
(
c
lo
c
k
c
y
c
le
)

Scenario 3 Scenario 4

1 2 3 4 5 6
0
200
400
600
800
1000
1200
Basis model
Predictive model
T
o
ta
l la
te
n
c
y
o
f
h
a
n
d
o
f
f
s
(
c
lo
c
k
c
y
c
le
)
Initializationnumbersof handoff

Scenario 5
Table 2: Handoff latencies of the proposed mechanisms.
Models Handoff latency
Model with reduced Scan 4 - 6 ms
Predictive model ~ 2 ms

Table 3: Synthesis results.

Number of
Slices
Number of Flip
Flops
Nb of 4 input
LUTs
Nb of bonded
IOBs
Frequency
(MHz)
Basis Handoff 782 604 1381 40 132
Reduced model 705 608 1313 40 124
Predictive
model
1165 995 816 41 150


9

5.2 Synthesis results
During the synthesis step, we have exploited FPGA xilinx
virtex II pro environment. This environment allows
implementing communication systems on programmable
circuits. The advantage of using FPGAs circuits is mainly
the system re-scheduling. For our application, RTL
synthesis is achieved using the ISE 8.1 of the Xilinx FPGA
virtex II pro environment. Synthesis results, of the three
approaches, are shown in table 3. These results should be
exploited in order to study their impact on the support of
the technological parameters specified in IEEE 802.11.
IEEE 802.11b Standard proposes a theoretical flow of
11 Mbps (5 to 7 Mbps in real case according to the
environment), thus, an average of 6 Mbps. In free
environment, the frequency band is the 2.4 GHz with three
available channels radios. Moreover, 802.11b proposes
MinChannelTime equal to 670 s, DIFS equal to 50 s
and SIFS equal to 10 s. Using these technological
parameters and the values of frequencies obtained in
table 3, our circuits reduce the handoff time for each model
as outlined in table 2. As result, the average of the handoff
latency is maintained between 4 and 6 ms for the reduced
scan model and slightly lower than 2 ms for the predictive
model.
Both schemes in [14] and [15] attempt to reduce the time
spent in the channel scanning phase when a handoff occurs.
By changing the APs and the clients, and by increasing
coordination between them, Neighbor Graphs achieves
handoff latency of about 40 ms, and SyncScan handoffs
take 2-3 ms. But, the technique requires periodic
suspension of communication that could last more than
10 ms, depending on hardware. Table 4 shows more details
allowing a simple comparison of the different handoff
mechanisms cited in this paper.
Table 4: Comparison of different handoff mechanisms.

Wireless
interface
Handoff
latency
Infrastructure
modification
Neighbor Graphs [14] 1 ~ 40 ms Yes
SyncScan [15] 1 2 3 ms Yes
MultiScan [16] 2 0 ms No
Reduced scan model 1 2 6 ms No
Predictive model 1 ~ 2 ms No
6. Conclusions
In this paper, IEEE 802.11 convectional handoff process
has been implemented on FPGA circuit using the high
level design technique and based on IEEE 802.11b
specifications. The convectional layer-2 handoff consumes
more time in the channel-scanning process. For this reason,
we have proposed two others handoff mechanisms for
mobility management. Our results show that the network,
using reduced scan or predictive model, may improve
suitable performances for MT transitions. Several
scenarios have been employed in order to evaluate handoff
latencies spent for each model compared with others
mechanisms for the conventional layer-2 handoff process.
The handoff steps are reduced from 184 to 41 clock cycles
becoming more suitable for real time applications.
We adopted the high level design for the implementation
of these models. In fact, we have used VHDL as high level
description language, ModelSim as a simulation tool to
check the behavior of each model at the RTL level and
ISE 8.1 of the FPGA xilinx environment for synthesis step.

References
[1] M. Gast, 802.11 Wireless Network: The Definitive Guide,
Second Edition, OReilly, 2005.
[2] Fayza A. Nada, On using Mobile IP Protocols, Journal of
Computer Science 2 (2), 2006, pp.211-217.
[3] H. Velayos and G. Karlsson, Techniques to reduce the IEEE
802.11b handoff time, IEEE ICC, 27(1), June 2004,
pp.38443848.
[4] Arunesh Mishra, Minho Shin, William Arbaugh, An
empirical analysis of the IEEE 802.11 MAC layer handoff.
ACM SIGCOMM Computer Communications Review
(ACM CCR), 33(2), April 2003, pp.93-102.
[5] Bob OHara, AI Petrick, IEEE 802.11 handbook a
designers companion, second ed., March 2005.
[6] IEEE Std 802.11-1997.Wireless LAN Medium Access
Control (MAC) and Physical Layer (PHY) specifications,
1997.
[7] A. Jain. Hando delay for 802.11b wireless LANs, Technical
report, University of Kentucky, 2003.
[8] Y. Kim, H. Jung, H. H. Lee & K. R. Cho, MAC
implementation for IEEE 802.11 wireless LAN, Router
Technology, Department, Electronics &
Telecommunications Research Institute, 2001.
[9] XILINX, Configurable LocalLink CRC Reference Design,
Nov. 2004.
[10] T. H. Meng, Design and implementation of an all-CMOS
802.11a wireless LAN chipset, Communication magazine,
IEEE, 41(8), Aug. 2003, pp.160-168.
[11] IEEE 802.11, IEEE wireless LAN medium access control
(MAC) and physical layer (PHY) specifications, Aug. 1999.
[12] IEEE 802.11b, Wireless LAN medium access control
(MAC) and physical layer (PHY) specifications: High-speed
Physical Layer Extension in the 2.4GHz Band, IEEE Std
802.11b-1999.
[13] Shiang-Chun Liou, Yueh-Min Huang; Trajectory prediction
in mobile networks. International Journal of Information
Technology, vol. 11, No. 11, 2005, pp.109-122.
[14] M. Shin, A. Mishra and W.A. Arbaugh, Improving the
Latency of 802.11 Hand-offs using Neighbor Graphs
Mobisys 2004 June, Boston, USA.
[15] I. Ramani and S. Savage,SyncScan: Practical Fast Handoff
for 802.11 Infrastructure Networks Proceedings of the
IEEE Infocom, March 2005.

10

[16] Vladimir Brik, Arunesh Mishra, Suman Banerjee,
Eliminating handoff latencies in 802.11 WLANs using
Multiple Radios, Applications, Experience, and Evaluation,
Internet Measurment Conference 2005: 299-304.

International Journal of Computers, Systems and Signals, Vol. 10, No.1, 2009 27
LOW POWER ASIC DESIGNS FOR FAST HANDOFF IN
IEEE802.11

Monji Zaidi, Ridha Ouni, Kholdoun Torki and Rached Tourki


CPM, 46, Avenue Felix VIALLAET 38031 GRENOBLE cedex, France

E-mail: Monji.Zaidi@fsm.rnu.tn

ABSTRACT
The Handoff process is a key problem in many wireless network processing applications. Current
implementations of this process using software implementation are time consuming and cannot
meet the gigabit bandwidth requirements. Implementing this process within the hardware
improves the search time considerably and has several other advantages like reducing power
consumption. In this paper we present an array based hardware implementation of this time
consuming process for network mobility. A new mechanism for mobility management to
minimize the handoff latency in IEEE 802.11 wireless local area network is also presented.
Compared to the basis model and at 1GHz, this new mechanism allows a profit of 60% in power
consumption and 20% in silicon area. Those two designs are described in VHDL at the RTL level
language and implemented on an ASIC (Application-Specific Integrated Circuit) and are evaluated
in terms of speed, area and power consumption.

Keywords: IEEE 802.11, Handoff, MAC, design, ASIC

1. INTRODUCTION

The world of WLANs is truly an exciting area, with major activity worldwide,
challenging traditional service providers and business models. Initially, WLANs were meant to be
an augmentation, not a replacement, of wired LANs and premises telephone systems. WLANs
were deployed in enterprise or corporate locations where there might be a number of factors that
limited or prevented wired systems from being installed. Today, we see much greater utility for
WLANs, as evidenced by the emergence of thousands of hotspots around the world. In some
cases, it is cheaper to deploy wireless in an office than to replace a crumbling old token-ring cable
plant with shiny new Ethernet. In general, WLANs operate over short ranges, anywhere from 10 to
500 feet (3 to 150 m), so their coverage areas are microcells or piccocells.
Mobility is the most important feature of a WLANs system. Undoubtedly the future
perspective of networking will demand to support mobile users traveling all over the world. In
addition to that, the number of portable devices that need access to the Internet is exponentially
increasing. Users on the other hand are no longer required to work in their company's home
network while they may be moving from place to place. Usually, continuous service is achieved
by supporting handover from one cell to another. Handover is the process of changing the channel
(frequency, time slot, spreading code, or combination of them) associated with the current
connection while a communication is in progress. It is often initiated either by crossing a cell
boundary or by deterioration in quality of the signal in the current channel.
In IEEE 802.11 based wireless LAN, when the Mobile Terminal (MT) changes its point
of attachment to the internet, it may produce a service interruption, since the MT cannot send or
receive any packets from the time at which it disconnects from one point of attachment to the time

ISSN 1608-5655 (Online), Category: Research articles, Publisher: IAAMSAD

28 International Journal of Computers, Systems and Signals, Vol. 10, No.1, 2009

at which it registers with a new point of attachment. Such an interruption would be unacceptable
for real time services such as voice-over IP, which demands a further optimization of the mobility
management in IEEE 802.11 based WLAN. Our motivation is to study the impact of the handoffs
for delay of the sensitive applications. Our evaluation results show that the conventional handover
is not totally effective for QoS requirements. So, we propose a hardware implementation for the
handoff process in WLAN networks, using an ASIC circuit.
The paper will be organized as follows: in section 2, an overview of related works is
provided. Section 3 presents an environment with a wireless communication based on
IEEE 802.11 standard. In section 4, we describe the existing layer-2 handoff mechanism.
Section 5, highlights the hardware implementation of handoff mechanisms (basis and proposed)
for campus wide networks. Results are dealt in Section 6. Finally, section 7 is the concluding part
of the paper.

2. RELATED WORKS

The literature contains several efforts proposing a mobility modeling and management
approaches aiming at improving the handoff management and at optimizing resource reservations.
In (Choi C.H. et. al., 2002), the authors suggest the use of an adaptive bandwidth
reservation based on a mobility graph and 2-tier cell structure to determine the amount of
bandwidth to be reserved in the cell. Another resource reservation approach is presented in (Islam
M.M. et. al., 2002) and (Islam M.M. et. al., 2003), where the authors propose a bandwidth
reservation scheme that uses a mobility parameter-based resource reservation estimator function
(RREF). This nonlinear function uses distance, direction, and velocity to calculate the probability
of a MT visiting a particular cell. The authors, in (Liu T. et. al., 1998), propose a predictive
mobility management scheme that models the MT, as a linear dynamic system driven by time
varying forcing functions that simulate subjective moving intentions from the user and objective
random perturbations from the environment. The mobility management model proposed in (Liang
B. and Haas, Z.J., 1999), is based on GaussMarkov model where an MTs velocity is correlated
in time to various degrees in order to predict the future location of a MT. In (Hou, J. and Fang Y.,
2001), authors initiated the idea of taking the mobility into account for call admission control
algorithms. They explore many important points for mobility-based call admission control. They
indicate that it is important to make the reservation at the appropriate time to save the reserved
bandwidth.. In this model, the authors did not use the mobility history of the users to enhance the
estimation accuracy. In (Yu, F. and Leung, V.C.M., 2002), a mobility-based predictive algorithm
for call admission control was presented. This algorithm is motivated by a computational learning
theory, which has shown that prediction is synonymous with data compression. The main
limitation of this approach is that it is not applicable for a soft handoff, which makes this approach
unsuitable for most of the existing and future mobile telecommunication. Moreover, this approach
requires every adjacent Access Point (APs) to reserve the channels for all the users.
Based on the observation that the movement behaviors of the majority of people are
performed repetitively process, the author in (Tabbane, S., 1995), models the user inter-cell
mobility as a time-dependent (location, probability) pair. This work uses the probability that the
user is found in a certain location during a given period of time as a base parameter. The (location,
probability) pair is derived from the long-time observation statistics. While providing a good
starting, enabling the regularity in a users daily movement, Tabbanes model does not reflect an
instantaneous movement behavior, and the (location, probability) pair is not totally efficient in
representing the itineraries which are usually present in a users movement pattern. As a possible
solution to improve the call connectivity, the Current implementations of this process using
software methods are time consuming and cannot meet gigabit bandwidth requirements. A few
works have addressed the hardware implementation of the mobility in WLANs networks. The
authors in (Chiang, M.H, 2006), proposed a design flow system and function module of active
scan in WLANs. In (Zaidi, M et. al., 2008), we have proposed the hardware transformation and
implementation of the handoff protocol from its initial software description. We have
implemented two models that reduce Scan phase during the handoff execution. These models have
been implemented on a FPGA circuit. Implementing this process in the hardware circuits
improves considerably the search time and has several other advantages like the reduction of
power consumption and the silicon area minimization.

3. HANDOFF PROCESS

The operation destined to change an association from one AP to another is known as a
handover. Original design of the IEEE 802.11 standard ( IEEE 802.11. Part 11, 1999), has just
considered the handoff signaling in the wireless part. Figure 1 shows the main elements involved
in a layer-2 handoff: the MT, the old AP, the new AP, and the distribution system (DS). It can be
observed that basic service sets (BSS1 and BSS2) must belong to the same extended service set
(ESS1). In the same way, radio channels of each cell (CHX, CHY) shall be none mutually
interfering channels. The handover procedure can be divided into three phases: discovery,
reauthentication, and reassociation ( IEEE 802.11. Part 11, 1999) and ( Mishra AS and Arbaugh
W, 2003).

Old AP New AP
CHX CHY
BSS1 BSS2
Handoff
ESS1
Distribution system
MT

Figure. 1 Involved elements in the layer 2 handover

The discovery process involves handoff initiation and scanning phases. As signal strength
and signal-to-noise ratio from a stations current AP get weaker, a MT loses connectivity and
initiates a handoff. Then the client is not able to communicate with its current AP, so the client
needs to find the other APs available. This scan function is performed at a MAC layer, and the
station can create the available AP list ordered by the received signal strength.
For the scan phase, a MT can perform scan operation either in passive or active mode. In
passive scan mode, using the information obtained from beacon frames, MT listens to each
channel of the physical medium to try and to locate an AP. In the active mode (the wireless NICs
do by default), as shown in Figure. 2, MT broadcasts additional probe packets on each channel and
receives responses from APs. Thus the MT actively probes for the APs, and the actual number of
messages varies from 3 to 11. Figure. 2 shows the sequence of messages typically observed during
a handoff process. The handoff process starts with the first probe request and ends with a
reassociation response from the new AP. The probe function follows the IEEE 802.11 MAC active
scan function and the standard specifies a scanning procedure as follows.
1. Using CSMA/CA, acquire the access right to the medium.
2. Transmit a probe request containing the broadcast address as destination, SSID, and
broadcast BSSID (Basic SSID).
3. Start a Probe Timer.
4. If medium is not busy before the Probe Timer reaches MinChannelTime, scan the next
channel. Otherwise, process all received probe responses.
5. Move to next channel and repeat the above steps.
After all channels have been scanned, the informations received from probe response are
scrutinized by a MT to select a new AP. Once the STA decides to join a specific AP,
authentication messages are exchanged between the STA and the selected AP, and after a
successful authentication, the STA sends a reassociation request and expects a reassociation
response back from the AP.

4. HANDOFF ARCHITECTURE

In this section, we describe the hardware implementation of the Handoff process. We
focus on the MAC receiver and transmitter part. Figure. 2 illustrates the system architecture. We
try to divide handoff functions to 5 parts: a controller module, a selection module, a receiver part,
a transmitter part and the SRAM with 256 words of 16 bits.

Figure. 2 Top level structure of the IEEE 802.11 Handoff

4.1. Existing layer-2 handoff mechanism
We are interested in using the handoff phases, according to models which offer a
transparent transition from one AP to another within a minimum delay. The IEEE 802.11 a/b/g
series offers a Wireless connectivity to the users at high rates. An AP provides connectivity for the
mobile users.

MT Active AP Old AP
New AP
Probe request
Probe response
Probe response
Probe response
Probe request
Probe response
Probe response
Probe response
...
.
.
.
C
h
a
n
n
e
l

1
C
h
a
n
n
e
l

N
Association request
Exchange
H
a
n
d
o
f
f

L
a
t
e
n
c
y
P
r
o
b
e

d
e
l
a
y
A
u
t
h
e
n
t
i
c
a
t
i
o
n

a
n
d
A
s
s
o
c
i
a
t
i
o
n

d
e
l
a
y

Figure. 3 Active discovery example
The 802.11 wireless devices allow the user to move freely between APs within, the
coverage area, commonly known as the hotspot. The operation of changing from one AP to
another AP is known as Handoff.
The conventional Handoff process is divided into three phases namely (a) Scan (b)
Authentication and (c) Association (OHara, B. and Petrick, A.I, 2005). The Scanning phase is
the dominating factor in handoff latency, accounting for more than 90% of the overall latency
(Arunesh, M. et. al., 2003). The probing process (or scanning process) finds a new available AP
with the best signal quality. Figure 3 illustrates the probing procedure as described in the IEEE
Standard 802.11. In this figure, N distinct channels are selected to probe. Once the channels to be
probed are determined, the MT switches to each selected channel.

4.2. MAC Frame Format
IEEE 802.11 MAC frame contains three fields:
A MAC header, which comprises frame control, duration, address, and sequence control
information.
A variable length frame body, which contains information specific to the frame type.
A frame check sequence (FCS), which contains an IEEE 32-bit cyclic redundancy code
(CRC).
The frame control field is composed of:
-Protocol Version: The Protocol Version subfield is two bits in length. For 802.11 standards the
value of the protocol version is 00.

Frame Control Duration Adress1 Adress2 Adress3 Ssquence control Adress4 frame Body FCS
Protocol Type Subtype To DS From DS More frag Retry Pwr mgt More data WEP Order
vesion
b0b1 b12 b11 b10 b8 b4b5b6b7 b2b3 b13 b14 b15 b9
2 2 6 6 6 2 6 0-2312 4
2 2 4 1 1 1 1 1 1 1 1
MAC Header
Octets:
Bits:

Figure. 4 Frame control field with MAC frame format

Type and Subtype: The Type and Subtype fields together identify the frame function.
There are three frame types: control, management and data frame. Each of the frame
types has several defined subtypes. We present, in the table 2 only the screens of control
and management frame, since the data frames are not discussed in this work.
To DS/From DS: It is described in Table 1.

Table 1: To DS/From DS combinations in data type frames

To/From DS value Meaning
To DS=0and From DS=0
A data frame moves from one STA to another STA within the same IBSS, as
well as all management and control type frames
To DS=1 and From DS=0 Data frame destined for the DS
To DS=0 and From DS=1 Data frame exiting the DS
To DS=1 and From DS=1
The wireless distribution system(WDS) frame being distributed from one AP
to another AP


Table 2: Valid type and subtype combination used in Handoff execution

Type value
b
3
b
2

Type description

Subtype value
b
7
b
6
b
5
b
4

Subtype description
00 Management 0000 Association Request
00 Management 0001 Association Response
00 Management 0010 Ressociation Request
00 Management 0011 Ressociation Response
00 Management 0100 Probe Request
00 Management 0101 Probe Response
00 Management 1000 Beacon
00 Management 1010 Disassociation
00 Management 1011 Authentication
00 Management 1100 Deautauthentication
01 Control 1011 RTS
01 Control 1100 CTS
01 Control 1101 ACK

More Fragments: If the value is 1, it means that there are still other fragments waiting
for transmission.
Retry: If the value is1, it means that the Data frame (or Management frame) is the
retried frame.
Power Management: A value of 1 indicates that the STA will be in a power-save mode.
A value of 0 indicates that the STA will be in the active mode.
More Data: The More Data field is set to 1 in broadcast/multicast frames transmitted by
the AP, when additional broadcast/multicast MSDUs, or MMPDUs, remain to be
transmitted by the AP during this beacon interval.
WEP: It is set to 1 if the Frame Body field contains information that has been processed
by the WEP algorithm.
Order: It is set to 1 in any data type frame that contains an MSDU, or fragment thereof,
which is being transferred using the Strictly Ordered service class.

5. HANDOFF PROCESS: DESIGN AND IMPLEMENTATION

In this section, we describe the implementation of handoff process. Five MAC
components have been used in order to design the proposed circuit. Figure 5 illustrates Handoff
system architecture divided into five parts. Those are controller module, receiver part, transmitter
part, selection component and memory (256 words of 16 bits).

5.1. Building blocks specifications

5.1.1 Receiver Module
The receiver module receives MPDUs from PLCP (physical layer) and decodes packages.
Table 3 lists MAC receiver part interface signals. The following notations are used to describe the
signal type: I: Input signal; O: Output signal and I/O: Bi-directional Input / Output signal

Table 3: Receiver part interface signals

CK I:bit Operation clock.
CN I:bit Sets receiver to receive data.
Recep_valiv I:bit Input from the controller module, it gives order to receive data.
Rts_recep I:bit Notifies transmitter that receiver got (RTS) frame.
Sig_level I[15:0]
Input from the physical layer, it indicates the link quality with the
corresponding AP.
In_recep I[15:0] Data from physical layer :( Probe Responses).
End_recep O:bit Notifies Controller that reception is complete.
Cts_recep O:bit Notifies that transmitter send Clear To Send (CTS) frame.
Write O:bit Output signal towards memory, it makes memory accessible in writing.
Out_recep O:[15:0]
Output signal towards memory, to save all Probe response in the
memory.
Adr_recep O:[7:0] Output signal towards memory, it selects the writing address.

5.1.2 Transmitter Frame Module
Transmit Frame Module is used as an interface to the physical layer that transmits frames.
In this Transmit Frame Module, the frames are stored in buffer first, and then in order to be
interfacied to Base-band module, used in physical layer. When transmission is completed,
Transmit Frame Module will issue an End_em signal to inform the Controller Module that the
transmission is finished. Table 4 gives more details about this component (input/ output signals).

Table 4: Transmitter part interface signals

CN I:bit Sets receiver to receive data.
Em_valiv I:bit Input from the controller module, it gives order to send data.
Cts_e I:bit
The transmitter notifies that the corresponding AP is ready to receive
data.
Sig_level I[15:0]
Input from the physical layer, it indicates the link quality with the
corresponding AP.
End_em I:bit Notifies Controller that the transmission is complete.
Rts_recep O:bit Notifies physical layer that the transmitter wants to send data.
Out_em O:[15:0] Data towards the physical layer :( Probe Request).

5.1.3. Control Module
All the actions are controlled or arranged by the Control Module in this design. The main
function of the Control Module is to handle information or data from the physical layer and to
coordinate all the other modules that include the receiver Module, the transmitter Module,
selection component and the memory.
The probing process (or scanning process) finds a new available AP with the best signal
quality. In this process, N distinct channels are selected to probe. Once the channels to be probed
are determined, the Controller informs the transmitter to switch to the selected channel and
broadcasts a probe request frame. After transmitting a probe request, APs respond with probe
response frames to the MT and the controller must inform the receiver to accept data from the
network, and it enables the memory to save data. This procedure is repeated until each selected
channel to be probed.
At the end of the probing phase, (a probe requests from APs are stored in the memory).
The controller activates the selection component to find the appropriate AP with the appropriate
SNR and the suitable channel. Once the good AP and channel are found, the controller
communicates again with the receiver and the transmitter in order to finish the authentication and
the association phases.

Transmitter, receiver and control module provide the interface to external memory as a
mailbox to exchange data. This memory has the capacity to store 256 words of 16 bits.

5.2. Handoff proposal: Model with reduced Scan phase
Using this model, the objective consists in reducing the handoff latency. We propose to
alleviate the scan delay since it takes the major part of the handoff latency. In fact, this solution
consists in transmitting the Probe requests on each scanning channels and stopping once a Probe
response indication is received with an adequate SNR. An SNR threshold level has been defined
to select an AP that provides a QoS guarantee. Figure 5 explains this approach based on reduced
scan phase.
During the Scan phase based on the basis handoff model, the MT must sweep the total
number (N) of the channels. The time allocated to scan each channel is called MaxChannelTim.
Thus, the time of Scan is given by the following equation.

canTime Number of Channels. MaxchannelTime

Where, MaxchannelTime is the time interval separating the first Probe Request and the last Probe
Response on each channel.
The scan time can be reduced when minimizing the channel number to be scanned. By
being unaware of negligible times, the idealized latency of this active Scan is given by the
following relation
ScanTime 1 pcMinchannelTime pc. MaxchannelTime
N

Where:
p(c) is the probability of one or more APs operating in the same channel (c), the Min Channel
Time and the Max Channel Time values are respectively 1 ms an 11 ms. The ideal latency should
extend from 11 ms to 110 ms. Taking the first equation account, it becomes unuseful to sweep all
the channels while the most adequate AP belongs to an already scanned channel. In other terms,
scanning the rest of the channels doesnt serve to find a useful AP, but it loses time which causes
higher scan latency. The implementation of this model takes into account the specific threshold
once reached, the MT stops the scan process and follows the rest of the basis handoff phases.
Then, it is obviously that the reduced scan model, described in figure 5, leads to a fast
communication establishment.

{
Scan a channel i
AP (Ch i) selection
SNR i > Thershold
Yes No
The rest of basis model phases
Transmit Probe request on channel i
Receive Probe response from channel i
N
e
x
t

c
h
a
n
n
e
l

Figure.5 Handoff with a reduced Scan phase.

6. DESIGN RESULTS

6.1 Logic Synthesis
The designs were synthesized based on the 130 nm CMOS technology by using the
Synopsys design_vision tool. We have written scripts that perform an automatic bottom-up
synthesis of the design. The Synthesis results of the proposed circuits are presented in Table 5.
These results show that these circuits can operate with 1 GHz, which makes it more suitable for
real time communications.

Table 5: Synthesis results.

Basis Handoff
Handoff with reduced
Scan: proposed
gain compared to
Basis Handoff
Estimated
area
(mm2)
Estimated power
(mW)
Estimated
area
(mm2)
Estimated
power
(mW)
Power
gain
(%)
Area
gain
(%)
500 MHz 0.114 22.8606 0.0917 18.2151 20.34 20.17
660 MHz 0.128 30.0922 0.0941 18.2972 39.20 26.56
1 GHz 0.133 46.1526 0.0986 18.3241 60.30 26.31

6.2 Clock Distribution: (Skew problem and clock times in the circuit)
The quality of the clock Distribution in a circuit plays a significant role in the
performances of a synchronous circuit. The majority of numerical applications are implemented in
synchronous logic, because the current tools for synthesis do not allow the automation for every
design. These tools for synthesis allow the automation for design with combinational or sequential
descriptions which rest on one or more clock. It is vital for a certain implementation that clock
must be known and fixed in all the physical circuit, and its geometrical propagation does not imply
distortion and dephasing. The solutions to guarantee a uniform clock distribution without skew
dephasing are multiple. The best solution used today to reduce the clock dephasing is based on
the concept of clock tree, it inserts on each level of hierarchy, buffer or reverser which rectifies the
clock signal skew locally. This solution has the advantage of producing a geographically
distributed and optimized consumption.

Figure. 6 Clock tree in silicon level

6.3. Layouts
Complete layout designs of the chips are performed by using the Cadence tools (Encounter)
at the frequency of 1 GHz. We have used the one-block approach in order to meet the timing
requirements and to generate the clock tree efficiently. The chip of basis handoff a circuit contains
77 signal pins in total, 0.0567 mm
2
for Combinational area and 0.073 mm
2
for No combinational
area. While, handoff with reduced scan contains 77 signal pins in total, 0.0314 mm
2
for
Combinational area and 0.0672 mm
2
for No combinational area.

Figure.7 a: layout of the basis Handoff b: layout of the modified Handoff

The IEEE 802.11b Standard proposes a theoretical flow of 11 Mbps (5 to 7 Mbps in real
case according to the environment), thus, an average of 6 Mbps. In a free space environment, the
frequency band is the 2.4 GHz with three available radio channels. Moreover, 802.11b proposes a
MinChannelTime, equal to 670 s, DIFS equal to 50 s and SIFS equal to 10 s. Using these
technological parameters and the values of frequencies obtained in table 8, our circuits reduce the
handoff time for each model as outlined in table 9. As a result, the average of the handoff latency
is maintained between 0.74 and 2.22 ms for the reduced scan model. Both schemes in (Shin, M. et.
al., 2004) and (Vladimir, B. et. al., 2005) attempt to reduce the time spent in the channel scanning
phase when a handoff occurs. By changing the APs and the clients, and by increasing coordination
between them, Neighbor Graphs achieve a handoff latency of about 40 ms, and SyncScan handoffs
take 2-3 ms. But, the technique requires periodic suspension of communication that could last
more than 10 ms, depending on the given hardware. Table 6 provides more details allowing a
simple comparison of the different handoff mechanisms cited in this paper.

Table 6: Comparison of different handoff mechanisms

Wireless
interface
Handoff latency
(ms)
Infrastructure
modification
Neighbor Graphs (Shin. M et. al., 2004)
1 ~ 40 yes
SyncScan (Ramani,I. and, and Savage, S.,
(2005)
1 2 3 yes
MultiScan (Vladimir B, et. al., 2005)
2 0 no
Basis Handoff (hardware)
1 ~ 2.2 no
Reduced scan model (Hardware)
1
0.74
2.22
no

In the following scenario, an MT must sweep three channels. On each channel, three
APs are active. The Handoff latency, according to the number of APs and the channels, is then as
follows:
First, a TM performs the scan of three channels. On each channel, it expects to receive
three answers from three APs. The signal level detected during the reception will be recorded.
After consulting the first channel, the TM changes its frequency and sends probe request on the
next channel to collect the parameters characterizing the access point communicating on the same
frequency (channel 2). This scenario is repeated for all channels to scan.
With reduced scan, the objective consists in reducing the time scan phase. So a TM stops the scan
since it detects a suitable signal with a guarantee of the QoS parameters.

Figure.8 Handoff latency vs. Channels and APs number

If the adequate signal level is received in the first channel or in the second channel; the
scan time can be reduced to 1 / 3 or 2 / 3 respectively.

7. CONCLUSION

In this paper, the IEEE 802.11 convectional handoff process has been implemented on an
ASIC circuit using the high level design technique and based on the IEEE 802.11b specifications.
The conventional layer-2 handoff consumes more time in the channel-scanning process. For this
reason, we have proposed an other handoff mechanism, using a reduced scan phase. Several
scenarios have been employed in order to evaluate the handoff latencies spent for each model
compared with others mechanisms for the conventional layer-2 handoff process. Reducing handoff
latency in WLAN becoming more suitable for real time applications. For this reason we adopt the
high level design for the implementation of these models. In fact, we have used VHDL as a high
level description language, ModelSim as a simulation tool to check the behavior of each model at
the RTL level. Synopsys and cadence tools are used for synthesis, place and route step
respectively.
Other interactions between the 802.11 MAC layer and IP protocols need further study.
Since 802.11 will be the dominant technology for WLANs, a fresh look at integrating the IP stack
and the wireless MAC must be justified to design a SOC (System on Chip) of IEEE 802.11 MAC
layer.

References

Arunesh, M., Minho, S. and William, A (2003), An empirical analysis of the IEEE 802.11 MAC
layer handoff, ACM SIGCOMM Computer Communications Review (ACM CCR),
33(2), pp.93-102.
Vladimir, B., Arunesh M. and Suman B. (2005), Eliminating handoff latencies in 802.11 WLANs
using Multiple Radios, Applications, Experience, and Evaluation, Internet Measurment
Conference, 299-304.
Chiang M.H. (2006), Implementation of IEEE 802.11 MAC using FPGA: Receiver part,
Departement of Electrical Engineering Tatuang University.


Choi C.H., Il Kim, M. and JoKim, S. (2002),Call admission control using the moving pattern of
mobile user for mobile multimedia networks, Proceedings of the 27th Annual IEEE
Conference on Local Computer Networks, 2002.
Hou, J. and Fang, Y. (2001), Mobility-based call admission control schemes for wireless mobile
networks, Wireless Comm. Mobile Comput. 1 (3), 269282.
IEEE 802.11. Part 11(1999), Wireless LAN medium access control (MAC) and physical layer
(PHY) specifications, IEEE Standard 802.11.
Islam, M.M., Murshed, M. and Dooley, L.S. (2002), A direction-based bandwidth reservation
scheme for call admission control, International Conference on Computers and
Information Technology2000 , Dhaka, Bangladesh, pp. 345349.
Islam, M.M., Murshed, M. and Dooley, L.S. (2003), New mobility based call admission control
with on-demand borrowing scheme for QoS provisioning, IEEE International Conference
on Information Technology: Coding and Computing2003 ( ITCC2003), Las Vegas,
Nevada, USA, pp. 263267.
Liang, B. and Haas, Z.J. (1999), Predictive distance-based mobility management for PCS
networks, IEEE INFOCOM99, New York.
Liu, T., Bahl, P. and Chlamtac, I. (1998), Mobility modelling, location tracking and trajectory
prediction in wireless ATM networks, IEEE J. Selected Areas Comm. 16.
OHara, B. and Petrick A.I. (2005), IEEE 802.11 handbook a designers companion, second ed,
Ramani, I. and Savage, S. (2005), SyncScan: Practical Fast Handoff for 802.11 Infrastructure
Networks, Proceedings of the IEEE Infocom.
Shin, M., Mishra, A., and Arbaugh, W.A. (2004), Improving the Latency of 802.11 Hand-offs
using Neighbor Graphs, Boston, USA.
Tabbane, S. (1995), An alternative strategy for location tracking, IEEE J. Selected Areas in Comm,
13, 880892.
Yu, F. and V.C.M. Leung (2002), Mobility-based predictive call admission control and bandwidth
reservation in wireless cellular networks, Computer Networks, 38 (5), 577589.
Zaidi, M., Bhar, J., Ouni, R. and Tourki, R. (2008), A new solution for micro-mobility
management in 802.11 Wireless LANs using FPGA, . SCS 2008. 2nd International
Conference on Signals, Circuits and Systems, Hammamet Tunisia.

Received: July 8
th
2009
Accepted in final form: January 5
th
2010 after two revisions

About the authors:

Monji Zaidi received the Dipl.-Ing. in electrical engineering for automation and processes control
in 2005 from the National engineers school of Sfax and the Mastere degree in Materials,
Nanostructures, devices and micro-electronics systems from the University of Monastir, Faculty of
Sciences of Monastir (FSM), Tunisia 2007. He is currently working toward the PhD degree in
electronic and communication in the Electronic and Micro- Electronic laboratory (EE) University
of Monastir. His research interests include Management of the WLAN technologies.

Rihha Ouni received his DEA in Matriaux et Dispositif pour l'lectronique and his PhD degree
in Physics (Electronics option) from the Science Faculty of Monastir, Tunisia, in 1997 and 2003,
respectively. Currently he is an assistant professor in the College of Computer and Information
Sciences (CCIS), King Saud University. His research interest is in the field of mobility
management in Wireless communication.

Kouldoun Torki received the Ph.D. degree from the INPG, Grenoble in 1990 and the DEA
microelectronics from INPG in 1986. Currently he is the Technical Director of CMP and Project
Coordinator for PhD students exchange with the University of Monastir (Tunisia).

Rached Tourki was born in Tunis, on May 13 1948. He received the B.S. degree in Physics
(Electronics option) from Tunis University, in 1970; the M.S. and the Ph.D. in Electronics from
Orsay Electronic Institute, Paris-south University in 1971 and 1973 respectively. From 1973 to
1974 he served as Microelectronics Engineer in Thomson-CSF. He received the Doctorat detat in
Physics from Nice University in 1979. Since this date he has been Professor in Microelectronics
and Microprocessors with the Physics department in the Faculty des of Sciences of Monastir. His
researches interests are digital signal processing and hardwaresoftware codesign for rapid
prototyping in telecommunications.

Handover Strategies Challenges in Wireless ATM
Networks
Jamila Bhar, Ridha Ouni, Kholdoun Torki, and Salem Nasri

AbstractTo support user mobility for a wireless network new
mechanisms are needed and are fundamental, such as paging,
location updating, routing, and handover. Also an important key
feature is mobile QoS offered by the WATM. Several ATM network
protocols should be updated to implement mobility management and
to maintain the already ATM QoS over wireless ATM networks. A
survey of the various schemes and types of handover is provided.
Handover procedure allows guarantee the terminal connection
reestablishment when it moves between areas covered by different
base stations. It is useful to satisfy user radio link transfer without
interrupting a connection. However, failure to offer efficient solutions
will result in handover important packet loss, severe delays and
degradation of QoS offered to the applications.
This paper reviews the requirements, characteristics and open
issues of wireless ATM, particularly with regard to handover. It
introduces key aspects of WATM and mobility extensions, which are
added in the fixed ATM network. We propose a flexible approach for
handover management that will minimize the QoS deterioration.
Functional entities of this flexible approach are discussed in order to
achieve minimum impact on the connection quality when a MT
crosses the BS.

KeywordsHandover, HDL synthesis, QoS, Wireless ATM.
I. INTRODUCTION
HE Asynchronous Transfer Mode (ATM) is a data
transport technology that supports a single high speed
infrastructure for integrated broadband communication
involving voice, video and data. ATM technology combines
some important features: short fixed-size packets or cells,
virtual circuits, statistical multiplexing, and integrated
services. All these concepts together provide a uniform
framework that guarantees traffic with quality of service
(QoS) [8].
Wireless ATM (WATM) is mainly considered as an
extension of ATM network issue. WATM will be
advantageous to support the seamless delivery of multimedia
flows with high Quality of Service (QoS). WATM must in all
cases allow the network to guarantee a connection continuity
of MT.
To support user mobility for a wireless network additional
protocols are needed and are fundamental. They mainly refer
to handover protocols, routing, and location management.
Thus, maintaining QoS guarantees demands to integrate
mobility support functions.
This paper introduces the mobility management solutions
characteristics of Wireless ATM network. It presents an
extended Handover technique for terminal mobility support in
WATM networks. WATM is able to deploy different
Handover types, which are intended to manage different
network event scenarios. For this purpose, several specific
functionalities and algorithms are proposed. In our approach,
improved backward and forward hard handover protocols
were been developed for switching MT active connections
from one base station to another. This approach aimed at
defining a solution with optimal method for applying
handover in WATM environment. The emphasis was
especially on deploying innovative process while maintaining
Qos parameters. So, some details of the proposed idea are
explained and performed by synthesis and validation phase.
The paper is organized as follows. First, we give an
overview of handover protocols. The second section outlines
the testing approach adopted for our architecture. Technical
challenges are presented and discussed. Then, Handover
buffering performances are analyzed and their requirements
are explained. Finally, description of our design, synthesis and
simulation phases are described and conclusions are given.
II. HANDOVER PROTOCOLS SPECIFICATIONS
Handover term refer to different approaches to supporting
mobility aspects. Distinctions between different propositions
can be made according to the performance characteristics,
diversity steps, state transitions, and control modes of
handover techniques. Generally, Handover can be defined as
the process by which an active MT changes its point of
attachment to the network, or when such a change is
attempted. The access network may provide features to
minimize the interruption to sessions in progress.
There are different types of handover classified according to
different aspects involved in the handover. We can then
identify handover types such as backward and forward. This
distinction refers to handover steps and to the BSs through
which the handover signaling information will be exchanged.
In backward handover, signaling messages are exchanged via
the old BS (the BS the MT has been attached to during the
recent past). Handover scenario, in this case is composed of
two stages: setup and handoff. In forward handover, the link
with the old base station is suddenly lost. Thus, the MT is
forced to seek connectivity through other, neighboring BSs.
In this paper, hard backward and forward handover will be
supported by WATM. Backward handover is usually used as
first solution. When the radio link is suddenly degraded, the
MT will be notified by the signal level, and a forward
handover procedure will be initiated to recover the connection.
T
INTERNATIONAL JOURNAL OF APPLIED MATHEMATICS AND COMPUTER SCIENCES VOL. 4 NO. 2 2007 ISSN 1305-5313
ENFORMATIKA VOL. 4 NO. 2 2007 ISSN 1305-5313 636 2007 WASET.ORG
In this case, the terminal interrupts the old connection and tries
to connect to a new BS. In the following section, we present
handover management strategy based on inband signaling.
III. HANDOVER MANAGEMENT STRATEGY
A. Proposal Architecture
The handover algorithm integration, in wireless ATM,
presents several technical challenges that need to be resolved.
This paper is focused on the handover decision making
process which satisfies the objectives mentioned above. This
process may lead to decide whether handover is necessary and
whether additional adaptation needs to be applied. Objectives
include satisfying the users devices preferability, supporting
important bandwidth for their respective applications and
avoiding data loss which may affect the applications.
The integrated architecture defines the following main
entities:
- Entity at base station: depending on a traffic situation, it can
play a role of old or new BS.
- Entity at switch node: it is responsible to collect traffic
parameters. It manages information from different BSs.
- Entity at mobile terminal: it makes final Handover decision.
Terminal decides about handover type and which BS is to be
selected for the handover. We use the MT because it is more
scalable to make decision by itself, especially in degraded
situation.
Handover entities are integrated at layer 2 when a handover
appears in the same clustered area (BSs attached to the same
switch). This type of handover is transparent to the routing
functionalities (it requires a simply link layer reconfiguration
without any mobility implications) [8].
Entities listed below are described with VHDL language.
Implementing handover mechanism with VHDL description
poses some specific constraints. The development of several
nodes of the network (BSs, MTs) requires traffic management
of simultaneous connexions. Other constraint that must be
considered is to provide to mobile terminal parameter for
handover decision at the right time, with most newly
information about traffic conditions. For this purpose, traffic
collected parameters are sent periodically to MT.
The steps for handover algorithm are detailed at the
following.

B. Phases of Handover Strategy
To define our architecture while maintaining optimal
resources to the mobile terminal, the following approaches are
adopted:
A handover approach adopts a mobile-initiated handover type.
The MT makes the initial decision to initiate the handover.
However, the WATM network participates to trigger handover
by informing MT about traffic management parameters in
order to select suitable BS candidates.
In a prediction phase, the mobile terminal monitors traffic
quality and controls signal strength indicators and error
probability of the channel. It is then periodically informed
with collected information using handover cell. When
receiving these parameters, MT actualizes few stored
parameters for handover decision algorithm. The MT initiates
Handover when the signal drops below than a threshold.
If the switch is informed through old BS about a new
handover procedure, it requests BSs about information
relevant to the Handover Decision. Collected information will
be filtered. The switch sends, to the MT, a BSs list prioritized
by the signal power of each one. The Handover Decision
process at MT entity is parameterized with this information. In
consequence, the MT sends via the old base station to the
switch, a Handover Setup Request message notifying the
intention to change to the selected BS. Otherwise,
communication efforts with the current BS will be wasted.
The BS list increases probability for handover successful.
However, it may cause additional delay.
From these steps, we can see that this process ensures that
the most up-to-date information is used for handover decision.
During handover, control messages are exchanged between
different components of the WATM network. They are
required to handle functions as begin, confirm, and end of the
connection with network nodes. They inform nodes also about
handover protocols type and QoS negotiation. Handoff
signaling enables wireless terminals to move seamlessly
between BSs while maintaining connections with their
negotiated QoS. Bad handovers signaling lead to degraded
power quality. They also have a deep impact on the transport
functions and band occupancy.
In our model, resource allocation is done after the switch
decision. In fact, it is not necessary to reserve resources in
each BS for the connection when they receive the HOR
message because only one BS will be finally chosen. Switch
needs to send only one Handover message to the selected BS
for resources allocation. This solution reduces the message
processing time of other BSs. The efficiency of the whole
network will thus be improved.
IV. ANOTHER ISSUES AND CHALLENGES IN WATM
A. Signaling Strategies
Handoff signaling enables wireless terminals to move
seamlessly between BSs while maintaining connections with
their negotiated QoS. Bad handovers signaling lead to
degraded power quality. They also have a deep impact on the
transport functions and band occupancy. [1] Presents a scheme
for handover provisioning in Wireless ATM networks based
on in-band signaling. Signaling information is carried using
fixed cell size equal to data cell. Handover signaling message
integrates control channel for some signaling functions.
Therefore, in proposal [8], the handover protocol is entirely
based on dedicated cells that are transmitted with the data
flow. The dedicated cells, termed Mobility Enhancement
Signaling (MES) cells, are Resource Management (RM) cells
similar to those used in the Available Bit Rate (ABR) ATM
transfer capabilities. We propose to use the in-band signaling
technique as explained in [1]-[8]. The goal is to efficiently
collect and manage the network information. This choice has
significant advantages such as modification requests
avoidance in WATM signaling functions. It also guarantees
in-sequence cell delivery over the connection during handover
procedure.
B. A Survey of Buffering Issue
The main complexity of WATM arises from the functions
and protocols for handover. Thus, an important issue in
WATM that needs further investigation is to maintain QoS
parameters for connections during handover. As WATM has
these critical characteristics, a main consequence is the need
for reducing data loss. An optimal handover procedure must
enable the network with a guaranteed level of QoS being
protected against cell loss, cell duplication, and loss of cell
sequence. An optimal design of handover should give a
lossless mechanism that also has low delay and delay
variation. [11] Consider that the main consideration during
handover is to maintain connection quality. Ensuring the
completion of handover procedure by preventing any cell loss
and avoiding cell duplication or cell reordering with very low
delay is of primary importance.
To ensure these conditions, Handover procedure should
guarantee an in-sequence cells delivery to terminals, with
desired QoS parameters. Since, the connections must be
handed over new BS while QoS requirements must be
satisfied. For this purpose, fixing optimal handover steps are
useful (Fig. 1).

Handover
type
HO measurement
-Signal power threshold
-Network parameters
SETUP: Check for
BSs resources
MT =>New BS=>SW
Backward
Handover
Forward
Handover
MT =>Old BS=> SW
Handover
Decision
Good traffic
conditions
T

Fig. 1 Handover stages

A promising approach to meet QoS requirements is based
on the storage of data cells in the selected BS buffer, while the
connection is being reestablished. Indeed, until the new
wireless link is created, cells cannot be transmitted between
the MT and the BSs. During this time, it may be necessary to
transfer stored data cells from old BS to new one. The reason
is that MT is disconnected from current BS and not yet
connected to the selected one. The old BS sends a handover
confirmation when their buffers are being emptied. Then the
switch notifies BSs that the handover is in final progress step.
It informs old BS to disconnect from the MT by sending the
HandOver End (HOE) message. New BS is also informed to
establish a new wireless link with the MT.
There are different choices with which MT should establish
the new connection. Discussions about the deployment of
efficient buffering mechanisms have taken place. In [5] data is
sent simultaneously to MT from old and new BS. This
approach minimizes buffer occupancy but it needs more
resources. This approach is called Make-before-break (MBB).
However, in [3] MT can wait for current BS to send all
buffered cells to new one. This is a Break-before-make (BBM)
approach. This is a good choice for connections without
critical timing requirements. However, when there is no end
buffering confirm message, data loss can appear. Thus, we
think that it is important to indicate the number of cell
buffered in the old BS. Else, a cell transfer in old BS to new
one may be not guaranteed. Here we assume that MT can wait
for current BS to send all buffered cells. Old BS transmits to
the switch end buffering information allowing the new BS to
start transmission so that the cell order is maintained (figure
2). Old BS informs the switch about the last quantity of the
data stored in the buffer. Data are measured in multiple NRM
user cells followed by a numbered RM cell. However, [7]
considers that MT should establish connection with new BS
before disconnecting from the old one. In this way, old BS can
be asked for another handover if MT cannot establish the new
radio link, thus ensuring a lossless handover. This approach
minimizes also buffer occupancy but it needs more resources.

HO Begin
5
6
4
3
2
2
2
1
Data transfert
MT
Up

BS2

Down
Up

BS1

Down
HO Begin
HOC (N cells in the
memory)
Switch

HO End
HO End
HO End
HO Confirm

Fig. 2 Handover buffer occupancy issue

In a successful Handover case, MT and new BS exchange
cells via the newly established wireless link. Otherwise, the
switch sends a HandOver Denied (HOD) message, and the
handover procedure can be re-initiated by the Terminal. One
of the disadvantages of an incremental re-establishment
handover algorithm is that all nodes in a network have to
support the augmented signalling protocol.
V. CONTRIBUTION OF THIS PAPER
This paper proposes a handoff solution for a WATM
environment. The proposed solution has for goal to meet the
challenges of wireless ATM and to reduce the number of
control signals required for handoff and the volume of
buffered information packets during handoff. It tries to
significantly improve the radio link transmission accuracy.
Performance results reported in the literature are mostly
obtained via analytical models [5]. Simulative study is planned
and it leads to characterize the effectiveness of the handover
protocols by testing several network situations.
In this work, each of the components is designed using a
hardware description language and synthesized to an FPGA. A
HDL description defines details of each of the different
components. Network architecture includes Switch, BSs and
MTs. The performance of the entire system is also dependent
on the interactions of these components when they are being
used by different traffic patterns. For this purpose, we
introduce a flexible architecture that supports dynamic
behavior of the system.
The objective of our methodology is to provide a rigorous
design flow for high-performance processing networks. For
accuracy, we have designed and implemented our components
using the VHDL hardware description language. To evaluate
system performances, we have synthesized our VHDL design.
The synthesis tool performs a detailed timing analysis and
reports a maximum clock frequency.
VI. SIMULATION AND SYNTHESIS RESULTS
Our approach of handover has been transposed on a concise
description which supports different Wireless ATM simulation
scenarios. The efficiency of this description for several
network situations evaluates the Handover algorithm
performances. Handover algorithm has been implemented in
an FPGA environment based on simulation and synthesis
tools. This algorithm has been integrated, in three parts, on the
base station, the MT and the switch according to their specific
tasks. Simulation environment contains two base stations

Fig. 3 Handoff from BS1 to BS2
Handover
Ready
Handover End
Handover Start
Handover End
Data cells from BS1 Data cells from BS2
Interrupted period

Fig. 4 Handoff Setup stage
BS1 informations
Data cells from BS1
Maintain connection
Handover Request
announcement
Request to information collect
(BS1, BS2), one mobile terminal (MT) and one switch (SW).
This architecture is based on two channels that could be
established from the switch to the MT. Our proposed handover
model is extensible for multi-channel architecture based on
multi-hierarchical level (MTs, BSs, SWs).
In Fig. 3 multiple Handover messages are omitted, we give
here an overview of critical sequences of Handover process.
Fig. 3 shows a network scenario that we consider for
performance evaluation of handover procedures. It presents a
BS switching through which a terminal receives data. The BS
converts cell data in order to adapt ATM and WATM cell
formats. Data cell transmission is initialized by Request To
Send (RTS) signal. Data are transmitted over (32 bits) port
accompanied with four control bits. A control bit is set to 1
to indicate delimiter and special characters and 0 to indicate
data bits. Fig. 4 shows that for some traffic condition,
handover setup algorithm can maintain the current base station
as more suitable to support the requested MT connection. In
this case, handover procedure is interrupted.

TABLE I
SYNTHESIS SUMMARY RESULT
MT BS SW
I/O put
ports
154 224 224
Number of
cells
6783 16208 15350
Total area
(mm
2
)
3.706 8.192 7.567

Fig. 5 Layout of BS entity
Table I summarizes implementation parameters for the
handover contribution in each system (MT, BS and SW). It
contains synthesis results obtained by synopsys tool.
Implementation parameters are explained with the needed
in/out put ports, the integrated equivalent cell number and the
occupied area. At this design level, a powerful tool for placing
and routing outlines the layout of the designed systems (Fig.
5).
Finally, results obtained from handover protocol
implementation reveal that implementing mechanism is
feasible, as it maintains the QoS characteristics and fails at
higher data rates. Since, it is essential to evaluate the handover
protocol in an environment with multiple mobile terminals. It
is also important to apply routing algorithm to provide a
robust handover mechanism.
VII. CONCLUSIONS AND FUTURE DIRECTIONS
This paper has presented a work for designing, synthesizing
and simulating networks service. By using a hardware design
flow, each component can be designed and characterized
separately. By using FPGAs technology, we have presented
performance results. The hardware synthesis tools provide a
maximum frequency of the device, and from simulations we
can determine the latency in terms of clock cycles.
This work has involved the impact of handover protocols in
a wireless ATM environment. For this purpose, developed
algorithms collect information to allow mobile nodes to
execute handover decisions in an optimal way. These
algorithms explain also a complexity of implementing
handover mechanism with VHDL description language.
To increase the efficiency of handovers, a survey of buffering
issue is considered when MT switches from BS to another.
The proposed architecture shows that necessary signalling
information must be available at the right place and at the right
time to reduce handover failure.
In fact, intelligent handover decisions are important in mobile
networks with different capabilities. Our solution ensures not
only that it handles diverse, and dynamic traffic situation but
also a hardware feasibility.

REFERENCES
[1] Carla-Fabiana Chiasserini, Renato Lo Cigno Handovers in Wireless
ATM Networks: In-band Signaling Protocols and Performance
Analysis, IEEE Transactions on Wireless Communications, Vol. l1, pp.
87-100, January 2002.
[2] Qing Wei et al Context-aware handover using active network
technology; Computer Networks; November 2005.
[3] Udo R.Krieger and Michael Savoric Performance evaluation of
Handover protocols for data communication in a Wireless ATM
network, International Teletraffic Congress (ITC16), vol. 3a & 3b, pp.
1261-1270, 1999.
[4] A. Pitsillides, F. N. Pavlidou A Survey of Wireless ATM Handover
Issues, Proceedings of the International Symposium of 3G
Infrastructure and Services, Athens, Greece, pp. 34-39, 2001.
[5] J.R. et al Radio channel emulation and multimedia communications
Handover support in an experimental WATM network, EUNICE99.
[6] Dimitrios D. et al Simulation, Modeling and Analysis of Path
Rerouting Algorithms for Handoff Control in Wireless ATM Networks,
Int. Conf. SCS2002.
[7] Fan Jiang Microcellular Handover in Wireless ATM, Proceedings of
the 32nd Hawaii International Conference on System Sciences, 1999.
[8] Marco Ajmone Marsan, et al Local and Global Handovers Based on In-
Band Signaling in Wireless ATM Networks, Wireless Networks, Vol. 7,
No.4, pp. 425-436, ISSN: 1022-0038, July 2001.
[9] Marco Ajmone Marsan, et al Performance Models of Handover
Protocols and Buffering Policies in Mobile Wireless ATM Networks,
IEEE Transactions on Vehicular Technology, Vol. 50, No.4, pp. 925-
941, July 2001.
[10] Qing Wei, Kroly Farkas, Christian Prehofer, Paulo Mendes, Bernhard
Plattner Context-aware handover using active network technology;
Computer Networks; 2006.
[11] Raymond R. Hoarea, Zhu Dinga, Shenchih Tunga, Rami Melhemb, Alex
K. Jonesa A framework for the design, synthesis and cycle-accurate
simulation of multiprocessor networks; J. Parallel Distrib. Comput. 65;
2005.
[12] Nasf Ekiz, Tara Salih, Sibel Kkner and Kemal Fidanboylu An
Overview of Handoff Techniques in Cellular Networks, International
Journal of Information Technology, vol. 2, No. 3, ISSN 1305-239X,
2005.

Jamila Bhar received her Engineering diploma in Electric and her DEA in
Communication system from the National School of Engineering of Tunis
(ENIT), Tunisia in 2001 and 2002, respectively. Currently, she is a PhD
student. Her research interests include protocol adaptation in heterogeneous
networks, traffic management and Quality of Service for high speed networks.
Her recent work has been in traffic control in WATM network.

Ridha Ouni received his doctoral degree in physic (2002) from the Science
Faculty of Monastir, Tunisia He is currently an assistant Professor at the
Preparatory Institute of Engineering Study of Monastir (IPEIM), Tunisia. His
research interests include computer networks, flow and congestion control,
interoperability and performance evaluation. He is interested in many areas of
hardware/software protocol verification and design for distributed systems.

Salem Nasri received his Doctoral degree in automatic control and computer
engineering from the National Institute of Applied Sciences of Toulouse
France, in June 1985. His research interests are in the fields of computer
networks, communication systems and multimedia applications. In May 2001
he obtained the diploma of Habilitation universitaire. Since then he has been
a professor. He developed collaboration with many laboratories in France
such as LSR (Grenoble), CRAN (Nancy), and some other laboratories in
Tunisia. Currently he is a professor in the Computer Science Department,
Qassim University in the Kingdom of Saudi Arabia.

CONFERENCES INTERNATIONALES

1. Thamer Alanazi, Ridha Ouni, Traffic Differentiation and Scheduling for QoS support in Vehicular
Sensor Networks, IEEE International Conference on Technological Advances in Electrical, Electronics
and Computer Engineering (TAEECE), Turkey, 2013 (ISBN: 978-1-4673-5611-4).
2. Jihed Khaskhoussi, Ridha Ouni, Enhanced-MPR-CDS approach for self-organization and routing in
WSNs, 1
st
Taibah University International Conference on Computing and Information Technology Al-
Madinah Al-Munawwarah, Saudi Arabia, 19-21 Rabi II 1433 Hijri (12-14 March 2012)
3. Monji Zaidi, Ridha Ouni, Jamila Bhar and Rached Tourki, A novel positioning technique with low
complexity in wireless LAN: hardware implementation, Proceedings of the World Congress on
Engineering, Vol II, July 6 - 8, 2011, London, U.K (ISSN: 2078-0966).
4. Monji ZAIDI, Jamila BHAR, Ridha OUNI and Rached TOURKI. Reducing Wi-Fi handover delay using a
new positioning process, International Conference on Communications, Computing and Control
Applications (CCCA), Hammamet, pp. 1-6, 2011 (ISBN.978-1-4244-9795-9).
5. M. Z. Hourani, R. Ouni, N. Hussain, Novel Data Harvesting Scheme for Efficient Data Aggregation,
ICW2941, WORLDCOMP'11, USA. 2011. (http://world-comp.org/p2011/ICW2941.pdf)
6. Monji ZAIDI, Rached TOURKI, Ridha OUNI, New geometric Approach to Mobile Position in wireless
LAN reducing complex computations, International Conference on Design & Technology of Integrated
Systems in Nanoscale Era, pp. 1-7, 2010.
7. Jamila BHAR, Ridha OUNI, Salem NASRI, Performance evaluation of compensation-reward
mechanism for resource allocation in wireless networks, The 15
th
IEEE International Conference on
Electronics, Circuits, and Systems, Malta 31
st
August to 3
rd
September 2008, pp 1195-1200.
8. Jamila BHAR, Mongi ZAIDI, Ridha OUNI, Salem NASRI, Performance Evaluation of Fuzzy
Controller for Traffic Stabilization, International Conference on Signals, Circuits & Systems (SCS08)
November 7-9, 2008, Hammamet (Tunisie).
9. Mongi ZAIDI, Jamila BHAR, Ridha OUNI, Rached TOURKI, A new solutions for micro-mobility
management in 802.11 Wireless LANs using FPGA, International Conference on Signals, Circuits &
Systems (SCS08) November 7-9, 2008 Hammamet (Tunisie).
10. Jamila BHAR, Ridha OUNI and Salem NASRI, Interoprabilit Ethernet/WATM, International
Conference: Sciences of Electronic, Technologies of Information and Telecommunications, Tunisia,
March 15-20, 2004.
11. Jamila BHAR, Ridha OUNI, Abdelhamid HELALI, Salem NASRI, Improvements of the ABR loop
performances in a wireless ATM network, International Conference on Microelectronic, IEEE
Conference, Gammarth, Tunisia, December 2004.

Traffc Differentiation and Scheduling for QoS
support in Vehicular Sensor Networks
JhumCIN^IunuZ
College of Computer and Information Sciences
King Saud University
Riyadh, KSA
thamermatar@hotmail.com
AhstractThe IEEE b2.1.4 is widely used next-generation
standard protocol in many applications utilizing wireless sensor
networks (WSNs) especially in vehicular sensor networks (VSNs).
However, currently differentiation and scheduling mechanisms
are not provided in IEEE b2.1.4 specifcation to improve the
quality of service (QoS) for delay-sensitive and critical events. In
this paper, multiple scheduling algorithms such as FIFO, Priority
queue, WRR and DWRR are integrated in compliance with
IEEE b2.1.4 to improve the throughput, enhance the
bandwidth utilization rate and perform fast processing and
delivery for urgent data trafc. we divide the standard trafc
into various types of data on the basis of different QoS
requirements such as real-time fxed data packets generation on
periodic basis, real-time variable size data packets generation on
periodic basis, non real-time variable size data packets and best
effort. The mandatory QoS service fow parameters are defned
for each type of trafc according to its time-criticality and delay
sensitivity.
Ke;ordsIehlcular sensor networks; scheduhng; Quaht; of
servlce;dgerentlatlon;1 02.J.4;ZlgBee.
I. INTRODUCTION
Sensor networks are recently rapidly growing research area
in wireless communication networks. Wireless sensors are of
small size and low-cost are deployed to establish a sensor
network. Adoption of IEEE 802.15.4 standard [7] for physical
and medium access control (MAC) layers and ZigBee
specifcation [8] for networking and application layers have
rapidly increased applications of Wireless sensor networks
(WSN s) in various areas such as industrial automation[ 1],
home and building automation, automatic meter reading,
embedded sensing and distributed process control systems [2,
3], machinery health monitoring [4, 5], smoke waring and
radiation leakage detection systems [6].
Vehicular networks are considered as mobile sensor
networks and characterized by several basic and special
characteristics such as no limited energy and storage capacity,
high node mobility and fast topology changes. The vehicular
sensor network can sense several types of data in its
surrounding area to provide wide variety of services like
traffc monitoring, crowded streets identifing, speed
controlling, lost vehicle locating and environmental
monitoring since it covers permanently a wide geographical
area [2,3,22].
C. J78J467356J382UJ3 LLL
dhu |un
King Saud University
Riyadh, KSA
rouni@ksu.edu.sa
For wireless sensor networks (WSNs), IEEE 802.15.4 is
used as de-facto standard. However, the behavior of
CSMA/CA, collision at heavy load, reduces the throughput
and energy consumption performance of WSN. These
problems demand MAC layer solutions to be proposed to
achieve the better performance of WSN. Scheduling aids in
providing quality of service (QoS) support to the prioritized
and categorized communication in wireless sensor networks.
This research aims to enhance QoS in a Vehicular Sensor
Networks (VSN) by integrating traffc differentiation and
scheduling mechanisms in order to reduce the end-to-end
delay, improve the throughput, enhance the bandwidth
utilization rate and perform fast processing and delivery for
urgent data trafc. We divide the standard trafc into various
data types on the basis of different QoS requirements as the
compulsory QoS service fow parameters are defned for each
type of traffc. The type of data may categorize as real-time
fxed data packets generation on periodic basis, real-time
variable size data packets generation on periodic basis, non
real-time variable size data packets and best effort.
The rest of this paper is structured as follows. Section 2 gives
a summary of related works and Section 3 gives a brief
overview of service differentiation and prioritization
methodology used in our scenario. Hence research constraints
used by our model and generated simulation results are
provided in Section 4. Finally, concluding remarks and future
work are presented in Section 5.
II. LITERATURE REVIEW
The MAC layer includes a very important processing level
since it rules the sharing of the medium which affects the
performance of all the upper layer protocols. MAC protocol
support QoS provisioning and determining the QoS support
performance by solving the medium sharing problems and
reliable communication.
Dif-MAC is a QoS aware MAC protocol based on
CSMA/CA access method to support hybrid prioritization and
differentiated services. Diff-MAC integrates an effective
service differentiation algorithm in order to increase the
channel utilization and provide fair and fast data delivery.
Diff-MAC is needed in WSN supporting QoS-constrained
heterogeneous trafc such as multimedia applications. To
34
provide QoS, Diff-MAC consists of (1) reducing the
retransmission using fagmentation of the long fames into
small manageable packets and transmitting them in form of
burst, (2) decreasing collisions and minimizing the packet
latencies by adjusting its contention window size as per traffc
requirements and (3) providing fair and reliable data delivery
among sensor nodes based on intra-queue prioritization feature
[19].
[5] Uses CSMA/CA as access protocol to provide service
differentiation in WSN. The Collect then Send burst Scheme
(CoSenS) is developed to facilitate implementation of
scheduling policies and primarily to handle its weaknesses. A
earliest deadline frst and fxed priority are implemented on
the top of CoSenS. The results present that the proposed
solution enhances reliability and end-to-end delay by adapting
traffc variations automatically. Authors claim that proposed
solution does not affect best effort trafc while meeting
deadline requirements for urgent trafc. Moreover, motes are
used for testing and implementation of CoSenS.
[6] Developed a novel cross-layer integrating an
asynchronous Energy Efcient and Fast Forwarding (EEFF)
protocol for WSNs is resulting to energy efciency and low
latency. EEFF implements new approaches improving
dynamic routing selection and low power listening which
leads to reducing the latency.
[7] Proposed a service differentiation algorithm with slight
modifcation on the protocol to enhance the achievement of
slotted CSMA/CA for time-critical events. The service
differentiation algorithms were particularly based on various
parameters such as the macHinE, aMBE and the Contention
Window (CW). They differently process the command and
data fames since they are affected by high and low priority
levels (service class), respectively. In other terms, different
attributes have been defned and assigned for different service
classes. This algorithm keeps slotted CSMAICA in its original
form and focuses on tuning related parameters effectively in
keeping the criticality of messages. Some existing works
[28,29,30] are interested in controlling over CW depending on
the changes in the network status. In [28], the Sensing Back
off Algorithm (SBA) has been addressed to maximize channel
throughput with impartial access to shared channel. When
packet collision occurs, it multiplies its back off interval by
D while on a successfl transmission, both sending and
receiving wireless sensors multiply their back off interval by
U, and the others overhearing(sensing) a successfl
transmission decreases their back off intervals by . D, U and
are defned in [28]. However, on the basis of p-persistent
CSMA/CA protocol, [29,30] addresses dynamic IEEE 802.11
wireless networks. Their approaches assume having a precise
number of the active wireless sensors, to estimate the network
state, while they do not consider QoS for real-time trafcs.
Node-based scheduling and level based scheduling,
proposed in [8], are two centralized heuristic scheduling
algorithms. The frst algorithm is inspired fom the classical
multi-hop scheduling using direct scheduling of the nodes
given in an ad hoc mode. The second algorithm uses a routing
tree to schedule the levels before scheduling the nodes. This
algorithm is more suitable for wireless sensor networks since
C. J78J467356J382UJ3 LLL
it supports many-to-one communication model. A nodes
distribution across levels affects the performance of these
algorithms.
In [9], the authors proposed at the MAC level a scheduling
algorithm that is able to support assorted connections with
different QoS necessities. At the physical (PRY) layer, each
connection utilize an adaptive modulation and coding (AMC)
scheme over wireless fading channels. The scheduling
algorithm assigns a certain priority level based on the QoS
requirements of each connection. Then, it adjusts dynamically
the priority level according to the channel and service status.
[25] Proposed a Real-Time Query Scheduling (RTQS)
algorithm for confict-fee transmission scheduling in order to
support real-time queries in WSNs. In this context, in confict
fee query scheduling [25] showed relatedness between
prioritization and throughput. Then, it proposed non
preemptive, preemptive and slack stealing query scheduling
algorithms as novel approaches for real-time scheduling. As a
result, the frst algorithm achieves a better throughput by
inverting priority. This problem has been solved by the second
algorithm with trade-off of reduced throughput. Finally, the
third algorithm combined the remuneration of preemptive and
non-preemptive scheduling algorithms to improve the
throughput and meet query targets.
In [26], the authors proposed scheduling algorithms that
are able to guarantee better processing and delivery especially
for data packets. These algorithms are namely the weighted
hop scheduling algorithm with Dynamic Source Routing
(DSR) and the weighted distance scheduling algorithm with
Greedy Perimeter Stateless Routing (GPSR) where the
scheduler lies above the MAC layer and between the routing
agent. In this context, these algorithms affect the data packets
with a higher weight in order to reduce the number of hops (or
geographic distances) towards their destinations and optimize
signifcantly the average delay without any additional control
packet exchange. They demonstrate that the average delay
reduces as the movement of nodes rises. The conventional
scheduling is considered which is typically used in mobile ad
hoc networks.
Current WSN applications generate different types of
trafc with various requirements such as delay-bounded,
bandwidth and reliable data delivery. Consequently, Quality
of-Service (QoS)-based mechanisms can improve efciently
the traffc delivery in WSNs. This work introduces new
differentiated service approaches and tasks accomplished by
scheduling disciplines and highlighting the impact of these
techniques on the QoS support in mobile sensor networks.
. SERVICE DIFFERENTIATION AND PRIORITIZATION
METHODOLOGY
Different types of traffc with various requirements such as
delay-bounded, bandwidth and reliable data delivery are
generated. Consequently, Quality-of-Service (QoS)-based
mechanisms can decrease end-to-end delay and improve
efciently the traffc delivery in a wireless sensor networks.
We used differentiated service approach with scheduling
mechanism to improve the overall network throughput.
35
. Problem formulation
A specifc problem that arises as a result of the collected
traffc diversity is how to differentiate and process the
diversifed trafc in a suitable way to their requirements. The
traffc diversity is caused by multidisciplinary supported
applications. It is controlled at the roadside unit (or base
station) acting as routers and coordinators. Traffc diversity
poses challenges that need to be resolved by integrating new
mechanisms to (a) classif packets according to their types of
service and (b) schedule them appropriately to their
requirements.
Grade of service is one of crucial parts of QoS in mobile
communications which involves outage probability and
blocking probability and scheduling starvation. Various
mechanisms such as mobility management, fair scheduling,
radio resource management, channel-dependent scheduling etc
are affected to measure the above said performance measures.
B. Possible Solutions
[t includes the use of message relay boxes for collection,
classifcation and scheduling messages and specifc roadside
gateways for proper data propagation. Moreover, maintaining
Quality-of-Service (QoS) in VSNs is challenging while nodes
are mobile.
IEEE 802.15.4 defmes unslotted CSMAICA channel
access protocol which enables contending wireless sensors to
access the shared channel without providing service
differentiation at the MAC layer. This lack of providing
service differentiation has hindered the development of service
differentiation model for rate-sensitive applications.
[n this paper, a suitable scheduling scheme among various
scheduler schemes is selected at MAC layer for assorted
connections with varied QoS requirements. Therefore, a
priority or weighted fnction is requested for every lim
established in the system and depending on wireless channel
quality, service priority across layers and QoS satisfaction
every connection is updated dynamically. The proposed
scheduling model is fexible, scalable, easily implementable,
guarantees QoS and utilizes the wireless bandwidth
efciently.
MAC layer controls medium sharing and all upper layer
protocols related to that for QoS provisioning. QoS cannot be
achieved at network, transport or higher layers without support
of MAC protocol.
The aim of this research consists of supporting Quality of
Service (QoS) in a vehicular sensor environment by
integrating traffc differentiation and scheduling mechanisms.
To address QoS provisioning, the research uses the model of
Service Differentiation. Service differentiation has two
stages: (i) assigning priority, and (ii) differentiation between
priority levels. The QoS is ensured using Queue Scheduling. A
better performance is achieved by assigning appropriate
priority to the trafc since higher priority is always served
frst.
L Diferentiation in VSN
The frst step for supporting Quality-of-Service (QoS) in
VSNs consists of including differentiation mechanism in the
C. J78J467356J382UJ3 LLL
MAC layer, since several types of events with different
signifcance and severity may happen in the roads. Moreover,
other non-related road traffc is to be supported by the sensor
network such as pollution control, urban application etc. The
differentiation mechanism will not retransmit packets as they
arrive but it consists of
Collecting and classifing data fom cars and other
neighbor platforms,
Marking and storing data in different queues
characterized with different priority levels.
D. Scheduling in VSN
The scheduling in VSN is achieved and tested using the
queuing methods such as F[FO, priority queue, WRR, DWRR.
Our proposed solution is evaluated by multiple scenarios using
OPNET simulation tool. The simulation results show the
proposed system improves the QoS when compared with
standard system.
The proposed system can achieve fast categorization of
incoming trafc at RSU fom the vehicles and treat them
according to their prioritization assigned for each trafc type.
The extensive simulation results frther justif the useflness
of proposed system to get better QoS in VSN.
IV. SERVICE DIFFERENTIATION AND PRIORITIZATION
METHODOLOGY
A low-cost and energy efcient IEEE 802. [5.4 radio
technology is used in nodes. These nodes communicate with
road side units positioned over small distances along road
side. In our simulation, we have tested FIFO, priority queue,
WRR, DWRR scheduling algorithms to determine how quality
of service can be enhanced.
. Research Constraints
The research has been simulated using OPNET on 6 lanes
with 8 Coordinators and 16 Routers comprising of [0 km road
length. For simplicity, straight roads are considered and tus,
comers and exits are omitted in our proposed high-way model.
Each vehicle in the system is assumed to be equipped with a
vehicle sensor system to send vehicle's information request to
the RSU. The trafc has been classifed into 5 Data Types
such as Audio, Video, SMS, Email and Interet. For thorough
testing the proposed scheme has been applied on packets of
different sizes such as 500, 1024, [500, 2500 as shown in
Table. 1.
We assume that vehicles running on the road with constant
driving behaviors, such as lane change, acceleration, and
overtaking, deceleration. Vehicles are moving in constant
speed and moving in their lane. Afer the distance d 1 ,d2,d3 is
reached, the vehicle may wait for constant time period for
signals on the road. Multiple scenarios are simulated
concurrently and compared.
The comparison includes the following statistics: end-to
end delay, media access delay, load and throughput. In this
scenario, DWRR, WRR, priority queue and F[FO scheduling
mechanisms are considered. the number and type of ZigBee
nodes in all scheduling mechanisms are same.
36
TALE I. ZIC HLLIARAML1LRS
aramclcrs alucs
Transmit Power 0.05
Transmit band 2.4 GHz
Ma. routers 16
Ma. coordinators 8
Packet Reception-Power Threshold -85
Packet Size 512,1024,1500,2500
Channel Sensing duration 0.1
we defne six trajectories where mobile nodes will travel as
shown in Fig. 1. if mobile node is out of its parent
transmission range, then it connects to the closer node and it
continues with transmission. The network structure of
simulated scenario using OPNET Modeler is shown in Fig. 2.
B. End-la-end delay
End-to-end delay is used to measure network delay faced
by every packet. it is measured as time interval fom message
transmission to the message complete delivery at receiving
end. Fig. 3 shows the end-to-end delay result of the simulated
scenario using differentiation and simulated mechanisms. The
DWR and WR mechanisms have less end-to-end delay as
compared to FIFO in this simulation.
c|os(1J cs 1i Uls|ncc(1J AMudc(1] lvc|:c11c OI0 w1nc AcOu1l1c |||ch|dcgtccs] .'(dcccs HOJ(dcgccsJ
S",od
1 0= 0= -+ -+ -t+ 00.00. Aoced ALtCed U-pccliCd
2 0=81027294 81.027293 0[((0)) 29.17, 0.213674 10_00. 39,17: AuOOod ALtO Umpedied
3 {I268302 206.056165 125,023133 0[((0)) 45.01, 6.213782 10.00. lm34.16. Autoc,-1 ALtocoed Un,pedied
4 0.5366(5 3)J.011827 6.21428S 2m28.00s Autoc,ed Au'cc,ed Umpeclied
5 0 26S3J2 457.45541S 6.2136&4 moos Autoc,e1 ALtoed Unspeclied
6 Ooo:roJ 581S476f mOO; 4m19.SO, Au'oc,cd ALtCed Umlied
.5366057C6,433 124.493530 0[((0)) 10.00s 3n14.32s Aut:O,ed ALtocunp<J.d Umpedied
'.07320'631.737126 125392 0[((0)) 45'" 6.213356 10_00; 6m0943. Autoeoed A<ocoed Un,peclied
1.341511 869.248 13.52: 6m32.95s Autoc,ed Au'cO,ed Umpeclied
Fig. I. Trajectory Information
Fig. 2. OPNET simulation
C. J78J467356J382UJ3 LLL
cc~
I
``

~~
M|CC|.fO.4Cr1OI!!iCCJCIWOr|
(LuSl|!|O-O!!|CJe!WOt|.fO.4r1O)
ZQM~C-UlC_O.1|_DV!h-LLS-1
OC1Oc4r1f'ff'cJwOrk
(LeS|ln|On- O!!|cJe!WO||Our1O)
Z|gOM~C-UlC_O.4|_h-DLS-1
Lb|CCfOuCruOI!!|CCJC|WOr|
(LuSl|!|O-O!!|CJe!WO|| mOO|l_uOdO_Uo)
Z|gOM~C-UlC_OJ|_rl-DLS-1
r
t{
~
r
-
`
`
Fig. 3. End-to-end delay
ZQC-Uu S|C_MhOu|r_Dhh-DbS-1
Z_|4AC-US|C_MhOur_T-LLS-1
L Z|gOAC-U|C_MhOur_HP-DLS-1
L Z|gOMAC-Uu S|C_MhOu|r_h-LbS-1
v|q(|nZ|gUB02_15_4 ACDly(C))
.
U
{
+~

t"
+
{
-
+
1
+
Fig. 4. Delay
{
r{
,
,
. r
{
~
t{'
+
tf
+
Delay is measured when packets of data take more time

than expected to reach destination. Fig. 4 shows the
measurement for overall global delay for FIFO, HPF, WR
and DWR scheduling schemes. Multiple factors contribute to
delay such as network congestion and packet processing at
each link till the fnal destination arives. Their effects can be
minimized by selecting a proper scheduling scheme. It is
observed that DWR and WR have a maximum value
approximately similar maximum values. whereas delay is
minimum in case of FIFO scheduling scheme.

``
LQL-US|C_MhOc4| _L-LLS-1
ZC-US|C_MhOc4| _lO-DLS-1
L LQM~L-S|C _MhOc4| _H|| -LLS-1
L LM~L-US|C_MhOc4| _\f-DLS-1
v|g(|nZQ02 15 4 ACWACCSSL|'(SC))
'
`
'
Fig. 5.
'
r{
Media access delay

-
'
r;
To analyze the performance of various sceduling

algorithms in providing QoS to users of vehicular sensor
network, Medium Access Delay is considered as an important
factor. During the evaluation of FIFO, HPF, WR and
DWR. Various scenarios, with same Physical and MAC
37
parameters, each implementing the different sceduling
scheme, were created. The results showed that the
performance of WRR and DWRR is better in providing QoS
for various types of trafc services as compared to FIFO and
HPF, because of their ability to differentiate and prioritize
various services.
L Data Tc Received
Data trafc received can be expressed as "number of bits
of the data received per unit time". Fig. 6 depicts the data
traffc received for the FIFO, HPF, WRR and DWRR
scheduling methodologies respectively in vehicular sensor
network. It noticeably point out that the data trafc received is
maximum in case of WRR scheduling scheme because each
packet fow or connection has its own packet queue in a
network interface card. WRR serves a amount of packets for
every nonempty queue.
ZQbeeMAC-BSC_ou|er _0JFH-0LS-1
_beaMAC-BSjouer_O-DLS-1
L|gbeaMAC-US|O_ ouer _H-DLS-1
L Z|gbeeMAC-US|c_ou|er _1Vh-DLS-1
,
,,,,, ~ . . MA C ' |c h 1 ))

. :.. :
.

:..:..| |
.|[
|
.. |
.
Fig, 6,
-+
-'
`
-'
Data Traffc Received

Also it is noted that data trafc received is minimum in
case of DWRR scheduling scheme as packets are held back
those exceed fom the packet length for the next round of
scheduler. Those packets exceeds fom packet length can be
calculated by subtracting maximum packet size number fom
packet length. Although DWRR scheduling scheme can
handle variable packet size without knowledge of their mean
size.
D. Data Tc Sent
Data trafc sent is expressed as the total number of bits
sent fom source to destination per unit time. Data traffc sent
includes all data bits irrespective of the condition whether
these bits reach the destination or not. Fig. 7 indicates the data
traffc sent for FIFO, HPF, WRR and DWRR scheduling
schemes.
It is noticed that data sent is maximum in case of DWRR
scheduling scheme as packets are held back those exceed fom
the packet length for the next round of scheduler. Those
packets exceeds fom packet length can be calculated by
subtracting maximum packet size number fom packet length.
Also DWRR scheduling scheme can handle variable packet
size without knowledge of their mean size. It achieves a better
generalized processor sharing (GPS) approximation without
prior knowledge of mean packet size of each connection.
Also it has been noticed that data traffc sent is minimum
in FIFO scheduling scheme as it organizes data relative to
time and does not perform manipulation of data on the basis of
C. J78J467356J382UJ3 LLL
prioritization. Moreover, it processes queue by ordering data
in frst come frst serve behavior, where each packet leaves the
queue in order they come.
|C-S|C_ouler_-0LS-1
|C-bS|C_ouler_O-DLS-1
L|gbeeMAC-bS|C_ouler_H||-0LS-1
L|gbeeMAC-bS|C_ot1er_JFh-DLS-1
Fig, 7, Data Tramc Sent
E. Throughput
Throughput is the actual amount of data transmitted
correctly starting fom the source to the destination within a
given time (seconds). The importance of analyzing this QoS
parameter is because the increased numbers of users of the
wireless medium is the reason for increased possibility of
interference. Throughput is quantifed with various factors
including packet collisions, barrier between nodes and the
differentiation and scheduling mechanism used. During the
simulation throughput as a global statistics has been measured
so any object could contribute to its value. It gives a general
idea of the overall throughput of the system. Figure 10 shows
that the maximum throughput is achieved using DWRR
scheduling mechanism, the WRR has second highest
throughput and the priority queue has third highest throughput
while FIFO scheduling mechanism has the lowest throughput.
The reason for this is because DWRR scheduling mechanism
is communicating more effciently as compared to other
mechanisms. Also in DWRR mechanism distributed total load
of the network among the ZigBee Routers as a result of which
collisions and packet drops are decreased.
|gbaaMAC-BS|c_Woi&ar_0\^Fh-0LS-1
/QaaMAC-Ba|c_Fo:|ar_O-DLS-1
E_AC-Ba|c_'4houar_H-0ES-1
L|gbaaMAC-Ba|c_|4hot1ar _t^H-DLS-1
,..,......... ' : +;
.....

.,... [

..,

..

,... /
,
== ================='
, [
=!'
- -' -' - -+
` * `
Fig, 8, Throughput
N. CONCLUSION
This work introduces new differentiated service
approaches and tasks accomplished by scheduling disciplines
and highlights the impact of these techniques on the QoS
support in mobile sensor networks. we compared the use of
38
different quality control algorithms for prioritizing and
scheduling of trafc received fom vehicles in ZigBee
environment. On the basis of our measurements and results,
we presented that DWR and WR have increased QoS by
decreasing the collision, packet drop rate and delay. This
research can be frther extended by implementing existing
modern priority and scheduling mechanism or by presenting
innovative new algorithm for particular scenario of vehicular
sensor networks.
VI. ACKNOWLEDGEMENT.
This work is supported by the research center of the
college of Computer and Information Sciences, King Saud
University, under the project number RC130396.
References
[I] U. Lee and M. Gerla, "Survey of urban vehicular sensing platforms",
Computer Networks, vol. 54, no.4, 2010, pp.527-544.
[2] K.C. Rama, "A Survey on Sensor Network", Journal of Convergence
Information Technology (JCIT), vol.Ol, Issue 01, 2010, pp.76-87.
[3] 1. Yick, B. Mukheee, and D. Ghosal, "Wireless Sensor Network
Survey", Computer Networks: The International Joural of Computer
and Telecommunications Networking, vol. 52, no. 12, 2008, pp. 2292-
2330.
[4]
[5]
I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, "A
Survey on Sensor Networks", IEEE Communications Magazine, vol. 40,
no. 8,2002,pp. 104-112.
B. Nefzi, Y.Q. Song, "QoS for wireless sensor networks: Enabling
service differentiation at the MAC sub-layer using CoSenS", Ad Hoc
Networks, vol. 10, no. 4, 2012, pp.680-695.
[6] T. Zhang, L. Chen, D. Chen, and L. Xie, "EEFF: A Cross-Layer
Designed Energy Effcient Fast Forwarding Protocol fr Wireless
Sensor Networks", In the proceedings of IEEE Wireless
Communications and Networking Conference (WCNC), pp. 1-6,2009.
[7] A. Koubaa, M. Alves, B. efzi, and Y.Q. Song, "Improving the IEEE
802.15.4 Slotted CSMAICA MAC for Time-Critical Events in Wireless
Sensor Networks", Workshop on Real Time Networks (RTN), vol. 6,
2006.
[8] S.C. Ergen and P. Varaiya, "TDMA scheduling algorithms for wireless
sensor networks", Wireless Networks, vol. 16, no. 4, 2010, pp.985-997.
[9] Q. Liu, X. Wang, and G.B. Giannakis, "A Cross-Layer Scheduling
Algorithm With QoS Support in Wireless Networks", IEEE trasaction
on vehicular technology, vo1.55, no.3, pp.839-847, 2006.
[10] P. Papageorgiou, "Literature Survey on Wireless Sensor Networks",
ACM SIGOPS Operating Systems Review, pp.6, 1999.
[11]
[12]
L. Song and D. Hatzinakos, "A Cross-Layer Architecture of Wireless
Sensor Networks for Target Tracking", IEEE/ACM transaction on
networking, voI.15, no.l, pp.145-158, 2007.
I. Demirkol, C. Ersoy, and F. Alagoz, "MAC Protocols fr Wireless
Sensor Networks: A Survey", IEEE Communications Magazine, vo1.44,
no.4, pp.115-121, 2006.
[13] K. Kredo II and P. Mohapatra, "Medium Access Control in Wireless
Sensor Networks", Computer Networks, vol. 51, no.4, pp.961-994,
2007.
[14] W. Ye and 1. Heideman, "Medium Access Control in Wireless Sensor
Networks", Wireless sensor networks, pp.73-91, 2004.
[15]
[16]
P. Buonadonna, D. Gay, 1.M. Hellerstein, W. Hong, and S. Madden,
"TASK: Sensor Network in a Box", In Proceedings of the 2nd IEEE
Europea Workshop on Wireless Sensor Networks and Applications
(EWSN), pp.133-144, 2005.
S. Madden, M.J. Franklin, J.M. Hellerstein, and W. Hong, 'The Design
of an acquisitional Query Processor fr Sensor Networks", In
Proceedings of International conference on Management of data ACM
SIGMOD, pp.491-502, 2003.
C. J78J467356J382UJ3 LLL
[17]
[18]
[19]
[20]
W. Ye, J. Heidemann, and D. Estrin, "An Energy-Effcient MAC
Protocol fr Wireless Sensor Networks", In Proceedings of the IEEE
Computer and Communications Societies INFOCOM, voI.3, pp. 1567-
1576,2002.
LAN MAN Standards Committee of the IEEE Computer Society,
"Wireless LAN medium access control (MAC) and physical layer
(PHY) specifcation", IEEE, NY, USA, IEEE Std 802.11,1999.
M.A.Yigitel, .I. Incel, and C. Ersoy, "QoS-aware MAC protocols for
wireless sensor networks: A survey", Computer Networks,voI.55, no.8,
pp.1982-2004,2011.
1. Chen, M. Dlaz, L. L1opis, B. Rubio, ad 1.M. Troya, "A survey on
quality of service support in wireless sensor and actor networks:
Requirements ad challenges in the context of critical infrastructure
protection", Joural of Network and Computer Applications, vol. 34, no.
4, pp. 1225-1239,2011.
[21] D. Chen and P.K. Varshney, "QoS Support in Wireless Sensor
Networks: A Survey", International Conference on Wireless Networks,
pp.227-233,2004.
[22] K. Lin, "Research on adaptive target tracking in vehicle sensor
networks", Journal of Network and Computer Applications, 2012.
[23] B. Nefzi and Y.Q. Song, "QoS for wireless sensor networks: Enabling
service differentiation at the MAC sub-layer using CoSenS", Ad Hoc
Networks, vol. I 0, no.4, pp.680-695, 2012.
[24]
[25]
[26]
[27]
[28]
[29]
C. Annadurai, "Review of Packet Scheduling Algorithms in Mobile Ad
Hoc Networks", International Journal of Computer Applications, vol.
15, no. I , pp.0975-8887, 20 II.
O. Chipara, C. Lu, ad G.C. Roman, "Real-Time Query Scheduling fr
Wireless Sensor Networks", 28th IEEE International Real-Time Systems
Symposium (RTSS), pp.389-399, 2007.
B.G. Chun and M. Baer, "Evaluation of Packet Scheduling Algorithms
in Mobile Ad Hoc Networks", ACM SIGMOBILE Mobile Computing
and communications Review, vol. 6, no. 3, pp. 36-49, 2002.
M.A. Yigitel, .I. Ince!, and C. Ersoy, "Design and implementation of
a QoS-aware MAC protocol fr Wireless Multimedia Sensor Networks",
Computer Communications, vol.34, no.16, pp.1991-2001, 2011.
Z.J. Haas and J. Deng, "On Optimizing the Backoff Interval for Random
Access Schemes", IEEE IEEE Transactions on Communications, vol.
51, no.12, pp. 2081-2090, 2003.
F. Cali, M. Conti, and E. Gregori, "IEEE 802.11 Protocol: Design and
Performance Evaluation of an Adaptive Backoff Mechaism", IEEE
Journal on Selected Areas in Communications, vol. 18, no.9, pp. 1774-
1786,2000.
[30] F. Cali, M. Conti, E. Gregori, "Perfrmance Modeling of an Enhaced
IEEE 802.11 Protocol", In Proceedings of IFIP ATM, vo1.99, 1999.
3J

Enhanced-MPR-CDS approach for self-organization and routing in
WSNs
Jihed KHASKHOUSSI
1
, Ridha OUNI
2

1
EE, Faculty of Sciences of Monastir, Tunisia
Jihed.khaskhoussi@tunisietelecom.tn
2
College of Computer and Information Sciences, King Saud University, KSA.
rouni@ksu.edu.sa

Abstract
Wireless sensor networks are poised to become
a very significant enabling technology in many
sectors. Already a few very low power wireless
sensor platforms have entered the marketplace.
Almost all of these platforms are designed to run on
batteries that have a very limited lifetime. Recent
advances in wireless sensor networks have led to
many new protocols specifically designed for sensor
networks where energy awareness is an essential
consideration. Most of the attention, however, has
been given to the self-organization and routing
protocols since they might differ depending on the
application and network architecture.
This work evaluates performances of two self
organization protocols: MPR and MPR-CDS in the
goal to propose new rules reducing the dominants
and relay sets. This optimization is based on the
probability of being covered by many of neighbor's
dominants. Those sets operate together to route data
from sender to receiver. We use a multi-hop routing
to reduce energy consumption and retransmission
rate.
Key words: WSN, Self-Organization, MPR, MPR-CDS.
1. Introduction
A wireless sensor network (WSN) is an
interconnected set of sensor nodes that monitor and
collect information about the environment and
transmit the collected data generally to sink node
which saves, analyzes and interprets data. Each
individual sensor node in the network consists of one
or more sensors, a radio transceiver, a
microprocessor and a small battery. During the last
decade, it has drawn extensive attention due to the
wide range of promising applications, such as
environmental monitoring, industrial sensing and
diagnostics, battlefield surveillance, target tracking,
search and rescue, and disaster relief [1, 2].
The lifetime of a network is a crucial feature in WSN
applications. Another main feature is reducing the
size of the routing table in each node. Most of the
techniques proposed, for extending the sensor
network lifetime and reducing the routing table size,
are based on hierarchical architectures or clustering
techniques [1].
The self-organization protocols are energy-
efficiently improved. They aim to select a limited set
to act as routers (active nodes) allowed message
retransmission. They avoid the blind flooding which
generates a lot of collisions that could possibly
prevent the broadcasting from being correctly
performed beside the significant energy consumption
by redundant messages. Many schemes of self
organization protocol have been proposed to replace
blind flooding, and they are classified in different
categories: simple flooding, probability based, area
based and neighbor knowledge methods.
In the present work, we are interested on the
neighbor knowledge based protocol. We aim to
optimize the Multipoint Relay-Connected Dominate
Set (MPR-CDS) scheme and propose a self enhanced
routing approach.
The paper is organized as follows. In section 2, we
present an overview of the main WSN concepts. We
also introduce and analyze the MPR-CDS self-
organization protocol. While, in section 3, we
develop and evaluate our optimized approach. Next,
section 4 describes the routing approach using an
intermediate field. Finally, section 5 provides some
remarks and concludes the paper.
2. Wireless Sensor Networks (WSN)
Primitive network is based on blind flooding
(Fig.1). The packet is retransmitted by all the
intermediate nodes in order to broadcast it in the
network. It is simple, easy to implement, and gives a
high probability that each node, which is not isolated
from the network, will receive the broadcasted
message. However, it consumes a large amount of
ICCIT 2012 875

bandwidth due to many unuseful redundant
retransmissions.

Source node.

Transmitted packet.

Fig.1. Flooding in WSN.

Many techniques are described in the literature to
reduce traffic flooding in WSNs. But, each technique
is developed for a target application and
characterized by its own advantages and
disadvantages. Here, we will discuss the multipoint
relaying mechanisms (MPR and MPR-CDS) as
possible solutions. These mechanisms are based on
neighbor knowledge.
The neighbor knowledge process uses HELLO
message defined by Mobile ad hoc Network
(MANET) [6] . These messages are broadcasted to
all neighbors at regular intervals. They contain
information about the neighbors and the link state.
Each input has an associated timeout, a guard time
and the type of link: asymmetric, symmetric or MPR.
After two rounds of beacons HELLO messages, the
entire network nodes have the lists of their 1-hop and
2-hop neighbors needed to calculate the MPR nodes.
The HELLO process is described in Fig.2.

Fig.2. HELLO message process.
2.1. Multi Point Relay (MPR) Protocol
Multipoint relay was presented as a technique to
reduce the number of redundant re-transmission in
the wireless sensors networks by electing a special
node set to cover the network based on 2-hop
neighbors knowledge. Several rules and algorithms
are proposed for this calculation. In this paper we
resort to the greedy [3] MPR set computation
described in the algorithm below.
MPR Algorithm [5]

1. Start with an empty multipoint relays set.
2. Add nodes which are the only neighbor of
some nodes in the 2-hop neighbors.
3. If there still exists some two-hop nodes
which are not yet covered, computes the
one-hop nodes degrees and chooses the
maximum one.
4. Repeat step 3 until all two-hop
neighborhoods are covered.
Fig.3 shows an example of a diffused message in the
network using the multipoint relays where only
9 transmissions are needed to reach all nodes.

Source node Dominant node
transmitted packet.

Fig.3. Deploying MPR protocol in WSN.
2.2. Multi Point RelayConnected
Dominating Set (MPR-CDS)
Adjih and al. proposed a novel extension of the MPR
to construct a small CDS source independent using
two simples rules based on the node ID and the
greedy algorithm [4].
In the present work, node is selected into the
dominating set of the network if:

1. It has the largest degree than all its
neighbors.

A
1

B
3

C
2
Hello (Empty)
Hello (Empty)
Hello (A, C) Hello (A, C)
1-hop: A,C
2-hop:
1-hop: B
2-hop: A
1-hop: B
2-hop: C
876

2. It is a multipoint relay selected by a
neighbor with the largest degree.

The multipoint relays are selected using the greedy
algorithm. Nodes with highest degrees calculate their
relays and call them to join the CDS at the third
round of HELLO messages. Fig.4 shows the MPR-
CDS application for a wireless sensor network
composed by 18 nodes where 8 transmissions are
required to cover the network.

Source node Dominant node
Transmitted packet
Fig.4. WSN deploying MPR-CDS protocol.
2.3. Protocol analysis
Dominated nodes or passive nodes listen, receive and
analyze packets until they were invited to react.
While, dominants or active nodes listen, analyze and
diffuse messages.
In both modes (active and passive), devices consume
the significant amount of energy on the radio
amplifier units. Table 1 provides a comparison of the
numbers of active and passive nodes, between
flooding, MPR and MPR-CDS protocols for the
same topology.

Table 1. Number of active and passive nodes.
Protocol Active nodes Passive nodes
Flooding 17 1
MPR 9 9
MPR-CDS 8 10

Table 2 outlined that transmission rate for MPR-CDS
is better than MPR. The MPR-CDS reduce the
number of transmitted packet to 53 % and the
received packet to 25 %. To this end, we focus on the
optimization of the MPR-CDS.
Table 2: Received and transmitted messages per cycle.
Trans. packets
Received
packets
Flooding 17 49
MPR 9 39
MPR-DS 8 37

3. Enhanced MPR-CDS (Enh-MPR-CDS)
model
Energy consumption in sensor networks is
proportional to transmitted and received packets
numbers. Some active nodes cause a redundancy i.e.
a message will be received several times without
being transmitted.
We aim to minimize redundancy by reducing the
number of relays involved in the dissemination
process. The challenge is how to choose nodes that
does not affect network connectivity.
I dea:
Relays surrounded by a large number of relays may
be a source of redundancy.

In the rest of this section, we focus to determine
conditions allowing the drop of active nodes from
the set based on neighborhood knowledge by
switching them from dominant to be dominated.
3.1. Simulation
To simulate network behavior, a simulator was
developed in MATLAB because we will not simulate
network parameters such as delay or throughput
which need a specific network simulator. Our
simulator generates a random topology in a
rectangular area with 40 nodes.
Global graph connectivity is checked by a simulation
of blind flooding. CDS connectivity check uses the
same process as global graph where only the
dominants have the right to diffuse.

For each node i, we define neighbors parameters:

- Card(i) is the number of node i neighbors.
- Cardm(i) is the number of node i neighbors
which are selected as dominant.

-
Every dominant node filling:
877

Dc(i) >Dc_optimal
Will be switched to passive mode (not allowed
packet forwarding). 1000 simulation graphs are
deployed to determine the optimal Dc_optimal value.
The simulation process for each Dc_optimal value is
described by the following chart:
graph generator
connex?
MPR-CDS
connex?
Drop nodes
N=1000?
Yes
Yes
No
No
No

Fig.5. Simulation flow chart.
3.2. Results
According to the simulation results, shown in Fig.6,
the optimal value of Dc_optimal is 0.79. Analysis of
the not optimized cases shows that failure is due to
the creation of new clusters. These clusters are
caused by switching some dominants, having at least
one neighbor having cardm(i)=1, to passive mode.
Therefore, we need to consider a new condition
based on the cardm parameter of the neighbors to
avoid isolated clusters.
Finally, all dominants filling the following rules
could be outdated from the dominant set:

Rule 1: Dc > 0.79.
Rule 2: min (cardm(ni)) > 1 (ni dominant
neighbor of node i ).
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
30
40
50
60
70
80
90
Dc
S
u
c
c
e
s

c
a
s
e
s

(
%
)

Fig.6. Optimization of MPR-CDS.
We call Enh-MPR-CDS the MPR-CDS
protocol after applying the new rules. The same
simulation process, as described earlier, is kept to
evaluate the impact of the Enh-MPR-CDS on the
WSN connectivity.
Results, outlined in table 4, show that within more
than 50 % of cases, the dominant set is optimized
and graph connectivity is guaranteed. Hence, new
approach could reduce the dominant set to 35 %.
Table 4. Optimization rate with Enh-MPR-CDS.
Graphs numbers Optimization rate
497
No dominant to be
removed
503
2 % to 35 % of dominant
to remove

Our new approach, by optimizing the dominant
set, affects directly the energy consumption because
the number of transmitted and received messages
will decrease (control and traffic).

Fig.7: WSN deploying E-MPR-CDS protocol.

Tables 2 and 5 and figures 4 and 7 illustrate the
differences between the classic MPR-CDS protocol
and our enhanced approach.

878

Table 5. Received and retransmitted packets with EMPR-DS.

Trans.
packets
Received
packets
MPR-DS 8 37
Enh-MPR-CDS 7 31

Our proposition is interesting especially for
dense networks. It saves energy by avoiding
unnecessary transmission and reception. The rest of
dominants (routers in the next section) will be able to
forward packets.
4. Routing in WSN
Sensor networks have restricted power and
computational resources. Therefore, sending data
from a source to a destination requires a routing
protocol. An efficient routing protocol guarantees
forwarding messages with less number of hops and
less data traffic [7,8].
Blind flooding guarantees the packet delivery but it
generates significant redundancy. We aim to
introduce the concept of intermediate hop based on
the Enh-MPR-CDS protocol. Dominants are
considered as routers. This approach starts with
empty routing entries. While transmitting packets it
will be enhanced without adding traffic.
An INTER field is used in the indexing the next hop
as described in the flow chart of figure 8.

Fig.8. Routing approach flowchart.
Receiving packets, the router checks the
INTER field value. If it is zero, so the sender (or
forwarder) does not have the destination address in
its neighborhoods tables (1-hop and 2-hop). It reacts
as follow:
If it finds destination ID in tables, it changes
INTER field value with the ID of the next hop.
Then, it adds entries to routing table that
containing the destination ID and the next hop
ID.
If it does not find the destination ID in tables,
it just forwards the packet.
While, when routers receive packet with INTER
values not zero:
If INTER = ID Address, the router checks
tables and changes INTER value with the
address of next hop and adds the entry to
routing table if does not exist or modified.
If INTER ID address, it just adds entry to
routing table.
At the end of the first cycle (if we consider that every
nodes transmit one message every cycle), the router
(or dominant) in 3-hop destination/sink neighbors
will have a new entries in routing table. This table
will be enhanced every cycle.
Considering a simple link between source and sink
node, Fig.9 shows the number of retransmitted
packets in the network at each cycle. It proves that
routing with INTER field reduces the number of
retransmitted packets every cycle until converging to
the optimal route (shortest path). This is due to the
enhancement of routing table every transmission
without introducing additional traffic or topology
control packets.
The energy gain in terms of transmitted and
received packets is performed every cycle (n) by
avoiding unnecessary transmissions of the n-hops.
The process continues until finding the optimal
routes.
1 2 3 4 5 6 7 8 9 10 11 12
5
10
15
20
25
30
35
40
45
Cycle Number
N
u
m
b
e
r

o
f

T
r
a
n
s
m
i
s
s
i
o
nEMPR-DS with new approach
OLSR
EMPR-DS

Fig.9. Number of transmitted packets using our routing approach.
5. Conclusions
In this paper, a new concept to optimize the
MPR-CDS protocol is introduced. This concept
consists of minimizing the dominant set in order to
reduce the number of retransmitted packet in WSNs.
Packet arrival
Inter = 0
Yes
No

Forward packet
Inter = Next-hop
Add Entries:
Src ID Dest ID
(next hop) (3
rd
hop)
Dest-ID
in table
No

Forward packet
Inter = 0
Yes
879

Our new protocol, Enh-MPR-CDS, was simulated in
differents scenarios and compared to MPR and
MPR-CDS. Results confirm that a significant
amount of dominants may be removed while
connectivity still guarantee. The new protocol is
more efficient for dense networks that are composed
by hundreds or even thousands nodes.
Furthermore, a new approach for routing in
WSNs has been described. This approach deploys an
intermediate field including the router ID which
serves the destination. This approach enhances
neighbors lists (3 hops and more) without adding
traffic. Based on simulation results, it performs
routes until finding the shortest path.
References
[1] Fangting Sun and Shayman, Mark,Prolonging
Network Lifetime via Partially Controlled Node
Deployment and Adaptive Data Propagation in
WSN, Information Sciences and Systems, 2007.
CISS '07. 41st , 2007, Page(s): 226 231.
[2] Wei Peng and Xi-Cheng Lu. On the reduction
of broadcast redundancy in mobile adhoc
networks. In Proc. First Annual Workshop on
Mobile and AdHoc Networking and Computing,
August 11-2000, pages 129130.
[3] Yongsheng Fu, Xinyu Wang, Wei Shi and
Shanping Li, Connectivity based greedy
algorithm with multipoint relaying for mobile
ad hoc networks, The 4th international
conference on mobile ad-hoc ans sensor
network, 2008.
[4] Cedric Adjih, philipe Jacquet and Laurent
Viennot, Computing connected dominated sets
with multipoint relays, Technical report,
INRIA, Oct.2002, www.inria.fr/rrrt/rr-
4597.html.
[5] C. Bettestetter, On the minimum node degree
and connectivity of a wireless multihop
network, in Proc. ACM MobiHoc, Lausanne,
Switzerland, Jun. 2002.
[6] T.Claussen, P.Jacquet, C. Adjih, A. Laouiti, P.
Minet,P. Muhlethaler,A. Quayyam and L.
Viennot, Optimized link state routing
protocol(OLSR), RFC 3626, Oct. 2003,
Network Working Group.
[7] I. Akyildiz,W. Su, Y. Sankarasubramaniam,
and E. Cayirci, A survey on sensor networks
Communications Magazine, IEEE, vol.40,
pp.102-114,Aug2002.
[8] K. Akkaya and M. Younis, A survey on routing
protocols for wireless sensor networks AdHoc
Networks, vol.3, no.3, pp .325-349, 2005.

880

AbstractNowadays, several positioning systems are
available for outdoor localization, such as the global
positioning system (GPS), assisted GPS (A-GPS), and other
techniques working on cellular networks, for example, Time
of Arrival (TOA), Angle of Arrival (AOA) and Time
Difference of Arrival (TDOA).However, with the increasing
use of mobile computing devices and an expansion of wireless
local area networks (WLANs), there is a growing interest in
indoor wireless positioning systems based on the WLAN
infrastructure. Wireless positioning systems (WPS) based on
this infrastructure can be used for outdoor / indoor
localization to determine the position of mobile users. An
important factor in achieving this is to minimize and simplify
the instructions that the mobile station (MS) has to execute in
the location determination process. Finding an effective
location estimation technique to facilitate processing data is
the main focuses in this paper. Therefore, in the wireless
propagation environment the Received Signal Strength (RSS)
information from three base stations (BSs) are recorded and
processed and they can provide an overlapping coverage area
of interest. Then an easy new geometric technique is applied
in order to effectively calculate the location of the desired MS.
Our new positioning method design was verified at the
algorithmic level using Matlab tool, described in Very-high-
speed integrated circuit Hardware Description Language
(VHDL) at the register transfer level (RTL) and it has been
synthesized using 7.1 version of FPGA Advantage for HDL
Design that evaluates the circuit in terms of speed, area and
power consumption.

Index TermsGeometric technique, Position estimation,
Wireless LAN, VHDL.

I. INTRODUCTION
Recently, the subject of mobile positioning in wireless
communication systems has drawn considerable attention.
With accurate location estimation, a variety of new
applications and services such as Enhanced-911, location
sensitive billing, improved fraud detection, intelligent

Monji Zaidi, Jamila Bhar and Rached Tourki are with Electronic and
Micro-Electronic Laboratory (EE, IT-06).FSM, Monastir, Tunisia
Ridha Ouni is with College of Computer and Information Sciences
(CCIS), King Saud University Riyadh, KSA

transport system (ITS) and improved traffic management
will become feasible [1]. Mobile positioning using
radiolocation techniques usually involves time of arrival
(TOA), time difference of arrival (TDOA), angle of arrival
(AOA), signal strength (SS) measurements or some
combination of these methods.
All of these methods are mainly based on trigonometric
computation. Comparisons and survey of these methods are
given in [2] and [3].
The TOA technique determines the MS position based
on the intersection of three circles. Two range
measurements provide an ambiguous fix, and three
measurements determine a unique position.
Given the coordinates of BSj, (j = 1, 2, 3) as (Xj, Yj),
and the distances dj between MS and BSj, the simplest
geometrical algorithm for TOA positioning (Figure. 1(a)) is
given in [2]. Coordinates of MS position (x,y) relative to
BS1 can be calculated as:
=
1
2

The simplest geometrical algorithm for TDOA
positioning (Figure. 1(b)) is given in [4]. There are two
estimated TDOA-s dj, 1 between BS1 and the jth base
station (j = 2, 3). Coordinates of MS position (x, y) relative
to BS1 can be calculated in terms of d1 as:
+
1
2

Where:

The AOA technique determines the MS position (x, y)
based on triangulation, as shown in (Figure. 1(c)). The
intersection of two directional lines of bearing with angles
and
defines a unique position, each formed by a

radial from a BS to the MS. The simplest geometric
Monji ZAIDI, Ridha OUNI, Jamila Bhar and Rached TOURKI
A Novel Positioning Technique with Low
Complexity in Wireless LAN:
Hardware implementation
Proceedings of the World Congress on Engineering 2011 Vol II
WCE 2011, July 6 - 8, 2011, London, U.K.
ISBN: 978-988-19251-4-5
ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
WCE 2011

solution can be derived using [5] with two AOA
measurements
and
:
x =
Y
+ X
tan(
) X
tan (
)
tan(
) tan (
)

y = Y
+ (x X
)tan(
)

BS1 BS2
BS3
MS
Hyperbola 1
Hyperbola 2
d 1
d 2
d 3
(b)
BS1
BS3
BS2
d1
d 2
d 3
MS
(a)
BS1 BS2
MS
2
1
(c)

Fig. 1. Position determination techniques: (a) TOA; (b) TDOA; (c)
AOA
Using any of the mentioned methods, the calculation can
be done either at the BS [network-based schemes] or at the
MS [mobile-based schemes]. Network-based schemes have
high network cost and low accuracy [3]. Mobile-based
location schemes are more interesting. However, since the
MS has limited energy source, in the form of the battery
pack, energy consumption should be minimized. An
important factor in achieving this is to minimize and
simplify the instructions that the MS has to execute in the
location determination process. The conventional
algorithms use complex computation methods that needed
relatively long execution time.
In this paper, we propose a novel wireless positioning
technique based on the WLAN infrastructure. The main
motivation for our approach is twofold: to improve the
accuracy of the location estimation and to minimize and
simplify the instructions that the mobile station (MS) has to
execute in the location determination process.
II. RELATED WORK
To improve the accuracy of the indoor positioning
system, several techniques demonstrate the viability of this
approach. Youssef et al. [6] show that the RADAR system
can be improved using the perturbation technique (joint
clustering technique) to handle the small-scale variations
problem. This technique can improve theRADARsystem
and provide location accuracy up to 3m.
The triangulation mapping interpolation system (CMU-
TMI) [7] performs a location calculation on the current
data, interpolates that data with the information in the
database, and then returns a location estimate based on this
interpolation. However, power consumption increases to
measure the signal strength on the client side.
The Ekahau Positioning Engine 4.0 [8], released in
October 2006, also uses an IEEE 802.11 network to
provide location information. It achieves an average
accuracy of 1m with at least three audible channels in each
location. This system requires site calibration up to 1 h per
1,200m
2
. While calibration-based efforts present good
accuracy results, there is still room for performance
enhancements. Due to the very dynamic nature of the RF
signal, the assumption that the radio map built in the
calibration phase remains consistent to the measurements
performed in the real-time phase does not hold in practice;
thus, at times, there is a need to rebuild the radio map. It
seems more reasonable to design a fully-automatic system
capable of acknowledging RSSI characteristics and
variations in both spatial and time domains.
Hitachi [9] released location technology based on TDOA
in March 2005. This system uses two types of access
points: a Master AP and a Slave AP. Slave APs
synchronize their clocks with that of a Master AP and
measure the arrival time from a mobile terminal; the Master
AP determines the location of the mobile terminal using the
TDOA between the signal reception times at multiple Slave
APs. While this technique has been found to achieve good
results in indoor environments, it requires specialized
hardware and fine-grain time synchronization, which
increases the cost of this type of solution.
Kanaan proposed a closest-neighbor with TOA grid
algorithm (CN-TOAG) [10]. This geolocation algorithm
presents a TDOA-based position detection technique to
improve location accuracy in the indoor environment by
estimating the location of the user as the grid point. This
technique is similar to the previous one [11], as it needs
specialized hardware and fine grain time synchronization,
which increase costs.
III. NEW GEOMETRIC LOCATION ALGORITHM
BASED ON THREE BSS
In the general geometrical triangulation location
researches, they assumed that the measured noise is
additive and the NLOS error is a large positive bias which
causes the measured ranges to be greater than the true
ranges [12].
BS
1

BS
2

BS
3

A
B
C

Fig. 2. Measured range circles and the associated intersected area
Under the assumption, the MS location will guarantee to
lie in the overlapped region (enclosed by points A, B and
C) of the range circles as shown in Figure. 2. Thus the MS
ISBN: 978-988-19251-4-5
WCE 2011

is necessarily located in the region formed by the points
BS
1
, BS
2
and BS
3
. But, it is noted that the intersection of
three circles may not be overlapped with the real
measurement results. Therefore, with the above assumption
we have to judge whether the three circles intersect or not
in our location algorithm.
If circles intersect as depicted in Figure. 3, then three
triangles can be drawn as: BS
1
MSBS
2
, BS
2
MSBS
3
and
BS
3
MSBS
1
.
Assumptions:
Different BSs are placed (two to two) at an equal
distance
The coordinates of BSs are known by the MS
The MS can inquire only on the received power coming
from the BSs (i.e. the distance which separates it from
each BS).
d
+ d
> ; d
+ d
> d
+ d
>
BS1 BS2
BS
3

MS d
1
d
2

d
3

23
21 12
13
32
31
) , ( 12 12 0 y x A
) , ( 23 23 0 y x B ) , ( 31 31 0 y x C
D
r1

Fig. 3. The associated triangles of the standard intersection of three
circles.
Note by:
D: The distance between tow BSs.
A
0
, B
0
and C
0
are the orthogonal projections of the MS
on (BS
1
BS
2
), (BS
2
, BS
3
) and (BS
3
BS
1
) respectively.
d1, d2 and d3 are the distances that separate the MS
from BS1, BS2 and BS3 respectively.

12
: is the geometrical angle between the MS-BS1 and
BS1-BS2. (Same things for the other angles).
We focus firstly on the triangle BS
1
MSBS
2
.
Based on the above assumptions and figure 2, we can write.
r
= d
cos

We can also write
d
= (D r
+(d
)
d
= (D d
cos
+ (d
cos
)
d
= D
+ d
2Dd
cos
= D
+ d
2Dr
=
D
+ d
2D

We define here the first factor q
by
q
=
r
D
=
D
+d
2D

Coordinates (x
, y
) of the point A
0
are given in [13] by
x
= q
+ (1 q
)X

y
= q
+ (1 q
)Y

Where:
(X
, Y
) and (X
, Y
) are the coordinates of BS

1
and BS
2
,
respectively.
Let the distance between BS
2
and B
0
be r
and the
distance between BS
3
and C
0
be r

As we described previously, we can get the coordinates of
points B
0
and C
0
as:
x
= q
+ (1 q
)X

y
= q
+(1 q
)Y

x
= q
+(1 q
)X

y
= q
+ (1 q
)Y

Where:
q
=
r
D
=
D
+ d
2D
=
r
D
=
D
+ d
2D

MS is then located in a new triangle A
0
B
0
C
0
, which is
smaller in terms of area compared to the starting triangle
BS
1
BS
2
BS
3
. In the other word we have just created three
new virtual BSs placed at A
0
, B
0
and C
0
.
It is very easy to calculate the distances between the MS
and the new points A
0
, B
0
and C
0
using the Pythagoras
formula. Thus
d(MS, A
) = d

d(MS, B
) = d

d(MS, C
) = d

Now, with the three new virtual BSs, MS can repeat the
same calculations as shown above. During this second
iteration, the orthogonal projections of MS on (A
0
B
0
),
(B
0
C
0
) and (C
0
A
0
) must be done to obtain new point A
1
, B
1

and C
1
that their coordinates may be determined as
previously. A
1
B
1
C
1
s area is smaller than the A
0
B
0
C
0
one.
At the i
iteration, the MS will be located in an A

i
B
i
C
i

triangle which is smaller than A
i-1
B
i-1
C
i-1
one. This A
i
B
i
C
i

triangle allows to designing the next triangle A
i+1
B
i+1
C
i+1
.
ISBN: 978-988-19251-4-5
WCE 2011

After a small number of iterations, the coordinates of
three vertices of the triangle (A, B and C) converge to the
actual coordinates of the MS. At the limit, the triangle
A
conv
B
conv
C
conv
with vertices A
conv
, B
conv
and C
conv
will be
considered as a point. So, it is possible to write:
x

We can then take the coordinates of the MS as:
x
=
x
+ x
+ x
3

y
=
y
+ y
+ y
3

The division by 3 implies that the MS is equivalent to the
gravity center of the A
conv
, B
conv
C
conv
triangle.
Figure 8 (section 5) shows the evolution and the
convergence of the three vertices coordinates for different
values of d
i
(d
1
, d
2
and d
3
).
IV. VHDL MODEL
The operation destined to calculate the coordinates of a
mobile terminal is known as a Location Process. Original
design of this process, has just considered.
Figure 4 shows the main elements involved in our new
mechanism: the MS, Three BSs, and the distribution system
(DS) or Ethernet.
It can mentioned that different BSs are placed (Two to
two) at an equal distance, and the MS can inquire only on
the position and the received time coming from the BSs
(i.e. the distance which separates it from each BS)
The equations for the x and y position of the mobile was
modeled using VHDL. The numeric_std package was used
to construct the VHDL model that was readily synthesized
into a low power digital circuit. The input signal of the
model are the x, y positions of the three BSs, i,j,k in meters,
and the signals TOA from the individual BS to the mobile
in nanoseconds. The input signal assignments are x
i
, y
i
,
TOA
i
, x
j
, y
j
, TOA
j
, x
k
, y
k
, TOA
k

Fig. 4. Involved elements in the Positioning process

Now, we describe the hardware implementation of the
location process. Figure 5 illustrates the system
architecture; we try to divide location process to 4 parts: a
process location algorithm, the square root component,
divider block and buffers to store data. The following
notations are used to describe the signal type: I: input
signal; O: output signal
TABLE I
PROCESS LOCATION PART INTERFACE SIGNALS
RST I:bit RESET system
TOA1 I:std_logic_vector Input from the BS1, it gives the time of
arrival value to travel the d1distance
X1 I:std_logic_vector BS1 abscissa
Y1 I:std_logic_vector BS1ordinate
arrival value to travel the d2 distance.
Y2 I:std_logic_vector BS2abscissa
arrival value to travel the d3 distance.
Y3 I:std_logic_vector BS3 abscissa
Xestim O:
std_logic_vector
Estimated abscissa of MS
Yestim O:
std_logic_vector
Estimated ordinate of MS

First, the main program (process location) receives data
from the external environment. Then, it calculates the
parameters r
, q
, r
, q
, r
, and q
as it was presented in
Section 3. During this stage the divider component is called
by the main program to perform the operations division.
Meanwhile the virtual coordinates
(x
, y
), (x
, y
) and (x
, y
) of points A
0,
B
0
and C
0
respectively, are determined.

Secondly, the distances d(MS, A
), d(MS, B
) and
d(MS, C
) are calculated using the square root operators

Fig. 5. Top level structure of the Location circuit

The implemented square root algorithm uses unsigned
integers, which have several advantages over floating-point
ISBN: 978-988-19251-4-5
WCE 2011

numbers in FPGA arithmetic. Operations on unsigned
integers are often simpler to implement, and they require
less chip area and resources. The square root operator
assumes that its input argument has already been converted
into an unsigned integer, which must be taken care of if an
application uses signed integers.
A symbol of the top-level VHDL design entity of the
square root operator with parameterizable input argument
width is presented in Figure 6.

Fig. 6. Top level diagram of the square root operator

Figure 7 shows the simulation of a fast Location
processing model. Optimal process latency is improved by
reducing the iterations number needed for convergence. So,
after a small number of iterations, the coordinates of three
vertices of the triangle (A, B and C) converge to the actual
coordinates of the MS. at this time, the triangle
A
conv
Bc
onv
C
conv
with vertices A
conv
, B
conv
and C
conv
will be
considered as a point and we can write:
x
= x
= x
= y
= y

We can then take the coordinates of the MS as:
(x
, y
)
conv
B
conv
C
conv
triangle.
The (x
, y
) Geolocalisation information, obtained

with a minimal cost has a very important role in several
applications. We are trying to take advantage of this
important information to develop a fast handover in an
environment with time constraints

Fig. 7. Simulation results of the hardware positioning
process
Now, it is necessary, as in any positioning method, to
evaluate the error or deviation (in m) between actual
(measured) and simulated values obtained by our method.
For this two cases have to be considered:
A. Line-of-sight (LOS) condition
This case occurs in open areas or in very specific spots in
city centers, in places such as crossroads or large squares
with a good visibility of BS. Sometimes, there might not be
a direct LOS signal but a strong specular reflection off a
smooth surface such as that of a large building will give
rise to similar conditions. The received signal will be strong
and with moderate fluctuations. Therefore, the extracted
distance from the received signal is correctly calculated.
In the table 1 we give some actual locations of the MS
(Actual x and y). Corresponding values of the true
distances d
1
, d
2
and d
3
which separate it from BS
1
, BS
2
and
BS
3
are calculated. Then the estimated position and
position error can be determined using our geometric
method.
A. Non Line-of-sight (NLOS) condition
This case will typically be found in Indoor environments.
This is a worst-case scenario since the direct signal is
completely blocked out and the overall received signal is
only due to multipath, thus being weaker and subjected to
marked variations. Under these conditions the geometric
method can be applied. However, the position error
increases significantly.

Fig. 8. Algorithm convergence with (d1, d2, d3) = (52 m, 76 m, 50 m)

The oboe simulation was done with the following BSs
coordinates.
BS
1
coordinates (in meters):(X
, Y
) = (0,0)
BS
2
, Y
) = (100,0)
BS
3
, Y
) = (50,86)

0 5 10 15 20
30
40
50
60
x(A)variations as function of iterations number
iterations number
x
(
A
)
0 5 10 15 20
20
40
60
80
x(B)variations as function of iterations number
iterations number
x
(
B
)
0 5 10 15 20
25
30
35
40
x(C)variations as function of iterations number
iterations numbers
x
(
C
)
0 5 10 15 20
0
20
40
60
y(A)variations as function of iterations number
iterations numbers
y
(
A
)
0 5 10 15 20
30
40
50
60
y(B)variations as function of iterations number
iterations number
y
(
B
)
0 5 10 15 20
30
35
40
45
y(C)variations as function of iterations number
iterations number
y
(
C
)
0 5 10 15 20
30
35
40
45
estimated abscissa of MS
iterations number
x
(
M
S
)
0 5 10 15 20
30
35
40
estimated ordinate of MS
iterations number
y
(
M
S
)
ISBN: 978-988-19251-4-5
WCE 2011

TABLE II
LOS MEASUREMENT AND POSITION DETERMINATION

xactua
(m)

yactua
(m)

d1
(m)

d2
(m)

d3
(m)

xestima
(m)

yestima
(m)

C.t
(i.n)

Error
(m)
20 10 22.
5
80.
5
82.
5
20.1
3
09.20 8 0.807
30 20 36 73 69.
5
29.8
3
19.64 8 0.395
40 30 50.
5
67 57.
5
40.3
0
29.70 8 0.424
40 40 57 72 47.
5
40.3
2
39.86 8 0.353
50 50 71 71 36 50.0
0
50.23 8 0.238
50 70 86.
5
86.
5
16 50.0
0
70.47 8 0.478
60 60 85.
2
72 28.
5
60.3
7
59.91 8 0.384
60 40 72 57 47.
5
59.6
7
39.86 8 0.353
70 30 76.
5
42.
5
60.
5
70.0
6
29.45 8 0.549
80 20 82.
5
27.
9
73 79.8
5
19.69 8 0.336
M0.4
C.t: Convergence time
I.n = Iterations number
M: mean
TABLE III
NLOS MEASUREMENT AND POSITION DETERMINATION
xactual
(m)
yactual
(m)
xestimated
(m)
yestimated
(m)
Error (m)
20 10 18.9700 8.4826 1.8340
30 20 29.0950 19.2922 1.1489
40 30 39.0763 28.7347 1.5666
40 40 41.4850 40.8110 1.6920
50 50 50.0000 51.0523 1.0523
50 70 48.2600 70.9884 2.0011
60 60 58.7450 60.6642 1.4199
60 40 61.1350 40.4331 1.2148
70 30 70.2300 28.0174 1.9959
80 20 79.7000 17.5872 2.4314
Mean1.6

B. Synthesis results
During the synthesis step, we have exploited FPGA
Xilinx virtex 5 environment. This environment allows
implementing communication systems on programmable
circuits. The advantage of using FPGAs circuits is mainly
the system re-scheduling. For our application, RTL
synthesis is achieved using the ISE 10.1 of the Xilinx
FPGA virtex 5 environment. A synthesis result, of the
proposed process location, is shown in table 3. These
results should be exploited in order to study their impact on
the support of the technological parameters specified in
IEEE 802.11. These results show that the circuit can
operate with 142 MHz, which makes it more suitable for
real time communications.
TABLE IV
SYNTHESIS RESULTS.
Number of
Slices
Number of
Flip Flops
Nb of 4
input
LUTs
Nb of
bonded
IOBs
Frequency
(MHz)
772 594 1371 30 142
V. CONCLUSION AND FUTUR WORKS
This paper presents new geometric oriented algorithm
that is based on three distances measurements to determine
the position of a mobile object. Provided that all operations
in our proposed algorithm are additions, subtractions and
multiplications based, the implementation is simplified
which reduces complexity.
Our results show that for a very reduced number of
iterations (k 8), the proposed method converges and
provides with a good accuracy the position of MS. Hence,
the major advantages of our algorithm are: implementation
simplicity, and low computation overhead.
We adopted the high level design for the implementation
of this model. In fact, we have used VHDL as high level
description language, ModelSim as a simulation tool to
check the behavior of the model at the RTL level and ISE
10.1 of the FPGA xilinx environment for synthesis step.
We obtained the exact solution for the two-dimensional
location of a mobile given the locations of three fixed base
stations in a cell and the signal TOA (Time of Arrival)
from each base station to the mobile device. Simulation
results for two different situations predict location of the
mobile is off by 0.4 m for best case and off by 1.6 m for
worst case.
In our future work we are ready to integrate the position
of the MS in the 4G Handover management.

REFERENCES
[1] J. Caffery, Jr., Wireless Location in CDMA Cellular Radio
Systems, Kluwer Academic Publishers, Boston, 2000
[2] Jami M., Ali R.F. Ormondroyd, Comparison of Methods of
Locating and Tracking Cellular Mobiles, Novel Methods of Location
and Tracking of Cellular Mobiles and Their System Applications,
(Ref.No.1999/046), IEE Colloquium, London UK, 1/1-1/6.
[3] Zhao Y. Standardization of Mobile Phone Positioning for 3G
Systems, IEEE Communications Magazine, No.4, Vol.40, (July
2002), pp.108-116.
[4] Y.T. Chan, K.C. Ho, A simple and efficient estimator for
hyperbolic location, IEEE Transactions on Signal Processing,
42(8) (1994).
[5] Alba Pages-Zamora, Josep Vidal, Dana H. Brooks, Closed-form
solution for positioning based on angle of arrival measurements,
in: Proc. of the 13th Sym. on Personal, Indoor and Mobile Radio
Communications, September 2002, vol. 4, pp. 15221526.
[6] Youssef, M. et al. WLAN location determination via clustering and
probability distributions, Proceedings of IEEE PerCom2003, Mar
2003.
ISBN: 978-988-19251-4-5
WCE 2011

[7] Smailagic, A., & Kogan, D. (2002). Location sensing and privacy in
a context-aware computing environment, IEEE Wireless
Communications, 9(5), 1017, Oct 2002.
[8] Ekahau, Inc., Ekahau Positioning Engine 4.0,
http://www.ekahau.com/. Accessed, Oct 2006.
[9] Yamasaki, R. et al. (2005), TDOA location system for IEEE
802.11b WLAN, Proceedings of IEEE WCNC05, pp. 23382343,
March 2005.
[10] Kanaan, M., & Pahlavan, K. (2004). CN-TOAG: , A new algorithm
for indoor geolocation, Proceedings of IEEE PIMRC04, 3, 1906
1910, Sep 2004.
[11] Biacs, Z., Marshall, G., Meoglein, M., & Riley, W. (2002). The
qualcomm/SnapTrack wireless-assisted GPS hybrid positioning
system and results from initial commercial deployments,
Proceedings of IOS GPS, pp. 378384, 2002.
[12] C. D. Wann and H.C. Chin, Hybrid TOA/RSSI Wireless Location
with Unconstrained Nonlinear Optimization for Indoor UWB
Channels, IEEE WCNC, 2007, March 2007, pp. 39403945.
[13] Chi-Kuang Hwang and Kun-Feng Cheng Wi-Fi Indoor Location
Based on RSS Hyper-Planes Method, Chung Hua Journal of
Science and Engineering, Vol. 5, No. 4, pp. 37-43 (2007)

ISBN: 978-988-19251-4-5
WCE 2011
Reducing Wi-Fi handover delay using a new
positioning process

Monji ZAIDI, Jamila BHAR, Ridha OUNI and Rached TOURKI
Monji.Zaidi@fsm.rnu.tn

AbstractMobility has now a crucial requirement for
wireless communication. Handover is one of the major tasks that
are used to support continuous transmission for a mobile
terminal into different radio coverage area. Optimizing the
existing handover protocol requires integrating new
functionalities. This work focuses on presenting and optimizing
handover algorithm. We analyze handover time in wireless local
area networks based on the IEEE 802.11b MAC protocol. In fact,
scan phase is the main contributor to the handover time. Then,
we propose a handover model which replaces a scan phase by a
positioning process. This model is able to select the suitable access
point (AP) based on the shortest distance far from the mobile
terminal (MT). Proposed Handover allows a mobile user to
reacting quickly to decide about to which access point to connect.
Simulation results show that the proposed model provides gains
in term of delays and Handover success in various scenarios
Keywords:IEEE 802.11, Handover, positioning, Latency.
I. INTRODUCTION
This research introduces characteristics of Wi-Fi (Wireless
Fidelity) architecture and particularities of its mobility.
Handover protocol has been attracting a lot attention from
researchers in order to provide continuity of communication to
mobile user. In order to reduce Handover procedure delay,
some steps of handover must be revised. Classical procedure
can then be optimized. Particularly, this work involves the
impact of positioning component for handover protocol in a
WiFi environment. It proposes fast handover protocol based on
positioning process in order to minimize amount of time for
handover procedure.
This paper is organized as follow. Section 2 presents the
IEEE 802.11 environment. In section 3, we describe the
classical handoff mechanism and we perform its constraints.
Section 4 presents proposed Handover algorithms in the
literature. In section 5, we detail our proposition to reduce
handover latency. The simulation, analysis and synthesis are
illustrated in Section 6. Section 7 concludes the paper.
Today, there are nearly pervasive WiFi delivers high- speed
Wireless Local Area Network (WLAN) connectivity to
millions of offices, homes, and public locations, such as
hotels, and airports. The integration of WiFi into notebooks,
handhelds and Consumer Electronics devices has accelerated
the adoption of WiFi to the point where it is nearly a default
feature in these devices [10]. WiFi is also characterized by
easiness to deploy and less cost than cellular service. WiFi has
the advantage to interoperate with others technologies as
WiMAX.
II. WIFI OVER VIEW
The IEEE 802.11 architecture consists of a number of
components. The Basic Service Set (BSS) is the basic building
block in the architecture, and the members of a BSS
communicate with each other or with the Internet hosts
through access points. Multiple BSSs can interconnect with
each other through a distribution system and form an Extended
Service Set (ESS) [5]. The BSS consists of a number of
mobile terminals. When MT is moving between different BSS,
it needs to switch its AP which functions as a bridge
permitting interconnection to the distribution System.

ESS
BSS
BSS
Distributed system
MT
AP
AP

Figure 1. IEEE 802.11 architecture component
IEEE 802.11 specification focuses on the two lowest layers of
the Open Systems Interconnection model (OSI model) that
incorporate both physical and data link component. The
physical layer consists of the radio and the radios shared
channel. The MAC layer maintains communications among
802.11 stations by managing the operation of the PHY and by
utilizing protocols that support and enhance communications.
III. HANDOVER SPESIFICATION
Handover term refer to different approaches to supporting
mobility aspects. Distinctions between different propositions
can be made according to the performance characteristics,
diversity steps, state transitions, and control modes of
handover techniques. Generally, Handover can be defined as
the process by which an active MT changes its point of
attachment to the network, or when such a change is
attempted. Handover initiation is, generally, based on signal
strength. The access network may provide features to
minimize the interruption to sessions in progress [3]. The
conventional Handover procedure comprises three phases
namely Scan, Authentication and Association or
Reassociation. Figure 3 illustrates Handover procedure as
described in the IEEE Standard 802.11. It explains the basis
handover model and its steps latencies [2]. More handover
procedure details can be seen in [9].
A. Scan phase
In the scan phase, a MT tries to find a new available AP with
the best signal quality. It selects channels to probe. Then, MT
sends a probe request frame in order to obtain information
from access point. For example, a MT would send a probe
request to determine which access points (APs) are within
range. When receiving a probe request frame, AP will respond
with a probe response frame containing capability information,
supported data rates, etc. The TM must explicitly scan each
channel (11 channels in 802.11b and 802.11g, and 8 channels
for 802.11a indoors) for potential access points. After probing
all selected channels, the AP candidate is determined from the
information received in the probe responses and their
associated Signal to Noise Ratio (SNR).
In its simplest form, the scan phase can be completely passive.
The MT switches to a candidate channel and listens for
periodic beacon packets generated by access points to
announce their presence (typically every 100 ms). However,
the latency incurred by this approach can be quite long since
the phase of beacon intervals is independent and a MT must
therefore wait the full interval on each channel [8]. Scanning
phase accounts for more than 90% of the overall latency. It is,
then, considered as the dominating factor in handoff latency
[1].
B. Authentication phase
Authentication phase depicts a process whereby the AP either
accepts or rejects the identity of a MT. The MT begins the
process by sending an authentication frame containing its
identity to the AP. With open system authentication (the
default), the MT sends only one authentication frame, to the
selected AP which could accept or reject the connection
through an authentication frame.
C. Reassociation phase
In Reassociation phase, we distinguish reassociation request
and reassociation response. In fact, if a MT roams away from
the currently associated AP and finds another AP having a
stronger SNR, it will send a reassociation frame to the new
AP. This last then coordinates the forwarding of data frames
that may still be in the buffer of the previous AP waiting for
transmission to the MT. An AP sends a reassociation response
frame containing an accept notice to the MT requesting
reassociation. Similar to the association process, the frame
includes information regarding the association, such as
association ID and supported data rates.
MT Active AP Old AP
New AP
Probe request
Probe response
Probe response
Probe response
Probe request
Probe response
Probe response
Probe response
...
.
.
.
C
h
a
n
n
e
l

1
C
h
a
n
n
e
l

N
Association request
Exchange
H
a
n
d
o
f
f

L
a
t
e
n
c
y
P
r
o
b
e

d
e
l
a
y
A
u
t
h
e
n
t
i
c
a
t
i
o
n

a
n
d
A
s
s
o
c
i
a
t
i
o
n

d
e
l
a
y

Figure 2. Handoff latency in IEEE 802.11 Networks [2].
IV. RELATED WORK
Handover should maintain connectivity to mobile terminal as
it moves from access point to another. Important issues related
to handover include selection of optimised access point,
initializing handover, handoff delay and routing. Different
approaches have tried to optimise handover procedure in WiFi
network to reduce handover time incurred by probing,
authentication and association phases.
Some researchers propose an authentication phase that is
designed to reduce the authentication delay during a WiFi
handover process.
Details of this approach can be found in [4, 5]. We are
interesting here of research that try to reduce a scan phase. In
fact, Syed S. Rizvi and all [11] verify that the active scanning
can reduce the overall handover time at MAC layer if
comparatively shorter beacon intervals are utilized for packet
transmission. In [11], mathematical model is proposed to be
used to effectively reduce the handover time of WLAN at
MAC layer. Simulation results verify that the utilization of
probabilistic approach with the active scanning yields lower
latency for each detection and search phases. Both simulation
and numerical results of this paper demonstrate that the
reduced handover time at MAC layer provides better load
balancing, high throughput, and minimum frame transmission
delay. The solution proposed in [2] consists of transmitting
Probe requests which the scanning channels, stops once a
Probe response indication is received with an adequate SNR.
An SNR threshold level has been defined to select AP that
provides QoS guarantee. Simulation results show that the
proposed model reduce handover time by 22,28%. In [6] the
handover was split into three phases, typically performed in
sequence: detection, search and execution. Proposed approach
has shown that the detection phase can be reduced to three
consecutive non-acknowledged frames when stations are
transmitting. A shorter beacon interval reduces the detection.
The idea in [7] is to monitor continuously the signal quality of
all access points in range. In this way terminal makes
decisions based on signal levels received from all access
points. If any available AP provides a significant better signal
quality or the actual associated AP has too weak signal level to
serve the MT with a specific link quality, it is necessary to
initiate a handover process. The whole handover process is
under control of the mobile node that is capable of performing
fast handovers. In [8] a technique called SyncScan is
described. This technique requires synchronizing short
listening periods at the MT with periodic transmissions from
each AP. SyncScan algorithm is implemented using
commodity 802.11 hardware. Proposed scheme allows better
handover decisions and seek to reduce the time spent in the
channel scanning phase when a handoff occurs. In fact,
implicit time synchronization is proposed to reduce the key
cost of discovering new wireless access points. By
synchronizing the announcement of beacon packets, a client
can arrange to listen to other channels with very low overhead.
As a result, handoff using this SyncScan approach is an order
of magnitude quicker than using the conventional approach.
V. HANDOVER USING POSITIONING PROCESS
In wireless environment, supporting continuous
communication with QoS guaranties is hard to attempt. This is
due to fluctuation of network conditions, long time of
signalling requirement. When moving into different radio
coverage area, Mobile Terminal must find a new access point
with which to associate. The access point should provide
sufficient signal strength. We propose in this work to choose a
nearest access points in order to increase handover success. In
consequence, a scan step is eliminated. A positioning process
offer advantage of focusing the choice of the access point
candidate to which a terminal has a shortest distance. The
positioning process is responsible for measuring distance
between terminal and each access point. The positioning
component calculates distance between mobile terminal and
access point. It generates different distances in order to choice
the nearest access point far from MT.
In this section, we outline the implementation of the
positioning process in the MAC layer of a TM. We explain a
modular architecture proposal of our contribution in the MAC
layer, based on receiver, transmitter component and
positioning process. Figure 4 details interaction between
different processes using various types of handshaking signals.
To determine the position of TM in a 2D space, we used our
new geometric approach to Mobile position in wireless LAN
reducing complex computations [12]. This new geometric
oriented algorithm is based on three distances measurements
to determine the position of a mobile object. Provided that all
operations in our proposed algorithm are additions,
subtractions and multiplications based, the implementation is
simplified which reduces complexity.
The proposed method converges and provides with a good
accuracy the position of MS using a very reduced number of
iterations (K<10). Hence, the major advantages of our
algorithm are: implementation simplicity, and low
computation overhead.
The following figure illustrates the system architecture; we
divided location process to 4 parts: a process location
algorithm, the square root component, divider block and
buffers to store data

Figure 3. Top level structure of the Location circuit
Developed algorithms give information to allow mobile
terminal to select an access point in an optimal way. These
algorithms explain also a complexity of implementing
positioning process with VHDL description language. In fact,
positioning process necessitates functions to the extraction and
the manipulation of traffic parameters. It needs also functions
to compute terminal position. Complex models including
arithmetic operators as addition, division and multiplication
are, then, required to be employed.
The equations for the x and y position of the mobile was
modeled using VHDL. The numeric_std package was used to
construct the VHDL model that was readily synthesized into a
low power digital circuit. The input signal of the model are the
x, y positions of the three APs, i,j,k in meters, and the signals
TOA from the individual AP to the TM in nanoseconds. The
input signal assignments are x
i
, y
i
, TOA
i
, x
j
, y
j
, TOA
j
, x
k
, y
k
,
TOA
k

After calculating its coordinates (x, y) and using a simple
calculation of Euclidean distances, the MT can easily compare
the distances which separate it from different available APs.
The shortest distance corresponds to the AP closest to the MT.
This closest AP will be then chosen to establish a new
connection
The following figure shows the selection of the closest AP
based on Euclidean distances

Figure 4. Min di (APi, MT) determination
The input signal of the model are the x, y positions previously
computed of the TM in meters, and the signals CLK and Rst
as operating clock and initialization circuit respectively. The
output signal is the shortest distance from N distances and
therefore the selected AP to establish the new connection.
VI. RESULTS AND OBSERVATION
A. Basis model
A MT broadcasts probe request over three channels. On each
channel, it expects to receive responses from three access
points. Response frames as well as their SNR are buffered and
then used to select the AP that satisfies the mobile
requirements
Figure 5 gives an example of an active scan simulation. It
shows also probe response frames identification, addresses
extraction and SNR measurement. The rest of the handoff
process reposes on authentication and association phases while
each one a frame is sent to the selected AP and a response is
received.
After probing all selected channels, the next access point is
determined from the information received in the probe
responses and their associated Signal to Noise Ratio (SNR).
The following algorithm details the process described above.

Algorithm: Full-scanning algorithm.
1. For each channel to probe do
2. Broadcast probe request on this channel
3. Start probe timer
4. while True do
5. Read probe responses
6. if MinChannelTime expires, then
7. break
8. end if
9. if MaxChannelTime expires, then
10. break
11. end if
12. end while


Figure 5. Simulation of the Scan Phase on the first channel (basis model).
B. Handover with positioning process
This work lies in the development, implementation and
discussion of handover protocol. It shows performances of
handover procedure in WiFi environment based on positioning
strategy. In fact the proposed model is fixed to reduce
handover time. Parameters values used to validation are
selected according to typical cases. Performance evaluation is

given by simulation. Our approach of handover is transposed
on a concise description which support different scenario in
WiFi environment. Handover algorithm is integrated on the
MT. It is implemented in an FPGA environment with
simulation and synthesis tools. The efficiency of this
description for several network situations evaluates the
Handover algorithm performances.

Figure 6. New Handover mechanism based on positioning phase
C. Comparison
With the basic model a mobile station broadcasts probe
request over three channels. On each channel, it expects to
receive responses from three access points. Response frames
as well as their SNR are buffered and then used to select the
AP, which satisfy the mobile requirements. Fig 5 gives an
example of an active scan timing diagram. It shows also probe
response frames identification, addresses extraction and SNR
measurement. The rest of the handoff process reposes on
authentication and association phases while each one a frame
is sent to the selected AP and a response is received.
However, with positioning process, optimal handoff latency is
improved by replacing the scan phase of the basic model with
a fast positioning process.
Simulation results show that the proposed model allows the
reduction of almost 60% of the basic handover. The following
figure shows the number of clock cycles required for each
model.
20
70
120
170
220
Handover with
positioning process
Basic Handover
(scan phase)
Association phase
Authentication phase
Scan phase

Figure 7. Handoff latency for new and basis models
D. Synthesis results
During the synthesis step, we have exploited FPGA xilinx
virtex 5 environment. This environment allows implementing
communication systems on programmable circuits. The
advantage of using FPGAs circuits is mainly the system re-
scheduling. For our application, RTL synthesis is achieved
using the ISE 10.1 of the Xilinx FPGA virtex 5 environment.
Synthesis results, of the two approaches, are shown in table 1.
These results should be exploited in order to study their impact
on the support of the technological parameters specified in
IEEE 802.11.
TABLE I. COMPARISON OF DIFFERENT HANDOFF MECHANISMS

Number of
Slices
Number of Flip
Flops
Frequency
(MHz)
Basis Handoff
782 604 132
Proposed
model
507 463 187

VII. CONCLUSION
This paper shows a model that focuses on reducing handover
time consumed by the channel scanning phase. Therefore, we
presented a work for designing, simulating and synthesizing a
positioning approach that replaces a scan phase. We discussed
how a positioning approach can decrease handover during.
Implemented positioning IP reduces handover latency by
finding out distance between mobile terminal and various
access points. Results presented in this paper show that MT
can select the optimal access point with low time. Component
is designed and characterized by using a hardware design
flow. The description was made with the high description
language VHDL. ModelSim was used to check the behaviour
of the system at the RTL level. It permits to determine the
latency in terms of clock cycles. Synthesis was undertaken
using the ISE 10.1 of the FPGA environment xilinx virtex 5,
in order to evaluate the performance of the circuit in terms of
surface area, critical time and frequency operandi.
The approach explained in this work has limitations that may
not be apparent in indoor network, but are inadequate outdoor
for network.
REFERENCES
[1] Monji ZAIDI, Jamila BHAR, Ridha OUNI, Rached TOURKI, A new
solutions for micro-mobility management in 802.11 Wireless LANs
using FPGA, International Conference on Signals, Circuits & Systems
(SCS08), November 7-9, 2008 Hammamet,Tunisia.
[2] Monji ZAIDI, Ridha OUNI, Jamila BHAR, Rached TOURKI, New
approaches reducing handoff latency in 802.11 wireless LANs, IJCSES
International Journal of Computer Sciences and Engineering Systems,
Vol.3, No.3, July 2009.
[3] Jamila BHAR, Ridha OUNI, Kholdoun TORKI, Salem NASRI,
Handovers strategies challenges in wireless ATM networks,
International Journal of Applied Mathematics and Computer Sciences, 4
(2): April 2007, pp 636-641.
[4] Andrs Bohk, Levente Buttyn, and Lszl Dra An authentication
scheme for fast handover between WiFi access points, WICON 2007,
October 22-24, 2007.
[5] Jidong,Wang and Lichun Bao, Mobile Context Handoff in Distributed
IEEE 802.11 Systems, Bren School of Information and Computer
Sciences, University of California,Irvine, CA 92697.
[6] Hctor Velayos, Gunnar Karlsson, Techniques to Reduce IEEE
802.11b MAC Layer Handover Time, TRITA-IMIT-LCN R 03:02 ISSN
1651-7717,ISRN KTH/IMIT/LCN/R-03/02--SE. April 2003.

[7] Norbert Jordan, Reinhard Fleck, Christian Ploninger Fast Handover
Support in Wireless LAN based Networks, Institute of Communication
Networks, Vienna University of Technology Favoritenstrasse 9/388, A-
1040 Vienna, Austria
[8] I. Ramani and S. Savage, SyncScan: Practical Fast Handoff for 802.11
Infrastructure Networks, Proceedings of the IEEE Infocom, March
2005.
[9] Arunesh Mishra, Minho Shin, William Arbaugh. An Empirical
Analysis of the IEEE 802.11 MAC Layer Handoff Process, CS Tech
Report NumberCS-TR-4395. UMIACS Tech Report Number
UMIACSTR-2002-75
[10] A. Saeed, Hafizal Mohamad, Borhanuddin Mohd. Ali & Mazlan Abbas,
Vertical Handover Algorithm for WiMAX/WiFi Interworking,
International Journal of Engineering (IJE), Volume (3) : Issue (5).2008.
[11] Syed S. Rizvi, Aasia Riasat, and Khaled M, A Elleithy
QUANTITATIVE ANALYSIS OF HANDOVER TIME AT MAC
LAYER FOR WIRELESS MOBILE NETWORKS, International
Journal of Wireless & Mobile Networks (IJWMN), Vol 1, No 2,
November 2009
[12] Monji ZAIDI, Rached TOURKI, Ridha OUNI A New geometric
Approach to Mobile Position in wireless LAN reducing complex
computations, 2010 International Conference on Design & Technology
of Integrated Systems in Nanoscale Era(DTIS 2010) , Hammamet
Tunisia March 23-25 2010.

WORLDCOMP'11-ICOMP'11- ICW2941
Novel Data Harvesting Scheme for Efficient Data
Aggregation
Mohammad Zuheir Hourani, Ridha OUNI
Computer Engineering Department
P.O.Box 51178, Riyadh 11543, KSA

AbstractThe basic idea behind intelligent transportation
system is how to deploy vehicular sensor network that have
many characteristic such as high computation power ,enough
storage space and mobile sensor node in order to design an
effective and efficient architecture for data collection and data
exchange. In this paper we will introduce an intelligent
transportation system with new network paradigm to collect
important information from the road environment based on the
vehicular sensor network (VSN). Data aggregate provides the
drivers by valuable information in order to make the road safer
and less congested. Our system framework consists of active
vehicular sensor node, passive vehicular sensor node and sink
node distributed according to the road segmentation for
collecting data from active vehicular sensor node passing by,
while active vehicular sensor node collect data from passive
vehicular sensor node in their segment using multihop data
harvesting. Our scheme aims to reduce broadcast storm and
avoid collision. Finally, the simulation using the OPNET
simulator shows the effectiveness of the proposed schema.
Keyword- ITS, VSN, IVS, data harvesting, hybrid architecture,
data aggregation.
I. INTRODUCTION
Significant advances in manufacturing technology equipment
and the advent of Micro-Electro-Mechanical Switches
(MEMS) has opened the way for the construction of
intelligent sensor nodes which are able to perform three
major functions: sensing, processing and wireless
communication. These wireless sensor nodes are
characterized by their intelligence, their small size, low cost,
battery powered, and easy to install and repair. These features
open doors to deploy WSNs in the future for a wide range of
applications because it greatly expands our ability to monitor
and control the physical environment from remote locations.
An interesting field where the use of WSNs proves
effectiveness is the field of Intelligent Vehicular Systems. An
Intelligent Vehicular System (IVS) uses technological
advances in computers and information technology to
improve the efficiency of both new and existing vehicular
systems.
Vehicular sensor networks (VSNs) is a technology where
sensors are deployed in the road side and in the vehicles to
sense various urban phenomenons and transmit information
for vehicular traffic control and monitoring. VSNs have
different characteristic from traditional sensor network (static
network), interns of mobility, computational, power supply,
memory storage and reliability. Moreover vehicular sensor
network VSN has a much more dynamic topology as
compared to the static WSN. It is often assumed that VSN will
move continuously in a random fashion, thus making the
whole network a very dynamic topology. This dynamic nature
of VSN is reflected in the choice of other characteristic
properties, such as routing, MAC level protocols and physical
hardware, beside this, dynamic topology of vehicular sensor
network VSN, communication links can often become
unreliable [1]. The previous characteristics allow deploying
vehicular wireless sensor network to design intelligent
transportation system.
In this paper we are interested to design an optimal system
architecture for such vehicular sensor network for vehicular
traffic control and monitoring. several assumption have been
made. First, we assume that vehicles communicate through a
wireless interface, implementing a CSMA/CA MAC layer
protocol that provides a RTS/CTS/DATA/ACK handshake
sequence for each transmission. Vehicular sensor network
adopt IEEE 802.11 as a cost efficient and widely deployed
solution for network communication. IEEE 802.11p is a draft
amendment to IEEE 802.11 standard to add wireless access
in vehicular environment. It supports data exchange for
vehicle to vehicle (V2V) and vehicle to infrastructure (V2I)
in the licensed ITS band of 5.9 GHz.
The number of sink nodes that are distributed beside the road
is very small in proportion to the number of vehicular sensor
nodes. So, we assume that sink node has a relatively fast
processor and a large storage device and has enough energy
resources. In addition, it has very large data base to store
information from the vehicles. However, vehicular sensor
node has lower storage space and low processor capability. It
is assumed that each vehicle has unique ID to identify the
vehicular node and included in each message sent.
Finally, we assume devices participating in vehicular networks
are highly mobile with a speed up to 30 m/s. But, their
mobility patterns are predictable due to the constrained
movement imposed by the road system and constrained speed
imposed by speed limits, traffic conditions and signals. In fact,
the mobility of vehicular sensors poses challenges to the
communication system. Mobility undermines the reliability of
communication and also causes the topology to continuously
change.
This paper focus on how to deploy WSN as an intelligent
transportation system over effective and efficient architecture
for data collection and data exchange allowing vehicle traffic
monitoring and control. The rest of the paper is organized as
follow. Section 2 provides a back ground and related work in
vehicular sensor networks. Section 3 introduces the proposed
scheme for data harvesting and dissemination. Section 4
evaluates the simulation results conducted with OPNET.
Finally, section 5 concludes the paper.
II. RELATED WORK
Recently, there is a strong interest from researchers in
deploying WSNs in VSNs in many applications that involve
constraints related to the traffic conditions such as traffic
monitoring and control, traffic estimation and monitoring
parking. Some research focus in moving vehicles to enable
wireless sensor communication between roadside and vehicle
or between vehicles. These applications aim to make roads
safer and less congested in order to save the time for people.
Its important to note that these applications encounter three
types of communications [2]:
Vehicle-to-Vehicle (V2V) communication: vehicles are
equipped with sensors in order to exchange information
that is crucial to avoid severe situations like traffic jam
avoidance.
Vehicle-to-Infrastructure (V2I) communications:
information flow from vehicles to sensors installed on
roadway infrastructure
Hybrid communication: uses both V2V and V2I
architecture
In [3] the author proposed a scheme based on the hybrid
communication. Vehicles will send all their sensed data to
infostations, where the data will be forwarded to
corresponding infostation based on the infostations
management area. Later, any vehicle requesting sensed data
can request these infostation, which is more of an indirect
form of vehicle to vehicle communication using relay nodes
forming another type of data harvesting protocol. However,
this technique requires installation of an infostation
infrastructure, which can be very costly and complex.
With the use of wireless sensor network, multiple sensory
devices can be networked together to share geographically
distributed information. In [4], the system consists of two
vehicle sensory nodes that are placed on each side of a two-
lane road. These two wireless sensory nodes will send the
collected data to the base node whenever a vehicle is
detected. The road traffic monitoring system consists of a 3-
tier structure. The system is made up of the end-node tier,
base node tier and lastly the PC tier. All data that is captured
by the end-nodes will be forwarded to the base node. The
base node will then perform pre-processing before
forwarding the message to the PC for analysis.
Communication via the PC to the end nodes is also carried
out through the services of the base node using a star
topology. The end-node tier is responsible for vehicle
detection and gathering all the data from its onboard sensors.
One of the biggest issues in realizing VSN is concerned with
data harvesting which is a technique where sensors create
data that summarize the characteristic of the data and send it
to the target. In [5], the author proposed a novel multi-hop
data harvesting (MDH) method for the V2I architecture.
MDH have two scheme proposed for VSN. The MDH
scheme using replicas (MDH-R) is proposed for requesting
data from single sensor node, while data aggregation scheme
is designed (MDH-RA) for cases when the request was made
to a geo cast region. Many applications in VSN may require
multi-hop data transmission to meet real-time constraints.
The author see multi-hop data dissemination capabilities may
become ideal for future researches in this area.
VSNs come out as new brand of vehicular networks, whose
propose is the real time gathering and diffusion of
information. In [2] the author used a Clustering Gathering
Protocol (CGP) that is across layered protocol based on
hierarchal and geographical data gathering, aggregation and
dissemination. The goal of CGP is to gather from all node in
the vehicular ad hoc network in order to offer different kind
ITS services, it allows telecommunication/service providers
to get valuable information about the road environment in a
specific geographical area, using V2V network to minimize
the high cost links usability and base station to gather
information from the vehicles
III. OVER VIEW
Our system framework is consisting of static road side node
(sink node), and mobile vehicular sensors. Road side nodes
are distributed according to the road segmentation for
collecting data from mobile vehicular sensors passing by and
to exchange data about traffic condition. While mobile
sensors on vehicles monitor the road condition and send this
information to active mobile neighbors when they are close
enough then to the road side sink node (see Figure 1). We
focus on vehicular mobility, collaboration between mobile
and static nodes, and information exchange among mobile
cars. Mobile cars can gather latest information spreading on
the map out of the reach of static node, whereas static node
can gather information from more active cars coming across,
where the connectivity between static and mobile nodes and
also between mobile and mobile nodes are most likely
meaningful and useful.

Figure 1: Vehicular Sensor Architecture
IV. ROAD SEGMENTATION
The roads are divided into small segments. On each road
segment, there are two road side node (sink node) located at
the both ends of the segment, as shown in Figure.2. Usually,
the road side nodes are placed on the roadside with different
distances ( + ) based on the road environment to collect
data from active cars passing by. So, cars can get the road
condition before entering this segment; while vehicular
mobile sensors, assisted by themobility of the vehicles, can
know the road information along their own path.
The road is divided into virtual segments with the different
length (Figure 2). In each segment an active node is selected
to gather data from all nodes in its segment, aggregate them,
and send the result to the sink node.

Figure 2: Road Segmentation
V. PROPOSED SCHEME
The proposed scheme consists of providing a feasible,
efficient and robust vehicular sensor network framework to
monitor road traffic and provide desired and reliable
information for users, particularly for drivers in automobiles.
In the context we decided to use active node based solution
for the V2V dissemination. The scheme will be divided into
three parts: Active vehicular sensor node selection phase,
data harvesting/dissemination phase, and the data sharing
phase.
ACTIVE VEHICULAR SENSOR NODE SELECTION
PHASE
Every road segment has two such sink near the ends. Every
vehicle enter the segment will send hello message to the sink
node at the beginning of each segment containing the vehicle
ID. Then, the sink node will store this information in the data
base. Using this information, the sink node will create an
active node based on two parameters as threshold; the
maximum number of vehicles (passive nodes) detected in the
road, and the elapsed time.
First, the sink node stores the data about each vehicle entering
the segment until reaching the maximum number of vehicles.
Second, the sink node will choose one of these cars randomly
to be an active node by broadcasting control information as
request including the ID of the vehicle. When collecting this
request, the other vehicle nodes (passive nodes) identify the
target node dedicated for forwarding their information. All
other nodes must know the active node in their segment. To do
so, the AVS will include its ID in the packet as new
destination and then diffuse reply to the sink node which will
be also received and processed by its neighbored vehicles.
The mobility of vehicular sensor network can affect the
topology of the network. Therefore, we also use the elapsed
delay to control when the sink sends request to create an active
node exactly before the group of vehicle leave the wireless
range of the sink. This time will be calculated using the
equation below.
=

Where is the distance that our wireless communication can
support (IEEE 802.11), and reflects the mobility of the
vehicles which is the velocity of the vehicular sensor node. As
a result, we need two counters in the sink node one for time
and the other for the number of vehicles. So, in this way we
guarantee that we create an active node for each group of
vehicles.

Figure 3: Active Node Selection

DATA HARVESTING AND DISSEMINATION PHASE
A road segment can be congested, or free. When it is
congested, it can be either heavily congested or lightly
congested. Moreover, the traffic condition is changing with
time passing. So, a simple but effective method is needed to
represent the road condition. For a better delivery ratio and to
reduce broadcast storms, a message has to be relayed by a
minimum number of intermediate active nodes to the
destination. To do so, nodes are organized on a set of
segments, in which one node or more (active node) gathers
data in its segment and later sends this data to the sink node.
Segment-based active node solutions provide less propagation
delay and high delivery ratio with also bandwidth fairness. In
[4] the authors use a distributed clustering algorithm to create
a virtual backbone that allows only some nodes to broadcast
messages and thus, to reduce significantly broadcast storms.
When the active node receives the data from the passive node
originally holding the data, it will process it in store & forward
fashion instead of sending directly to the sink node when the
next sink node is far. Similarly, the passive node keeps the
data in its memory during a parametric time, and waits for
active node to be closest enough from it. An example is given
in figure 4, where the node (F) cannot reach the active node.
In this case, it will store its data till it reaches the active node
or wait another active node.
All nodes in the segment unicast their sensed data to the active
node, using a mechanism similar to DCF (Distributed
Coordination Function, presented in IEEE 802.11).
Each node wait a random bounded back-off time,
At the end of the back-off time, a node send a request to
send to the active node,
The active node acknowledges the reception by sending a
Clear to Send message,
The node sends its data to the active node.
An example is shown in figure 4.

Figure 4: Data Harvesting Model
Passive vehicular sensor nodes will monitor the traffic
condition by measuring the speed of the vehicle and send this
data to active node in its segment. The active vehicular sensor
node will store the data message and count the number of
vehicles in its segment, because it has limit number of
vehicles. This data will be forwarded to sink node ( + 1) that
is located at the end of this segment. In turn, the sink node
( + 1) will send this data to the preceding sink node () that
to update its data about traffic condition in this segment and
floods it to new coming vehicles that wish to enter this
segment.
When the vehicle is going to enter a new road segment, the
sink node at the near end will communicate with this vehicle.
So, the vehicle can know the road condition of this segment in
advance. There is no needs to place more sink node in the
middle of one segment, because even if the vehicle get
information at the middle of a segment, drivers still cannot
change their direction or change the route trip.
DATA SHARING PHASE
In some areas, data traffic may increase dramatically due to
many vehicles requesting for data at the same time. In this
case, there is a high probability that more than one vehicle is
requested to be an active vehicular node from the sink node.
When the active node reach its limit from the passive node due
to the congestion that involve many vehicle in the segment, in
this situation the sink node send request to create new active
node to provides fairness which is very important in a sensor
network where every node has to send its data. It also reduces
significantly broadcast storms and thus avoids collisions.
another case that we can have new active node when we have
two groups of cars and there is a time between them and one
of them reach the end of wireless communication range of
sink but the maximum number of passive node still not
complete. In this case, the sink node will request new active
node from the coming group based on time factor in order to
ensure that each group have active node to send information to
it. It can be seen from figure 5, when there are many vehicles
have data to be sent about the traffic condition which means
congestion occurs on a specific segment. In this case, the
active node sends to the sink node message in order to create
new active node.

Figure 5: Sharing Phase
VI. SIMULATION
The network topology was built in OPNET Modeler with the
following design considerations. There are three scenarios in
which we compare the traffic behavior based on congestion
and active node availability. For the first scenario, we have
one active node sidelined with 30 passive nodes and the
second scenario deals with congestion by increasing the traffic
generation by dramatically doubling the passive nodes.
Finally, the third scenario involves a mitigation step from the
sink by requesting an additional active node to accommodate
the situation for this congestion.
The environment for the research is set as, CSMA-CA with
minimum back-off exponent set to 3 and having 4 as
maximum number of back-offs with channel sensed every 0.1
seconds and operating in 2.4Ghz frequency band and the
transmit power is set to 0.002, with ACK wait duration set to
0.05 seconds for the participating active and passive nodes.
First of all the traffic received at the active agent is of utmost
importance to study the performance of the system. In figure
6 the traffic received is indicated as in the three scenarios.
When there is no congestion in the network the average
traffic received converges to 4300 b/s which decreases to
3500 b/s as soon as there is congestion in the network due to
extra management and control information besides data
traffic and finally having the additional active node will help
in having more data traffic. Secondly, we will consider the
delay incurred for transmission in the three scenarios,
average values are considered to study the effect of
congestion on the network. In this first scenario where we
have just one active node the average delay is around 0.013
seconds where as it rises to 0.014 seconds when congestion
occurs with the inclusion of additional passive nodes in the
network.

Figure 6: Data traffic received
After this the sink requests for an additional active node to
corner out the increase in the delay and with the inclusion of
this active node the load is distributed between the two active
nodes hence decreasing the delay incurring at each node
which is illustrated in figure 7.
Finally, the delay to access media is shown in figure 8 which
reveals the fact that without congestion in the network the
channel was accessed on an average of 0.0014 seconds but
with the inclusion of congestion the delay rises to a value of
0.021 seconds approx. and with the second active node
sharing the load reduces this value to 0.018 seconds which is
in the desirable range.

Figure 7: MAC delay

Figure 8: Media access delay
So, in all we can say that including extra active node is only
required in scenarios where the congestion in the network
increases and this will distribute the load among the
participating active nodes which are used for harvesting of
the information to the sink.
VII. CONCLUSION
In this paper, a scheme for data harvesting and data exchange
based on active vehicular sensor node is proposed. We
provide a collaborative hybrid method to deliver important
information to particular drivers effectively. We use road side
sink and vehicular sensor nodes to restore and exchange data,
then we use OPNET simulator to study our novel scheme
which illustrates that during the time of congestion in the
network, it is better to have additional active node beside the
old one and have many advantage as we see from the result.
VIII. REFERENCES
[1]. Munir, S.A.; Biao Ren; Weiwei J iao; Bin Wang; Dongliang
Xie; Man Ma , Mobile Wireless Sensor Network: Architecture
and Enabling Technologies for Ubiquitous Computing,
Advanced, pp:113 120, 08 August 2007.
[2]. I. Salhi, M. O. Cherif, S. M. Senouci, A New Architecture for
Data Collection in Vehicular Networks, IEEE International
Conference on Communications ICC'09, pp.16, August 2009.
[3]. U. Lee, E. Magistretti, B. Zhou, M. Gerla, P. Bellavista,
A.Corradi. Efficient data harvesting in mobile sensor
platforms, Proceedings of the 4th annual IEEE International
Conference on Pervasive Computing and Communications
Workshops PERCOMW06, pp. 352, 2006.
[4]. H. Ng, S.L. Tan, J . G. Guzman, Road traffic monitoring using
a wireless vehicle sensor network, International Symposiumon
Intelligent Signal Processing and Communications Systems,
ISPACS08. pp. 14, Feb. 2009.
[5]. K.W. LimY.-B. Ko, Multi-hop data harvesting in vehicular
sensor networks, Communications, IET, Vol.4, no7, pp.768
775, April 2010.

A New geometric Approach to Mobile Position in
wireless LAN reducing complex computations
Monji ZAIDI, Rached TOURKI
Monji.Zaidi@fsm.rnu.tn
Ridha OUNI
College of Computer and Information Sciences (CCIS),
King Saud University Riyadh, KSA
rouni@ksu.edu.sa
AbstractPositions estimation from Time of Arrival (TOA),
Time Difference of Arrival (TDOA), and Angle of Arrival (AOA)
measurements are the commonly used techniques. These
approaches use the location parameters received from different
sources and they are based on intersections of circles, hyperbolas,
and lines, respectively. The location is determined using standard
complex computation methods that are usually implemented in
software and needed relatively long execution time. An important
factor in achieving this is to minimize and simplify the
instructions that the mobile station (MS) has to execute in the
location determination process. Finding an effective location
estimation technique to facilitate processing data is the main
focuses in this paper. Therefore, in the wireless propagation
environment the Received Signal Strength (RSS) information
from three base stations (BSs) are recorded and processed and
they can provide an overlapping coverage area of interest. Then
an easy geometric technique is applied in order to effectively
calculate the location of the desired MS.
Keywords: Received Signal Strength (RSS), Wireless, Position
estimation, geometric technique
I. INTRODUCTION
Mobile location estimation has attracted a significant
amount of attention in recent years. The network-based
location estimation schemes have been widely adopted based
on the radio signals between the mobile device and the base
stations. Currently, given that many buildings are equipped
with WLAN (Wireless Local Area Network) access points
(shopping malls, museums, hospitals, airports, etc.) it may
become practical to use these access points to determine user
location in these indoor environments.
A variety of wireless location techniques have been studied
and investigated [1], [2], [3]. Network-based location
estimation schemes have been widely proposed and employed
in wireless communication systems. These schemes locate the
position of the MS based on the measured radio signals from its
neighborhood BSs. The representative algorithms for the
network-based location estimation techniques are the Time-Of-
Arrival (TOA), the Time Difference-Of-Arrival (TDOA), and
the Angle-Of-Arrival (AOA). The TOA scheme estimates the
MSs location by measuring the arrival time of the radio signals
coming from different wireless BSs, whiles the TDOA method
measures the time difference between the arriving radio
signals. The AOA technique is conducted within the BS by
observing the arriving angles of the signals coming from the
MS. The equations associated with the network-based location
estimation schemes are inherently nonlinear.
In this paper, an efficient geometry location estimation
algorithm is proposed to obtain the estimated position of the
MS, under Line-of-sight (LOS) and/or Non-Line-of-sight
(NLOS) environments. The MSs position is obtained by
confining the estimation based on the signal variations and the
geometric layout between the MS and the BSs. Both the 2D
and 3D locations of the MS can be estimated using the
proposed technique scheme. Reasonable location estimation
can be acquired within some of computing iterations even with
the existence of NLOS errors. The remainder of this paper is
organized as follows:
Section 2 describes related work for wireless location
estimation. The proposed algorithm is explained in Section 3
for the 2D location estimation. The simulation and analysis are
dealt in section 4. The performance evaluation of the proposed
scheme is conducted in Section 5. Section 6 draws the
conclusions and the future works.
II. RELATEDWORKS
Different locations estimation schemes have been
proposed to acquire the MSs position. Therefore various types
of information (for example, the signal traveling distance, the
received angle of the signal, and the Receiving Signal Strength
(RSS)) are involved to facilitate the algorithm design for
location estimation. The primary objective in most location
estimation algorithms is to obtain higher estimation accuracy.
Given the coordinates of BSj, (j = 1, 2, 3) as (Xj, Yj), and
the distances dj between MS and BSj, the simplest geometrical
algorithm for TOA positioning (Figure. 1(a)) is given in [4].
Coordinates of MS position (x,y) relative to BS
1
can be
calculated as:
x
y

The simplest geometrical algorithm for TDOA positioning
(Figure. 1(b)) is given in [5]. There are two estimated TDOA-s
dj, 1 between BS
1
and the j
th
base station (j = 2, 3). Coordinates
of MS position (x, y) relative to BS
1
can be calculated in terms
of d
1
:
x
y

X

Where:
2010 International Conference on Design & Technology of Integrated Systems in Nanoscale Era
-1- 978-1-4244-6340-4/10/$26.00 2010 IEEE
K

Inserting this intermediate result into the following equation
with j gives a quadratic equation in terms of d
1
.
u
x Y
y x

Taking its positive root and substituting it into (*) results in
the final solution. The AOA technique determines the MS
position (x, y) based on triangulation, as shown in (Figure.
1(c)). The intersection of two directional lines of bearing with
angles
and
defines a unique position, each formed by a

radial from a BS to the MS. The simplest geometric solution
can be derived using [6] with two AOA measurements
and
:
x
Y
tan
tan
tan
tan

y Y
x X
tan

BS
1
BS
2
BS
3
MS
Hyperbola 1
Hyperbola 2
d
1
d
2
d
3
(b)
BS
1
BS
3
BS
2
d
1
d
2
d
3
MS
(a)
BS
1
BS
2
MS
2
1
(c)

Figure 1. Position determination techniques: (a) TOA; (b) TDOA; (c) AOA

Using any of the mentioned methods, the calculation can be
done either at the BS [network-based schemes] or at the MS
[mobile-based schemes]. Network-based schemes have high
network cost and low accuracy [7]. Mobile-based location
schemes are more interesting.
However, since the MS has limited energy source, in the
form of the battery pack, energy consumption should be
minimized. An important factor in achieving this is to minimize
and simplify the instructions that the MS has to execute in the
location determination process. The conventional algorithms
use complex computation methods that needed relatively long
execution time.
III. NEWLOCATIONALGORITHMBASEDONTHREE
BSS
In the general geometrical triangulation location researches,
they assumed that the measured noise is additive and the NLOS
error is a large positive bias which causes the measured ranges
to be greater than the true ranges [8].
Under the assumption, the MS location will guarantee to lie
in the overlapped region (enclosed by points A, B and C) of the
range circles as shown in Figure. 2. Thus the MS is necessarily
located in the region formed by the points BS
1
, BS
2
and BS
3
But, it is noted that the intersection of three circles may not
be overlapped with the real measurement results. Therefore,
with the above assumption we have to judge whether the three
circles intersect or not in our location algorithm.
BS
1
BS
2
BS
3
A
B
C

Figure 2. Measured range circles and the associated intersected area
If circles intersect as depicted in Figure. 3, then three
triangles can be drawn as: BS
1
MSBS
2
, BS
2
MSBS
3
and
BS
3
MSBS
1
.
Assumptions:
Different BSs are placed (two to two) at an equal
distance
The coordinates of BSs are known by the MS
The MS can inquire only on the received power
coming from the BSs (i.e. the distance which separates
it from each BS).
u
B u
B anu u
B
-2-
Note by:
D: The distance between tow BSs.
A0, B0 and C0 are the orthogonal projections of the
MS on (BS1 BS2), (BS2, BS3) and (BS3 BS1)
respectively.
d1, d2 and d3 are the distances that separate the MS
from BS1, BS2 and BS3 respectively.
12: is the geometrical angle between the MS-BS1 and
BS1-BS2. (Same things for the other angles).
We focus firstly on the triangle BS
1
MSBS
2
.
Based on the above assumption and the following figure, we
can write.
i
cos

BS1 BS2
BS
3
MS d
1
d
2
d
3
23
21 12
13
32
31
) , ( 12 12 0 y x A
) , ( 23 23 0 y x B ) , ( 31 31 0 y x C
D
r1
Figure 3. The associated triangles of the standard intersection of three
circles.
We can also wiite
u
B i
B u
cos
cos
Bu
cos
Bi
B

We define here the first factor q
by
q
B

B

The range of the parameter q
can determine the shape of

the triangle BS
1
MSBS
2
.For example.

If q

BS
2
MS
d
1
d
2
D BS
1
If q

D BS
2
BS
1
MS
d
1
d
2
If q

BS
1
BS
2
D
d
1
d
2
MS
Coordinates x
of the point A
0
are given in [9] by
x
X

y
Y

Where:
X
and X
are the coordinates of BS

1
and BS
2
,
respectively.
Let the distance between BS
1
and B
0
be i
and the distance

between BS
3
and C
0
be i
As we described previously, we can get the coordinates of

points B
0
and C
0
as:
x
X

y
Y

x
X

y
Y

Where:
-3-
q
B

B
B

B

MS is then located in a new triangle A
0
B
0
C
0
, which is
smaller in terms of area compared to the starting triangle
BS
1
BS
2
BS
3
. In the other word we have just created three new
virtual BSs placed at A
0
, B
0
and C
0
.
It is very easy to calculate the distances between the MS
and the new points A
0
, B
0
and C
0
using the Pythagoras formula.
Thus
uNS A
,
uNS B
,
uNS C
.
Now, with the three new virtual BSs, MS can repeat the
same calculations as shown above. During this second iteration,
the orthogonal projections of MS on (A
0
B
0
), (B
0
C
0
) and (C
0
A
0
)
must be done to obtain new point A
1
, B
1
and C
1
that their
coordinates may be determined as previously. A
1
B
1
C
1
s area is
smaller that the A
0
B
0
C
0
one.
At the i
iteration, the MS will be located in an A

i
B
i
C
i
triangle which is smaller than A
i-1
B
i-1
C
i-1
one. This A
i
B
i
C
i
triangle allows to designing the next triangle A
i+1
B
i+1
C
i+1
. After
a small number of iterations, the coordinates of three vertices
of the triangle (A, B and C) converge to the actual coordinates
of the MS. At the limit, the triangle A
conv
B
conv
C
conv
with
vertices Ac
onv
, B
conv
and C
conv
will be considered as a point. So,
it is possible to write:
x

We can then take the cooiuinates of the NS as
x

conv
B
conv
C
conv
triangle.
The following figure (section 4) shows the evolution and
the convergence of the three vertices coordinates for different
values of di (d
1
, d
2
and d
3
),
IV. SIMULATIONRESULTS
BS1 coordinates (in meters):X



Figure 4. Scenario 1:u
m u
m anu u
m
0 5 10 15 20
30
40
50
60
iterations number
x
(
A
)
0 5 10 15 20
20
40
60
80
iterations number
x
(
B
)
0 5 10 15 20
25
30
35
40
iterations numbers
x
(
C
)
0 5 10 15 20
0
20
40
60
iterations numbers
y
(
A
)
0 5 10 15 20
30
40
50
60
iterations number
y
(
B
)
0 5 10 15 20
30
35
40
45
iterations number
y
(
C
)
0 5 10 15 20
30
35
40
45
iterations number
x
(
M
S
)
0 5 10 15 20
30
35
40
iterations number
y
(
M
S
)
-4-

Figure 5. Scenario 2:u
m u
m anu u
m
V. PERFORMANCE ANALYSIS
The reception of a tuple of signal strengths does not lead
directly to the position of the device. A conversion of this
tuple of received signal strengths into a position is required
The following model introduces some wall attenuation
factors to describe more closely the environment. The walls
materials must be characterized, and their properties must be
introduced in the model, leading to the following
approximation [10]:
Where
is the signal strength received by the

mobile at distance d,
the signal strength

received at the known distance
from the AP, and a

coefficient modeling the radio wave propagation in the
environment For example, in free path loss environment, we
have 2. In indoor environments, this factor will be closer
to 3 [11].
is the number of walls of different nature,
is
the number of walls having an attenuation of
. It is clear that
the received power is always sullied with errors. Therefore
errors on the distance and on the position of MS can take
place. Those errors appear because the propagation models are
too simple in comparison to the complex indoor RF
propagation.
Now, it is necessary, as in any positioning method, to
evaluate the error or deviation (in m) between actual
(measured) and simulated values obtained by our method. For
this two cases have to be considered:
A. Line-of-sight (LOS) condition
This case occurs in open areas or in very specific spots in
city centers, in places such as crossroads or large squares with a
good visibility of BS. Sometimes, there might not be a direct
LOS signal but a strong specular reflection off a smooth
surface such as that of a large building will give rise to similar
conditions. The received signal will be strong and with
moderate fluctuations. Therefore, the extracted distance from
the received signal is correctly calculated.
In the table 1 we give some actual locations of the MS
(Actual x and y). Corresponding values of the true distances d
1
,
d
2
and d
3
which separate it from BS
1
, BS
2
and BS
3
are
calculated. Then the estimated position and position error can
be determined using our geometric method.
B. Non Line-of-sight (NLOS) condition
This case will typically be found in Indoor environments.
This is a worst-case scenario since the direct signal is
completely blocked out and the overall received signal is only
due to multipath, thus being weaker and subjected to marked
variations. Under these conditions the geometric method can be
applied. However, the position error increases significantly.
0 5 10 15 20
60
62
64
iterations number
x
(
A
)
0 5 10 15 20
55
60
65
iterations number
x
(
B
)
0 5 10 15 20
40
50
60
70
iterations numbers
x
(
C
)
0 5 10 15 20
0
50
100
iterations numbers
y
(
A
)
0 5 10 15 20
58
60
62
64
iterations number
y
(
B
)
0 5 10 15 20
50
60
70
80
iterations number
y
(
C
)
0 5 10 15 20
55
60
65
iterations number
x
(
M
S
)
0 5 10 15 20
40
50
60
70
iterations number
y
(
M
S
)
-5-
TABLE I. LOS MEASUREMENTS AND POSITION ESTIMATION
xactual (m) y
actual
(m) d
1
(m) d
2
(m) d
3
(m) x
estimated
(m) y
estimated
(m)
C.t
(I.n) Error (m)
20 10 22.5 80.5 82.5 20.1300 09.2035 8 0.8070
30 20 36 73 69.5 29.8350 19.6409 8 0.3952
40 30 50.5 67 57.5 40.3063 29.7057 8 0.4248
40 40 57 72 47.5 40.3250 39.8619 8 0.3531
50 50 71 71 36 50.0000 50.2384 8 0.2384
50 70 86.5 86.5 16 50.0000 70.4782 8 0.4782
60 60 85.2 72 28.5 60.3752 59.9144 8 0.3848
60 40 72 57 47.5 59.6750 39.8619 8 0.3531
70 30 76.5 42.5 60.5 70.0600 29.4535 8 0.5498
80 20 82.5 27.9 73 79.8552 19.6961 8 0.3366
Mean=0.4321
C.t = Convergence Time.
(I.n) = Iterations number.
TABLE II. NLOS MEASUREMENTS AND POSITION ESTIMATION
x
actual
(m) y
actual
(m) x
estimated
(m) y
estimated
(m) Error (m)
20 10 18.9700 8.4826 1.8340
30 20 29.0950 19.2922 1.1489
40 30 39.0763 28.7347 1.5666
40 40 41.4850 40.8110 1.6920
50 50 50.0000 51.0523 1.0523
50 70 48.2600 70.9884 2.0011
60 60 58.7450 60.6642 1.4199
60 40 61.1350 40.4331 1.2148
70 30 70.2300 28.0174 1.9959
80 20 79.7000 17.5872 2.4314
Mean=1.6357
VI. CONCLUSION
This paper presents new geometric oriented algorithm that
is based on three distances measurements to determine the
position of a mobile object. Provided that all operations in our
proposed algorithm are additions, subtractions and
multiplications based, the implementation is simplified which
reduces complexity.
Our results show that for a very reduced number of
iterations (k , the proposed method converges and
provides with a good accuracy the position of MS. Hence, the
major advantages of our algorithm are: implementation
simplicity, and low computation overhead.
The very fast growth of modern VLSI technology offers a
hardware realization of an ever-growing share of mathematical
means, so, in our future works, the proposed algorithm for
location determination will be implemented in hardware using
for example, a simple field programmable gate array (FPGA)
chip.
-6-
REFERENCES
[1] Y. Zhao, Standardization of Mobile Phone Positioning for 3G
Systems, IEEE Comm. Magazine, vol. 40, pp. 108-116, July 2002.
[2] H. Koshima and J. Hoshen, Personal Locator Services Emerge, IEEE
Spectrum, vol. 37, pp. 41-48, Feb. 2000.
[3] J.H. Reed, K.J. Krizman, B.D. Woerner, and T.S. Rappaport, An
Overview of the Challenges and Progress in Meeting the E-
911Requirement for Location Service, IEEE Comm. Magazine, vol. 36,
pp. 30-37, Apr. 1998.
[4] I. Jami, M. Ali, R.F. Ormondroyd, Comparison of Methods of Locating
and Tracking Cellular Mobiles, Novel Methods of Location and
Tracking of Cellular Mobiles and Their System Applications (Ref. No.
1999/046), IEE Colloquium, London UK,1/1-1/6.
[5] Y.T. Chan, K.C. Ho, A simple and efficient estimator for hyperbolic
location, IEEE Transactions on Signal Processing, 42(8) (1994).
[6] Alba Pages-Zamora, Josep Vidal, Dana H. Brooks, Closed-form solution
for positioning based on angle of arrival measurements, in: Proc. of the
13th Sym. on Personal, Indoor and Mobile Radio Communications,
September 2002, vol. 4, pp. 15221526.
[7] Y. Zhao, Standardization of mobile phone positioning for 3G systems,
IEEE Communications Magazine 40 (4) (2002) 108116.
[8] C. D. Wann and H.C. Chin, Hybrid TOA/RSSI Wireless Location with
Unconstrained Nonlinear Optimization for Indoor UWB Channels,
IEEE WCNC, 2007, March 2007, pp. 39403945.
[9] Chi-Kuang Hwang and Kun-Feng Cheng Wi-Fi Indoor Location Based
on RSS Hyper-Planes Method, Chung Hua Journal of Science and
Engineering, Vol. 5, No. 4, pp. 37-43 (2007)
[10] Y. Chen and H. Kobayashi, Signal strength based indoor
geolocation,in Proceedings of the IEEE International
Conference on Communications (ICC 02), vol. 1, pp. 436439,
New York, NY, USA, April-May 2002.
[11] R. Vaughan and J. B. Andersen, Channels, Propagation and
Antennas forMobile Communications, ElectromagneticWaves
Series 50, The Institution of Electrical Engineers, London, UK,
2003.
-7-
Performance evaluation of
compensation/reward mechanism for resource
allocation in wireless networks

Jamila BHAR, Ridha OUNI and Salem NASRI
Faculty of Sciences of Monastir (FSM)
Monastir, Tunisia
jamilabhar@yahoo.fr

AbstractIn this paper we evaluate an algorithm to
distributing resources provided by the Access Point at
the MAC layer in wireless networks. We consider a
TDMA/FDD-based mechanism. System model is
designed and tested for a wireless ATM networks
architecture. In fact, the QoS characterising ATM
technology provides the ability to well transmit multiple
traffic types and treat every one according to some
generic parameters. The basic idea used in this work is
to provide slots reassignment, and to dynamically adjust
present parameters. Simulation tests show that
developed algorithms provide better results and permit
to decrease blocking probability for new connections,
particularly to CBR connections.

I. INTRODUCTION
In this paper, we are interested to evaluate a frame
reservation strategy to allow efficient transmission of
multi-service traffic over TDMA/FDD channels in a
WATM networks. This scheme is based on dynamic
allocation of bandwidth among connections carrying
different types of traffic. The system strategy is to
reserve bandwidth (which is changed dynamically)
for each type of traffic during each frame-time. The
distribution of bandwidth among the corresponding
VCs depends on parameters of each traffic type.
Compared to classic TDMA mechanism, the
simulation results using TDMA based on a
compensation/reward method (TDMA-CR) shows
more efficient characteristics of such resource
allocation scheme.
The performances of the proposed schemes are
evaluated (in terms of connection blocking probability
and traffic load) for various data traffic models. The
paper is organised as follows. We first present the
protocol aspect of TDMA scheme and we invoke the
WATM architecture. The second section is devoted to
the proposed algorithm particularity and experimental
design. Regarding the impact of resource allocation
procedure we present our simulation model and the
derived performance results. Finally we summarize
the findings of our study and we give some
concluding remarks.

II. TDMA SCHEME DESCRIPTION
Typical dynamic TDMA protocol is always selected
for resources allocation in WATM networks. It
provides QoS services due to its superiority and
suitability for real-time multimedia traffics [9]. In a
TDMA mechanism, the bandwidth allocation assumes
the form of time-slot allocation and leads to link
scheduling.

Fixed frame duration
.
.
.
.

Frame
header Signalling
mini-slots
Dynamic
allocated
ABR, VBR
and UBR slots
Fixed
allocated
CBR slots
Idle
slots

Figure 1. Dynamic TDMAUPLINK access control frame format
The TDMA-frame begins by a frame header
transporting information for synchronisation. The
frame can be divided in three sessions (figure 1).
They consist of a signalling period, a data
transmission period and an idle period. Source
terminal does not always have data to send. In
consequence, the data transmission period is variable.
The number of data slots for each connexion depends
on characteristics of each connexion type. In this
paper, we assume that the slot size is equal to a
WATM cell. When a signalling period achieved, the
AP knows all the terminals that have data to transmit
and calculate the slots number to be assigned for each
978-1-4244-2182-4/08/$25.00 2008 IEEE. 1195
connexion. The system begins then the data
transmission period. If there are no data to send, the
system proceeds directly to an idle period, which lasts
until the next session. When a frame period finishes,
the system begins the next round and the same
procedure is repeated. The duration of each frame is
fixed.

III. SYSTEM ARCHITECTURE
Asynchronous transfer mode (ATM) based
technology can provide high speed wireless
multimedia communications. In fact, the fine-grain
multiplexing provided by ATM due to the fixed small
cell size is well suited to slow-speed wireless links
since it leads to lower delay jitter and queuing delays
[1]. The wireless ATM protocol architecture is based
on incorporation of wireless access and mobility
related functions into the standard ATM stack. A high
speed but low complexity wireless access technique is
critical for providing bandwidth-on-demand
multimedia services to mobile terminals. Typical
target bit rates for the radio physical layer of wireless
ATM are around 25 Mbps and a modem must be able
to support burst operation with relatively short
preambles consistent with transmission of short
control packets and ATM cells [10].

Wireles
s
Control
+
Signali
ng
W-Data Link
Control
W-Medium
Access
W-Physic
control

Standard
ATM
Physical
Layer

User service
ATM Network Layer
ATM
Adaptation
User service
ATM Network Layer
ATM Adaptation Layer
W-Physic
control
W-Medium Access
Control

Wireles
s
Control
+
Signali
ng
W-Data Link
Control
WATM
NNI
ATM-Switch
Wireless
Terminal
Access
Point

Figure 2. Design architecture of WATM network
For efficient sharing of the available wireless
bandwidth among multiple wireless terminals, a radio
MAC layer is required. For our study, a TDMA
approach is adopted for medium access control where
several virtual circuits are multiplexed in a single
radio channel. The TDMA frame structure supports
constant bit rate (CBR), available bit rate (ABR),
variable bit rate (VBR) and unspecified bit rate
(UBR), services within each access point transmission
cell area.

IV. DESIGN MODEL
A. Wireless Terminal MAC Model
A Wireless Terminal (WT) describes a CBR, VBR,
ABR and UBR traffic models. The ATMFs Traffic
Management specification defines four cell-based
traffic parameters namely the Peak Cell Rate (PCR),
Sustainable Cell Rate (SCR), Maximum Burst Size
(MBS) and Minimum Cell Rate (MCR) [13]. The
PCR is a maximum rate at which the user will
transmit cells. Its inverse, the minimum cell inter
arrival time (1/PCR), may be easier to measure in
practice and it is useful to evaluate performance. The
SCR is an upper bound on the possible conforming
average rate for an ATM connection, where the
average rate is simply the number of cells transmitted
divided by the connections duration. For ideal
Constant Bit Rate (CBR) traffic, the PCR equals the
SCR. For Variable Bite Rate (VBR) traffic, the SCR
is typically less than the PCR.
For CBR VCs, slots are allocated according to their
demanded bit rates. A CBR traffic convention
includes the PCR and the Cell Delay Variation
Tolerance (CDVT) factors. For VBR source model,
we consider an on-off that transmits a number of
cells at its SCR. Then, slots allocated to different
sources depend on traffic model types parameters.

B. Access Point Mac model
The crucial networking algorithm is placed at the AP.
It includes receiving packet (data/signalling), FIFOs
control, resources managements, etc. Hence,
signalling and data WATM cells are multiplexed and
treated according to resources allocation scheme. The
access point controls the uplink bandwidth allocation
for ATM cells from each Wireless Terminal (WT),
taking into account the number and the type of active
connections and their bandwidth requirements.
The medium bandwidth of WATM network is divided
into two separate channels: uplink and downlink. The
uplink channel transfers information from WT to the
AP. Each of those channels is further partitioned into
several sub frames, carrying different classes of
traffic. A set of buffer per-VC cell scheduling
schemes are used as first-come first-served (FCFS)
(figure 3). In general, FCFS cell-scheduling algorithm
is the simplest method. It has the advantages of low
costs and it is easy to implement, but it performs
worse in the scheduling efficiency and the slot
utilization.
The entity of radio resource manager, located at the
access point, takes part in the connection admission
1196
control (CAC) process for a WATM terminal
originated or terminated connection. It performs the
wireless connection admission control (WCAC) and
is responsible for the long-term allocation of
bandwidth to ATM connection over the radio
interface.
.
.
.
Scheduling Module
Data
Cells

n
1
Signalling
Packet
Signalling
Module

.
.
.
ATM
Interfa

Control
Unit
n-1
PHY EMIS
PHY RECE

Figure 3. AP radio medium architecture in WATM
The entity of scheduler is responsible for scheduling
the traffic transmitted through the wireless medium.
This component decides the time an ATM cell will be
transmitted [15]. Buffering at the access point is
required for queuing data cells. As a result, we have
to control CDV parameter.

V. TDMA COMPENSATION/REWARD (TDMA-CR)
PROPOSAL FOR DYNAMIC RESOURCE ALLOCATION
In the proposed system design, radio spectrum is
divided into time slots which are assigned to different
connections where a user application can send data
only in its own dedicated slots. Due to the FDD
duplexing technique utilized in the proposed MAC
protocol, two distinct carrier frequencies are used for
the uplink and downlink channels. Uplink consists of
slotted signalling mini-slots, followed by allocated
CBR, ABR, VBR and UBR data slots. The signalling
period is used by the wireless terminals to send their
bandwidth requirements to the access point. Then,
each TDMA round begins by a signalling phase
during which, all terminals keep their radios on. In
fact, when a WT needs to communicate with any
other, initially it sent a connection request message to
the AP. According to the QoS requirements of this
connection request, the AP assigns adequate number
of time slots for this connection using registered
parameters. These network parameters are
manipulated with an algorithm based on ATM service
type characteristics. If there are not enough slots for
the request, a connection can not be established with
required QoS guarantees. However, AP tries, in such
case, to adjust system resource by the resources
allocation mechanism proposed to apply a
compensation/reward process. In fact, to avoid reject
of CBR connection, the proposed algorithm access to
a number of slots from ABR and VBR connection
without degrading performances of these lasts. This
algorithm stops if the desired resources are satisfied.

no
Yes
no
Yes
no
Yes
.
.
no
no
Yes
Transmission
Disconnection and
parameters
actualization
- Save connection
- Accept notification
-Network parameters
update

no
Yes
- Connection Reject
- Release
notification
Packet
arrival
Connection
request
Disconnection
request
Data
Satisfying service
class requirements
by unused slots.
Satisfying service
class requirements
by unused slots.
Satisfying service
class requirements
by compensation.
- Satisfying the
minimum service
class
requirements
Satisfying service
class requirements
by compensation.
VBR
CBR

Figure 4. AP MAC layer algorithm
ABR and VBR connections can reward theirs slots (or
at least some of them) during next frames according
to slots availability. As a result, a number of data slot
allocated to connected terminals differ from one
frame to another.
TDMA frame
ABR1 CBR1 CBR2 CBR3 ABR2 Idles
Slots
ABR1 CBR1 CBR2 CBR3 ABR2
+ +
Disconnection:
CBR3
New connection:
ABR3
New connection:
CBR4
ABR1 CBR1 CBR2 CBR4 ABR2 Idles
Slots
ABR1 CBR1 CBR2 CBR4 ABR2 ABR3

Figure 5. Compensation/Reward protocol proposal
1197
Figure 5 shows an example of proposed protocol
scenario. When, the access point resources cannot
satisfy a new CBR connection, it tries to compensate
these resources from ABR and VBR connections.
Else it rejects the CBR demand. Considered ABR and
VBR connections can reward its resources if there are
remained slots. Contrary to CBR source, ABR or
VBR can begins its connection with a rate lower than
the required rate and it can increase it during next
frames.
Each WT (source) sends its data to the AP over its
allocated slot-time. If it has no data to send, the
terminal operates in idle mode. When a frame
finishes, the next frame begins and the distribution of
slots time is actualized. The AP collects the network
informations from all the terminals and forwards them
acquitting cells. Formats of signalling and data
WATM cells are described in figure 6. Signalling
packet is used for connection or disconnection request
and as acknowledgment packet. The same format is
maintained and only the signification of
PRM1(PaRaMeter1), PRM2 and PRM3 parameters is
modified. These lasts incorporate service class
parameters, transmitted from WT to the AP, useful to
calculate number of time slots specific to each
connection. Such, CBR connection requires SCR
(typically equal to PCR) rate and CDV constraints
and ABR connection requires MCR and PCR rates as
parameters.

Frame Header
AP Identifier
TDMA Frame No
Down signaling slots
Up signaling slots
Down data cells
frame size
CRC
Reserved (2 bytes)
CLP
Type
CSN
Control
MVCI
MVCI
HEC
CRC
DATA (48 bytes)
GFC
PT
CLS
MVCI
MVCI
CRC
PRM3 (2 bytes)
PT
PRM2 (2 bytes)
PRM1 (2 bytes)
WATM Data cell
Signaling packet

Figure 6. WATM packets formats
Proposed protocol functionalities are defined in
following. Initially, the process will identify the type
of the request message. If it is a disconnection
demand, the system actualizes its available resources.
If it is a connection request, AP scheme first registers
the corresponding VCs as active and stores the
reservation information. Then it estimates resource
needed by WT to decide if the system resource is
sufficient to permit WT into system. The AP
transmits slot allocation information (number of slots
allocated to each user and their positions in the
frame).

VI. SIMULATION RESULTS AND PERFORMANCE
ANALYSIS
Different extension schemes of TDMA are proposed
in the literature for a fair and efficient operation of the
MAC protocol. Research strategies focus on resolving
difficulties to distribute carefully AP system resources
between WTs. Difficulties are related to various
traffic conditions like buffer occupancy, connection
parameter requirements. Diverse solutions are
proposed in the literature. [11] Proposes to inform
access point about the status of the input queues.
Then, information about wireless terminals queues
must be sent in signalling packets over a special short
control slots. It will be useful to distribute the suitable
number of slots for each WT. We mention also that a
scheduling technique, used for multiplexing terminals
data, impacts the buffer size.
Another challenge studied in this context is the
improvement of handoff performances. Then, buffer
management and optimal reservation of radio
resources allows ameliorating handoff efficiency.
Many admission control strategies have been
discussed in literature to give priorities to handoff
requests compared to the new connection requests as
shown in [3, 5]. When a mobile terminal moves from
one area to another, the access point in the new area
must allow sufficient resources to this handoff
connection. The premature termination of established
connections, in absence of sufficient resources, is
usually more objectionable than rejection of a new
connection request.
In this work, we evaluate the proposed protocol to
ameliorate some traffic conditions. Figure 7 shows
the proportionality between output and input flow for
different channel utilization cases. Channel utilization
is defined in [1] as the ratio of the number of slots
used for WATM cell transmission to the total number
of slots available for transmission. Figure 7 depicts
the influence of connected terminals number on the
output percentage. This percentage decrease when the
input flow is important. This is explained by a huge
traffic of terminals that leads to a degradation of
available resource at the AP. Consequently, the use of
1198
an efficient resource allocated protocol is essential to
avoid traffic performances degradation.
0
10
20
30
40
50
60
70
80
15 30 50 70 90
%input
%

o
u
t
p
u
t
8 MTs
35 MTs

Figure 7. Variation of traffic in function of terminals number
Moreover, TDMA-CR is a resource allocation
protocol witch affect the connection reject
probability. Figure 8 shows the effect of the proposed
resource allocation protocol on the reject probability.
It compares, for the same considered traffic scenario,
a reject probability between classical TDMA and
TDMA-CR schemes. We mention here that the big
difference between the two cases at a first three
seconds is explained by the great effect of
compensation method. In the remained simulation
time, the difference of reject probability variation
between the two methods becomes small. This result
is justified by the limit of resources compensation for
each traffic class. This means that the efficiency of a
proposed scheme can decrease when resources are
limited or a great number of terminals need to be
served.
0
0,05
0,1
0,15
0,2
0,25
0,3
0,35
0
,
0
6
0
,
6
8
1
,
3
1
,
9
1
2
,
5
3
3
,
1
6
3
,
7
9
4
,
4
5
,
0
3
5
,
6
4
6
,
2
7
6
,
8
9
7
,
5
1
Time (sec)
R
e
j
e
c
t

p
r
o
b
a
b
i
l
i
t
y
TDMA
TDMA-CR

Figure 8. Connection reject probability
Results presented by figure 8 show that a
compensation mechanism does not decrease a reject
connections number only, but it also increase the
percentage of resources utilisation as shown in figure
9. The improvement of resources utilisation explains
that compensation approach avoid resources wasting
and increase a number of served connections. In fact,
if compensation method decreases the number of
rejected connections, it means that this method allows
connecting new terminals with remainder and
compensated resources.
0
10
20
30
40
50
60
70
80
90
1 2 3 4 5 6 7 8
time (sec)
%
i
n
p
u
t
TDMA
TDMA-CR

Figure 9. Resources utilisation percentage
0
0,1
0,2
0,3
0,4
0,5
0,6
0
0
,
2
8
0
,
5
6
0
,
8
3
1
,
1
2
1
,
4
1
,
6
7
1
,
9
5
2
,
2
3
2
,
5
1
2
,
7
8
3
,
0
7
3
,
3
4
3
,
6
2
Time (sec)
C
D
V

(
m
s
)
TDMA
TDMA-CR

Figure 10. CDV of CBR connection
The cell delay variation (CDV) is a QoS
parameter which depends on the type of service class.
CDV should take on almost constant value especially
for CBR classes. Figure 10 compares a cell delay
variation (CDV) of CBR connection using classical
and modified TDMA method. We conclude that a
modified TDMA scheme let CDV more stable. It
gives a sufficient resources distribution strategy that
arrange all connections in an equilibrium way.

VII. CONCLUDING REMARKS
It is necessary to provide QoS guarantees subject to
ensuring delay constraints, prescribed data rate and
loss for the various applications such as voice, video
and high quality multimedia services. The MAC is
necessary for QoS based bandwidth allocation to
multiple ATM traffic classes. In this works, we have
proposed an approach to support wireless multimedia
applications using TDMA/FDD MAC protocol based
WATM. The design algorithms use TDMA/FDD-
based access scheme, with a compensation/reward
process. The TDMA mechanism traits all signalling
information of connected terminals and executes the
suitable process to compute the allocation of slots in
next frame. Simulation results show the efficiency of
such compensation/reward process to ameliorate
traffic condition. We mention that this algorithm can
be adopted for others networks protocols.
1199
REFERENCES
[1] Ekram Hossain and Vijay K. Bhargava; Link-Level Traffic
Scheduling for Providing Predictive QoS in Wireless
Multimedia Networks; IEEE TRANSACTIONS ON
MULTIMEDIA, FEBRUARY 2004, VOL. 6, NO. 1.
[2] Xiaohua Li; Contention Resolution in Random-Access
Wireless Networks Based on Orthogonal Complementary
Codes; IEEE TRANSACTIONS ON COMMUNICATIONS,
JANUARY 2004, VOL. 52, NO. 1.
[3] parameswaran ramanathan, Krishna M. sivalingam, prathima
agrawal and shalinee kishore; Dynamic resource allocation
schemes during handoff for mobile multimedia wireless
networks; IEEE journal on selected areas in communications,
July 1999, VOL XX.
[4] Bih-Hwang Lee Hsin-Pei Chen Su-Shun Huang; Dynamic
Resource Allocation for Handoff in WATM Networks;
Proceedings of the 2005 11th International Conference on
Parallel and Distributed Systems (ICPADS'05) IEEE 2005.
[5] Maria C. Yuang, Po L. Tien, and Ching S. Chen; A
Contention Access Protocol with Dynamic Bandwidth
Allocation for Wireless ATM Networks; IEEE 2000, p149-
153.
[6] D. Raychaudhuri, L.J. French, R.J. Siracusa, S.K. Biswas, R.
Yuan, P. Narasimhan & C. Johnston; WATMnet: A
PROTOTYPE WIRELESS ATM SYSTEM FOR
MULTIMEDIA PERSONAL COMMUNICATION; IEEE J.
Selected Areas in Communications, Janvier 1997, pp.83-95.
[7] Tarek Bejaoui1, Vronique Vque, Sami Tabbane; Combined
Fair Packet Scheduling Policy and Multi-Class Adaptive CAC
Scheme for QoS Provisioning in Multimedia Cellular
Networks; International Journal Communication System
2005; Copyright 2005 John Wiy & Sons. Ltd.
[8] Jaime Snchez, Jorge Flores Troncoso, Jos R. Gallardo;
RETRANSMISSION ALGORITHM BASED ON POWER
PRIORITIES FOR WIRELESS NETWORKS; PIMRC 2002;
IEEE2002.
[9] Celal Ceken, Ismail Erturk, and Cuneyt Bayilmis; A New
MAC Protocol Design for WATM Networks; ADVIS 2004,
LNCS 3261, pp. 564574, 2004; Copyright Springer-Verlag
Berlin Heidelberg 2004.
[10] H. KIM, S.K. BISWAS, P. NARASIMHAN, R. SIRACUSA
and C. JOHNSTON; Design and Implementation of a QoS
Oriented Data-Link Control Protocol for CBR Traffic in
Wireless ATM Networks; Wireless Networks 7, 531540,
2001; Copyright Kluwer Academic Publishers 2001.
Manufactured in The Netherlands.
[11] Rolf Sigle, Thomas Renger; Fair Queueing Wireless ATM
MAC Protocols; Computer Networks 31(9-10): 985-997
(1999)
[12] Wojciech Burakowski, Halina Tarasiuk, Andrzej Beben and
Marek Dabrowski; EuQoS IST project: Overview of the QoS
framework for EuQoS; IST-1999
[13] Peter Sholander, Luis Martinez, Lawrence Tolendino, Bruce
A. Mah; The Effects of User Mobility on Usage Parameter
Control (UPC) in Wireless ATM Systems;
[14] Michael Wolf, Rolf Sigle; Medium Access in Wireless ATM
Systems for Industrial Applications: Requirements and
Solutions, Proceedings of the 1998 IEEE International
Performance, Computing, and Communications Conference,
p18-22, ISBN: 0-7803-5794-9.
[15] jongho bang, sirin tekinay, nirwan ansari; A novel capacity
maximisation scheme for multimedia Wireless ATM systems
providing QoS guarantees for handoffs; VTC2000, copyright
IEEE2000.

Author biography
Jamila Bhar- received her Engineering diploma in Electric and her
DEA in Communication system from the National School of
Engineering of Tunis (ENIT), Tunisia in 2001 and 2002,
respectively. Currently, she is a PhD student. Her research interests
include protocol adaptation in heterogeneous networks, traffic
management and Quality of Service for high speed networks. Her
recent work has been in traffic control in WATM network.
Ridha Ouni- received his doctoral degree in physic (2002) from
the Science Faculty of Monastir, Tunisia He is currently an
assistant Professor at the Preparatory Institute of Engineering Study
of Monastir (IPEIM), Tunisia. His research interests include
computer networks, flow and congestion control, interoperability
and performance evaluation. He is interested in many areas of
hardware/software protocol verification and design for distributed
systems.
Salem Nasri- received his Doctoral degree in automatic control
and computer engineering from the National Institute of Applied
Sciences of Toulouse France, in June 1985. His research interests
are in the fields of computer networks, communication systems and
multimedia applications. In May 2001 he obtained the diploma of
Habilitation universitaire. Since then he has been a professor. He
developed collaboration with many laboratories in France such as
LSR (Grenoble), CRAN (Nancy), and some other laboratories in
Tunisia. Currently he is a professor in the Computer Science
Department, Qassim University in the Kingdom of Saudi Arabia.

1200
Performance Evaluation of Fu::y Controller for
Traffic Stabili:ation

Jamila BHAR, Mongi ZAIDI, Ridha OUNI and Salem NASRI
Faculty oI Sciences oI Monastir (FSM), Tunisia
jamilabharyahoo.Ir

$EVWUDFW- This paper proposes a useful algorithm approach
to hybrid control systems combining fuzzy logic and predictable
control techniques for network infrastructure. The given model
employs two rules to Fuzzy logic rate prevention based on queue
length and buffer frequency. The learning rule for this network
must then be modified accordingly. The implementation of the
fuzzy model was carried out in simulation phase. The simulation
was tested with several scenarios. Obtained results proved
favourable comparing to ideal models.
I. INTRODUCTION
Fuzzy theory has been used successIully in many applications
area |3, 5, 7|. This study shows that it can be used to improve
traIIic perIormance Ior diIIerent network architectures. There
are many advantages oI Iuzzy control, one oI which is the
ability to handle imprecise data. Variable traIIics are known
with their perIormance distortion in bad network conditions.
By combining some Iuzzy techniques and classical traIIic
control algorithms, a more eIIicient network results can be
achieved.
In this paper, a hybrid method oI rate controller and Iuzzy
logic has been proposed and applied. In Iact, usually allowed
source rate is estimated Iunction oI network condition
collected Irom destination messages. The consideration oI
calculated Iuzzy parameter can give better perIormances.
Simulation results show that this idea is useIul to stabilize
traIIic conditions and avoid Iluctuation oI traIIic parameters.
The paper is structured as Iollows. In section II, the basic
concepts oI the Iuzzy theory are described. Section III gives
the structure oI the proposed Iuzzy system. Several details oI
our proposed control model are also discussed. Finally, we
provide simulation results and conclusions.
II. MOTIVATION
Actually, networks are designed to support high-bandwidth
and to give transIer Ilexibility oI multimedia traIIic with
varying bit rates and Quality oI Service (QoS) requirements.
Transporting services, with widely diIIerent characteristics
and optimal utilization oI the network resources, requires
complex, nonlinear and distributed control structures. In order
to achieve its potential, these networks will need to
accommodate several interacting control mechanisms, such as
admission control, Ilow and congestion control, input rate
regulation, routing, bandwidth allocation, queue scheduling,
and buIIer management.
In consequence, designing eIIective control strategies may be
diIIicult Ior dynamic traIIic parameters particularly in wireless
networks. Additionally, rate must be able to be controlled
eIIectively even in the presence oI a large round trip time
(RTT). Consequently, some researchers are looking at Iuzzy
systems to be applied to control system design.
In Iact, Researches demonstrate that the introducing oI Iuzzy
logic in the control systems helps to achieve good dynamical
response, disturbance rejection and low sensibility to
parameter variations and external inIluences. Hence, Fuzzy
controller is considered as an important control approach
because oI the variety oI advantages that it oIIers over the
classical control techniques. The objective oI the use oI Fuzzy
logic is to consider best attributes to provide better response
than iI we utilize classic controller.
All Iuzzy evaluations are based on the rules in the symbolic
representation |1|. Linguistic variables, membership Iunctions
and Iuzzy operations are reIerenced by their respective
symbolic names. Symbolic representation oI the rules allows
easy understanding oI the rules and simpliIies the debugging
process.
III. FUZZY CONTROL SYSTEMS
Recently Iuzzy logic as a novel robust control design method
has shown promising results. The emphasis in Iuzzy control
design is around uncertainties in the system parameters and
operating conditions.
To control signals, we design data base and rule base. Then
logic decision is made. Finally, deIuzziIication phase is
applied. In Iact, Fuzzy logic controller is based on three well
known blocs: FuzzyIication bloc, inIerence bloc and
deIuzzyIication bloc. The IuzzyIication stage transIorms crisp
values Irom a process into Iuzzy sets. FuzziIication also means
the assigning oI linguistic value, deIined by relative small
number oI membership Iunctions to variable. InIerence
Iunction expresses relations between the input Iuzzy sets oI
linguistic description rules and the output Iuzzy set. We
mention that as well as inputs conditions change, action
2008 International Conference on Signals, Circuits and Systems
978-1-4244-2628-7/08/$25.00 2008 IEEE -1-
recommended Ior the outputs must be observed.
However, the Fuzzy rule bases are deducted Irom the
command strategy. They are Iormulated by a sentence as:
II condition Then decision. There are several methods oI
inIerence in the literature: Max-min, max-prod or sum-prod
rules. One can gather in only one rule. Several linguistic
variables introduce OR and AND operators. The
deIuzzyIication stage transIorms the Iuzzy sets in the output
space into crisp control signals.
In order to get control schemes that would be less sensitive to
parameter variations than traditional controllers, we consider
the hybrid controller structure shown in Fig 1. We don`t
consider here a switching between the two techniques, but its
combination. The eIIiciency oI this approach is good Ior a
traIIic characterized by Iast changing conditions and
Iluctuations cases.
IV. DESIGN OF PROPOSED FUZZY CONTROLLER

This paper proposes a robust Iuzzy controller Ior output
regulating traIIic rate. Based on the introduced modelling
scheme, the main idea consists oI combining Iuzzy regulator
and measured rate. Fig 2 shows how to obtain output rate. By
supervision oI FIFO occupancy and its Iilling Irequency,
Fuzzy logic gives system inputs. Each linguistic variable has
an arbitrary number oI associated membership Iunctions: each
membership Iunction is represented by a triangle. Triangle
membership Iunctions allow simple implementation oI the
most widely used Iorms: rectangle, triangle and trapezoid.
Output membership Iunction was expressed in the Iuzzy
singletons.

Hybrid
control
Rate
controller
Fuzzy logic
controller
_

Figure 1. Hybrid controller

Membreship
functions
Fuzzy rules
Rate correction
Outputs
Inputs
Filling frequency
FIFO length

Fuzzyfication

Defuzzyfication

Inference

Figure 2. Proposed Fuzzy controller scheme

TABLE I. LINGUISTIC RULES
F
Q
IDVW
UHGXFWLRQ
VORZ
UHGXFWLRQ
QXOO
VORZ
LQFUHDVH
IDVW
LQFUHDVH
HPSW\
very
strong
strong strong middle middle
PLGGOH strong strong middle low low
IXOO middle low very low very low very low

Fuzzy linguistic terms oI input and output variables are shown
in Iig 3, 4 and 5 respectively. The values oI Iuzzy membership
Iunctions parameters were assigned and reIined through
preliminary computer simulation. Fuzzy rules participating in
this control action are reported in Table 1.
In order to normalize linguistic variables, we proceed in the
Iollowing way. II the variation Iield oI the entry variable 'y is
|a, b|, it can be converted to |-1, 1| using the Iollowing linear
relation:
a b
a b y
J

=
2

The triangular and trapezoidal membership Iunctions are,
generally, selected with the deIined variation Iield oI each
system entry (2N1). N can be Iixed accordingly.
In this work, we consider as preliminary data the FIFO
average variation and the Iilling Irequency. These input
parameters will be used as linguistic variables. Their variation
states are explained in the Iollowing Iigures:
FIFO: (empty, middle, Iull)
q
2 1 0 -1 -2
full
empty
middle
(q)

Figure 3. FIFO linguistic variable
Filling Irequency: ((a)Iast reduction, (b)slow reduction,
(c)null, (d)slow increase, (e)Iast increase).

f
-1 -1,5 -1,5 1 1,5 0,5 0
(f)
(d) (c) (b)
(a) (e)

Figure 4. Filling Irequency linguistic variable
In the output, the rate can oscillate according to Iive variation
levels as: (very low (1), low (2), middle (3), strong (4), very
-2-
strong (5)). Fig 5 illustrates the variation oI the output
linguistics variables.

(3) (2) (1) (5) (4)

Figure 5. Output linguistic variable
The proposed Iuzzy algorithm employs three units. The Iirst
stage is updated aIter every interval T. T is called here
command slot. It represents the time interval between
consecutive Ieedbacks. It should be properly chosen according
to considered parameters. T can be assumed with simulations
experiments. It indicates the beginning oI two parallel
processes. In Iact, during the Iirst stage, Iuzzy algorithm
calculates linguistic variables considered in this work: FIFO
average length and FIFO Iilling Irequency. AIter normalizing
these variables, it deIines their associated membership
Iunction (j
k
(Q
n
), j
k
(F
n
)).

Wait for Command slot
Calculate FIFO average
length (Q)
Calculate FIFO filling
frequency (F)
normalize 'F`
min max
max min
n
2
F F
F F F
F

=

normalize 'Q`
min max
max min
n
2
Q Q
Q Q Q
Q

=

For 1k5
Calculate k(Fn)
For 1k3
Calculate k(Qn)
For 1j3
Calculate the weight of each
linguistic rule by using
min_max method

Calculate Z by using centroid
method
=
=
=
5
1
5
1
f
k
f
K k
P
Z P
Z
Correct Rate

Figure 6. Fuzzy control algorithm
A second stage is based on the maxmin Method to apply
inIerence Iunctions. It is based on carrying out the operator
AND by minimum law. The operator OR is carried out by
maximum law. In consequence, we obtain and store values oI
the output variable to be sent to the deIuzziIication stage.
DeIuzziIication process is used to compute the Iinal Iuzzy
output Fig 6 shows some details oI the developed Iuzzy
algorithm.
V. SIMULATION RESULTS

In this paper, Iuzzy logic is applied with ABR (Available Bit
Rate) service in wireless ATM (WATM) network. The ABR
service developed Ior ATM network need to be ameliorated
Ior use in wireless ATM environment |10|. Indeed, there are
supplementary constraints such as the diIIiculty to predict the
traIIic characteristics Ior mobile terminals, bandwidth loss
caused by the supplementary Iields in the header, interIerence
phenomenon and important bit error rate on a wireless
channel. For these reasons and to provide uniIormity oI end-
to-end quality oI service (QoS) guarantees by wireless ATM,
the presented work Iocus on the development oI a robust
adaptive congestion control algorithm in Wireless ATM
networks.
In Iact, ACR (Allowed Cell Rate) is determined by a
combination oI classical ABR estimation and Iuzzy output
parameter. The proposed hybrid Iuzzy logic controller is
described using VHDL language. SpeciIic developed packages
let possible conversion oI real variables into integer or Iloat to
be inserted in signalling messages.

Fig 7 shows parameters obtained Ior a traIIic scenario. Q and
F represent measured linguistic variables. The generation oI
Iuzzy rules implies in total 15 combinations (see table 1). The
application oI maximum operator gives 5 weights (p1, p2, p3,
p4 and p5). Finally, deIuzziIication stage calculates rate
corrector based on centroid method.

Figure 7. Simulated Fuzzy parameters
-3-
To evaluate the developed Iuzzy control algorithm, we
consider diverse traIIic conditions Ior wireless ATM nodes.
The source terminal is inIormed to moderate its rate according
to traIIic Ilow and congestion control algorithm. Fig 8 shows
simulation results oI the Iuzzy hybrid controller. It
demonstrates that the developed algorithm guarantees an
automatic control strategy that stabilizes traIIic conditions and
avoids Iluctuation oI traIIic parameters. Comparing to ACR
variation estimated by classical ABR service (Iig 9), the
proposed Iuzzy algorithm gives more eIIicient results. In Iact,
the Iuzzy hybrid controller attempts to stabilize the ACR rate
regarding to ER (Explicit Rate) in Iluctuated traIIic conditions.

0
10000
20000
30000
40000
50000
60000
70000
5
5
5
1
1
0
7
1
2
2
0
8
7
3
3
1
0
3
4
4
1
1
9
5
5
1
3
5
6
6
1
5
1
7
7
1
6
7
8
8
1
8
3
9
9
1
9
9
1
E
+
0
6
1
E
+
0
6
1
E
+
0
6
1
E
+
0
6
Time (ns)
R
a
t
e

(
C
e
I
I
s
/
s
)
ER
ACR

Figure 8. ACR and ER variation Ior Fuzzy hybrid controller
0
10000
20000
30000
40000
50000
60000
70000
5
5
5
1
1
9
8
9
2
3
9
2
3
3
5
8
5
7
4
7
7
9
1
5
9
7
2
5
7
1
6
5
9
8
3
5
9
3
9
5
5
2
7
1
E
+
0
6
1
E
+
0
6
1
E
+
0
6
1
E
+
0
6
2
E
+
0
6
Time (ns)
R
a
t
e

(
C
e
I
I
s
/
s
)
ER
ACR

Figure 9. ACR and ER variation Ior classical ABR service
0
10000
20000
30000
40000
50000
60000
70000
5
5
5
1
8
6
1
9
3
7
1
8
3
5
5
7
4
7
7
4
3
1
1
9
2
8
7
5
1
E
+
0
6
1
E
+
0
6
1
E
+
0
6
2
E
+
0
6
2
E
+
0
6
2
E
+
0
6
2
E
+
0
6
2
E
+
0
6
3
E
+
0
6
Time (ns)
R
a
t
e

(
C
e
I
I
s
/
s
)
ER
ACR

Figure 10. ACR and ER variation Ior VS/VD ABR service
0
10000
20000
30000
40000
50000
60000
70000
2
0
4
0
4
2
1
8
3
2
2
3
2
6
0
2
4
6
8
8
2
6
1
1
6
2
7
5
4
4
2
8
9
7
2
3
0
4
0
0
3
1
8
2
8
3
3
2
5
6
3
4
6
8
4
3
6
1
1
2
Time (ns)
R
a
t
e

(
C
e
I
I
s
/
s
)
ACR(Classic)
ACR(VSVD)
ACR(Fuzzy)
ER

Figure 11. Behavior oI Rate vs time
Results oI proposed Iuzzy algorithm are also compared with
VS/VD (Virtual Source/Virtual Destination) concept. |10|
shows that VS/VD model stabilises source rate as consequence
oI control loop reduction. Fig 10 illustres the eIIect oI ABR
loop rapidity on ACR variation. However, the rapidity oI the
reaction does not eliminate a rate Iluctuation.
Fig 11 explains the ACR variation Ior the three cases
(Classical ABR service, VS/VD concept and hybrid Iuzzy
controller). This Iigure demonstrates that pics oI rate are
disparate and Iluctuations still in moderate interval Ior ACR
variation using Fuzzy controller.
VI. CONCLUSION
The main objective oI this work is to develop a hybrid
network controller that is able to adjust traIIic rate. It is based
on its own parameters and even on rate value given by
classical traIIic controller. This hybrid system will, according
to the network architecture in which it works, satisIy traIIic
stabilization. In Iact, the developed control algorithm is robust
because it can be adapted to diIIerent network architecture.
Simulation results have demonstrated that the hybrid Iuzzy
controller gives satisIactory perIormance.
REFERENCES
|1| Shih-Hsu HUANG and Jiang-Yuan LAI; A High Speed Fuzzy InIerence
Processor with Dynamic Analysis and Scheduling Capabilities; IEICE
Trans. InI. And Syst., VOL E88-D, NO 10, p2410-2416, october 2005
|2| S. A. Taher, and A. Shemshadi; Design oI Robust Fuzzy Logic Power
System Stabilizer; XXI. International ConIerence on Computer,
Electrical, and Systems Science, and Engineering; CESSE 2007 2007
WASET.ORG
|3| Tung-Sheng CHIANG, Chian-Song CHIU, ana Peter LIU; Robust
Fuzzy Integral Regulator Design Ior a Class oI AIIine Nonlinear
Systems; IEICE Trans. Fundamentals, VOL E89A, NO 4, APRIL 2006
|4| Keun-Chang KWAK, ana Dong-Hwa KIM, Adaptive Neuro-Fuzzy
Networks with the Aid oI Fuzzy Granulation; IEICE Trans. InI. & Syst.,
VOL E88D, NO 9, SEPTEMBER 2005
|5| M. Zerikat, and S. Chekroun;Design and Implementation oI a Hybrid
Fuzzy Controller Ior a High-PerIormance Induction Motor;
Internationnal Journal oI Applied Mathematics and Computer Sciences,
VOL 4, NO 2, 2007 ISSN 1305-5313
|6| Lachlan L. H. ANDREW and KUAN Su-Hsien, Improving the
robustness oI FUZZY logic ATM ABR rate control Ior large round trip
times; Internationnal ConIerence on Intelligent Signal Processing and
Communications, Melbourne, Australia, Nov 1998
-4-
|7| Farag Azzedin, Ahmad Ridha, and Ali Rizvi; Fuzzy Trust Ior Peer-to-
Peer Based Systems; Proceedings oI World Academy oI Science,
Engineering and Technology, VOL 21, MAY 2007 ISSN 1307-6884
|8| SiegIried Gottwald; Matimatical Fuzzy Control. A Survey oI Some
Recent Results; Logic Journal oI the IGPL, VOL 13, NO 5, p525-541,
2005
|9| Ajith Abrahama, Ravi Jainb, Johnson Thomasc, Sang Yong Hana; D-
SCIDS: Distributed soIt computing intrusion detection system; Journal
oI Network and Computer Applications, VOL 30, p8198, 2007.
|10| Jamila BHAR, Ridha OUNI, Abdelhamid HELALI, Salem NASRI;
Improvements oI the ABR loop perIormances in a wireless ATM
network; proceeding oI International ConIerence on Microelectronics
(ICM), VOL 6, NO 8, p295-298, Dec. 2004.

-5-
A new solution for micro-mobility management
in 802.11 Wireless LANs using FPGA
Monji ZAIDI, Jamila BHAR, Ridha OUNI and Rached TOURKI
Electronic and Micro-(OHFWURQLFODERUDWRU\(()
Faculty of Sciences of Monastir (FSM), Tunisia
elmongi_zaidi@yahoo.fr
Abstract Seamless connectivity in wireless access networks
is critical for time sensitive applications requiring Quality-of-
Service guarantees with bounded data transmission delay.
Currently, the latency inherent in the handoff process can
preclude the successful delivery of such applications by
introducing delays up to several seconds. As the mobile clients
are moving from one access point to another, the convectional
layer-2 handoff consumes more time in the channel-scanning
process. We present a novel WLAN handoff architecture using a
hardware method. This new handoff procedure reduces the
discovery phase according two models extended from the basis
model. There are two general methods to implement MAC
functions in normal case. The first method is CPU-based. It uses
software for protocol analysis and CPU, such as DSP, for process
management. It is more flexible in design stage and easy to
modify, but its disadvantages are processing speed too low and
higher cost. The other method means that all functions are
processed by hardware circuits. The advantage of this method is
circuit reconfiguration and processing speed very high, but it
needs long developed time. We purpose the last method to
implement handoff functions in the MAC layer.
I. INTRODUCTION
IEEE 802.11 [1] based wireless LANs have seen a very fast
growth in the last few years. Voice over IP [2] (VoIP) is one of
the most promising services to be used in mobile devices over
wireless networks. One of the main problems in VoIP
communication is the handoff latency [3] introduced when
moving from one Access Point (AP) to another. Then, the
amount of time needed for the handoff [4] in the 802.11
environment is too large for seamless VoIP communications.
We were able to reduce the handoff latency using extended
handoff procedure, with modifications being limited to mobile
devices and compatible with standard 802.11 behaviors.
In this work, we address the problem of handoff latency at the
MAC layer. As a first solution, we propose to alleviate the
scan delay since it takes the major part of the handoff latency;
the mobile nodes can limit the number of scanned channels
and take informed decision about the most appropriate access
point to be associated with.
The second solution consists of selecting/predicting [5], the
future access point before handoff setup. Then, the AP is
selected in an advanced step in order to allocate resources
needs useful for the next MT hop and to minimize handoff
latency. The proposed mechanisms are compared with
standard process for the conventional layer-2 handoff process.
We show that handoff latency can be significantly reduced.
This paper is organized as follow. Section 2 presents a
wireless communication environment based on IEEE 802.11
standard. In section 3, we describe the existing layer-2 handoff
process in WLAN and related works. In section 4, we detail
handoff mechanisms proposed for campus wide networks. The
simulation, analysis and synthesis are dealt in Section 5.
Section 6 concludes the paper.
II. IEEE 802.11 STANDARDS
There are currently three IEEE 802.11 standards [6]: 802.11
a, b and g. The 802.11a standard operates in the 5 GHz ISM
band. It uses a total of 32 channels of which only 8 do not
overlap. Both 802.11b and 802.11g standards operate in the 2.4
GHz ISM band and use 11 among the 14 possible channels.
While 802.11b can operate up to a maximum rate of 11
Mbit/sec, the 802.11g and 802.11a standards can operate up to
a maximum rate of 54 Mbits/sec. The 802.11g standard is
backwards-compatible with the 802.11b standard while the
802.11a standard, because of the different ISM band, is not
compatible with the two other.
A. Wireless LAN architecture
We assume that the reader is familiar with the 802.11 standard.
Thus, we briefly summarize the network functions and services
of the original standard, in order to better clarify the Handoff
Process and the innovative contributions. We outline this
architecture as described in [7].
ESS
BSS BSS
Distribution Network
AP AP
MT MT
Figure 1. IEEE 802.11 architecture.
978-1-4244-2628-7/08/$25.00 2008 IEEE -1-
A traditional WLAN architecture can be considered as a type of
cellular architecture, where each cell is called Basic Service Set
(BSS) and is controlled by a base station called Access Point
(AP). Each network has a name called Service Set Identifier
(SSID) which is advertised by the AP in special control
messages called beacons. When two or more APs using the
same SSID are connected via a broadcast layer 2 network,
called Distribution System (DS), an Extended Service Set
(ESS) is created (fig 1).
The standard also defines a set of networking services, which
are categorized into station services and distribution services.
Specifically, station services are the Authentication,
Deauthentication, Confidentiality, and Mac Service Data unit
(MSDU) Delivery services, while the distribution services
include Association to the access point, Disassociation,
Reassociation, Distribution in the whole ESS, Integration
towards non-802.11 networks. Additional MAC services are
defined in the standard and in some extensions (e.g. 802.1le,
802.11i, 802.1lf) for optimizing and protecting the use of the
wireless resources through rate adaptation, quality of service
differentiation, encryption/decryption, and so on.
The IEEE 802 family consists of a series of specifications for
local area network. IEEE 802.11 specification focuses on the
two lowest layers of the OSI model because they incorporate
both physical and data link component and the data link layer is
partitioned into the logical link control (LLC) and the media
access control (MAC). All 802.11 networks have both a MAC
and a physical component. The PHY layer consists of the radio
and the radios shared channel. The MAC layer maintains
communications among 802.11 stations by managing the
operation of the PHY and by utilizing protocols that support
and enhance communications over the radio medium
B. IEEE 802.11 Management Frames
The IEEE 802.11 management frames enable stations to
establish and maintain communications. The following are
common IEEE 802.11 management frame subtypes, with the
description quoted from [8].
- Probe request: A Mobile terminal (MT) sends a probe
request frame when it needs to obtain information from another
station. For example, a MT would send a probe request to
determine which access points are within range.
- Probe response: A MT will respond with a probe response
frame, containing capability information, supported data rates,
etc., after it receives a probe request frame.
- Authentication request: The 802.11 authentication is a
process whereby the access point (AP) either accepts or rejects
the identity of a MT. The MT begins the process by sending an
authentication frame containing its identity to the access point.
With open system authentication (the default), the MT sends
only one authentication frame, and the access point responds
with an authentication frame as a response indicating
acceptance (or rejection).
- Reassociation request: If a MT roams away from the
currently associated access point and finds another access point
having a stronger beacon signal, the STA will send a
reassociation frame to the new access point. The new access
point then coordinates the forwarding of data frames that may
still be in the buffer of the previous access point waiting for
transmission to the MT.
- Reassociation response: An AP sends a reassociation
response frame containing an acceptance or rejection notice to
the MT STA requesting reassociation. Similar to the
association process, the frame includes information regarding
the association, such as association ID and supported data rates.
III. LAYER-2 HANDOFF PROCESS AND RELATED WORKS
In WLAN, a handoff can be defined as the process of leaving
the basic service set of an access point to enter a new one. A
handoff is triggered by a degradation of the signal quality
which falls below a predefined threshold. The handoff can be
the result of either excessive noise, interference or user
mobility (decrease of the signal intensity due to the increasing
distance to the associated access point).
At the MAC layer, the handoff process as defined in the IEEE
802.11 Standard [9] can be decomposed into three phases:
scanning, authentication and reassociation (Fig 2).
1 Scanning: In order to discover on which channel the
surrounding access points are transmitting, a mobile node
needs to scan all the channels. Two scanning methods are
described in IEEE 802.11. Passive Scanning entails
determining the presence of access points by successively
listening to all the channels and waiting for the reception of
beacon messages identifying the access point. This method,
while offering the advantage of low overhead, presents the
drawback of introducing a significant delay. In order to
alleviate this problem, an active scanning method has been
defined. The mobile node broadcasts a Probe Request on each
channel and waits a minimum period MinChannelTime for
any Probe Response. After the scan of all the channels and the
processing of all the beacon messages or Probe Responses
received (according to the implemented scanning process), the
mobile node can take an informed decision on the most
appropriate access point (with the best channel quality).
2 Authentication: The Authentication process involves
establishing the identity of the mobile node and authorizing its
access to the basic service set of the access point.
3 Reassociation: The Reassociation process consists in
transferring an association between an access point and a
mobile node to another access point. The operations between
the old AP and the new AP are defined by the Inter-Access
Point Protocol [10]. The scanning process has been identified
as the principal source of delay in the handoff mechanism [4]
[11]. Few works have been conducted aiming at reducing the
latency at the MAC Layer [4] [11]. They essentially adopt the
same approach by optimizing the waiting time of the mobile
node during the active scanning process. Indeed, IEEE 802.11
defines two parameters: MinChannelTime, the minimum
waiting period before considering that the channel is idle; and
MaxChannelTime, the maximum waiting period after a Probe
Response has been successfully received. However, no exact
value of these parameters has been explicitly set. The
suggested optimal values deduced from previous experiments
-2-
are approximately 6.5ms for MinChannelTime and 11ms for
MaxChannelTime.
We are interesting to use handoff phases according to models
which offer transparent transition from one AP to another
within a minimum delay. The IEEE 802.11 a/b/g series offer
Wireless connectivity to the users at high rates. An access point
(AP) provides connectivity for the mobile users. The wireless
nature of 802.11 devices allows the user to move freely
between APs within the coverage area, commonly known as
the hotspot. The operation of changing from one AP to another
AP is known as Handoff. The handoff may be within the same
subnet (intra-subnet) or between two different subnets (inter-
subnet). In the intra-subnet Handoff, the IP address of MT
remains the same. Additional processes are required if the new
AP is not connected to the same subnet as the old AP of the
MT. These further operations are known as layer-3 Handoff.
Fig 2 illustrates the probing procedure as described in the IEEE
Standard 802.11. In this figure, N distinct channels are selected
to probe. Once the channels to be probed are determined, the
station switches to each selected channel and broadcasts a
probe request frame. We call this latency the Channel Switch
and Transmission overhead (CS&T). After probing all selected
channels, the next access point is determined from the
information received in the probe responses and their
associated Signal to Noise Ratio (SNR). The following
algorithm details the process described above.
Algorithm: Full-scanning algorithm.
1: For each channel to probe do
2: Broadcast probe request on this channel
3: Start probe timer
4: while True do
5: Read probe responses
6: if Medium is idle until MinChannelTime expires then
7: break
8: end if
9: if MaxChannelTime expires then
10: break
11: end if
12: end while
During the transition phase (authentication and association), a
MT identifies a suitable candidate AP, breaks its association
with the current AP and then reassociates with the targeted AP.
The MT performs the following transition steps (see fig 2)
1. The MT stops data transmission to its current AP.
2. The MT switches its radio to the channel used by the
targeted AP.
3. The MT completes a Reassociation Request / Response
exchange with the targeted AP.
4. If Reassociation succeeds, the MT authenticates and
performs 802.11i key management with the new AP, to secure
its new link.
5. The MT requests that the new AP allocates bandwidth to
maintain the quality of service (QoS) required by its
applications.
6. When these steps are completed, data flow resumes between
MT and the infrastructure, now via the new AP. The
introduction of IEEE 802.11i security and the negotiation of
QoS using 802.11e have increased this transition time from a
few ms to a few seconds.
MT Active AP Old AP
New AP
Probe request
Probe response
Probe response
Probe response
Probe request
Probe response
Probe response
Probe response
...
.
.
.
C
h
a
n
n
e
l

1
C
h
a
n
n
e
l

N
Association request
Exchange
H
a
n
d
o
f
f

L
a
t
e
n
c
y
P
r
o
b
e

d
e
l
a
y
A
u
t
h
e
n
t
i
c
a
t
i
o
n

a
n
d
A
s
s
o
c
i
a
t
i
o
n

d
e
l
a
y
Figure 2. Handoff process in the IEE 802.11 Standard.
Existing handoff mechanism is based on Scanning channels,
authentication and association or reassociation phases. This
model is described with a hardware description language using
a Finite State Machine (FSM). In this model, handoff initializes
when SNR (Signal Noise Rate) drops than a specific threshold.
Fig 3 explains, with more details, the basis handoff model and
its steps latencies.
IV. ARCHITECTURE DESIGN: NOVEL APPROCHES
A. Model with reduced Scan phase
In this work, the objective consists of reducing handoff latency.
As a first solution, we propose to alleviate the scan delay since
it takes the major part of the handoff latency. In fact this
solution consists of transmitting Probe requests on each
channel stops once a Probe response indication is received with
an adequate SNR. An SNR threshold level has been defined to
select AP that provides QoS guarantee. Fig 4 explains the first
approach based on reduced scan phase.
During Scan phase of basis handoff model, the MT must sweep
the total number of channels (N). The time allocated to scan
each channel is called MaxChannelTime. Thus, the time of
Scan is given by the following equation.
) 1 ( * ChannelTim number Channels Scantim
-3-
MaxcannelTim is the time interval separating the Probe
Request and the last Probe Response on each channel. The scan
time can be reduced when minimizing the channel number to
scan. By being unaware of negligible times, the idealized
latency of this active Scan is given by the following relation
) 2 ( * ) ( * )) ( 1 (
1

NumChannel c
c
Max c p Min c p Scantim
Where P(c) is the probability of one or more access points
functioning in the same channel c. whether Min Channel Time
and Max Channel Time values are respectively 1 ms an 11 ms
ideal latency should extend from 11 ms to 110 ms. Based on
(1) it becomes unuseful to sweep all channels while the most
adequate AP belongs to a channel already scanned. In others
terms scanning the rest of channels dont serves to find useful
AP, but it loses time which causes to higher scan latency.
Scan Channel i
AP(Ch i) selection
SNR i >Thershold
The rest of basis
model phases
N
e
x
t

c
h
a
n
n
e
lTransmit probe Request on channel i.
Receive probe Response from channel i.
Yes
NO
Figure 3. Second model: reduced Scan phase.
The implementation of this model in account the specific
threshold once reached, the MT stops the scan process and
follows the rest of the basis handoff phases. Then, it is
obviously that this implementation leads to a fast
communication establishment.
B. Predictive model
The third model, based on predictive approach [5], can be
adopted in two hypotheses. The first one occurs when
supported applications meet temporal constraints or significant
QoS level. Then, the Handoff process should be achieved
without (a) generating communication rupture at the
applicative level and (b) degrading QoS guaranties. The second
situation occurs when mobile stations follow predefined
trajectories. The solution consists of selecting the future access
point before handoff setup. Then, the AP is selected in an
advanced step in order to allocate resources needs useful for the
next MT hop and to minimize handoff latency. This solution is
a probabilistic based approach since we predict AP. However,
it takes in account several arguments and parameters to outlines
decision for the probable AP that can be chosen to maintain an
established connection.
The mobile station achieves its first attachments using the basis
model. For each transition from one cell to another, the MT
records AP addresses which can be used to determinate its
trajectory. The future AP can be predicted using addresses of
the old APs present in the direction MT. As well as the number
of AP meted by the MT becomes higher, so its trajectory is
defined rigously. We propose here that three successive
attachments or AP addresses allow discovering the MT
trajectory. We represent the MT displacement into FSM,
described as following.
connection established
(N<3)
SNR drops
Thmax
Basis Handoff
(N>=3)
SNR drops
Thmin
Prediction
SNR drops
Thmax
Authentication
Association
Yes
Yes
Yes
Figure 4. Predictive model state diagram.
V. IMPLEMENTATION
Our contribution for reducing handoff latency reposes on the
lower layers of the 802.11 networks (MAC and physical layer
[12, 13] is divided into two sub layers: PLCP (Physical Layer
Convergence Procedure) and (Physical Medium Dependent).
We assume that a specific data was done by PLCP.
The PLCP is the bridge between MAC and radio transmission
layers it translates MPDU to PMD frames. The PMD is
responsible for transmitting any bits it receives from the PLCP
to the wireless medium by using antenna. The major efforts of
the handoff processing is almost in the MAC layer, which we
did consider new approaches to be integrated in this level.
In this section, we outline the implementation of the MAC
layer in a mobile terminal. A modular architecture proposal of
the MAC layer, based on receiver and transmitter component,
is described in fig 6. It details interaction between Mac layer
and both LLC and PLCP layers. Various types of handshaking
signals are integrated to manage and control transmission in
both directions. Table 1 outlines the interfacing signals and
briefly indicates their operations.
The receiver component receives PMDUs from PLCP and
decodes them into various types of paquets [14, 15, 16]. First it
identifies the type of frames (signalization, data or control) and
next processes them accordingly. For the handoff initialization
and operation, the main task of the MAC receiver component
consists of processing probe response frames and controls SNR
level.
TABLE I. RECEIVER/TRANSMITTER COMPONENT INTERFACE SIGNALS.
Name Description
Clk Operation clock
Canal_val Sets Channel value to physical layer
Order Used to order the answers in a well defined
order
Phy_start Physical layer notifies receiver start to
receive data
Sig_level Indicates the signal force coming from AP
Prob_resp MT receives the Probe Response frames
coming from APs
Phys_data Data signals from physical layer
Prob_reqst Notifies transmitter send probe request frame
-4-
Ack Notifies transmitter send acknowledge frame
Scan_fini Indicates host that all channels has been
scanned completely
The transmitter component allows (a) events detecting (b)
parameters buffering and (c) message generating. Otherwise, it
maintains a permanent interaction with the receiver component
in order to manage events (SNR, New connection) and
satisfy requirements (service classes, addresses). The
transmitter is mainly responsible first, for handoff initialization
by generating probe requests over different channels. Second it
allows automatising handoff phases according to the specific
approaches described in the section (4).
P
h
y
_
s
t
a
r
t
P
r
o
b
_
r
e
s
p
S
i
g
_
l
e
v
e
l
P
h
y
s
_
d
a
t
a
P
r
o
b
_
r
e
q
s
t
A
c
k
C
a
n
a
l
_
v
a
l
O
r
d
e
r
S
c
a
n
_
f
i
n
i
Physical Layer:PLCP and PMD
MAC
LLC: Logic Link Control
C
l
k
Transmitter Part Receiver Part
1
6
424
1
6
Figure 5. Modular architecture proposal of the MAC layer.
A. Simulation models and results
1) Basis model
Figure 6. Simulation of Scan Phase on the first channel.
A mobile station broadcasts probe request over three channels.
On each channel, it expects to receive responses from three
access points. Response frames as well as their SNR are
buffered and then used to select the AP, which satisfy the
mobile requirements. Fig 7 gives an example of an active scan
timing diagram. It shows also probe response frames
identification, addresses extraction and SNR measurement. The
rest of the handoff process reposes on authentication and
association phases while each one a frame is sent to the
selected AP and a response is received.
2) Handoff with reduced Scan
Fig 8 shows the simulation of a fast handoff processing model.
Optimal handoff latency is improved by reducing the scanning
phase, which finishes once an adequate AP is detected. A
higher detected SNR avoids scanning the other channels and
allows to join with the corresponding AP. Authentication and
association phases are similar than those of the basis model.
This approach remains probabilistic opposite of the adequate
AP order while an other AP may be detected within the rest of
channels. In worst case, this approach improves the basis
model performances
Probe Request Frame(First channel) Probe Response Frames Authentication frame Association frame
Association response Adequate thershold on the first channel Authentication response
handoff finished
Figure 7. Reduced Scan.
3) Predictive model.
The predictive handoff model is similar than the basis model
during the first three transitions. This period serves as training
phase in order to discover the MT trajectory over an historical
setting created with signalization messages between the MT
and the meted AP. This model could predict the future AP
since the fourth handoff execution. The MT authenticates with
this AP without scanning step.
Fig 9 shows the predictive model simulation at the 5the
handoff initialization. At this level, the MT trajectory is
discovered and the future AP is predicted according to the
SNR drops
5th handoff initialisation
Adresses of the three last APs which
accomodated the MT
Old connection New connection
Handoff finished
...
Figure 8. Attachment within predictive model.
-5-
historical transitions. When the SNR drops, the MT
authenticates and joins with this AP without scanning.
This approach improves reducing handoff latency compared to
the two others models as outlined in fig 10.
This study enables to conclude that the handoff latency
depends on scan period or in other terms on the number of AP
which operates under several channels. In fact, minimizing one
among N channel to be scanned allows reducing almost 1/N of
the total handoff latency
The handoff latency measures the time between a probe
requests is sent until an association reply is received. The basis
model requires to seen all channels. It needs 184 clock cycles
Without counting the Channel Switch and Transition overhead
(CST). The second model provides, in general three responses
according to the probe response order of the adequate AP. we
consider hear six scenarios for six handoff execution.
Scenario1, scenario2 and scnario5: we suppose that the
adequate signal is received during Scan of the first channel
(Half-time); In this case, the scan of the third and the second
channel is not necessary.
Scenario3, scenario4 and scnario6: means that the adequate
signal is received on the second channel (Half-time). In this
case, the scan of the third channel is not necessary.
It is clear that there is significant decrease of handoff latency
by proposed models.
1 2 3 4 5 6
0
50
100
150
H
a
n
d
o
f
f

l
a
t
e
n
c
y

(
c
l
o
c
k

c
y
c
l
e
)
Basis model
Model withreduced scan: Proposed
Predictive model: Proposed
Figure 9. Handoff latency for the three models.
B. Synthesis results
In this context, we have exploited FPGA xilinx virtex II pro
environment. This environment allows the implementation of
communications systems programmable circuits. The
advantage of using FPGA circuits is essentially the possibility
of re-scheduling of circuits. For our application, synthesis RTL
is made using the ISE 8.1 of the Xilinx FPGA virtex II pro
environment. We achieved the synthesis of the three,
developed approaches. The results of synthesis are shown in
the following table.
TABLE II. SYNTHESIS OF MODELS.
Number
of Slices
Number
of Flip
Flops
Nb of 4
input
LUTs
Nb of
bonded
IOBs
Frequenc
y MHz
Basis
Handoff
782 604 1381 40 132
Reduced
model
705 608 1313 40 124
Predictive
model
1165 995 816 41 150
VI. CONCLUSION
Supporting user mobility in WLAN remains a challenging
task, especially with the QoS requirements of applications
necessitating bounded transmission delay. The handoff
process is a complex mechanism, involving significant delay,
which can be detrimental to QoS guarantees and the
convectional layer-2 handoff consumes more time in the
channel-scanning process. Based on these observations, this
paper aims at reducing the handoff delay in WLANs using a
hardware method. By maintaining information on the
surrounding access points and by delivering this information
to the mobile nodes upon request, significant improvements
can be achieved. Our results show that when the MT requests
network using reduced scan or predictive approach, it may
establish connection with more performance. The proposed
mechanisms are compared with standard process for the
conventional layer-2 handoff process. The handoff latency is
significantly reduced. This is more suitable for real time
applications
We adopted the high level design for the realization of the
systems. The description was undertaken with the high
description language: VHDL. ModelSim was used to check the
behaviour of the system at the RTL level. Synthesis was
undertaken using the ISE 8.1 of the FPGA environment xilinx
virtex II pro, in order to evaluate the circuit performances.
REFERENCES
[1] M. Gast; 802.11 Wireless Network: The Definitive Guide, Second
Edition, OReilly, 2005
[2] Fayza A. Nada; On using Mobile IP Protocols, Faculty of Computers
and Information; Suez Canal University; Ismailia; Egypt; Journal of
Computer Science 2 (2): 211-217, 2006
[3] H. Velayos and G. Karlsson; Techniques to reduce the IEEE 802.11b
handoff time; In IEEE ICC, vol. 27, no. 1, June 2004, pp. 38443848.
[4] Arunesh Mishra, Minho Shin, William Arbaugh; An empirical analysis
of the IEEE 802.11 MAC layer handoff. ACM SIGCOMM Computer
Communications Review (ACM CCR), vol. 33, no. 2, April, 2003.
[5] Bob OHara, AI Petrick; IEEE 802.11 handbook a designers
companion, second ed., March 2005.
[6] IEEE Std 802.11-1997; Wireless LAN Medium Access control (MAC)
and Physical Layer (PHY) specifications, (1997).
[7] A. Jain. Hando delay for 802.11b wireless LANs; Technical report.
University of Kentucky, 2003.
[8] Y. Kim, H. Jung, H. H. Lee & K. R. Cho; MAC implementation for
IEEE 802.11 wireless LAN, Router Technology, Department,
Electronics & Telecommunications Research Institute, 2001.
[9] IEEE. Wireless LAN medium acces control (MAC) and physical layer
(PHY) specifications, 1999.
-6-
[10] IEEE. IEEE trial-use recommended practice for multi-vendor acces point
interoperability via an inter acces point protocol across distribution
systems supporting IEEE 802.11 operation, 2003.
[11] H. Velayos and karlsson G. Technique to reduce the IEEE 802.11b
handoff time. In swedish National computer networking Workshop,2003
[12] IEEE 802.11, IEEE wireless LAN medium access control (MAC) and
physical layer (PHY) specifications, Aug. 1999
[13] IEEE 802.11b, Wireless LAN medium access control (MAC) and
physical layer (PHY) specifications: High-speed Physical Layer
Extension in the 2.4GHz Band, IEEE Std 802.11b-1999
[14] Y. Kim, H. Jung, H. H. Lee & K. R. Cho, MAC implementation for
IEEE 802.11 wireless LAN, Router Technology, Department,
Electronics & Telecommunications Research Institute, 2001.
[15] XILINX, configurable Locallink CRC Reference Design, Nov, 2004
[16] T. H. Meng; Design and implementation of an all-CMOS 802.11a
wireless LAN chipset, Communication magazine, IEEE, Vol.41, No.8,
Aug. 2003.
SETIT 2004
International Conference: Sciences of Electronic,
Technologies of Information and Telecommunications
March 15-20, 2004 TUNISIA

Interoprabilit ETHERNET/WATM

J amila BHAR; Ridha OUNI; Salem NASRI
Laboratoire dElectronique et de Micro-Electronique (EE),
Facult des Sciences de Monastir, Tunisie
jamilabhar@yahoo.fr

Rsum Parmi les besoins actuels dans le domaine de tlcommunication, Nous distinguons le recourt
limplantation des applications multimdias sur diverses architectures. Dans cet environnement htrogne, une
multitude de normes sinterfre interdisant linterconnexion directe des rseaux. La complexit de linterconnexion
dpend de lincompatibilit protocolaire des rseaux dfinis par la dissymtrie chaque niveau de leurs modles en
couches. Dailleurs, les passerelles sont les systmes les plus complexes permettant linterconnexion des rseaux.
Cet article se dirige dans le cadre de linteroprabilit des rseaux htrognes. La contribution principale de ce
travail est le dveloppement dun mcanisme de conversion dinformations pouvant tre changes lors de
linteroprabilit Ethernet/WATM. Linterface propose se charge de la gestion des procdures lies la
segmentation et au rassemblage des donnes.
La dmarche suivie consiste en premire tape ltude et la spcification de linterconnexion
Ethernet/WATM. La deuxime tape a pour but de dfinir le flot de conception et la description des fonctions
dinterconnexions. Dans la dernire tape, nous prsentons les rsultats acquis et nous mentionnons la possibilit
dentrevoir des nouveaux horizons pour optimiser lapproche propose. Afin dimplmenter les fonctions
lmentaires dinterconnexion, nous avons utilis le langage de description VHDL et les outils de simulation et de
synthse V_system et FPGA_Adv.

Mots Cls: Rseau, Interconnexion, Ethernet, WATM, VHDL.
1 INTRODUCTION
La diversit des rseaux, lorientation vers les
rseaux sans fils et le besoin de lintgration des
applications sur des architectures htrognes ont
favoris la tendance de la conception des systmes
dinterconnexion. Ces derniers permettent de fournir
une large gamme de services multimdias et visent
homogniser lenvironnement de communication.
Les tendances des recherches actuelles dans le
domaine des rseaux sorientent lexploitation de
la diversit des rseaux. Ceci permet de profiter des
avantages offerts par chacun et de dvelopper des
applications bases sur diverses architectures.
Ltude et la contribution linteroprabilit des
rseaux rpandus avec ceux sans fils sont fortement
demandes. Dans ce contexte, linterfonctionnement
Ethernet/WATM donne un exemple concret compte
tenu des caractristiques et des performances de
deux infrastructures.
Lors de linteroprabilit entre les deux rseaux,
de nombreux problmes sont poss par
lhtrognit de lenvironnement de
communication. Les donnes transitant entre les
deux rseaux utilisent des formats incompatibles.
Les dbits et les mcanismes de transmissions ainsi
que les politiques de connexion sont propres
chaque rseau. Ce travail est une contribution la
conception dun systme dinterconnexion
Ethernet/WATM. Nous nous intressons la
spcification de linterface cible ainsi que la
conception dun mcanisme de conversion de format
de linformation.
Dans cet article, nous introduisons
lenvironnement de lapplication. Puis, nous
prsentons les systmes dinterfaage entre rseaux
htrognes. Nous dcrivons, ensuite, le flot de
SETIT2004
conception suivre pour le dveloppement de ces
systmes. Enfin, nous prsentons le mcanisme
dadaptation de format de donnes dvelopp ainsi
que les rsultats de simulations et de synthse
obtenues.
2 ENVIRONNEMENT DE LAPPLICATION
Chaque nouvelle technologie cherche
satisfaire les perspectives dsires par lusager.
Lobjectif du dveloppement des rseaux (haut
dbit) WATM et Ethernet est de fournir une large
gamme de services multimdias. En effet, Les
applications, bases sur les technologies de rseau
sans fil large bande telles que WATM, suivent une
importante croissance. Un grand nombre de
prototypes de WATM a t dvelopp. Plusieurs
protocoles ont t excuts afin de garantir une
meilleure qualit de service et une bande passante
maximale. Linspiration de WATM partir de
linfrastructure ATM permet de bnficier des
avantages de ce dernier, ce qui est trs utile pour des
applications multimdias. Limplmentation de lIP
sur WATM permet de bnficier des avantages de IP
et de WATM en mme temps. Dune part,
l'adressage IP est ais et fortement rpandu. Dautre
part le rseau Ethernet permet linterconnexion dun
nombre lev de systmes informatiques. Il est
devenu le rseau le plus populaire utilis par la quasi
totalit des entreprises utilisant un rseau LAN. Le
rseau Ethernet offre des performances de scurit,
une capacit importante de transmission et une
facilit dutilisation[12].

Ethernet

Serveur
Passerelle
WATM

Figure 1. Interfonctionnement Etherne/WATMt

Linterconnexion de ces rseaux haut dbit
permet le support de plusieurs applications
htrognes avec des performances suffisantes.
Toutefois, linterconnexion de ces deux rseaux pose
des problmes de ralisation. Ces difficults
proviennent des diffrences des architectures de
communication des rseaux communicants. Nous
nous intressons, dans ce travail, lhtrognit
entre Ethernet et WATM en terme de format des
paquets de donnes. Dans les deux paragraphes qui
suivent, nous prsentons les formats de paquets
dinformation de deux rseaux interconnecter.

2.1 Format des cellules WATM
La transmission des cellules WATM se fait
laide du MPDU (MAC Protocol Data Units). Le
MPDU est compos dun ensemble des donnes
utiles prcdes dune entte. Des groupes de
recherche telles que ATM Forum et PRISM ont
propos et essay des prototypes utilisant la structure
de la figure 2 [9][4][7].
Nous nous intressons dans ce travail
maintenir le choix du format de cellule ayant une
taille de 56 octets. Cette structure permet de
conserver la taille 48 octets de la charge utile utilise
dans le rseau ATM. Lajout de champ CRC permet
la protection contre les erreurs qui sont plus
important pour une liaison sans fil. Linsertion de
deux champs Cell_seq_Num et contrle est trs
importante. Le premier champ permet de contrler la
squence des cellules. Il contribue, au niveau de la
couche WDLC, la dtection et correction des
erreurs. Le champ contrle est utilis par la couche
WMAC pour le contrle et la supervision des
fonctions lies la couche daccs au mdia.
Toutefois, la taille de cellule reprsente un
inconvnient en cas de perte de trame de donnes car
le rseau sera oblig de retransmettre une quantit
dinformation plus importante que sil utilise des
pico cellules. Pour assurer un transfert fiable des
informations WATM, la couche dadaptation AAL5
joue un rle trs important. Elle traite des paquets de
donnes de tailles variables et multiples de 48 octets
et assure la gnration des cellules WATM [2].
La figure 2 montre le format des cellules WATM.

Figure 2. Format dune cellule WATM

2.2 Format des trames Ethernet
Chaque message, mis par une station source, est
diffus sur tout le rseau. Seule la station destinatrice
reoit, stocke en mmoire puis traite le message. Ce
dernier est dcoup en trames Ethernet de taille
variable de 46 1500 octets de donnes utiles [1].
Chaque trame comporte aussi deux champs
dadresses de destination et de source, appeles
encore adresses physiques ou MAC. Plusieurs
Entte
watm
2 octets
Entte
ATM
4 octets
Donnes utiles
48 octets
Crc
watm
2 octets
Type
GFC
Cell_seq_N
Control

VCI
PLT
CLP

HEC
SETIT2004
protocoles sont vhiculs sur Ethernet. Le champ
type permet didentifier le protocole au niveau des
couches suprieures et par consquent le format des
donnes. La queue de la trame Ethernet est un
champ CRC de 4 octets permettant le contrle
derreurs de transmission. Ethernet supporte
plusieurs protocoles sur ses couches suprieures : IP
sur sa couche rseau et UDP, XTP et TCP sur sa
couche transport. TCP/IP sadapte convenablement
avec Ethernet permettant une grande capacit de
transmission et un contrle de trafic et de
congestion.
La structure de la trame Ethernet est reprsent par
la figure 3.

Figure 3. Format de la trame Ethernet

Pour ladaptation des formats des donnes
changer, des procds de segmentation et de
rassemblage sont ncessaires. Ces mcanismes de
conversion de format des donnes sont bass sur des
fonctionnalits assurant dans un sens de trafic la
transformation des donnes Ethernet en un ensemble
de cellules WATM. Dans lautre sens, lopration
inverse est ralise pour convertir des cellules
WATM sous forme dune trame Ethernet.
3 SPECIFICATION DE LINTERCONNEXION
La diversit des rseaux en terme darchitecture
et de protocoles de communication ncessite
lutilisation des systmes dinterconnexion pour
permettre leurs interoprabilits. Ces systmes
appels passerelles, sont charges de la conversion
de protocoles, de contrle de flux et de
lacheminement des paquets. La complexit de ces
fonctionnalits dpendent de la dissymtrie des
protocoles mis en jeu chaque niveau du modle en
couche des deux rseaux.
Le systme dinterconnexion utilis est constitu
par un ensemble de mcanismes chargs des
oprations dmission et de rception des donnes.
La conversion de format de trame ncessite
lintgration dun ensemble de mcanismes faisant
intervenir des mmoires et des fonctions de
segmentation, de rassemblage et deux modules
dcriture et de lecture des donnes. La figure 4
montre une adquation architecture-algorithme pour
le mcanisme dadaptation. La conception du
mcanisme de conversion, devra respecter certaines
contraintes telles que les dlais de transit et la taille
de mmoire mis en jeu. Les mmoires sont utilises
pour stocker temporairement des informations reues
lors de la transaction des donnes entre les deux
rseaux. Les units de donnes sont mises la
disposition de ce systme pour la segmentation ou le
rassemblage avant dtre mises.

Figure 4. Architecture modulaire du mcanisme de
conversion

La mmoire reprsente un lment essentiel
dans la conception dune passerelle
dinterconnexion. Elle joue le rle dintermdiaire
entre les deux rseaux interconnecter et participe
labsorption de lincompatibilit en terme de format
de donnes et de dbit. La DPRAM 128x32 utilise
est double ports. Cette caractristique offre la
possibilit du fonctionnement en parallle des tats
dfinies par le mcanisme. Les donnes peuvent tre
transfres sur des ports de 32 bits laide des
signaux DO (Data Out) et DI (Data In). Cette taille
de port permet daugmenter le taux de transfert de
donnes. Deux signaux internes de statu sont
dfinies pour indiquer les adresses de dbut et de fin
de stockage. Lemploi de ces signaux facilite la
gestion du mcanisme de rassemblage. Chaque
signal est dfini par 14 bits (figure 5) faisant
intervenir les adresses dbut et fin de la plage
mmoire utilise.

Adresse dbut
14 bits
Adresse fin
14 bits

Figure 5. Format du registre Flag.

Les diffrents procds intgrs dans ce systme
sont souvent complexes. Dans le cas gnral, ils sont
trs typiques chaque application dune organisation
protocolaire.

Entte
Ethernet
22 octets
Donnes
De 46 1500 octets
CRC
4 octets
Pram-
bule
7 octets
Dlimit
-eur
1 octets
Adr_D
6 octets
Adr_S
6 octets
Type
2 octets
Trame
Ethernet
Cellules
WATM
Fin
Fin
Lect-
ure
Segm-
entation
Mmoires
Lect-
ure
Rass-
emblage
SETIT2004
4 ALGORITHME DADAPTATION

Lincompatibilit de point de vue de format de
paquet exige au systme dinterconnexion de
transmettre les donnes au destinataire sous format
comprhensible. Ceci ncessite la conversion des
donnes. Le mcanisme de conversion peut tre
dcrit selon trois tats principaux : repos, rception
(lecture/criture) et conversion. Dans ltat de repos
linterface reste en tat dattente et de test en
permanence de lactivation dune demande de lun
des deux rseaux. Ltat de lecture/criture consiste
recevoir ou transmettre des donnes partir ou
vers un autre rseau. Dans ltat de conversion, le
mcanisme dadaptation permet les modifications
ncessaires pour transmettre des donnes dans le
format adquat du rseau destinataire.
Lalgorithme dvelopp est organis en un
ensemble de processus charg chacun deffectuer
une tche bien dfinie. Ces processus
(squentiels/concurrents) peuvent communiquer
travers des paramtres de sensibilits.

Figure 6. Algorithme dadaptation de
format de linformation

La figure 6 montre lalgorithme de conversion
de format faisant intervenir plusieurs MEFs. En
effet, les trames de donnes mises partir du rseau
Emetteur sont reues, puis stockes temporairement
dans la mmoire. A la dtection de la fin du message
indiqu par le EOP (end Of Packet) la passerelle
initialise une demande dmission de donnes vers le
rseau Destinataire. Ds quelle reoit une
confirmation, elle active un signal pour le dbut de
segmentation ou de rassemblage de linformation
stocke. Cette information est prsente dans la plage
mmoire limite par deux adresses affectes chacune
sur 14 bits du registre flag (figure 5). Pour indiquer
le dbut ou la fin de lmission ou de la rception
nous avons utilis les signaux courte dure de
transmission RTS et CTS. Il est noter que grce au
fait que le RTS et le CTS sont des trames courtes, le
nombre de collisions est rduit, puisque ces trames
sont reconnues plus rapidement par le destinataire.
Pour permettre une meilleure gestion de la
communication et maintenir les performances de
chaque rseau, une mthodologie de conception est
ncessaire. les concepteurs proposent dutiliser
diverses approches mthodiques pour automatiser le
flot de dveloppement.
Nous prsentons dans le paragraphe suivant
lenvironnement de conception adopt pour ce
travail.
5 ENVIRONNEMENT DE CONCEPTION
5.1 Flot typique de conception
Actuellement, la plupart des travaux se ralisent
sur des niveaux permettant de rduire la complexit
et le temps de conception. Ceci est d au grand
dveloppement dans le domaine de la
microlectronique, de linformatique et des
tlcommunications. Cette volution est
accompagne dune apparition de nouvelles
techniques de conception de plus en plus
performantes. Ces techniques interviennent le long
du chemin des donnes depuis le rseau, travers
des entits de traitement. Ce type dapproche de
conception seffectue selon une mthode
hirarchique descendante, appele aussi conception
top-down. Lapproche descendante part du systme
en circuits puis sous circuits et volue ainsi jusquau
schma compos de transistors en passant par
diffrents niveaux dabstraction. Labstraction dun
composant est une description succincte qui
supprime les dtails dimplantation [6] [8][10].
Un flot de la mthode descendante est complt
par lajout des tapes de validation. Le circuit
modlis peut tre valid par mode de simulation
tous les niveaux de description. Cette opration est
possible partir dun fichier de test nomm test-
bench dans le cas de lutilisation du langage VHDL.
En effet, nous pouvons dfinir le flot de conception
partir de trois phases principales : il sagit de la
spcification, de la vrification et de la synthse.
Dabord, la spcification du systme dans son
environnement traduit, laide dun langage de
description matriel ou HDL (Hardware Description
Language), sa description un niveau dabstraction
bien dtermin. La simulation du systme permet de
vrifier son comportement. Enfin, la synthse permet
le passage dun niveau dabstraction un niveau
plus bas, jusquau niveau "portes logiques" avant
lintgration sur FPGA ou ASIC [3].

5.2 Langage de description
La slection dun langage se rsume
gnralement un compromis entre plusieurs
critres de computation : la puissance dexpression
Repos
Demande de rception
Rception
Demande de
conversion
Reass-
emblage
Segm-
entation
SETIT2004
du langage, les possibilits dautomatisation fournies
par les modles travers le langage et les capacits
supportes par loutil. Pour valuer les performances
du modle propos, le langage de description
darchitecture VHDL (Vhsic Hardware Description
Language) a t choisi [11]. Le langage VHDL
permet de dcrire les diffrents tats du mcanisme
de conversion. Lutilisation de ce langage vise la
rduction de la complexit et du cycle de
dveloppement, travers les possibilits de
description (RTL, Structurelle), de vrification
tous les niveaux dabstraction et dautomatisation de
la synthse.

5.3 Outils de simulation et de synthse
La vrification au niveau RTL(Register
Transfer Level) aussi bien quau niveau "portes
logiques" justifie le comportement du systme
pendant une phase avance de la conception. Le but
de la vrification du systme est de voir la raction
de tous les signaux de sortie du systme en fonction
des diffrents tats pris par les signaux lentre. Le
concepteur pourra alors dterminer si le
comportement observ est en accord avec le cahier
des charges.
Les outils de simulation (V-system) et de
synthse (FPGA_Adv) permettent aux concepteurs
dvaluer lespace entier de conception : puissance
des techniques utilises, qualit dimplmentation,
atteindre une implmentation optimale. Ceci est
atteint en recherchant automatiquement le meilleur
tablissement des Machines dEtats Finies (MEFs) et
lallocation des units de conception. Quelque soit le
niveau dabstraction, la synthse prend en compte
diffrents objectifs doptimisation, tels que la
minimisation du dlai, de la surface [5] ou bien plus
rcemment, de la consommation du circuit.

6 RESULTATS DE SIMULATION

Les rsultats de simulation de notre description
VHDL montre le comportement des diffrents
signaux de donnes, de dialogue et de contrle lors
de lchange des informations entre les deux profils
de communication Ethernet et WATM. Le transfert
des donnes entre les deux rseaux peut se drouler
dune manire squentielle ou concurrente.

6.1 Processus de Segmentation
La figure 7 montre la simulation de deux
processus de lecture et de segmentation. En premier
lieu, la figure montre lacquisition dune trame
Ethernet initialis par les signaux RTS et CTS. Les
signaux CS_a1, WR_a1, AD_a1 et DI_a1 traduisent
lcriture en mmoire de la trame lue sur le bus
DI_a1.
En second lieu le processus actif est celui de
segmentation de la trame et son mission en cellules
WATM. Ces deux processus sont rgis par plusieurs
tats concurrents et squentiels. Lmission de
chaque cellule est prcde par une demande et une
confirmation travers les signaux RTS_Watm et
CTS_Watm. La gnration de la dernire cellule
transmettre au rseau WATM seffectue selon la
taille de donnes utiles qui reste. En gnral, elle est
complte par des bits de bourrages afin de respecter
le format utilis. Les derniers octets de cette cellule
contients ncessairement les champs (CPCS-UU,
CPI, Length et CRC) afin de permettre au niveau de
la destination le reassemblage dune unit de
donnes de type AAL5.

6.2 Processus de rassemblage
La figure 8 montre lacquisition des donnes
WATM et le droulement de la phase de
rassemblage des cellules reues. La procdure de
rception des cellules sinitialise par le dialogue
entre le rseau WATM et la passerelle, gre par les
signaux RTS_WATM et CTS_WATM. Durant cet
tat, les cellules WATM sont stockes en mmoire.
Les charges utiles de linformation stocke sont
ensuite rassembles pour les envoyer ensuite au
rseau Ethernet (DO_Eth). La passerelle gre la
mmoire en lecture afin dextraire la charge utile de
la trame mettre avec le rythme dhorloge
CLK_Eth.
Dans certain cas, le processus de rassemblage
passe ltat de bourrage qui permet de respecter le
format de la trame Ethernet. Ce processus sachve
par lenvoi du champ de contrle CRC_Eth.
7 SYNTHESE DU SYSTEME

Les rsultats de synthse sont tablis dans
lenvironnement de dveloppement FPGAdvantage.
Cet environnement permet limplmentation des
systmes de communication sur circuits
programmables. Lavantage de lutilisation des
circuits FPGAs est essentiellement la possibilit de
la re-programmation du circuit. Ceci permet
damliorer les performances de lapplication
lissu des tests dexprimentation. Les rsultats de
synthse montrent aussi le degr de complexit du
systme concevoir. La synthse est ralise en
imposant des contraintes fin de minimiser le dlai
de propagation (temps critique) ou la surface
dintgration. Le dlai (cibl sur la technologie
AMS(Austria Mikro Systeme)) peut tre optimis
pour amliorer les performances de la passerelle en
SETIT2004
terme de rapidit ou frquence de fonctionnement
atteinte.

Figure 7. Simulation du processus de la segmentation dune trame Ethernet

Figure 8. Simulation du processus de rassemblage des cellules WATM

Rception de cellules WATM
Dmarrer rassemblage (signal interne)
Gnration de len-tte Ethernet
Emission trame Ethernet
Stockage de donnes Ethernet en mmoire
Emission dune cellule WATM
Dialogue avec lEthernet (demande-rponse)
Dialogue avec WATM (demande-rponse)
SETIT2004
Pour notre application, la synthse RTL est faite
laide de loutil LenordoSpectrum de
lenvironnement FPGAdvantage. les rsultats de
synthse donnent une estimation sur les
performances du circuit cible. Ces performances
sontvalues en termes de nombre de gnrateurs de
fonction (Function generator FG). Les rsultats de
synthse donnent une ide sur les possibilits
dimplmentation sur FPGA. Nous remarquons par
exemple que limplmentation sur un seul FPGA de
type 4036xlBG352 est possible. Pour les autres
types, limplmentation ncessite plusieurs circuits
FPGA dont leur capacit permet la programmation
du systme conu. Ce ci dpend notamment du
nombre de module introduit par lalgorithme. Cette
solution de partitionnement peut engendrer des
dlais supplmentaires causes des interconnexions
entre les diffrents FPGAs.

Table 1. Rsultats de synthse en vue dintgration
sur circuit programmable
(Xilinx famille 4000)

FGs HFGs IOs
Circuit
Disp Utilis Disp Utilis Disp Utilis
4010P
Q208
800
438.00
%
400
183.75
%
160 94.38%
4013xl
PQ160
1152
208.68
%
576
155.73
%
112
134.82
%
4013P
Q208
1152
299.65
%
576
126.56
%
192 78.65%
4036xl
BG352
2592 92.75% 1296 69.21% 256 58.98%

Il est donc trs important de fixer le bon choix
de la technologie permettant dobtenir des rsultats
plus intressants et davoir plus en dtails les
caractristiques (frquence, surface, consommation
etc..) pour une solution de prototypage.
8 CONCLUSION

Dans cet article, nous avons propos des
algorithmes permettant ladaptation de format des
informations pouvant transiter entre les deux profils
de communications Ethernet et WATM. Ces
algorithmes intgrent des mcanismes de traitement
des donnes savoir la rception, lmission, la
segmentation dans un sens du trafic et le
rassemblage dans lautre sens. Nous avons adopt
la conception de haut niveau pour la ralisation du
systme. En effet, la description t entreprise avec
le langage de haut niveau VHDL. ModelSim de
lenvironnement FPGAdvantage et V-system ont t
utilis afin de vrifier le comportement du systme
au niveau RTL. Ltape de synthse a t entreprise
laide de loutil LeonardoSpectrum de
FPGAdvantage afin dvaluer les performances du
circuit.
Ce travail se prsente comme la brique de base
dans la conception dune passerelle dinterconnexion
Ethernet-WATM. La fonction ralise actuellement
permet les conversions de format des paquets
changs. Plusieurs travaux sont en perspectives
permettant des fonctions lies la fiabilit de
transfert et lacheminement des paquets.

REFERENCES
[1] Histoire et volution dEthernet ; Rseaux hauts dbits ;
Herv GILBERT ; Universit J ean Monnet Saint Etienne
Novembre 1999.
[2] An Adaptative Data Link Layer Protocol for Wireless ATM
Networks ; Sunil J agannath ; Department of Electrical
Engineering and Computer Science; University of Mysore
India 1994.
[3] FPGA et Parallelisme ; Frdric Magniette, Olivier Martin ;
DEA Informatique Universit Paris XI Orsay ; J anvier
1998.
[4] Design and Performance of Radio Access Protocols in
WATMnet, a Prototype Wireless ATM Network; P.
Narasimhan, S.K. Biswas, C.A. J ohnston, R.J . Siracusa &
H. Kim; C&C Reserch Labs, NEC USA.
[5] Interoperability of ATM Ethernet Interworking System:
Design and Congestion Control. R.Ouni, S.Adel, N.Salem
and all ; Laboratoire dEE, Facult des Sciences de
Monastir, Institut National Polytechnique de Grenoble
(INPG). ECUMN2000 Institut Universitaire de technologie
de Colmar.
[6] Thse : Conception dun systme dinterface pour
ladaptation de protocoles de communication entre rseaux
locaux : Application la conception dune passerelle
Ethernet-ATM ; OUNI Ridha ; Universit du centre Facult
des sciences de Monastir Fvrier 2002.
[7] A protocol Aided Concatenated Forward Error Control for
Wireless ATM. J ong Park, Departement of Electrical
Engineering Korea Military Academy; J ames Caffery,
Departement of ECECS University of Cincinnati; IEEE-
2002.
[8] Thse : Dveloppement dune Mthodologie de Conception
Matriel Base de Modules Gnriques VHDL/VHDL-
AMS en Vue dune Intgration de Systmes de Commande
Electriques ; Youssef KEBBATI ; Universit Louis Pasteur
Strasbourg ; 2002.
[9] Implementing distributed and dynamic resource allocation
in WATM; Christian Sinner*; Michael Wolf**; *Siemens,
Research and Predevelopment; **DaimlerChrysler,
Research and technology; Germany; Computer Networks
1999.
[10] Thse : Intgration des fonctionnalits multimdia dans les
systmes rpartis temps rel : Application au systme de
communication industrielle ; SOUDANI Adel ; Universit
du centre Facult des sciences de Monastir Fvrier 2003.
[11] Modlisation de systmes intgrs Numriques :
Introduction VHDL ; Notes de cours option 2
me
cycle
version 2001 ; Alain Vachoux.
[12] Gigabit Ethernet; Vijay Moorthy; http//www.cis.ohio-
state.edu/~jain/cis788-97/gigabit_ethernet/index.htm2000.

Improvements of the ABR loop performances in a wireless ATM network

Jamila BHAR, Ridha OUNI, Abdelhamid HELALI, Salem NASRI
Electronic and Micro-Electronic (EE) Laboratory,
Faculty of Sciences of Monastir, Tunisia
{jamilabhar; ridha_ouni}@yahoo.fr

Abstract

There were a number of research activities to promote
multimedia and wireless applications rate. Efficient flow
and congestion control mechanisms are needed in high-
speed networks to meat the new constraints and to offer
best quality of service. Mechanisms founded on the rate
adjustment had a particular interest. Wireless ATM
technology, based on a number of wired ATM protocols
and services, is considered a solution for these
requirements. It provides multimedia services to mobile
users and promises a support of quality of service (QoS).
Otherwise, the WATM performances depend on the
adopted mechanisms for flow and congestion control. So,
several factors can influence this mechanism such as error
probability on the wireless link and Handover procedure
type. The ABR service, standardized by the ATM forum, is
the most widespread traffic congestion control mechanism
permitting to provide fair sharing resources. ABR
specifications were designed to enable efficient bandwidth
allocation and low cell loss ratio. This paper addresses the
interaction among the flow control algorithm, the
Handover procedure, and the resulting QoS
characteristics. We present a novel ABR congestion
control algorithm which is adaptive to changing network
conditions and robust to terminal mobility. The objective
of this algorithm is to achieve a high utilization factor and
to maintain the compatibility between the wired and
wireless link constraints. Therefore, it was important to
develop the necessary mobility enhancements, such as
handovers and radio resource management. This
approach guarantees the development of an effective
control strategy for the congestion control mechanism.
Keywords Wireless ATM, ABR, Handover, QoS

1. Introduction

The goal of the ABR congestion control mechanism is to
fairly share the bandwidth, left over from high priority
services, by the ABR connections and to minimize cell
loss ratio. The ABR service developed for ATM network
cannot be the same used in the wireless ATM environment
for various reasons. Indeed, there are supplementary
constraints such as the difficulty to predict the traffic
characteristics for mobile terminals, bandwidth loss caused
by the supplementary fields in the header, interference
phenomenon and important bit error rate on a wireless
channel. For these reasons and to provide uniformity of
end-to-end quality of service (QoS) guarantees by wireless
ATM, it is necessary to focus on the improvement of the
existing ABR flow-control algorithms performances.
In this paper, we present a robust adaptive congestion
control algorithm for ABR service in Wireless ATM
networks. The principal design objective was to show that
this algorithm performs better with various criteria related
to ABR loop characteristics and Handover procedure
decision.
We consider wireless ATM network environment. The
underlying WATM architecture of our study is derived
from WAND project prototype [10]. We specify the ABR
service class. The flow-control algorithm is specified with
two control loops spanning the links between a mobile
terminal and its adjacent base station as well as the latter
and a data source. We study the impact of handover
schemes on this algorithm. Handoff procedure must be
capable of dynamically reestablishing virtual circuits to
new base stations (BSs) while ensuring minimal loss of
data.
This paper proceeds as follows. The first section presents
the general implementation of ABR flow-control
algorithms in a Wireless ATM network. The WATM
protocol architecture is also proposed. The next section
deals with different types of handovers proposed in the
literature, and gives handover solution to support ABR
loop performances. The last section shows and evaluates
simulation results. Conclusions are stated and issues for
future research are discussed.

2. ABR flow-control in a Wireless ATM
network

The network supervises without interruption the traffic and
provides the feedbacks for the source terminal system.
There are many rate-based congestion schemes. ERICA is
used as the well-known algorithm [5]. For an ABR
connection, information carried in ABR Resource
0-7803-8656-6/04/$20.00 2004 IEEE
< Back

Figure 1. Structure of VS/VD node

Management (RM) cells is primarily generated by ABR
source end systems. The destination end system turned
then the information received. The feedback information
can be modified from the destination or/and switch
traversed. Indeed, a switch along the path can specify
explicitly the maximal supported cell transmission rate
using the explicit rate (ER) field of an RM cell. For this
purpose, it evaluates the bandwidth requirements of all
assigned ABR connections, calculates a fair share of the
available bandwidth and distributes it among all active
ABR connections traversing the corresponding link.
Before returning RM to the source, the ER in the RM cell
is compared with the calculated ER value. As a
consequence of the feedback mechanism, the source can
adjust its rate according to this information. When the
source receives the RM cell, it sets its current cell rate
(CCR) to the ER in the RM cell. This last is periodically
inserted into the stream of ABR data cells. RM cells that
flow from source to destination are referred to as Forward
RM (FRM) cells and RM cells that flow from destination
to source are referred to as Backward RM (BRM) cells.
The control cells FRM or BRM have the same cell data
format. If the switch or the destination is congested they
can send a BRM cell without waiting for the FRM cell. In
this case, this cell is called OOR (Out Of Rate). The goal
is to inform the ABR sources much faster about the
congestion. It is distinguished by the BN (Backward
Notification) field of the RM cell. Under circumstances
such a WATM it is advantageous to divide the control
loop into several coupled smaller loops by introducing
virtual sources (VS) and virtual destinations (VD). The
virtual sources and destinations have the following
properties: Each control loop behaves in the same manner
as a separate ABR loop. This VS/VD approach has the
advantage to improve the response of the controls subject
to changing ABR capacity.
Wireless ATM network is composed by Mobile Terminals
(MTs), Base Stations (BSs) and switch. As shown in
figure 2, BS is specified with mobility enhanced UNI/NNI
and radio interface capabilities that provide the interface
between the wired and the wireless portions of the
network. Switch and BSs are wired together and represent
the fixed network segment. Each BS communicates with
all the MTs in its corresponding macro area. A number of
BSs can be connected to the same switch. In the protocol
stack, radio interface between the BS and the MT
comprises the Radio Physical Layer (RPHY), the Radio
Multiple Access Control Layer (RMAC) and the Radio
Logical Link Control Layer (RLLC). It globally referred
to as the Radio Access Layer (RAL). The end terminals
require AAL Layer functions for end to end virtual
connection. However, BSs and Switch support the ATM
layer and can extract or insert handover protocol cells
from/into the connection data flow [8; 10].
For the segmented ABR loop applied in wireless ATM
environment, the base station acts as a Virtual Source (VS)
and Virtual Destination (VD) pair. The base station
consists of the part of deciding how to calculate ACR of
loop2 and ER of loop1 (figure 1) [7]. In proposed
mechanism, the BS with VS/VD system generate FRM
cells to terminals and determines the ACR according to
the CI/NI bits but also according to rules bellows:
ACR is function of (CI, NI, ER, ACR)

If (CI='0' and NI='0') then ACR=min(ER,
ACR+RIF*PCR, PCR)
If (CI='1' and (NI='0' or NI='1')) then ACR=min(ER,
ACR-ACR*RDF, MCR)
If (CI='0' and NI='1') then ACR=min(ER, ACR)

The adjustment of ER value is based on different
parameters such as ER in BRM, CCR but also ACR of
loop 2. The VD sends an OOR RM cell to ABR source
when receiving a congestion indication from loop 1.

Figure 2. Protocol Stacks of Network Nodes
Wireless Channel

ABR
Source
Base
station

ABR
Destination
1
st
Loop 2
nd
Loop
3. The Handover solution

In systems such WATM, terminals will frequently change
their attachment access point in the network due to
movement of the WT(Wireless Terminal). When this
occurs, the WTs active connections must be transferred to
the new base station, so that connectivity is maintained. In
order to maintain desired quality of service (QoS), this
work is focused on the handover scheme that allows a
reliable control of connections transfer. Handover
concerns all activities that can appear following the
mobility of the terminal between the different zones
covered by base stations such as :
The measure of the wireless channel parameters for
the current transmission.
The Handover procedure initiation.
The selection and the measure of the new wireless
channel parameters.
Different handover schemes have been proposed for
WATM. There are many works that have studied various
aspects of the WATM handover requirements. Evaluations
of the different proposals are though closely related to find
the handover procedure that suit better with WATM
constraints. The following is a description of handover
algorithm working. Some details are omitted, and may be
found in [1, 2, and 3].
a. Soft Handover: In this case, the handover can be
predicted and MT initiates the handover procedure via the
new BS. The mobile terminal connections are passed to
the new base station without interrupting communication
with the old base station. Indeed, data and control
messages are transferred on the connection through the
new BS before the old one is released. After a short time
period the old connection is released.
b. Hard Handover: In this case, the handover occurs
when the connection to the current BS is broken. The
connections are interrupted at the old base station and re-
established through the new base station. Indeed, the new
BS is contacted first before the handover initiation is
finished. The connection held by the old BS is released
before the connection to the mobile through the new BS is
established.
Soft and Hard handovers are specific for each situation.
They will be introduced according to sums criteria. The
choice of handover type is then based on the network
conditions. Regarding data communication, the
corresponding handover procedures have to guarantee the
sequence integrity and loss-free delivery of WATM cells
during the transition process. In fact, we study the
handover procedures impact on the efficiency of the ABR
flow control protocol.
During handover control messages are exchanged between
base station and mobile terminals and between base
stations themselves. These messages handle functions as
connection and disconnection state transfer during
handover. Handovers are typically initiated by MTs, but
WATM networks should also be able to trigger handovers
for network management purposes. The mobile terminal
announces the beginning of its handover a short period
before the real handoff procedure occurs. In this stage, an
OOR-RM cell is sent to the virtual source to update the
ER. These recent information permit the adaptation of rate
in the new BS to avoid congestion.
In this paper, we specified the source, the destination and
the switch behaviours. We defined how a virtual source
should behaves in response to BRM information and also
how ABR loop is coupled with the flow control factors as
well as the handover procedure for mobile terminals. Our
strategy is to select one handover procedure and compare
performance result on the ABR control protocol for
different proposal schemes. This proposed solution is
chosen for its flexibility and fairness properties.
Consequently, the rate-based flow control algorithm is
formulated as a dynamic problem where the goal is to
evaluate if the ABR feedback control scheme ensures its
QoS objective and improve flow control performances.

4. Results

The most important criteria that a successful handover
protocol for an indoor wireless ATM must fulfil are:
Minimal Service Interruption, No Cell Reordering or
Duplication and Minimal Cell Loss, QoS Maintenance,
Maintaining the Cell Sequence, Robustness and Stability.
This outcome is expected since the fast response of this
scheme is guaranteed by sending BRM cells with recent
information about the status of the wireless link between
the new BS and the MT. This information is useful to set a
connection as soon as the MT can reach its new BS during
a handover.
The performed simulation experiments allowed evaluating
the efficiency of the ABR flow-control protocols ERICA
with and without the VS/VD concept for a WATM
environment. Particularly, the impact of different
handover schemes (hard and soft handover) and the BS
behaviour on ABR loop is evaluated. The ability of
algorithms developed to achieve fairness in some
interesting situations is investigated.
For simulation, we used a dual port memory
DPRAM128x32. To guarantee write and read operations
managing, we developed a control mechanism for
concurrent actions in the memory. We are mainly
interested in the stability of the data flow and the
influences of combined handover event in the ABR loop.
Figure 3 shows the behavior of virtual source and virtual
destination. Simulation results show that the BS with
VS/VD behaviour informs the source with recent
information faster than with one ABR loop. If we compare
the hard and soft handover procedure on ABR loop, we
note that there is a difference on the procedure during and
on the delay to inform the BS to allocate the necessary
terminal resources. Selecting one handover type is related
to degree of network condition degradation. This strategy
permits to establish the efficient reaction with minimal
loss.

5. Conclusion

In this paper we proposed and evaluated an ABR flow
control algorithm for a Wireless ATM environment. This

Data cells Reception of FRM cell

Figure 3. VS/VD behavior of BS

algorithm ameliorates the congestion control performances
by adding new constraints to the ABR loop. Results
obtained justify that it is important to consider suitable
handover protocols in order to maintain QoS parameters.
The proposed mechanism supports optimum Handover
procedure when required. Thus, it minimizes the total
execution time of handovers. An additional advantage is
focused on the same actions performed during soft and
hard handovers. This characteristic facilitates the
implementation of the required functional entities. In
conclusion, the impact of handover protocols on WATM
performances over ABR service class is considerable and
provides an alternative approach that requires further
studies.
The aim of handoff signalling is to enable wireless
terminals to move seamlessly between BSs while
maintaining connections with their negotiated QoS. For
this purpose, there is a close inter-working with the
connection management and rerouting functions of the BS.
The format of signalling messages exchanged at handover
procedure may be also optimised to guarantee Qos
requirements. Therefore, the employment of ARQ scheme
will reduce the cell loss rate due to the error characteristics
of the wireless link. The future work will adopt a rerouting
scheme for proposed handoff procedure. The proposed
handover mechanism can also be developed to support
inter-switch handovers.

References

[1] Adaptation of the Fuzzy Explicit Rate Marking to ABR
flow-control in wireless ATM Netwok; Michael Savoric,
Department of Electrical Engineering Technical University of
Berlin; Udo R.Krieger, Deutsche Telekom AG
Technologiezentrum Am Kavalleriesand Germany. MMB 99,
Trier, September 1999.

[2] The Impact of Handover Protocols on the performance of
ABR flow-Control Algorithms in a Wireless ATM Network;
Michael Savoric, Adam Wolisz, Department of Electrical
Engineering TU Berlin Germany; Udo R.Krieger, T-Nova
Deutsche Telekom, Am Kavalleriesand Germany. European
Transactions on Telecommunications (ETT) J ournal, J uly-
August 2000.

[3] Performance evaluation of Handover protocols for data
communication in a Wireless ATM network; Udo R.Krieger and
Michael Savoric; ITC16, Edinburgh, J une 1999.

[4] ABR Traffic Congestion Control Mechanism with VS/VD
in Wireless ATM Networks; Moonsik Kang, Sangmin Lee,
Department of Electronic Engineering, Kangnung National
University Kangwon Kangnung Korea; 2000.

[5] ERICA Switch Algorithm : A Complete Description, ATM
Forum/96, Raj J ain, Shiv Kalyanaraman, Rohit Goyal, Sonia
Fahmy, Ram Viswanathan; The Ohio State University
Department of CIS Columbus; http://www.cis.ohio-
state.edu/~jain/

[6] Source Behavior for ATM ABR Traffic Management: An
Explanation; Raj J ain, Shivkumar Kalyanaraman, Sonia Fahmy
and Rohit Goyal; Department of Computer and Information
Science; The Ohio State University; and Seong-Cheol Kim;
Samsung Electronics Co. Ltd.Chung-Ang Newspaper Bldg; 1996

[7] Design considerations for the virtual source/virtual
destination; VS/VD feature in the ABR service of ATM
networks; Shivkumar Kalyanaraman; Raj J ain; J ianping J iang;
Rohit Goyal; Sonia Fahmy; Department of Computer and
Information Science, The Ohio State Uniersity, Columbus,
USA; J une 1998

[8] Local and Global Handovers Based on In-Band Signaling in
Wireless ATM Networks; M.AJ MONE MARSAN;
C.F.CHIASSERINI; A.FUMAGALLI; R.LO CIGNO and
M.MUNAF; Wireless Networks,2001

[9] Microcellular Handover in Wireless ATM; Fan J iang ; Timo
Kkl; Proceedings of the 32nd Hawaii International
Conference on System Sciences 1999

[10] Implementation of the wireless ATM access terminal; J uha
Ala-Laurila; Jussi Lemilainen; Nokia Wireless Business
Communications, Computer Networks; 1999.
Data cells memorized
VS
VD
RM cell filtered
Emission of BRM cell
Emission of FRM cell
Reception of BRM cell
< Back

3.rapport Activite Pub

Încărcat de

Informații document

Descriere originală:

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

3.rapport Activite Pub

Încărcat de

Drepturi de autor:

Formate disponibile

Papiers publis dans des

and KHOLDOUN TORKI

College of Computer and Information Sciences,

Faculty of Sciences of Monastir, Tunisia

CMP, INPG, Grenoble, France

Corresponding author. Tel.: +216 22691815; fax: +216 73501785.

Fig. 17. Various membership functions.

Delay is measured when packets of data take more time

Media access delay

To analyze the performance of various sceduling

Data Traffc Received

defines a unique position, each formed by a

) are the coordinates of BS

iteration, the MS will be located in an A

) are calculated using the square root operators

) Geolocalisation information, obtained

defines a unique position, each formed by a

can determine the shape of

are the coordinates of BS

and the distance

As we described previously, we can get the coordinates of

iteration, the MS will be located in an A

is the signal strength received by the

the signal strength

from the AP, and a

is the number of walls of different nature,

S-ar putea să vă placă și