Content uploaded by Badia Bouhdid
Author content
All content in this area was uploaded by Badia Bouhdid on Mar 20, 2024
Content may be subject to copyright.
Enhanced Fuzzy logic-based Cluster Stability in
Vehicular ad hoc Network
Mohamed Aissa
University of Nizwa, Nizwa
Sultanate of Oman
Oman
m.aissa@unizwa.edu.com
Badia Bouhdid
ESPRIT School of Engineering
Hana-Lab, ENSI, University of Manouba
Tunisia
badiaa.bouhdid@ensi-uma.tn
Adel Ben Mnaouer
Department of C.E.C.S
Canadian University Dubai
17781, UAE
adel@cud.ac.ae
Abstract—VANET nodes are characterized by their high
mobility and they exhibit different mobility patterns. There-
fore, VANET clustering schemes should take into consideration
the mobility parameters among neighboring nodes to produce
relatively stable clustering structure. This paper proposes a
novel cluster head selection Fuzzy Logic-based k-hop distributed
clustering scheme for VANETs. This scheme considers the safe
inter-distance between vehicles as one of important metrics for
cluster head selection. Our contribution deals better with the
scalability and stability issues of VANETs and can achieve a high
stable cluster topology compared to other schemes. These features
make the proposed scheme more suitable for VANETs networks.
Simulation proves the effectiveness of the new proposed scheme
to create stable clusters by reducing re-clustering overhead and
prolonging cluster lifetime compared to other existing clustering
schemes for VANET.
Index Terms—Vehicular networks, V2V, clustering, Fuzzy logic
system, stability.
I. INTRODUCTION
Vehicular Ad hoc Networks (VANETs) is considered as a
subclass of Mobile Ad Hoc Networks (MANETs). VANET
provides safety and comfort to the road users and consequently
it is one of the influencing areas for the improvement of Intel-
ligent Transportation System (ITS). It assists vehicle drivers
to communicate, to exchange information and to coordinate
among themselves in order to avoid critical situations through
Vehicle to Vehicle communication e.g. road side accidents,
traffic jams, speed control, free passage of emergency vehicles
and unseen obstacles etc. As an ad-hoc network, a VANET
cannot rely on specialized infrastructure to support network
stability. Each node in a VANET is required to maintain its
own connectivity to other nodes in the network. With the large
number of nodes and the lack of routers, a flat routing scheme,
where each node acts as a router, may cause serious scalability
and hidden terminal problems.One possible solution to these
problems is hierarchical clustering.
Clustering is a technique to group nodes into several
clusters. Each node in the cluster structure plays one of
three roles: Cluster Head (CH), Cluster Gateway (CG), and
Cluster Member (CM) [1]. The CH in a cluster plays the
roles as coordinator and backbone. It is in charge of all the
communications inside a cluster, managing medium access and
allocating the resource to cluster members [2]. A CG is a
border node of a cluster that can communicate nodes belonging
to different clusters [1].
The clustering scheme has been well investigated in wireless
ad-hoc networks in recent years. Due the high mobility of
VANET, clustering schemes should take into consideration
mobility factors among neighboring nodes to produce rela-
tively stable clustering structure.
In this work, we introduce a new clustering approach with
the aim of increasing the stability of the network topology. A
new Fuzzy logic based cluster head selection method is pro-
posed to handle the variety of environment terms and provide
flexibility for the cluster structure. These proposed scheme
overcomes some inefficiencies detected in others clustering
schemes. It considers the safe inter distance between vehicles
as one of main metrics for cluster head selection. The proposed
metrics should select cluster heads which provide safe clusters
and avoid collisions with adjacent vehicle nodes and intend to
create stable clusters by reducing re-clustering overhead and
prolonging cluster lifetime. For the purpose in our scheme we
add a new election metric which represent the safe distance.
Nodes having higher number of stable neighbors, maintaining
safe distances to their stable neighbors, and having closer
speed to the average speed of their stable neighbors are more
qualified to be elected as cluster-heads. Furthermore, this
approach takes the speed difference, in addition to the link
connectivity duration and direction, into consideration during
the clustering process.
Section 2 deals with description of the system overview
and fuzzy logic-based cluster head selection. In section 3, a
presentation of the proposed scheme structure. In section 4
we conducted intensive experimental analysis to investigate
the performance of our proposed technique and compare it to
the best well-known algorithms (such as PPC [7] and CMAC
[6]). Our simulation results reveal that SFVC proved to have
the best performance, has good flexibility and is able to form
stable cluster structure. Security and other performance issues
should be considered in the future work.
II. SY ST EM OV ERVIEW AND FUZZY LOGIC-BA SE D
CL US TE R HE AD E LE CTION
As illustrated in figure 1, we propose a new Fuzzy Logic
Controller for calculation of the Fit Factor actor value. In
designing a fuzzy inference system, the first step is determi-
nation of input and output variables, and their fuzzy set of
membership functions.
Fig. 1: Proposed fuzzy logic scheme structure
This is followed by designing fuzzy rules for the system.
Furthermore, a group of rules are used to represent the
inference engine (knowledge base) for articulating the control
action in linguistic form [3]. The priority of a node becoming
a CH is determined by its FitFactor. Hence, first the nodes
start calculating their FitFactor to become a CH and broadcast
messages containing their FitFactor values.
A. Metrics computation step
We consider the average distance, relative velocity, and link
connectivity duration as the three metrics for CH selection.
1) Average distance (ADi): : For VANET clustering or
routing algorithms, authors use the following formula (equa-
tion 1) to calculate the overall average absolute distance
between all vehicles that are directly connected to a current
vehicle i:
ADi=P|Ni|−1
j=1 Dij
|Ni| − 1(1)
With |Ni|is the nodal degree of a node vi.
The major drawback of this average distance is that it
is blind and passive and does not differentiate the types of
neighbors for a current node i(Figure 2).
Our contribution is to boost delegating CHs which are
situated in a safe distance (SDi,j ) and prevent those which
are situated in a critical distance at the front and rear vehicles.
The new overall average absolute distance ADibetween all
vehicles that are directly connected to vehicle iis deduced in
equation 2:
ADi=1
|Ni| − 3
|Ni|−3
X
j=1,j6=i±1
Di,j +1
2(SDi−1,i +S Di,i+1)(2)
Fig. 2: Highway topology
2) Link connectivity duration Ci:We are motivated by the
analyses provided in [4] to calculate link connectivity duration
which refers to the interval during which the vehicle is in a
cluster head state until it shifts into a cluster member or cluster
undecided state. We extend this formula to calculate the link
stability factor between the ith vehicle and all its connected
neighbor vehicles, which can be formulated as given in the
following equation 3:
Ci=1
vi
|Ni|−1
X
j=1
Di,j (3)
3) Relative Velocity (RVi): In every time interval, each
vehicle, will have information about all vehicles within its
communication range and hence will calculate its average
velocity difference from all other vehicles as given in equation
4:
RVi=
1
|Ni|−1.Pj=1
|Ni|−1|vj−vi|
vi
(4)
The relative velocity of the node with its neighbors means
how long these neighbors node have spent their time beside
this node. To build relatively stable cluster structure, we
consider the vehicles with better neighborhood degree (|Ni|).
Therefore, in the FitFactor evaluation, these metrics should be
considered jointly.
Wi=w1.RV i +w2.ADi+w3.Ci(5)
B. Fuzzy logic-based cluster head selection
1) Fuzzification step: The fuzzy membership function of
the average absolute distance is defined as in Figure 3(a). The
sender node uses the membership function and the average
absolute distance to calculate which degree the distance factor
belongs to Small, Medium, Large. The fuzzy membership
function of the average velocity is defined as in Figure 3(b).
The sender node uses the membership function and the average
velocity to calculate which degree the average velocity belongs
to Slow, Medium, Fast. The fuzzy membership function of the
link connectivity duration is defined as in Figure 3(c). The
sender node uses the membership function and the link con-
nectivity duration to calculate which degree the link connec-
tivity duration belongs to Short, Medium, Long. Figure 3(d)
depicts the correlation behavior between two inputs (relative
velocity and average distance) and the output variables.
Fig. 3: Fuzzy Membership Functions
Fig. 4: An example of output on the output membership
function
2) Rules mapping step: Once the fuzzy values of average
absolute distance, average velocity and link connectivity du-
ration have been calculated, Fuzzy Inference Engine maps
these fuzzy values to the IF/THEN rules defined in Table
1 and contained in the Knowledge Rule Base to calculate
the fit factor for each node. The Fuzzy Inference system is
designed based on 27 rules as presented in Tables 1 and
2. The linguistic variables of the fit factor are defined as
Perfect, Very Good, Good, Acceptable, Bad, Very Bad. For
example, in Table 2, Rule3 may be expressed as follows. IF
Average Distance is Small Large and Relative Velocity is Slow
and Link Connectivity Duration is Long THEN Fit Factor is
Perfect.
3) Defuzzification step: Defuzzification is the process of
producing a numeric result based on an output membership
function and corresponding membership degrees. The output
membership function is defined as in Fig. 4. Here we use the
Center of Gravity (COG) method to defuzzify the fuzzy result.
More specifically, we cut the output membership function (Fig.
4) with a straight horizontal line according to the correspond-
ing degree, and remove the top portion. Then, we calculate
the Centroid of this shape.
TABLE I: Knowledge rule base Explanation.
Rule avd ω φ Weight
1. Small Slow Short Good
2. Small Slow Medium Very Good
3. Small Slow Long Perfect
4. Small Medium Short Acceptable
5. Small Medium Medium Good
6. Small Medium Long Very Good
7. Small Fast Short Bad
8. Small Fast Medium Acceptable
9. Small Fast Long Good
10. Medium Slow Short Acceptable
11. Medium Slow Medium Good
12. Medium Slow Long Very Good
13. Medium Medium Short Bad
14. Medium Medium Medium Acceptable
15. Medium Medium Long Good
16. Medium Fast Short Very Bad
17. Medium Fast Medium Bad
18. Medium Fast Long Acceptable
19. Large Slow Short Bad
20. Large Slow Medium Acceptable
21. Large Slow Long Good
22. Large Medium Short Very Bad
23. Large Medium Medium Bad
24. Large Medium Long Acceptable
25. Large Fast Short Not Acceptable
26. Large Fast Medium Very Bad
27. Large Fast Long Bad
III. CLUSTERING PROCESS AND PROTOCOL
STRUCTURE
This section describes the Safe Vehicular Clustering Algo-
rithm (SFCA ) algorithm. SFCA can be divided into two main
steps:
•CH selection process
•Cluster Maintenance
A. Cluster head selection process
In SFCA , all nodes are divided into three types:
•Cluster Head (CH),
•Cluster Member (CM),
•Free node (FN).
Fig. 5: State transitions of a vehicle in clusters
The CH selection process in our algorithm is based on
the calculation of an utility function using (14), which has
the following objectives: cluster formation and cluster main-
tenance, extend the clustering phase and reduce the number
of re-clustering. In our SFVC algorithm, only vehicle nodes
travelling in the same direction are taken into consideration.
We describe here our CH selection algorithm:
1) In the initial phase (that means there is no cluster
established in the system), each vehicle node begins
with an initial state: FN and broadcasts a Hello Message
to all its neighbours containing its speed.
2) After receiving the Hello message from all its neighbors
each node calculates:
a) The neighbors (degree) of each node i using (2).
b) The average velocity difference using (3).
c) The relative velocity using (4).
d) The overall average absolute distance adiusing
(12).
e) Link stability factor using (15)
3) Each node employs fuzzy logic to calculate a fit factor
value Wi(combined weight) using (16)
Wi=w1ωi+w2avdi+w3φi
4) The election of CH is based on the comparisons of
adjacent nodes’ combined values Wi. The vehicle node
with the smallest Wiamong its one-hop neighbours
becomes CH,
5) The CH sends an invitation message to its direct neigh-
bours which turn into its CMs.
6) The same procedure is repeated for all the remaining
nodes.
Figure 5 illustrates the sequence diagram of the proposed fuzzy
logic-based CH selection system.
1) FN joins an existing cluster to become CM: As it was
mentioned previously, a vehicle’s default clustering state is
Free Node (FN) when it joins the highway from the very
beginning or from an entrance. At this moment, the FN
broadcasts HELLO message periodically. If a nearby CH
hears FN’s HELLO message, it will response a WELCOME
message, and allows this FN to join its cluster. If two or more
CHs have the same distance to FN, the one with less combined
weight will be picked.
2) CM leaves current cluster to become FN: CMs are
required to report periodically their moving information to
their CH inside a cluster. If the CH does not hear from a
specific CM, then this CM is supposed to be outside the
transmission range and that it has left current cluster. In this
case, the left CM will keep the FN state until it joins other
clusters.
3) CH left current cluster: As it was mentioned previously,
CH and CM should exchange periodically messages to men-
tion their presence. Consequently, if a CM did not hear the
response from a CH for a period of times, then this CH is
believed to leave current cluster. All CMs in the cluster will
turn into FNs, and begin broadcasting HELLO messages to
start a new round of CH election.
4) Cluster merge: In the highway, at a certain time, we
can have a situation where two CHs are too close to each other
that they are almost in the same transmission range. In this
case, these two clusters will be merged. The CH having the
TABLE II: Network Parameters and Values
Parameters value
Transmission range (r) 100 m to 300 m
velocity 60 km/hour to 120 km/hour
Simulation time 600 seconds
Vehicle number 400
Length of highway 20 km
Data rate 6 Mbps
Size of message 100 bytes
MAC protocol IEEE802.11 [?]
CWmin (collision window) 15
CWmax (collision window) 1023
Slot time 16 us
SIFS 32 us
DIFS 64 us
most neighbours (from old and new cluster) is automatically
elected as the CH in new cluster, and the second CH will
become a CM. After the cluster merge, some CM can be
outside the signal range of new CH. In this case, these CMs
will be transformed into FNs and begin to broadcast HELLO
messages to find another cluster to join.
IV. PERFORMANCE EVALUATION
A. Simulation Settings and Performance Metrics
Extensive simulations are performed using NS2 Simulator
[5]. In this section, we evaluate our proposed scheme, SFVC,
and compare it to the Cluster-based Medium Access Con-
trol protocol (CMAC) [6] and the Position-based Prioritized
Clustering (PPC) method [7]. A brief description of these two
algorithms was introduced in section 1. In the simulation, we
assigned all CMAC method parameters equal weights. In the
position-based prioritized clustering (PPC) method, the prior-
ity of the node is calculated based on the eligibility function.
A node having longer travel time has higher eligibility value,
and this value decreases as the velocity of the node deviates
largely from the average speed.
The mobility model is built based on the car following
model presented in [8], where, the mobility behavior of the
car is developed based on the relation with respect to the car
ahead. In this model, the velocity of the vehicle is computed
based on different factors, such as the safe distance, to the
front vehicle and the speed of both vehicles at time t such
that a safety distance is maintained. The network parameters
used in the simulation are depicted in Table 3:
Given that VANETs are typically constrained in terms of
vehicle mobility and driver behavior, addressing these con-
straints leads to review the cluster stability, the size and the
number of clusters in the performance evaluation of the three
simulated clustering algorithms (CMAC, PPC and SFVC).
B. Performance evaluation of the proposed fuzzy clustering
scheme
1) Cluster stability: In a high dynamic VANET, vehicles
keep joining and leaving clusters along their travel route.
Consequently their status changes dynamically and depends on
the clustering algorithms being used. In fact a vehicle CM can
leave its cluster and form a new one to be CH, or joins a nearby
(a) (b) (c)
Fig. 6: Average number of status change per vehicle
(a) (b) (c)
Fig. 7: Average number of formed clusters
(a) (b) (c)
Fig. 8: Average cluster lifetime
cluster. Also a vehicle CH can merge with a nearby cluster to
be a CM or to change its status to FN. For each vehicle, the
sum of all these transition events defines the number of status
change of the vehicles over its existence in the network system.
The number of status changes of the vehicle represents
a key factor to evaluate the cluster stability. In fact, an
efficient clustering algorithm should minimize the number of
status changes of the vehicle by minimizing vehicle transitions
between clusters.
Figure 6, illustrates the Number of Status Change (NSC) in
relation with the velocity and the transmission range, where
the velocity varies from 60 km/hour to 120 km/hour and the
transmission range from 150 m to 300 m.
Increasing the transmission range, whenever in low (Fig.
5.a), medium (Fig. 6.b) or high speed (Fig. 6.c) have a sharp
influence in the NSC. In fact NSC decreases with the increase
of the transmission range for the three simulated methods and
this is because increasing the transmission range R, increases
the probability that a vehicle stays connected with its cluster-
head. However increasing the velocity of vehicle degrades the
performance of both algorithms CMAC and PPC, where the
NSC increases with the increase of average speed. Fortunately,
for SFVC, this average speed increase has little impact on the
NSC per vehicle.
2) Number of clusters: The high dynamic nature of VANET
leads to the production of clusters over time and cluster
structure variation. Due to the mobility of the nodes, reducing
the rate at which clusters are created is a challenging task.
The average number of clusters in the system is calculated as
the total number of clusters generated by algorithm execution
by all vehicles. Figure 7 illustrates the average total number
of clusters generated over all simulation runs for different
transmission ranges and velocity. According to our simulation
results, we can deduce that whenever increasing the transmis-
sion range or velocity, the number of clusters generated by
SFVC algorithm is always smaller compared to those produced
by CMAC and PPC algorithms. Furthermore, we can observe
that this number decreases as the transmission range increases.
3) Cluster life time: The cluster-head lifetime is directly
related to its task lifetime which is defined as the time period
from the moment when a vehicle becomes a cluster-head to
the time when it is merged with a nearby cluster or it becomes
FN.
As illustrated in Fig. 8, the average cluster lifetime produced
by SFVC, PPC and CMAC is affected by velocity variations.
In fact, the cluster life time decreases as the velocity of the
vehicles increases which also related to the NSC of the vehicle
and the cluster stability. However, from these figures, we can
see also that the cluster life time performed by SFVC method
is important compared to those performed by CMAC and
PPC methods. Fig. 8.c illustrates that SFVC increases the
average cluster lifetime by 20 −30% compared to the PPC
and CMAC. Increasing the cluster lifetime, by maintaining the
current cluster structure stable as much as possible, enhances
the cluster stability and this can enhance consequently the
system performance by reducing the communication overhead
and traffic generation related to re-clustering process
V. CONCLUSION
This work proposes a novel CH selection Fuzzy Logic-based
k-hop distributed clustering scheme for VANETs algorithm.
The proposed scheme considers the safe inter distance between
vehicles as one of main metrics for cluster head selection.
SFVC provide safe clusters and avoid collisions with adjacent
vehicle nodes and intend to create stable clusters by reducing
re-clustering overhead and prolonging cluster lifetime. Finally,
a simulation study was conducted which its illustrative results
demonstrate superiority of the performance of SFVC archi-
tecture and its clustering scheme in comparison with other
existing schemes.
REFERENCES
[1] Lo, Shou-Chih, Yi-Jen Lin, and Jhih-Siao Gao. ”A multi-head cluster-
ing algorithm in vehicular ad hoc networks.” International Journal of
Computer Theory and Engineering, pp. 242-247, 2013.
[2] Ding, Ranran, and Qing-An Zeng. ”A clustering-based multi-channel
vehicle-to-vehicle (V2V) communication system.” Master Thesis, Uni-
versity of Cincinnati, 2009.
[3] C.J. Huang, I.F. Chen, K.W. Hu, H.Y. Shen, Y.J. Chen, and D.X. Yang. A
load balancing and congestionavoidance routing mechanism for teal-time
traffic over vehicular networks. Journal of Universal Computer Science,
pp. 2506–2527, 2009
[4] Wang, Sheng-Shih, and Yi-Shiun Lin. ”PassCAR: A passive clustering
aided routing protocol for vehicular ad hoc networks.” Computer Com-
munications, pp. 170-179, 2013.
[5] Issariyakul, T. and Hossain, E., “Introduction to Network Simulator
NS2,” Springer, 2009
[6] G Yvonne, W Bernhard, G Hans Peter, Medium access concept for
VANETs based on clustering, in Proceedings of the 66th IEEE Vehicular
Technology Conference (VTC), vol. 30. Baltimore, pp. 2189–2193,
2007.
[7] W Zhiagang, L Lichuan, Z MengChu, A Nirwan, A position-based
clustering technique for ad hoc intervehicle communication. IEEE Trans
Man Cybern,pp. 201–208, 2008.
[8] S Krauß, Microscopic modeling of traffic flow: investigation of collision
free vehicle dynamics, PhD thesis, Mathematisches Institute, Universitat
zu Koln, 1998.
[9] IEEE Standard 802.11-1999, Wireless LAN medium access control
(MAC) and physical layer (PHY) specifications,1999.