4
A Novel Enhanced Weighted Clustering Algorithm for Mobile Networks Chang Li, Yafeng Wang, Fan Huang and Dacheng Yang, Member, IEEE Wireless Theories and Technologies Lab (WT&T) Beijing University of Posts and Telecommunications Beijing, P. R. China [email protected]  Abstract  —Many applications of mobile Ad Hoc networks (MANETs) adopt hierarchical structure for the scalability and simplify of management. Clustering is the most popular method to impose a hierarchical structure in MANETs. As a newly proposed weighing based clustering algorithm, the Weighted Clustering Algorithm (WCA) has improved performance compared with other previous clustering algorithms. However, the high mobility of nodes will lead to high frequency of re- affiliations which will increase the network overhead and minimize the network lifetime. To solve this problem, we propose an improved weight based clustering algorithm, namely enhanced weighted clustering algorithm (EWCA). The goals of the algorithm are maintaining stable clustering structure, minimizing the overhead for the clustering set up, maximizing lifespan of mobile nodes in the system, and achieving good end- to-end performance. Through simulations we have compared the performance of the proposed algorithm with that of WCA in terms of the number of clusters formed, number of re-affiliations, and the lifespan of the network. The results demonstrate the superior performance of the proposed algorithm.  Keywords-ad hoc networks; clustering; cluster head; Weighted Clustering Algorithm; lifetime I. I  NTRODUCTION  Current wireless cellular networks solely rely on the wired  backbone by which all base stations are connected, implying that networks are fixed and constrained to a geographical area with a pre-defined boundary. However, ad hoc is a self- organized and self-configuring multi-hop wireless network. It does not rely on a fixed infrastructure and works in a share wireless media. So it plays a critical role in places where a central backbone is neither available nor economical to build. Due to the nature of the network, it can be widely used in temporary and emergency scenarios. For example, ad hoc networks can be applied to battlefield, disaster recovery situations, emergency search-and-rescue operations, and so on[1,2]. Furthermore, the major problems of mobile devices are gathered in the fluidity of move and limited energy. Consequently, a multi-clusters network architecture for wireless systems should to be able to dynamically adapts itself with the changing network configuration. By definition, the clustering consists in dividing the network in several groups named “clusters”. One node in each cluster is appointed as a “cluster head (CH)” and is given some responsibilities including the designation of the members of the cluster and the maintenance of the cluster. Clustering algorithm is usually  performed in two phases: clustering formation and clustering maintenance. In the clustering formation phase, elect CH among the nodes in the network. After electing CHs, some of the nodes are move out from the current cluster and attached to another cluster, this leads to the second phase, the clustering maintenance. We believe a good clustering scheme should  preserve its structure as much as possible when nodes are moving and the topology is slowly changing. Otherwise, re- computation of CHs and frequent information exchange among the participating nodes will result in high computation cost overhead. Therefore, the proposed algorithm aims to avoid excessive computation in the cluster maintenance, and the current cluster structure should be preserved as much as  possible. In this paper, we propose an enhanced weighted clustering algorithm (EWCA) to maintain stable clusters by keeping a node with weak battery power from being elected as a CH, minimizing the number of clusters, and minimizing the overhead for the clustering formation and maintenance. The rest of this paper is organized as follows. In Section 2, we review relevant several clustering algorithms proposed  previously and its limitation. Section 3 presents the proposed algorithm for mobile ad hoc networks. The analysis of  performance and simulation results of the proposed algorithm is given in Section 4. Finally, we give conclusions of this paper in Section 5 II. PREVIOUS WORK Clustering algorithms can be based on criteria such as energy level of nodes, their position, degree, speed and direction. Probably the most crucial point when dealing with clustering is the criterion how to choose the CH. The number of CHs strongly influences the communication overhead, latency, inter- and intra-cluster communication design as well as the update policy.  Several clustering mechanisms have been  proposed, namely, Lowest-ID [6], Highest-Connectivity [5], and Weighted Clustering Algorithm (WCA) [3]. The Highest-Degree uses the degree of a node as a metric for the selection of CHs. The degree of a node is the number of neighbors each node has. The node with maximum degree is chosen as a CH, since the degree of a node changes frequently, the CHs are not likely to play their role as CHs for very long. This work is supported by National High Technology Research and Development Program of China (863 Program) (Grant No.2007AA01Z211). 978-1-4244-3693-4/09/$25.00 ©2009 IEEE

05301654 2009

Embed Size (px)

Citation preview

Page 1: 05301654  2009

8/6/2019 05301654 2009

http://slidepdf.com/reader/full/05301654-2009 1/4

A Novel Enhanced Weighted Clustering Algorithm

for Mobile Networks

Chang Li, Yafeng Wang, Fan Huang and Dacheng Yang, Member, IEEE

Wireless Theories and Technologies Lab (WT&T)Beijing University of Posts and Telecommunications

Beijing, P. R. China

[email protected]

 Abstract   —Many applications of mobile Ad Hoc networks

(MANETs) adopt hierarchical structure for the scalability and

simplify of management. Clustering is the most popular method

to impose a hierarchical structure in MANETs. As a newly

proposed weighing based clustering algorithm, the Weighted

Clustering Algorithm (WCA) has improved performance

compared with other previous clustering algorithms. However,

the high mobility of nodes will lead to high frequency of re-

affiliations which will increase the network overhead andminimize the network lifetime. To solve this problem, we propose

an improved weight based clustering algorithm, namely

enhanced weighted clustering algorithm (EWCA). The goals of 

the algorithm are maintaining stable clustering structure,

minimizing the overhead for the clustering set up, maximizing

lifespan of mobile nodes in the system, and achieving good end-

to-end performance. Through simulations we have compared the

performance of the proposed algorithm with that of WCA in

terms of the number of clusters formed, number of re-affiliations,

and the lifespan of the network. The results demonstrate the

superior performance of the proposed algorithm.

 Keywords-ad hoc networks; clustering; cluster head; Weighted 

Clustering Algorithm; lifetime

I.  I NTRODUCTION 

Current wireless cellular networks solely rely on the wired  backbone by which all base stations are connected, implyingthat networks are fixed and constrained to a geographical areawith a pre-defined boundary. However, ad hoc is a self-organized and self-configuring multi-hop wireless network. Itdoes not rely on a fixed infrastructure and works in a sharewireless media. So it plays a critical role in places where acentral backbone is neither available nor economical to build.Due to the nature of the network, it can be widely used intemporary and emergency scenarios. For example, ad hoc

networks can be applied to battlefield, disaster recoverysituations, emergency search-and-rescue operations, and soon[1,2].

Furthermore, the major problems of mobile devices aregathered in the fluidity of move and limited energy.Consequently, a multi-clusters network architecture for wireless systems should to be able to dynamically adapts itself with the changing network configuration. By definition, theclustering consists in dividing the network in several groupsnamed “clusters”. One node in each cluster is appointed as a“cluster head (CH)” and is given some responsibilities

including the designation of the members of the cluster and themaintenance of the cluster. Clustering algorithm is usually

 performed in two phases: clustering formation and clusteringmaintenance. In the clustering formation phase, elect CHamong the nodes in the network. After electing CHs, some of the nodes are move out from the current cluster and attached toanother cluster, this leads to the second phase, the clustering

maintenance. We believe a good clustering scheme should  preserve its structure as much as possible when nodes aremoving and the topology is slowly changing. Otherwise, re-computation of CHs and frequent information exchange amongthe participating nodes will result in high computation costoverhead. Therefore, the proposed algorithm aims to avoidexcessive computation in the cluster maintenance, and thecurrent cluster structure should be preserved as much as

 possible.

In this paper, we propose an enhanced weighted clusteringalgorithm (EWCA) to maintain stable clusters by keeping anode with weak battery power from being elected as a CH,minimizing the number of clusters, and minimizing the

overhead for the clustering formation and maintenance.The rest of this paper is organized as follows. In Section 2,

we review relevant several clustering algorithms proposed previously and its limitation. Section 3 presents the proposedalgorithm for mobile ad hoc networks. The analysis of 

 performance and simulation results of the proposed algorithmis given in Section 4. Finally, we give conclusions of this paper in Section 5

II.  PREVIOUS WORK  

Clustering algorithms can be based on criteria such asenergy level of nodes, their position, degree, speed anddirection. Probably the most crucial point when dealing with

clustering is the criterion how to choose the CH. The number of CHs strongly influences the communication overhead,latency, inter- and intra-cluster communication design as wellas the update policy. Several clustering mechanisms have been

  proposed, namely, Lowest-ID [6], Highest-Connectivity [5],and Weighted Clustering Algorithm (WCA) [3].

The Highest-Degree uses the degree of a node as a metricfor the selection of CHs. The degree of a node is the number of neighbors each node has. The node with maximum degree ischosen as a CH, since the degree of a node changes frequently,the CHs are not likely to play their role as CHs for very long.

This work is supported by National High Technology Research andDevelopment Program of China (863 Program) (Grant No.2007AA01Z211).

978-1-4244-3693-4/09/$25.00 ©2009 IEEE

Page 2: 05301654  2009

8/6/2019 05301654 2009

http://slidepdf.com/reader/full/05301654-2009 2/4

Moreover, as the number of ordinary nodes in a cluster isincreased, the throughput drops and system performancedegrades.

The Lowest-ID chooses the node with the lowest ID as CH,the system performance is better than Highest-Degree in termsof throughput. However, a major drawback of this algorithm isthat it is solely based on the ID of nodes without consideringthe qualifications of a node possibly being elected as a CH.Also, those CHs with smaller ID suffer from the batterydrainage, resulting short lifetime of the system.

Chatterjee et al [3], propose the WCA which worksdifferently from the algorithms described above since it is onlyinvoked on demand by isolated nodes. To determine the CHnodes, the algorithm considers the ideal number of nodes that acluster can handle, the mobility, the distance between a nodeand its neighbors and the battery power, WCA assigns weightsto these different parameters, a node is selected to be the CHwhen it has the minimum weighted sum of four indices. TheCH election procedure is invoked at the time of systemactivation, and also when the current dominant set is unable tocover all the nodes. After the election, all the nodes are inclusters with a CH in each cluster and each node has a listconstituted by its neighbors and the set of all the CHs.

Due to the dynamic nature of the system considered, thenodes as well as the CHs tend to move in different directions,thus the system has to be updated from time to time. Theupdate may result in formation of new clusters. It may alsoresult in nodes changing their point of attachment from one CHto another within the existing dominant set, which is called re-affiliation. If a node detaches itself from its current head-cluster and attaches itself to another CH, then involved CHsupdate their member list instead of invoking the electionalgorithm. If the node goes into a region not covered by anyCH, then the CH election algorithm is invoked and the new

dominant set is obtained.

As a result, the overhead induced by WCA is very high. If anode moves into a region that is not covered by any CH, thenthe cluster set-up procedure is invoked throughout the wholesystem. This triggers re-affiliations and unnecessary overheadfor the well operating clusters. To solve this problem, we

  propose an improved weight-based clustering algorithmEWCA which can enhance the stability of the network bytaking the overhead of re-election CH and its energy intoconsideration.

III.  THE PROPOSED ALGORITHM 

As mentioned above, the overhead induced by WCAcontrol messages is very high, since it uses a large part of 

  bandwidth for building and maintaining the dominant set(discovery of neighbors, election process, signal strengthmonitoring), which cannot be used for useful datatransmissions. In this section, we present the proposedEnhanced Weighted Clustering Algorithm (EWCA). The

 proposed algorithm is an enhanced version of WCA to achievedistributed clustering set up and to extend lifetime span of thesystem. EWCA consists of the clustering set up and clusteringmaintenance phases.

 A.  Clustering set up

In this stage we need the position of nodes, thus it isnecessary that each node broadcasts its ID to all its neighborsin the same transmission range. Each neighbor that received the

 broadcasted message can estimate its distance from the strengthof the signal received. Global Position System (GPS) can beanother solution, however it has the disadvantage of moreenergy consumption.

Step 1: Find the neighbors of each node i (i.e., nodes within its

transmission range). This gives the degreei

d  of this node.

( ){ }∑≠∈

<=ii I i

rangeitxiidist d 

'',

', , (1)

Whererange

tx is the transmission range of  i .

Step 2: Compute the degree-difference δ  −=Δii

d  for each

node i , where δ   is the number of nodes (pre-definedthreshold) that a CH can handle ideally.

Step 3: For every node, compute the sum of the

distancesi

 D with all its neighbors as

( )∑==

n

 j jii dist  D

1, , (2)

( ) ( ) ( ) ,22

, ji ji jiy y x xdist  −+−= (3)

Where ( ) jidist 

,is the distance between node i and node j ,  x  

and  y defines the position of the node i or   j .

Step 4: Calculate the mobility of the node. The mobility isevaluated periodically in order to expect the future state of thenetwork, it is defined as follows

( ) ( ),1

t t  Dt  Dt 

M iii

Δ+−Δ

= (4)

( ) ( )( ),1

1

,∑=

=n

 j

 jiit dist 

n

t  D (5)

Wherei

M  is the relative speed of the node versus the other 

nodes, )(t  Di

is the average distance between the node i and

all its neighbors at time t , n is the number of neighbors of 

node i .

Step 5: Calculate the consumed energy of a node i ,

∑=

=q

ik ied  E 

1

, (6)

Where q is the times period which a node i acts as CH and

ik d  is the degree of a node i act as a CH at thk  times,

i E   

implies how much battery power has been consumed during a

node i act as CH, e indicates how much battery power has been consumed since we assumed that consumption of battery power is more for a CH than for an ordinary node.

Step 6: After this, we calculate the quality normalized to 1 for each node by pondering each parameter by a coefficient. Thequality of the node is a measure of its suitability as a cluster head.

Step 7: Calculate a combined weight

iiiiiE wM w Dwww

4321+++Δ= , (7)

14321

=+++ wwww , (8)

Page 3: 05301654  2009

8/6/2019 05301654 2009

http://slidepdf.com/reader/full/05301654-2009 3/4

For each node i , The coefficients321

,, www and4

w are the

weighing factors for the corresponding system parameters.

Step 8: Choose the node with a minimumi

w , to be the CH. All

the neighbors of the chosen CH can no longer participate in theelection algorithm. If a node is an isolated node, it becomes aCH automatically.

Step 9: Repeat Steps 2 to 7 for the remaining nodes not yetassigned to any cluster. According to step 6, it is a non-overlapped cluster.

The CH election takes place at the start of the simulationand when a node can no longer be covered by the dominant set.

 B.  Clustering maintenance

The second phase is the clustering maintenance. There aretwo situations that invoke the clustering maintenance. One isnode movement to the outside of its cluster boundary and theother is excessive battery consumption as a CH. When anordinary node moves to the outside of its cluster boundary, it isrequired to find a new CH to affiliate with. If it finds a newCH, it hands over to the new one. If a node moves out of itscluster and does not receive packets from any other node for aspecified period of time, it declares itself as CH. This strategymake the cluster is more stable than WCA. Fig. 1 shows thedetail of the strategy, if a node moves out of the transmissionrange of its current CH, it broadcasts a find CH message. If thenode does not receive any CH Ack message within a giventime period, it declares itself as a CH by sending a Cluster message. Fig. 1 shows an example of the normal node

 becoming a CH case.

Figure 1. the normal node becoming a clusterhead

Because a CH plays a role as coordinator in its cluster, it isfeasible to assume that a CH must process more tasks and thusneeds to consume more battery power than ordinary node, itshould work as a CH in turn. Each CH updates the amount of consumed battery power when it sends and receives packets. If 

the amount of consumed battery power becomes more than a  pre-defined threshold value, the CH resigns and the CH  becomes an ordinary node, and it broadcasts a CH resignmessage. Then all the nodes in the cluster need to determine anew CH. Each node calculates its weight value and broadcastsit using a weight information message. Then each node builds anew set and resorts to the cluster set-up algorithm given in Fig.2. After choosing an appropriate CH, other nodes affiliate withthe newly elected CH. Fig. 1 shows an example of change CHcase.

Figure 2. election of a clusterhead due to battery power consumption

IV.  PERFORMANCE ANALYSIS 

In this section, we present the performance of the proposedalgorithm EWCA obtained by simulation. The measured

 performance of the proposed algorithm was compared with thatof WCA. The simulation parameters have been listed in table 1.In the simulation experiments, N was varied between 10 and70, and the transmission range was varied between 10 and 70m. At every time unit, the nodes are moved randomlyaccording to the random way model in all possible directions in

100× 100 meters square space with velocity distributed

uniformly between 0 and maximum displacement along eachof the coordinates. This behavior is repeated for the duration of 

the simulation. We assumed a predefined threshold for eachCH which can handle (i.e. cluster size) is 10 nodes (ideal

degree). The weight values used are1

w = 0.7,2

w = 0.2,3

w =

0.05 and4

w = 0.05, parameters1

w and2

w are higher than

3w and

4w because we want properties of connectivity and

distance with neighbors to be more important for a good CHthan low mobility and battery energy. In order to study theeffect of the network density on the resulting topologies and toevaluate the cluster maintenance algorithm, we varied thenumber of the nodes inside the terrain and the power transmission range parameter.

TABLE I. SIMULATION PARAMETERS 

Parameter Meaning Value

  N Number of nodes 10-70

X*Y Size of the network  100×100

Speed Speed of the nodes 3-20 m/s

R Transmission range 10-70 m

Idea degreeIdeal number of nodes

for each cluster 10 nodes

duration Time of simulation 500 sec

w1,w2,w3,w4 Weighing factor 0.7,0.2,0.05,0.05

Fig. 3 compared the average number of clusters of EWCAwith that of WCA [3]. For this, N is set to 20 and 60 using bothalgorithms. The transmission range is varied as described

above, the number of clusters decreased as the transmissionrange increased. The results show that the proposed algorithm

  produced less clusters than WCA. As mentioned above,reducing the number of CHs strongly influences thecommunication overhead, latency, inter-cluster and intra-cluster communication design as well as execution of reorganization of clusters. As a result, our algorithm gave

 better performance in terms of the number of clusters when thenode density in the network is high.

Page 4: 05301654  2009

8/6/2019 05301654 2009

http://slidepdf.com/reader/full/05301654-2009 4/4

10 20 30 40 50 60 700

5

10

15

20

25

30

transmission range

  a  v  e  r  a  g  e  n  u  m

   b  e  r  o   f  c   l  u  s   t  e  r  s

EWCA:20 nodes

EWCA:60 nodes

WCA:20 nodes

WCA:60 nodes

 Figure 3. Average number of CHs

We simulate a system of 30 nodes on a 100 × 100 grid. Therelationship between re-affiliation count and transmission

range is illustrated in Fig. 5. The nodes can move in all possible directions with velocity varying from 0 to a maximumvalue 20m/s. Fig. 5 indicates that re-affiliation per unit time of EWCA is lower obviously in transmission range 30 to 50,compared with the WCA. EWCA and WCA have similar re-affiliation per unit time in other intervals. We explain thereason as follows: When transmission range is small, every CHonly manages few nodes, so the re-affiliation frequency is nothigh. While the transmission range becomes large, one cluster can cover a large area. Thus, it is not easy for a node to moveout of the transmission range of its CH. The benefit of decreasing number of re-affiliations mainly comes from thelocalized cluster maintenance in our algorithm.

10 20 30 40 50 60 700.5

1

1.5

2

2.5

3

3.5

transmission range

  r  e  a   f   f   i   l   i  a   t   i  o  n  c  o  u  n   t

EWCA

WCA

 Figure 4. Average number of re-affiliations

Fig. 5 shows the mutation of the minimum lifespan of nodes with respect to the number of nodes with different nodespeeds. We observe that when the mobility of nodes increased,nodes consumed more battery power. Therefore, the minimumlifespan of nodes decreased in both EWCA and WCA. As wecan see in figure 5, with the number of nodes 70 and the speed

20 km/s, the proposed algorithm produced about 43.5% moreminimum lifespan of a node than WCA. These resultsdemonstrate an important fact that more stable clusteringarchitecture may also lead to a longer network lifetime. This is

  because in our clustering algorithm, most nodes with lower   battery power will become ordinary nodes, which have fewer tasks and consumes lower battery power, thus the lifetimes of ordinary nodes will be longer.

10 20 30 40 50 60 7040

50

60

70

80

90

100

110

120

130

140

number of nodes

  m   i  n   i  m  u  m    l   i   f  e

  s  p  a  n  o   f  n  o   d  e  s

EWCA(5m/s)

EWCA(20m/s)

WCA(5m/s)

WCA(20m/s)

 Figure 5. The minimum lifespan of nodes

V.  CONCLUSION 

In this paper we have presented an improved weight basedclustering algorithm EWCA that can be applied in MANETs toimprove their stability. EWCA mainly focuses on reducing thefrequency of re-affiliation, the lifetime of the node and thenumber of cluster. Also, it has a feature to control battery

 power consumption by switching the role of a node from a CHto an ordinary node. The performance of the proposed EWCAdemonstrated that it outperforms WCA in term of re-affiliationcount, lifetime of the network and better end-to-end

 performance with less overhead.

R EFERENCES 

[1]  Yu. J. Y, Chong. P. H. J, “A survey of clustering schemes for mobile adhoc networks,” Communications Surveys & Tutorials, Vol. 7, 2005, pp.32–48.

[2]  Bricard-Vieu. V, Nasser. N, “A Weighted Clustering Algorithm UsingLocal Cluster-heads Election for QoS in MANETs,” GlobalTelecommunications Conference, 2006. pp. 1–5

[3]  Chatterjee. M., Das. S., and Turgut. D., “WCA: a weighted clustering

algorithm for mobile ad hoc networks,” Journal of Cluster Computing(Special Issue on Mobile Ad hoc Networks), 5, 2002, pp. 193–204.

[4]  Bricard-Vieu.V, Nasser. N, and Mikou, N, “A Mobility Prediction-basedWeighted Clustering Algorithm Using Local Cluster-heads Election for QoS in MANETs,” Wireless and Mobile Computing, Networking andCommunications, 2006, IEEE International Conference on. pp. 24–30

[5]   Nocetti, F. G., Gonzalez, J. S., and Stojmenovic, I, “Connectivity Basedk-Hop Clustering in Wireless Networks,” Telecommunication systems,Vol. 22, 2003, pp. 205–220.

[6]  A. Ephremides, J.E. Wieselthier and D.J. Baker, “A design concept for reliable mobile radio networks with frequency hopping signaling,”Proceedings of IEEE, Vol. 75(1), 1987, pp. 56–73.