Content uploaded by Li Zhang
Author content
All content in this area was uploaded by Li Zhang on Nov 26, 2015
Content may be subject to copyright.
9781424428007/09/$25.00©2009IEEE ICIEA2009
ANovelClusteringAlgorithmforAdHocNetwork
LiGao
1
,DejunMu
2
,YuexianWang
2
1
CollegeofComputer
NorthwesternPolytechnicalUniversity
Xi’an,P.R.China
gaoli@nwpu.edu.cn
GuoqingZhang
2
,LiZhang
2
2
CollegeofAutomation
NorthwesternPolytechnicalUniversity
Xi’anP.R.China
gniq@mail.nwpu.edu.cn
Abstract— In recent years, various types of ad hoc routing
protocols have been studied in the mobile ad hoc networks.
Specifically,the clustering hierarchical routing algorithms have
beendevelopedtoincreasethesystemperformance.Hierarchical
structure has inevitably brought some drawbacks, maintaining
the hierarchical structureneeds morecomplicated clusterheads
selection algorithm, whichmay resultin the cost of maintaining
cluster structure. This paper explores a novel clustering
algorithm for ad hoc network. This algorithm is based on the
higher stability of the cluster structures and the lower cost of
maintainingtheroute,andtheconceptof“ExceptionDegree”is
introducedintothealgorithmwhichcanjudgewhenevertostart
to adjust cluster structures in terms of the exception degree.
Analysis and experiments demonstrate the features that the
frequencyof changing clusterheads is lowerand the stabilityis
higher.
IndexTerms—Adhocnetwork,clustering,routingalgorithm,
stability
I. INTRODUCTION
Ad hoc network is a mobile wireless selforganizing
networkwhichisalsoinfrastructureless.This kindofnetwork
is characteristic of being distributing, dynamic, autonomous
andmobile.Adhocnetworkhastwosystemstructures:flatand
hierarchical structure. In flat structure, the functions of all
nodesaresimilar,andtheirstatusesarethesame.Thenetwork
hasstrongerrobustness,butwhennodesareincreasingrapidly,
especiallywhenthenodesaremoving,therearisesomedefects,
such as lower processing quality, higher costs of control,
frequentlyinterruptedroutingandsoon,whichleadstosharply
decline on network quality, so it is mainly used in medium
sized and smallsized networks.Hierarchical structure applies
clustering to divide the whole network into several clusters.
Theclusterisregardedasacentralpointtoensurethestability
and recombination of the cluster structure. The backbone
networki scomposed of cluster heads, which can achieve the
communicationacrosscluster.
Thehierarchicalstructurehassomeadvances,suchasbetter
expandedqualityandunlimitedscaleofnetwork.Byincreasing
the number of clusterh eads or network hierarchy to improve
the capacity of network, on the same condition of network
scale, the route and control costs of the hierarchical structure
arelowerthanthatofflatstructure,andhierarchicalstructureis
easier to achieve the mobile management and network local
synchronization.However,hierarchicalstructurehasinevitably
broughtsomedrawbacks,maintainingthehierarchicalstructure
needs more complicated cluster heads selection algorithm,
which may result in the cost of maintaining cluster structure.
When topology is changing, especially when the nodes are
moving intensely, updated frequency of cluster structure is
climbingsharply,whichbringsaboutanumberofmaintaining
costsandcausesadeclineofnetworkquality.
Focusingontheproblemthatroutingcostishigh,thispaper
proposesanovelclusteringalgorithmbasedonMinIDwhich
also includes the advantage of MinID algorithm comprising
cluster simply, and the concept of “exception extent” is
introduced into the algorithm which can judge whenever to
start toadjust clustering structures according to theexception
degreeinordertokeepnetworkoriginalstructure.Forexample,
originalcluster retainmore nodes, and changingfrequency of
clusterh eads in cluster structure isl ower, andthe stability of
clusteringstructure can be improved, andthe control costs of
maintainingclusterstructuredropdramatically,allofthesecan
improvenetworkqualityeffectively.
The reminder of this paper is organized as follows. In
Section II we investigate the related work. Our proposed
improved clustering algorithm is discussed in Section III.
Section IV showsthe performance evaluation results. Finally,
SectionVconcludesthispaper.
II. RELATED WORK
A. MinIDClusteringAlgorithm
MinID clustering algorithm is a simple clustering
algorithmwhichdistributesanonlyIDtoeachnodeandselects
theMinIDnodeto beclusterheads.Thisclusteringalgorithm
workload is lower. It can be achieved conveniently with a
quicker convergence. However, when the nodes are moving
intensely, we need to frequently implement clustering
algorithm to maintain topology structure, which brings high
costs especially when two clusterh eads move to each other’s
communication range and clusters combination occurs.
AccordingtothepropertyofMinIDclusteringalgorithm,this
may cause that nodes in the whole cluster give up current
cluster heads, and reimplementinspection of forming cluster
and algorithm process. As fundamental algorithm of routing
protocol,thiscostofrebuildingclusterstructureisnotallowed.
B. HighestDegreeClusteringAlgorithm
Aiming at the problem that MinID clustering algorithm
may lead to change frequently in cluster structure, Lim and
Gerla propose a MinID clustering thought thatis to improve
445
adjustment mechanism of cluster structure. When cluster
structureischanging,wecouldnotbaseonMinIDclustering
algorithmanymoretorecluster,butkeepnodeswhichishigh
connectiondegreeand onehopneighbornodesintheoriginal
cluster.Eachnodedetectswhetherlocalhostisintherangeof
one hopof thehi ghest degree nodesor not, ifnot, we dothe
“leavecluster”operation.
The advantage of this algorithm is that the number of
clusterinthenetworkisfewer,thatisaveragenumberofhops
between source nodeand destination node is fewer, so it can
decrease timelapse of packet going back and forth. This
algorithm can improve the stability of clustering structure
adjustment in principle, but the precondition itrequires does
not have inevitable feasibility in practice. It shows in two
aspects:
If there are two or several highest degree nodes in the
cluster structure, Auxiliary judging criteria should be
introducedinthetestingandjudgingoftheleavingcluster.As
distributed algorithm, it proposes much higher requests on
synchronization of information and accuracy of judgment
amongnodes.
Iftherearecommunicationrangewherenodesdepartfrom
othernodesinthecluster,especiallyifthereareseveralnodes
departing from each other in the same time, nodes cannot
acquire or speculate the degree of other nodes, and the
judgment on highest degreehas to be unsuccessful, so nodes
cannotdealwithclusterstructureadjustmentcorrectly.
III. IMPROVED CLUSTERINGALGORITHM
A. Algorithmdesign
The improved clustering algorithm falls into cluster
formingalgorithmandclusterstructuremaintainingalgorithm.
Owning to the advantages of MinID clustering algorithm
which comprises cluster quickly and is simple, efficient and
doesnotrelyonanyouterauxiliarycondition,westillusethe
thoughtofMinIDclusteringalgorithminthestageofforming
cluster to be facing implementation algorithm. When cluster
structurestartsto change,theimprovedalgorithmbeginswith
the improving stability of cluster structure and declining the
costsofmaintainingclusterstructure,andkeepsnodesasmany
aspossibleintheoriginalclusterinorderthattheaveragetime
of cluster heads serving as is longer and the frequency of
changing cluster heads is lower. In the stage of maintaining
clusterstructure,weapplythemaintainingalgorithmbasedon
the exception degree. According to the exception degree we
judgewhatnodesappearexceptionlinks,thatis,wedealwith
the nodes which are not in the original cluster to make them
leave cluster, and keep cluster heads and other nodes in the
originalcluster.Sowemaintainthestabilityofclusterstructure
and decrease the maintaining costs which are caused by
changingclusterheads.
B. DefinitionConcerningAlgorithmandHypothesiss
The flat ad hocn etwork which consists ofn freemoving
nodes is abstracted to be a connected digraph G=(V,E). V
represents a set of network nodes, and E represents a set of
twoway link among nodes. The distribution of network has
random property. Because of short distance the nodes in the
samesmallzonehavephysicspositioncorrelation,andwecan
usethephysicspositionofthenodestobuildasubnet.
Definition 1 In G=(V,E),n odes x,y V, if there is a side∈
betweennodexandy,thatmeansthereisaninfinitetransport
linkbetweennodexandy.
Definition2Thedistancebetweentwonodesd(x,y)isthe
minimumnumberofhopsbetweennodexandy.
Definition 3 A cluster Ci(Ci<V ,i=1,2…) is consist of
nodes, Regarding 2 random nodes x,y Ci, d(x, y)≤2 and∈
V= Ci.∪
Definition4Ifx,y Ciandd(x,y)≥3(ifnot,thedistance∈
is255hops),Thereisaonehopexceptionlinkbetweennodes
xandywhichareinthesamecluster.
Definition5Ifnodex V,ED(x)representsthenumberof∈
hop of exceptionlink, whose number is exception degree. If
anynodeincertainclusterx Ci,ED(x)=0,wecansaythatthe∈
clusterstructureisstable.
C. TheDscriptionofAlgorithmEquations
1) Algorithmofselectingclusterheads
Theselectionofclusterheadsconsistsofstageofforming
cluster algorithm and cluster structure maintaining algorithm.
In the stage of forming cluster, wer ecommend theminimum
ID node which adheres to d(x,y)=1 as the cluster heads, and
dividetheflatadhocnetworkwhichconsistsofnfreemoving
nodesinton clusters,V= Ci,and therelationshipamongthe∪
clustersiscrosslap.
Withthemovingofnodes,clusterstructurestartstochange
andcometothestageofmaintainingclusterstructure.Thenwe
calculate the node’s exception degree. If there is a node x
whose exception degree is ED(x)>0 in the cluster Ci, we
believethatxisa exceptionnodeanditisnecessarytoadjust
clusterstructure, or thecluster structure comes tostable stage
anditisnotnecessarytoadjustclusterstructure.
There are two approaches in the algorithm to acquire
exception link information. One is exception link flooding.
When anode finds that there are exception links from local
nodetoothernodesintheneighbornodeslist,itbroadcaststhis
messageright now.The lifecycle ofthis flooding massageis
maximum number of hop in the exception link of originator
node except 255(a broken circuit); theother is exception link
peculation. When a node findsthat there are links which are
brokencircuitsfromtheexceptionlinkmessagewhichis stored
bylocalhosts,wecanknowanendofthislinkmustdisconnect
to thelocal hosts. We speculate on the exception link of this
nodeand gather statisticson all exceptionlinks of destination
whichisthisnodeinordertoacquireasubsetorarealproper
setoftheexceptionlinkofthisnode.
2) Principleonadjustingclusterstructure
a) Theleavingclusteroperationonexceptionnodes
If ED(x) of node x is maximum exception degree in all
nodes of the cluster, we make x do the “leaving cluster”
operation.Ifxisthisnode,wechangethestateofnodexinto
noclusterheads andclean neighbor node lists, exceptionlink
listsand exceptiondegreelists;if xisnotthis node, weclean
nodexfromlistsofnodeswhichareinthesamecluster,andat
446
thesametimediscardtheitemcorrespondingthenodexfrom
exceptionlinklistandupdateexceptiondegreelist.
Ifmaximumintheexceptiondegreelistisnotuniqueand
thereisnoothernodesexceptthesemaximumnodes,thereare
generally broken circuits in the cluster, and we make all
maximumdisconnectnodesdotheleavingclusteroperation.
b) Treatmentoftwoclusterscombination
When two clusters come into mutual one hop
communication range, these clusters will be combined. The
cluster which hasmore nodes will be new cluster heads. The
other cluster will flood combination message, and when the
nodesinthisclusterreceiveclustercombinationmessagethey
will change cluster heads into new cluster heads. New cluster
detectsexceptiondegreeanddocorrespondingtreatment.
c) Thetreatmentofaddingnewnodesinthecluster
When anode x which does not confirm its cluster heads
entersa certainclusterCi,wecalculateonanynodefirstly,if
y Ci, and d(x, y)<3, this node and any other node in the∈
clusterare in the two hopsrange, andthe node x whichjoins
cluster Ci would not result in another adjustment on cluster
structure,thenwedealwithxenteringclusterCi.
3) Selectionalgorithmongateway
Generally speaking, the communication between two
neighborclusterheadscarriesouttwohopcommunicationvia
Gateway,ifnot,weadoptdistributedGateway.Thealgorithm
principleonGatewayselectionismainlybasedon“firststate,
firstlywin”.
TherearetwotableskeepingintheGatewaynodes.Oneis
GatewayTable;the otherisCHTablewhichkeepsalladdress
ofclusterheadsnodeswhereGatewaycanarrivedirectly.
For two neighbors cluster heads which can arrive within
two hops range, as Fig. 1 shows. Node D is the member of
cluster heads A. If D receives broadcasting information from
clusterheadsBandjudgethatitisintheonehoprange,firstly
DchangethenodestatusintoGW_READY,thenitsearchesin
theGatewayTableoflocalhosttofindwhetherotherGateway
nodesofBhavesignedasGatewaynodeswhichhavearrived
at cluster heads B, such being the case, node D gives up
becomingGatewayan dchange its statusinto ORDINARY to
beordinary status;otherwisenodeDbecomesGatewaynode,
setting its status asGATEWAY, updatingits GatewayTable,
andsendingGatewayinformationtotheGatewaynodeswhich
D can arrive at. The cluster heads node adds D to its
GatewayTable.
Fortwoneighborclusterheadswhichcannotarrivewithin
twohopsrangeandhavethreehopsavailableroute,theywill
use distributed Gateway to exchange data. Cluster B and C
cannotcarryoutexchangingdatawithintwohops,andnodeE
belongs to Cluster B. We searchGateway nodes signedas B
and C in the GatewayTable of local nodes, if there are such
nodes,itshowsthattherehasbeenavailableGatewayandnode
B gives up taking on the responsibilities to inspection on
Gateway; if not, node E sets its Status as
DISTRIBUTE_READY.WhennodeFfindsthestatusofnode
E has changed into DISTRIBUTE_READY, F searches
whether there are other nodes whose status are
DISTRIBUTE_READYinthelocalcluster,ifnot,nodeFalso
change status into DISTRIBUTE_READY, and these two
nodes sign themselves as DISTRIBUTE_Gateway to be
distributed Gateway of cluster B and C, then broadcast this
informationtothecluster.
Figure1. Gatewayselectionfigure
IV. ALGORITHM ANALYSISAND SIMULATION
A. AlgorithmAnalysis
Fig.2isacomparisonclusterreconfigurationresultofMin
ID clustering algorithm and improved algorithm. Fig. 2(a)
generates initialized cluster structure based on MinID
clusteringalgorithm.Whenclusterheadsnodesstarttomove,
althoughthesenodesarestillmutuallyconnective,accordingto
MinIDclusteringalgorithmthenodesin the same clusterare
redivided into three new clusters shown as in Fig. 2(b). If
node 5 moves to and fro, the cluster structure will change
intensively.
For cluster structure shown asin Fig. 2(a),th e exception
degreeofeachnodeis0.However,inFig.2(b),becausethere
are exception links when nodes are moving. Making node 9
whose exception degree is maximum leave cluster, cluster
structureadjusts to two clusters shown as in Fig. 2(c). Most
nodes are kept in the original cluster and cluster structure
maintainsstabilitytoacertainextent.
Figure2. ClusterreconfigurationreusltofMinIDclusteringalgorithm
447
Figure3. Averageclusterheadchangingfrequency
Figure4. Timeasclusterhead
B. ExperimentSimulation
For cluster structure shown asin Fig. 2(a),th e exception
degreeofeachnodeis0.HWeusenetworksimulationtoolNS
2tocarryout simulationrespectivelyon MinIDDclustering
algorithm and improved algorithm. We select 1000m×1000m
square zone as simulation scene. There are 50 points. This
simulationadjustscommunicationdistanceofnodesviatesting
RXThreshholdsetinPhysicallayerinscenarioandchangethe
connectionamongnetworknodes.Themovingmodelofnodes
is “random waypoint”, themoving speed range sets020m/s,
andthewholesimulationtimeis400s.Becausewemainlytest
the stability of clustering algorithm, we add a CBR data
businessinsceneandcarryoutthesimulationbetweenanytwo
nodes. In addition, in this protocol the interval time of
periodicallybroadcastingnodeinformationis 5s.Thecontents
whichareinspectedbythesimulationaretheaveragenumber
ofchangingclusterheadsandtheaveragetimeofeachcluster
headsoccupying.ThesimulationresultsareshownasinFig.3
andFig.4.
As we can see from the figures, compared with MinID
clustering algorithm, in the improved algorithm the average
frequency of cluster heads changingis lower andth eaverage
time of each cluster heads occupying is longer. The former
shows that the number of nodes leaving original subclusters
and joining new subclusters is low; the later shows that the
clusterstructureisstable.If the nodecommunicationrangeis
largerorsmaller,thedifferencebetweentwoalgorithmsisless
thanthatinordinaryrange.Consideringthatthemoveofnodes
itself has less influence on nodes connection if the node
communicationrange is largerorsmaller,thisphenomenonis
reasonable. Experiment data shows that the improved cluster
algorithmindeedhasmorestablequality.
V. CONCLUSION
This paper proposes a new improved algorithm based on
MinIDwhichalsoincludestheadvantageofMinIDalgorithm
forming clustersimply, and theconcept of“exception extent”
wasintroducedintothealgorithmwhichcanjudgewheneverto
start toadjust clustering structures according to theexception
degree.Thealgorithmenhancesthestabilityofclusterstructure
by improving adjustment mechanism of cluster structure
changing, and decreases the time and costs of distant rout
discovering. The algorithm analysis and experiment indicate
thattheaveragenumberofclusterheadschangingislower,the
averagetimeofclusterheadsoccupyingislonger,andstability
ofclusterstructureisimproved.
ACKNOWLEDGMENT
We thank Wu Chen for providing us with the simulator
andhissupportofourwork.
REFERENCES
[1] C.C.Chiang, H.K.Wu, W.Liu a nd M.Gerla. Routing in Clustered
Multihop, Mobile Wireless Networks with Fading Channel [A],
ProceedingsofSICON,1997
[2] C. R. Lin, GERLA M. Adaptive Clustering for Mobile Wireless
Networks [J]. IEEE JOURNAL ON SELECTED AREAS IN
COMMUNICATIONS,1997,Vol.15,NO.7
[3] Yu J, Chong P.A survey of clustering schemes for mobile Ad hoc
networks[J].IEEECommunicationsSurveys,2005,7( 1):32 48.K.
[4] PBasu,NKhan,TDCLittle.AMobilityBasedMetricfor Clusteringi n
MobileAdHocNetworks[A].ProcofIEEEICD CS2001Workshopon
WirelessNetworksandMobileComputing[C].2001.413418
[5] Perkins,C,BeldingRoyer,EandDa s,S. Ad hoc OnDemand Distance
Vector(AODV)Routing.IETFRFC3561.2003.
[6] M Chatterjee, S K Das, D Turgut. WCA: A Weighted Clustering
Algorithm for Mobile Ad Hoc Networks [J]. Journal of Clustering
ComputingIEEE,2002,5(2):193204
[7] S Basagni. D istributed Clustering for Ad Hoc Network s[ A]. Proc of
Int’l Symp on Parallel Architectures, Algorithms a nd Networks[C]
.1999.310315.
[8] Shyan Hwang, Chang Chieh Liu, Chiung Ying Wang. Link Stability
Based Routing and Clustering in Ad Hoc Wireless Networks Using
FuzzySetUsingFuzzySetTheory[J].InternationalJournal ofWireless
InformationNetworks,2002,9(3):201212
[9] M Chatterjee, S K Das, D Turgut. WCA: A Weighted Clustering
Algorithm for Mobile Ad Hoc Networks [J]. Journal of Clustering
ComputingIEEE,2002,5(2):193204.
[10] Xu Ha o, MU De jun, LI Li xin .On D emand Cluster ing Routing
algor ithm in Ad hoc networks .Computer Engineer ing and
Applications,2007,43(14):3 6.
448