A Novel Trajectory Clustering technique for selecting cluster heads in Wireless Sensor Networks
Wireless sensor networks (WSNs) suffers from the hot spot problem where the sensor nodes closest to the base station are need to relay more packet than the nodes farther away from the base station. Thus, lifetime of sensory network depends on these c…
Authors: Hazarath Munaga, J.V.R. Murthy, N.B.Venkateswarlu
International J ournal of Recen t Tre nds in Engineering, Issue. 1, Vo l. 1, May 2009 A Novel Traject ory Clustering technique for selecting cluster heads in Wireles s Sensor Networks Hazarath Munaga 1 , J.V.R. Murthy 1 , and N.B.Venkateswarlu 2 1 University Colle ge of Engine ering, Dept. of C SE, J.N. T.University Ka kinada, A.P, India Email: {hazarath. munaga, mjonnalagedda}@ gmail.com 2 AITAM, Dept o f CSE, Tekkali, A.P, India Email: venkat_ritc h@yahoo .com Abstract —Wireless sensor netw orks (WSNs) s uffers from the hot spot proble m w here the sensor nodes closest to the base station are n eed to relay m ore packet than the nodes farther away from the base stat ion. Thus, li fetime of sensory network depends on these closest nodes. Clustering methods are used t o extend the lifeti me of a wireless sensor network. However, current cl ustering algorithms usually utilize tw o techniques; selecting cluster heads w ith more residual energy, and rotating cluster heads periodically to distribute the e nergy consumption a mong nodes i n each cluster and lengthen the netw ork lifetime. Most of the algorithms use random selection fo r selecting the cl uster heads. Here, we propose a novel trajectory clustering technique fo r s electing the cluster heads in WSNs. O ur algorithm selects the cl uster heads based on traffic and rotates per iodically. It p rovides the firs t trajectory based clustering technique fo r selecting the cluster heads and to extenuate the h ot spot problem by prolonging the network lifetime. Index Terms —Trajectory c lustering, Wireless sensor networks, Network life time, Cluster head I. I NTRODUCTI ON Wireless sensor networks (hereinafter, WSNs) are networks of wireless nodes that ar e d eployed o ver an area for the purpose of monitoring certain pheno mena of interest. To keep spe cific ar eas under observation, WSNs deploy h undreds or thousands of integrated sensor nodes to sample data from observe d envir onment. The nodes perform certain m easurements , process the m easured data and transmit t he proc essed data to a b ase station over a wireless channel. The base station collects data fro m all the nodes, and analyzes this data to draw conclusion s about the activity in the area of interest. In practice , due to the large q uantit y of sens or node s, it is i nfeasible to recharge the batteries in WSNs. T herefore, sensor network lifeti me is a pri mary concern in sensor net work design. In literature many resear chers c oncerning pr otocols for WSNs have been prop osed to improve the e nergy consumption and the network lifetime. Those protocols can be categorize d into three classes: routing p rotoco ls, sleep-and-awake scheduling protocols, and clustering protoco ls. The ro uting pr otocols [1] [2] d etermine the energy-efficient multi -hop paths fro m eac h node to the base station. In sleep-a nd-awake scheduling protocols [3- 5], every node in the schedule ca n sleep , in order to minimize energ y consumptio n. I n cl ustering p rotoco ls [ 6] [7] data aggregatio n ca n b e used for reducing e nergy consumption. Da ta aggregatio n, also known as data fusion, ca n c ombine multiple data packets received fro m different sensor nod es. It r educes the size of the data packet by eliminating the red undancy. Wireless communication cost is also decreased by the reduction in the d ata packets [ 8]. Therefore, clustering protoc ols improve the energ y consumpti on and the net work lifeti me of the WSNs. Clustering [9 ] is a co mmonly ad opted approach in sensor networks to m anage po wer efficiently. In clustering, senso rs in the monitoring area ar e grouped into clusters; all sensor nodes within the same cluster send t heir data to t he cluster head, which then for wards the a ggregated data to the b ase station. Therefore, cluster heads “typicall y die at a n ear ly stage” [1 0]. This is sometimes calle d as the hot sp ot pro blem [11]. Without adding extra nodes or red istributing the available energ y, this p roble m i s hard to solve. For ex ample, [10] have shown that var ying the tran smission po wer of nodes, e ven considering unlimited tra nsmission ranges, d oes not so lve the hot spot problem. At the sa me time, it is al so envisioned that sensor nodes will become “extremely inexpensive” [ 12]. W hile beyond a certain node density, adding additional nodes doe s not provide any improvement regard ing se nsing, co mmunication or coverage [ 13], adding nod es might obvio usly help to increase the l ifetime of a se nsor network while providing the same service to its users, i. e. leveraging sensor value s from the same n umber of nod es. Ref. [ 14] p roposed LEACH, a well-known cluster ing protoco l for WSNs. LEACH includes d istributed clu ster formation, local processing to reduce global communication and ra ndomized rotation of cluster head s among all t he node s in the ne twork. E ach cluster select s a cluster head, which is respo nsible for aggregating collected data and sending data to base station. LEACH provides a good model that he lped to reduce infor mation overload and provides a reliable data to the end us er. Together, these features allo w LEACH to ac hieve t he desired p roperties. Ref. [ 15] the p roblem of finding a n e nergy-bala nced solution to d ata p ropagation in WS Ns using a probabilistic al gorithm was c onsidered for the first time. The lifespan o f t he networ k is m aximized b y ensurin g International J ournal of Recen t Tre nds in Engineering, Issue. 1, Vo l. 1, May 2009 that the energy consumptio n in each slice is the sa me. Sensors are assumed to be randomly distributed with uniform di stribution in a circular region or, more generally, the sec tor o f a disk. Data have to be pr opagated by t he WSN towards a sink loc ated at the center of t he disk, a nd it is sho wn that e nerg y balance can be achieved if a recurrence relation between the p robab ilities that a slice ejects a message to the si nk is satis fied. Ref. [16] proposed clustering-based routing protocol called base station co ntrolled dynamic cl ustering proto col (BCDCP), w hich utilizes a hi gh e nergy base statio n to set up cluster heads and perfor m other energy-intensive tasks, can noticea bly enhance the lifetime o f a network. Ref. [17 ] pro posed two ne w algorith ms under the na me PEDAP, which are near optimal minimu m spanning tree based wi reless ro uting sche me. T he perfor mance of t he PEDAP was compared with LE ACH a nd PEG ASIS, and showed a slightl y better networ k lifetime than PEGASIS. Ref. [ 18] p roposed a new ro uting scheme (SH ORT) , to achieve higher energy e fficienc y, network life time, and more throughput than PEGASIS, and PEDAP-PA protoco ls. T his scheme used the centralized algorit hms and r equired the p owerful b ase station. T he performance results sho wed t hat S HORT can achieve b etter “ener gy X delay” performance than the exi sting chain based data aggregation pro tocols. Ref. [19] prop osed EECR, which is an energy e fficient clustering routing algorithm. T he p erformance of t he EECR was co mpared w ith LE ACH, and showed a slightly better net work lifeti me than LE ACH. However, the unsolved pro blem of co nsiderable energ y consumption on the cl uster formation still exists. Here, we consider t he path follo wed by the node to transfer d ata to the base station a s the “tra jectory” . We used our proposed novel trajectory clustering algorithm for clustering such paths and obtained “representative trajectory” is u sed to as sign the c luster heads. T hese obtained cluster heads will be used for co mmunicating data to the base station. I n this paper, w e co ncentrated o n the rotatio n of c luster head s among all se nsor nod es to improve the li fetime of t he network b ased on the tra ffic density. We tested our p roposed method a nd found that this method enha nces the lifeti me of the net work. II. N OVEL A LGORIT HM This section consider s the WSNs consisting o f hundreds or thousands of de ployed se nsor nodes in t he sensing field. O n the basis of [20][ 16], it is assumed b y the follo wing pr operties of the WSNs to si mplify the network model. • The base statio n is located far away fro m the sensors, • The nodes have uni form initial e nergy allocation and all sensor nodes have eq ual capab ilities (data processing, wirele ss communi cation, batte ry power). • All sensor node s have various transmission po wer levels, and each nod e can cha nge the po wer level dynamically. • Each node senses t he environmen t at a fixed r ate, and • All nodes are i mmobile. The sensor nodes are geographicall y grouped into clusters and cap able of ope rating in two ba sic modes: the sensing m ode and t he cluste r head mode [ 20]. In the sensing mode, the n ode senses the task a nd sends the sensed data to its cl uster hea d. In cluster head mode, a node gathers d ata from its cl uster members, perfor ms data fusion, and transmit s t he data to the b ase station. T he base station in turn perfor ms the key task of cluster head selection. A. Cluster Head Selection Initially t he nodes will trans mit a he llo packet to the base station. After receiving hello packets fro m the nodes, usi ng the Traje ctory Clusteri ng algorith m, the b ase station computes the re presentative traj ectory b y clustering the traj ectories (here the trajectory is nothing but the pat h used b y t he nod e to tr ansfer its data to the base statio n). T he nodes of the obtained representative trajecto ry are considered as the cluster heads. Then t he base station splits the net work into c lusters (equal to the number of nodes i n the rep resentati ve traj ectory), and identifies the nodes in the repr esentative trajecto ry as t he correspo nding cluster heads. Then, the base station broad cast a m essage to the network m entioning about th e nodes and their corresp onding cl uster heads . Subsequently the nod es will use its cluster head s to transmit data . T his pr ocess will b e pe rformed per iodicall y and the cluster head s will c hange based o n the traffic. Cluster head selection ro utine contains the following stages:- 1. Base station computes the cluster heads using proposed T rajectory Clusteri ng algorith m; 2. Split the net work into N clu sters; and 3. Bro adcast message to all nod es mentioni ng cluster members and their co rrespo nding cluster heads B. Trajectory Clustering The success o f any cl ustering algorithm depe nds on t he adopted dissimilarity measure. Fo llowing section explains about the ad opted dissimilarity mea sure. Ref. [21], proposed th e us age of Eu clidean distance between ti me series of equal length a s t he measure o f their similarity. T he idea has been generalized in [2 2] for subsequence matching. In a similar way [23] used Discrete Wavelet T ransfor m and [24] used Pr incipal Component Analysis for measuring t ime series si milarity. Another approach whic h is b rought fro m i mage processing is Time Warping techniqu e and it is used in [25] to match signals in spee ch rec ognition. A si milar technique is used to find lon gest co mmon subseq uence (LCSS) of t wo sequence s using fast prob abilistic algorithms to co mpute the LCSS, a nd then def ine the distance using the le ngth o f this subsequence [26]. Here w e adop ted Hausdo rff distance [2 7] for calculating dissimilarity between trajecto ries. T he following are so me of t he definitions used in our algorithm. International J ournal of Recen t Tre nds in Engineering, Issue. 1, Vo l. 1, May 2009 Definition 1 : A trajectory (t) is represented as trj(t id ,u 0 ,u 1 ,u 2 ..,u n ) where (t id ) is a unique trajec tory id (data packet), and (u 0 ,u 1 ,u 2 ,..,u n ) is a seque nce o f nodes reflecting the spatial p osition of the nod e. Definition 2 : We d efine the spatial dissimilarity function bet ween two traj ectories t 1 and t 2 as the maximum of one way distances between t wo traj ectorie s. The one w ay distance from a trajectory t 1 to another trajecto ry t 2 is d efined as the integral of t he Hausdorf f distance between po ints of t 1 to trajecto ry t 2 divided b y the number of poin ts in t 1 ( |t 1 | ). dist ow (t 1 ,t 2 ) = dp t p d t t p h ∫ ∈ 1 ) , ( | | 1 2 1 The Hausdorff dista nce f ro m a trajectory p oint p to another traj ectory t 2 is defined as d(p, t 2 ) = min q ∈ t2 {d(p,q)} . T he dis tance bet ween trajector ies t 1 and t 2 is the maximu m of their one way distances, i.e., dist(t 1 ,t 2 ) = max{ dist ow (t 1 ,t 2 ),dist ow (t 2, t 1 ) } Clearly the di st ow (t 1 ,t 2 ) is not symmetric but dist(t 1 , t 2 ) is s ymmetric. Note that dist ow (t 1 ,t 2 ) is the inte gral of t he shortest distance s from points in t 1 and t 2 . 1) Trajecto ry Cluster Ro utine Trajec tories are groupe d into clusters usi ng the threshold . Here the threshold is considered as a maximu m value, such t hat all traj ectories are grouped into a single cluster. The trajectory cluster routine co ntains the following stage s: 1. Dissimilarit y matrix for traj ectories w ill be computed using t he Hausdor ff distance, 2. Using following Initialization Algorithm trajecto ries are groupe d into initial clusters; a. T ake first sa mple a s first cluster. Classify all the remainin g trajectorie s into this cluster if the y are within the th reshold. b. Take a traj ectory (seq uentiall y) which i s not alread y classified into any of the cluster and co nsider it as a new c luster. Take all the o ther trajectories which are not kept in any o f the clusters and keep in this clu ster if t hey s atisf y the threshold limit. c. Repeat step b till no ne w clust er is adde d. 3. Using t he follo wing R epTraj Algorithm representative traj ectories ar e computed. a. For each T rajector y o f cluster C calculate cumulative dissimilarity w ith all other trajecto ries of the same clust er C. Select the trajecto ry which i s havin g minimum cumulative d issimilarit y and take this as representative traj ectory of that cluster. 4. By co nsidering the traj ectories received fro m step 3, a s initial clus ter ca nters, using t he follo wing Re- cluster Algorithm re co mpute clusters and their representative traj ectories until there is no c hange in the represe ntative traje ctories. a. For each Trajec tory calculate dissimilarity with all the K r epresentative trajecto ries a nd classi fy to th e cluster for which dissimilari ty is lo w. b. Re-calculate repr esentative trajecto ries using RepTraj Algorithm. C. Data co mmunica tion phase There are three steps during the data co mmunication phase: data collection, data fusion and data trans mission. Initially eac h se nsor nod e tr ansmits t he se nsed information to its clus ter head at the time slot assigned b y its cluster head. In o rder to save itself ener gy, the node will close transmit part durin g the time slot, which is not required to it. Once data from all sensor nod es hav e been received, the cluster h ead performs data fusion o n the collected data and reduces the a mount of raw data that need to send to the ba se stati on. Once the data gat hering and data fusion are completed , the cluster head sends the compressed d ata to the base station. As mention pr eviously, a ll the nodes c an work as a cluster head s. Due to this, an y nod e can be come a clu ster head or a cluster member . At ea ch turn the cluster head calculates available po wer and compares with the cluster members. W henever the cluster heads p ower b ecomes less than the minimum p owe r holding, then the cluster heads infor ms to it s cluster members and a ssigns the maximum p ower holding cl uster member a s t he c luster head and in tur n co mmunicates to the base station . Whenever the cl uster head is changed base station r epeats the cluster findi ng process a nd modifies the cl usters. III. E XPERIMENTAL W ORK To e valuate the perfor mance o f o ur algorit hm, it has been simulated and co mpared its performance with energy efficie nt clustering r outing (hereafter , EECR). Before the si mulation and r esults are introd uced, the radio model and so me important p ara meters [19] used i n simulation have b een describe d. A. The ra dio model We have used bo th the free-space propagation model and the two-ra y ground p ropagation model to appro ximate path loss susta ined because o f wireles s channel tra nsmission. Given a threshold transmission distance of d 0 , the free-space m odel i s used wh en d < do, and the two-ray model is applied for cases when d ≥ d 0 . Using these two models, the transmit energy costs for the transfer of a b-bit data message between two nodes separated b y a distance of d meters is give n: if d
Original Paper
Loading high-quality paper...
Comments & Academic Discussion
Loading comments...
Leave a Comment