A Fault Tolerant Trajectory Clustering (FTTC) for selecting cluster heads inWireless Sensor Networks

International Journal of Computational Intelligence Research. ISSN 0974-1259 V ol.4, No.X (2008), pp. XXX–XXX c  MIR Labs http://www .ijcir .com A F ault T olerant T rajectory Clustering (FTTC) for selecting cluster heads in W ireless Sensor Networks Hazarath Munaga 1 , J. V . R. Murthy 1 and N. B. V enkateswarlu 2 1 Department of Computer Science and Engineering, UCEK J.N.T .U Kakinada, Andhra Pradesh, India { hazarath.munaga,mjonnala gedda } @gmail.com 2 Department of Computer Science and Engineering, AIT AM, T ekkali, Andhra Pradesh, India venkat ritch@yahoo.com Abstract : Wir eless sensor networks (WSNs) suffers from the hot spot problem where the sensor nodes closest to the base station ar e need to relay mor e pack et than the nodes farther away from the base station. Thus, lifetime of sensory network depends on these closest nodes. Clustering methods are used to extend the lifetime of a wireless sensor network. However , current clustering algorithms usually utilize two techniques; selecting cluster heads with more residual energy , and rotating cluster heads periodically to distribute the energy consumption among nodes in each cluster and lengthen the network lifetime. Most of the algorithms use random selection for selecting the cluster heads. Here, we propose a Fault T olerant T rajectory Clustering (FTTC) technique for selecting the cluster heads in WSNs. Our algorithm selects the cluster heads based on traffic and rotates periodically . It provides the first F ault T olerant T rajectory based clustering technique for selecting the cluster heads and to extenuate the hot spot problem by prolonging the network lifetime. Keyw ords : Fault T olerant T rajectory Clustering, T rajectory clustering, Wireless sensor networks, Network life time, Cluster head. I. Introduction W ireless sensor networks ( hereinafter , WSNs) are networks of wireless nodes that are deployed over an area for the pur- pose of monitoring certain phenomena of interest. T o keep specific areas under observation, WSNs deploy hundreds or thousands of inte grated sensor nodes to sample data from observed en vironment. Although, these devices are not very accurate and reliable indi vidually , their deployment in large number enhances their accuracy and reliability . In addition WSNs can provide area coverage that was not possible with other wireless networks. The y can be deployed in extremely hostile en vironments, such as near volcanically activ e sites, inside a chemical industry , or probable disaster areas or in mainly inaccessible en vironments. WSNs hav e numerous advantages. They are easier and faster to deploy than any wired network. The y hav e a large cov erage area and longer range. They hav e higher degree of fault tolerance than wireless networks, because failure of one or more sensors or nodes does not ef fect the operation of the network and mainly they are self-configuring. The nodes perform certain measurements, process the measured data and transmit the processed data to a base station ov er a wireless channel. The base station collects data from all the nodes, and analyzes this data to draw conclusions about the activity in the area of interest. In practice, due to the large quantity of sensor nodes, it is infeasible to rechar ge the batteries in WSNs. Therefore, sensor network lifetime is a primary concern in sensor network design. In literature man y researchers proposed various protocols to reduce the energy consumption and improve the network lifetime of WSNs. Those protocols can be categorized into three classes: routing protocols, sleep-and-a wake scheduling protocols, and clustering protocols. The routing proto- cols [1] [2] [3] [4] determine the energy-ef ficient multi-hop paths from each node to the base station. In sleep-and-aw ake scheduling protocols [5] [6] [7], every node in the schedule can sleep, in order to minimize energy consumption. In clustering protocols [8] [9] [10] [11] [12] [13] [14] [15], data aggregation can be used for reducing ener gy consumption. Data aggregation, also known as data fusion, can combine multiple data packets received from dif ferent sensor nodes. It reduces the size of the data packet by eliminating the redundancy [16] [17] . W ireless communication cost is also decreased by the reduction in the data packets. Hence, by reducing the energy consumption clustering protocols increases the network lifetime. Clustering [18] is a commonly adopted approach in sen- sor networks to manage power efficiently . In clustering, sensors in the monitoring area are grouped into clusters; all sensor nodes within the same cluster send their data to the cluster head, which then forwards the aggregated 1 data to the base station. In this way clustering reduces the ov erhead and increases the network lifetime; on the other hand, the disadvantage is due to heavy usage of cluster heads “typically cluster heads die at an early stage” [19]. This is sometimes called the hot spot problem [20]. W ithout adding extra nodes or redistributing the av ailable energy , this problem is hard to solve. F or example, Ref. [19] has shown that, with v arying transmission power of nodes and even considering unlimited transmission ranges does not solve the hot spot problem. At the same time, it is also en visioned that sensor nodes will become “extremely inexpensi ve” [21]. While beyond a certain node density , adding additional nodes does not pro vide any improvement regarding sensing, communication or cov erage [22], adding nodes might obviously help to increase the lifetime of a sensor network while pro viding the same service to its users, i.e. leveraging sensor values from the same number of nodes. Ref. [14] proposed the randomized clustering algorithm to organize sensors into clusters in a wireless sensor network. Computation of the optimal probability of becoming a cluster head was presented. Ref. [23] defined the maximum cluster-lifetime problem, and they proposed distributed, ran- domized algorithms that approximate the optimal solution to maximize the lifetime of dominating sets on wireless sensor networks., [24] considered the k-domatic partition problem, and the y proposed three deterministic, distrib uted algorithms for finding large k-domatic partitions. Ref. [8] proposed LEA CH, a well-kno wn clustering protocol for wireless sensor netw orks. LEA CH includes distributed cluster formation, local processing to reduce global commu- nication and randomized rotation of cluster heads among all the nodes in the network. Each cluster selects a cluster head, which is responsible for aggreg ating collected data and sending data to base station. LEA CH provides a good model that helped to reduce information overloaded and provides a reliable data to the end user . T ogether , these features allow LEA CH to achie ve the desired properties. Also, an improved scheme of LEA CH was proposed, named LEA CH-C [9]. In LEA CH-C, a centralized algorithm at the base station makes cluster formation. Ref. [25] deals with the problem of finding an energy- balanced solution to data propagation in WSNs using a probabilistic algorithm was considered for the first time. The lifespan of the netw ork is maximized by ensuring that the ener gy consumption in each slice is the same. Sensors are assumed to be randomly distributed with uniform distri- bution in a circular region or, more generally , the sector of a disk. Data have to be propagated by the WSN to wards a sink located at the centre of the disk, and it is shown that energy balance can be achiev ed if a recurrence relation between the probabilities that a slice ejects a message to the sink is satisfied. Ref. [10] proposed PEGASIS. PEGASIS makes a commu- nication chain using a T raveling Sales Person heuristic. In PEGASIS, nodes are organized into a chain using a greedy algorithm so that each node transmits to and recei ves from one of its neighbors. A randomly selected node from the chain will forward the aggregated data to the base station, thereby reducing per round energy expenditure compared to LEA CH. Ref. [26] proposed clustering-based routing protocol called base station controlled dynamic clustering protocol (BCDCP), which utilizes a high ener gy base station to set up cluster heads and perform other energy-intensi ve tasks, can noticeably enhance the lifetime of a network. Ref. [27] proposed two ne w algorithms under the name of PED AP , which are near optimal minimum spanning tree based wireless routing schemes. The performance of the PED AP was compared with LEACH and PEGASIS, and showed a slightly better network lifetime than PEGASIS. Ref. [28] proposed a new routing scheme; called SHOR T , to achiev e higher energy ef ficiency , network lifetime, and more throughput than PEGASIS, and PED AP-P A protocols. This scheme used the centralized algorithms and required the powerful base station. The performance results showed that SHOR T can achiev e better “energy X delay” performance than the existing chain based data aggre gation protocols. Ref. [11] proposed HEED, by extending LEA CH and considering range limits of the wireless communication and intra-cluster communication cost. The probability for each node to become a tentativ e cluster head depends on its residual energy , and all the tentati ve heads in which are competing for becoming the final cluster heads. The final cluster heads are selected according to the intra-cluster communication cost. HEED terminates within a constant number of iterations, and achie ves fairly uniform distrib ution of cluster heads across the network. Ref. [29] proposed EECR, which is an energy ef ficient clustering routing algorithm. The performance of the EECR was compared with LEA CH, and sho wed a slightly better network lifetime than LEA CH. Howe ver , the unsolved problem of considerable energy con- sumption on the cluster formation still exists. Here, we con- sider the path follo wed by the node/sensor to transfer data to the base station as the “ tr ajectory ”, and using the proposed trajectory clustering [30] [31], we cluster the trajectories and obtain the “ r epresentative trajectory ”. Then the nodes lies in the representativ e trajectory are considered as the cluster heads and the obtained cluster heads will be used for commu- nicating data to the base station. Moreov er , we concentrated on the rotation of cluster heads among all sensor nodes to im- prov e the lifetime of the network based on the traf fic density . W e tested our proposed method and found that this method considerably enhances the lifetime of the network. II. Novel Algorithm This section considers the wireless sensor networks consist- ing of hundreds or thousands of deployed sensor nodes in the sensing field. On the basis of [10] [26], it is assumed by the follo wing properties of the wireless sensor networks to simplify the network model. • The base station is located far a way from the sensors, • The nodes hav e uniform initial ener gy allocation and all sensor nodes have equal capabilities (data processing, wireless communication, battery power). • All sensor nodes hav e various transmission power lev- els, and each node can change the power le vel dynami- cally . • Each node senses the en vironment at a fixed rate, and • All nodes are immobile. The sensor nodes are geographically grouped into clusters and capable of operating in two basic modes: the sensing mode and the cluster head mode [10]. In the sensing mode, the node senses the task and sends the sensed data to its clus- ter head. In cluster head mode, a node gathers data from its cluster members, performs data fusion, and transmits the data to the base station. The base station in turn performs the key task of cluster head selection. A. Cluster head selection Initially the nodes will transmit a hello packet to the base station. For calculating the shortest path, we can consider various Quality of service (QoS) parameters like spatial distance, processing delay , av ailable bandwidth, allocated buf fer space etc., in this phase of research we consider only spatial distance. Then, we used Dijkstra’ s [32] algorithm for finding the shortest path of the hello packet. After receiving hello packets from the nodes, using the T rajectory Clustering algorithm, the base station computes the repre- sentativ e trajectory by clustering the trajectories (here the trajectory is nothing b ut the path used by the node to transfer its data to the base station). The nodes of the obtained representativ e trajectory are considered as the cluster heads. Then the base station splits the network into clusters (equal to the number of nodes in the representativ e trajectory), and identifies the nodes in the representative trajectory as the corresponding cluster heads. Then, the base station broadcast a message to the network mentioning about the nodes and their corresponding cluster heads. Subsequently the nodes will use its cluster heads to transmit its data. This process will be performed periodically and the cluster heads will change based on the traffic. Cluster head selection routine contains the following stages:- 1. Base station computes the cluster heads using TC algo- rithm; 2. Split the network into N clusters; and 3. Broadcast message to all nodes mentioning cluster members and their corresponding cluster heads B. T rajectory Clustering The success of any clustering algorithm depends on the adopted dissimilarity measure. Follo wing section explains about the adopted dissimilarity measure. Agrawal et al., [33], proposed the usage of Euclidean distance between time series of equal length as the measure of their similarity . The idea has been generalized in [34] for subsequence matching. In a similar way [35] used Discr ete W avelet T ransform and [36] used Principal Component Analysis for measuring time series similarity . Another approach which is brought from image processing is T ime W arping technique and it is used in [37] to match signals in speech recognition. A similar technique is used to find longest common subsequence (LCSS) of two sequences using fast probabilistic algorithms to compute the LCSS, and then define the distance using the length of this subse- quence [38]. In [39] suggested this technique to measure the similarity of time-series data in data mining. Here we adopted Hausdorff distance [40] for calculating dissimilarity between trajectories. The follo wing are some of the definitions used in our algorithm. 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 trajectory id (data packet), and ( u 0 , u 1 , u 2 ,.., u n ) is a sequence of nodes reflecting the spatial position of the node. Definition 2 : W e define the spatial dissimilarity function be- tween two trajectories t 1 and t 2 as the maximum of one way distances between two trajectories. The one way distance from a trajectory t 1 to another trajec- tory t 2 is defined as the integral of the Hausdorff distance between points of t 1 to trajectory t 2 divided by the number of points in t 1 ( | t 1 | ). dist ow (t 1 , t 2 ) = 1 | t 1 | Z p ∈ t 1 d h ( p, t 2 ) dp The Hausdorf f distance from a trajectory point p to another trajectory t 2 is defined as d ( p, t 2 ) = min q ∈ t 2 { d ( p, q ) } .The distance between trajectories t 1 and t 2 is the maximum of their one way distances. dist(t 1 , t 2 ) = max { dist ow (t 1 , t 2 ) , dist ow (t 2 , t 1 ) } Clearly the dist ow ( t 1 , t 2 ) is not symmetric b ut dist ( t 1 , t 2 ) is symmetric. Note that dist ow ( t 1 , t 2 ) is the integral of the shortest distances from points in t 1 and t 2 . 1) F ault T olerant T rajectory Cluster Routine T rajectories are grouped into clusters using the thr eshold . Here the threshold is considered as a maximum value, such that all trajectories are grouped into a single cluster . The tra- jectory cluster routine contains the following stages: 1. Dissimilarity matrix for trajectories will be computed using the Hausdorff distance, 2. Using following Initialization Algorithm trajectories are grouped into initial clusters; (a) T ake first sample as first cluster . Classify all the remaining trajectories into this cluster if they are within the threshold. (b) T ake a trajectory (sequentially) which is not al- ready classified into any of the cluster and con- sider it as a new cluster . T ake all the other tra- jectories which are not kept in any of the clusters and keep in this cluster if they satisfy the threshold limit. (c) Repeat step b till no ne w clusters are added. 3. Using the follo wing RepT raj Algorithm representative trajectories are computed. (a) For each T rajectory of cluster C calculate cumula- tiv e dissimilarity with all other trajectories of the same cluster C. Select the trajectory which is ha v- ing minimum cumulati ve dissimilarity and take this as representativ e trajectory of that cluster . 4. By considering the trajectories receiv ed from step 3, as initial cluster centers, using the follwing Re-cluster Al- gorithm re compute clusters and their representati ve tra- jectories until there is no change in the representative trajectories. (a) For each T rajectory calculate dissimilarity with all the K representativ e trajectories and classify to the cluster for which dissimilarity is low . (b) Re-calculate representati ve trajectories using Rep- T raj Algorithm. C. F ault T olerant Routine T o make the system f ault tolerant, instead of selecting only one optimal representativ e trajectory , routine is asked to se- lect ne xt p number of optimal trajectories, based on their pri- ority le vels, and there corresponding cluster heads will be selected for communicating data to the base station. This list will keep by the base station, in any case of fault occurs with the initial cluster heads, as a choice it can go for other alter- nativ es. D. Data communication phase There are three steps during the data communication phase: data collection, data fusion and data transmission. Initially each sensor node transmits the sensed information to its cluster head at the time slot assigned by its cluster head. In order to sa ve its energy , the node will close transmit part during the time slot which is not required to it. Once data from all sensor nodes ha ve been recei ved, the cluster head performs data fusion on the collected data and reduces the amount of raw data that need to send to the base station. Once the data gathering and data fusion are completed, the cluster head sends the compressed data to the base station. As mentioned pre viously , all the nodes can work as cluster heads. Due to this, any node can become a cluster head or a cluster member . At each turn, the cluster head calcu- lates available po wer and compares with the cluster mem- bers. Whenev er the cluster heads po wer becomes less than the minimum po wer holding, then the cluster heads inform to its cluster members and assigns the maximum po wer holding cluster member as the cluster head and the same inform to the base station. Whene ver the cluster head is changed, base station repeats the cluster finding process and modifies the clusters. III. Experimental W ork T o ev aluate the performance of the algorithm, it has been simulated and compared its performance with ener gy effi- cient clustering routing (hereafter, EECR). Before the simu- lation and results are introduced, the radio model and some important parameters [29] used in simulation have been de- scribed. A. The radio model W e hav e used both the free-space propagation model and the two-ray ground propagation model to approximate path loss sustained because of wireless channel transmission. Gi ven a threshold transmission distance of d0, the free-space model is used when d < d 0 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 be- tween two nodes separated by a distance of d meters is giv en: if d < d 0 , E T ( b, d ) = E T x b + E amp ( d ) b = E T x b + ε 1 d 2 b − (1) if d ≥ d 0 , E T ( b, d ) = bE B F b + ε 2 d 4 b − − − (2) W ith regard to the energy cost incurred in the receiv er of the destination node, we giv e in Eq. (3): E T ( b ) = E Rx b − − − (3) W e have summarized the dif ferent meanings and values for energy terms in T able 1. Energy consumed during data ag- gregation in the cluster head E da , is also taken into account. B. The number of clusters W e assume that N nodes are distributed in the area of A*A randomly . If there are M clusters, then there are N/M nodes in each cluster on an a verage. Every cluster head recei ves the sensed data from its cluster nodes, aggregates all the data, and sends it to the base station. The total ener gy spent on transmitting a frame for e very clus- ter head can be expressed as: E 1 = bE T x N M + bE da N M + bε 2 d 4 1 − − − (4) where d 1 is the distance between cluster head and base sta- tion. In one frame, the cluster nodes transmit the sensed data mes- sages to its cluster head. The ener gy spent for each cluster member is as below: E 2 = bE T x + bε 1 d 2 2 − − − (5) where d 2 is the distance between the member node and its cluster head. If the cluster head is in the center of the cluster , the density of ev ery cluster is ρ = M / A2 , then d 2 can equate to d 2 = r 1 2 π A 2 M − − − (6) T able 1 : Summarizes meaning of each term and typical value S.No T erm Meaning V alue 1 E da Consume energy for data aggre gation 5nJ/bit 2 E T x , E Rx Radio Electronics Energy 50nJ/bit 3  1 T ransmit applied for free space 10pJ/(bit* m 2 ) 4  2 T ransmit applied for two way model 0.0013 pJ/(bit* m 4 ) The energy spent for each cluster member is modified as: E 2 = bE T x + bε 1 1 2 π A 2 M − − − (7) The energy dissipation in a cluster can be e xpressed as: E 2 = bE T x + bε 1 1 2 π A 2 M − − − (8) The total energies dissipated in all the clusters can be ex- pressed as: E = M E c = b  2 E T x N + E da N + M ε 2 d 4 1 + ( N − M ) ε 1 1 2 π A 2 M  − − − (9) if, ∂ E ∂ M = 0 , we can get the follo wing Eq. (10) M = A s N 2 π ε 1 ( ε 2 d 4 1 − E T x ) − − − (10) In our simulation, we consider N = 100, A = 100 m 2 and d 1 = 90m, and for v arious number of clusters i.e., from six to twelve. 1) Results It has been simulated that 100 nodes randomly located in the sensing field of 100 X 100 m 2 with the base station located at least 90 m away . All sensor nodes periodically sense ev ents and transmit the data packet to the base station. All sensor nodes start with an initial ener gy of 2 J and the data message size is fixed at 516 bytes, of which 16 bytes represent the weight value. W e choose three different coefficients C 1 = 0.5, C 2 = 0.4 and C 3 = 0.1. T o e valuate the performance of our algorithm, we compare its performance with EECR. Performance is measured by the number of rounds aliv e and the total data messages successfully deli vered. Follo wing fig- ures 1 & 2 shows the simulated results. It is obvious that our algorithm outperform EECR in the number of rounds the nodes aliv e. The nodes that remain aliv e in EECR are a maximum of 175 rounds, whereas with our proposed method rounds alive are 350 ( see Fig. 1). If the system life time is defined as the number of rounds alive, with our proposed technique system life can increase 90%. Subsequently the number of packets deliv ered at the base station during the number of rounds of activity is increased from a maximum of 40000 (EECR) to 70000 ( see Fig. 2). C. F ault T olerant system T o design the system as fault tolerant, here the lists of cluster heads on their priority le vel are being pro vided; which are 5 6 7 8 9 10 348 350 352 354 356 358 360 362 364 366 368 370 Rounds Alive Number of Clusters Figure. 1 : Number of rounds aliv e 30000 40000 50000 60000 70000 0 20 40 60 80 100 Nodes Alive Packets Delivered Figure. 2 : Packets deli vered obtained from the ne xt representati ve trajectories. This list will keep by the base station, in any case of fault occurs with the initial cluster heads, as a choice it can go for other alter- nativ es. Follo wing T able 2 shows the list of cluster heads, the number of nodes handled by cluster head and their expected life time. IV . Conclusions In this paper , a no vel F ault T olerant T rajectory based cluster - ing solution is presented for selecting cluster heads in wire- less sensor networks. T rajectory clustering algorithm enables sensor nodes to reduce data pack ets by data aggre gation. The wireless communication cost is decreased by reduction of data packets, and thus the clustering technique extends the lifetime by reducing the ener gy consumption of the netw ork. The simulation results demonstrated that our proposal sig- nificantly improves the lifetime and reduce the ener gy con- T able 2 : List of Cluster heads based on Priority Priority List of Cluster Heads No. of nodes handles by each cluster head Expected Life time (in terms of rounds) 1 0, 34, 69, 39, 33, 68 28, 11, 30, 15, 7, 9 368 2 0, 34, 69, 39, 33, 68 36, 6, 27, 15, 7, 9 368 3 1, 68, 0, 34 51, 14, 16, 9 351 4 0, 87, 34, 68 60, 6, 11, 23 351 sumption of wireless sensor networks compared with exist- ing clustering protocols. W e assume that the nodes are error free. Howe ver , error will arise due to the noise in the real network environments. As a future work, we plan to extend the method to increase its robustness. Acknowledgments Hazarath Munaga, w ould like to thank Prof. Garimella Rama Murthy , Associate Professor, Purdue University , U.S.A for his suggestions during initial stage of this work. References [1] R. Shah, J. Rabaey . Energy aware routing for lo w energy ad hoc sensor networks, in: Proc. IEEE W ireless Communications and Networking Confer- ence(WCNC), 2002 [2] Y . Y u, R. Go vindan, D. Estrin. Geographical and energy aware routing: a recursiv e data dissemination protocol for wireless sensor networks, 2001 [3] J. Chang, L. T assiulas. Energy conserving routing in wireless ad-hoc networks, in: IEEE Infocom, pp.22– 31, 2000 [4] J.-H. Chang, L. T assiulas. Maximum lifetime routing in wireless sensor networks, IEEE/A CM T ransactions on Networking, 12(4), pp. 609–619, 2004. [5] F . Y e, G. Zhong, S. Lu, L. Zhang. PEAS: A robust energy conserving protocol for long-liv ed sensor net- works, in: 3rd International Conference on Distributed Computing Systems (ICDCS 03), 2003 [6] C. Gui, P . Mohapatra. Power conservation and qual- ity of surveillance in target tracking sensor networks, in: Proc. of the 10th annual international conference on Mobile computing and networking (Mobi-Com 04), A CM Press, New Y ork, NY , USA, pp. 129–143, 2004 [7] J. Deng, Y .S. Han, W .B. Heinzelman, P .K. V arshney . Balanced energy sleep scheduling scheme for high den- sity cluster-based sensor netw orks, Else vier Computer Communications Journal, Special Issue on ASWN04 28, pp. 1631–1642, 2005 [8] Heinzelman W , Chandrakasan A, Balakrishnan H. En- ergy efficient communication protocol for wireless microsensor networks. In Proc. of the 33rd Annual Hawaill International Conference on System Sciences, Jan 4–7, 2000, Maui, HI, USA. Los Alamitos, CA, USA: IEEE Computer Society , p: 223, 2000 [9] W . Heinzelman, Application-specific protocol archi- tectures for wireless networks, Ph.D. thesis, Mas- sachusetts Institute of T echnology . [10] S. Lindsey , C. Raghavendra, K. Si valingam, Data gath- ering algorithms in sensor networks using energy met- rics. IEEE T ransactions on Parallel and Distributed Sys- tem, 13(9), pp. 924–935, 2002 [11] O. Y ounis, S. Fahmy , Heed: A hybrid, ener gy-efficient, distributed clustering approach for ad hoc sensor net- works, IEEE T ransactions on Mobile Computing, 3(4), pp. 366–379, 2004 [12] C. Li, M. Y e, G. Chen, J. W u, An energy-ef ficient unequal clustering mechanism for wireless sensor networks, in: Proc. of the 2nd IEEE International Conference on Mobile Ad-hoc and Sensor Systems (MASS05), 2005. [13] M. Demirbas, A. Arora, V . Mittal, Floc: A fast lo- cal clustering service for wireless sensor networks, in: W orkshop on Dependability Issues in W ireless Ad Hoc Networks and Sensor Networks (DIW ANS/DSN), 2004. [14] S. Bandyopadhyay , E.J. Coyle, An energy-ef ficient hi- erarchical clustering algorithm for wireless sensor net- works, in: IEEE INFOCOM, 3pp. 1713–1723, 2003 [15] C.-Y . W en, W .A. Sethares, Automatic decentralized clustering for wireless sensor networks, EURASIP J. W irel. Commun. Netw ., 5(5), pp. 686–697, 2005 [16] C. Intanagonwiwat, R. Govindan, D. Estrin, Di- rected diffusion: a scalable and robust communication paradigm for sensor networks, in: Mobile Computing and Networking, pp. 56–67, 2000 [17] B. Krishnamachari, D. Estrin, S.B. W icker , The im- pact of data aggregation in wireless sensor networks, in: Proc. of the 22nd International Conference on Dis- tributed Computing Systems (ICDCSW 02), pp. 575– 578, 2002 [18] T . Kwon and M. Gerla, Clustering with power control, Proc. of IEEE Military Communications Conference (MILCOM), vol. 2, Atlantic City , NJ, pp. 1424–1428, 1999 [19] M. Perillo, Z. Cheng, and W . Heinzelman, On the prob- lem of unbalanced load distribution in wireless sensor networks, In IEEE GLOBECOM W ireless Ad Hoc and Sensor Networks, 2004. [20] S. Ahn and D. Kim, Proacti ve context-aw are sensor net- works. In European W orkshop on W ireless Sensor Net- works, Zurich, Switzerland, 2006 [21] D. Estrin, R. Govindan, J. S. Heidemann, and S. K u- mar . Next century challenges: Scalable coordination in sensor networks. In Mobile Computing and Network- ing (Mobicom99), pp. 263–270, 1999 [22] N. Bulusu, D. Estrin, L. Girod, and J. Heidemann. Scal- able coordination for wireless sensor networks: Self- configuring localization systems. In International Sym- posium on Communication Theory and Applications (ISCT A), 2001 [23] T . Moscibroda, R. W attenhofer , Maximizing the life- time of dominating sets, in: 19th International Parallel and Distributed Processing Symposium, 2005 [24] S.V . Pemmaraju, I.A. Pirwani, Energy conservation in wireless sensor networks via domatic partitions, in: MOBIHOC, 2006 [25] C. Efthymiou, S. Nikoletseas, J. Rolim, Energy bal- anced data propag ation in wireless sensor networks, in: Best papers of the Fourth International W orkshop on Algorithms for W ireless, Mobile, Ad Hoc and Sensor Networks, 2004 [26] S. Murug anathan, Ma DCF , PI. Bhasin, A centralized energy-ef ficient routing protocol for wireless sensor networks, IEEE Communications Magazine, 43(3), pp. 8–3, 2005 [27] H.O. T an, I. Korpeoglu, Power efficient data gathering and aggregation in wireless sensor networks, Issue on Sensor Network T echnology , SIGMOD Record. [28] Y . Y ang, H.-H. W u, H.-H. Chen, Short: Shortest hop routing tree for wireless sensor networks, in: Proc. of IEEE ICC 2006, 2006 [29] LI Li, SONG Shu-Song, W en Xiang-ming. An energy efficient clustering routing algorithm for wireless sen- sor networks, The Journal of China Uni versities of Posts and T elecommunications,13(3), 2006 [30] Hazarath Munaga, Lucio Ieronutti, Luca Chittaro, CAST - A Novel Trajectory Clustering and V isualiza- tion tool for Spatio T emporal Data, in the First Interna- tional conference on Intelligent Human Computer In- teraction (IHCI-2009), Springer , pp. 169–175, 2009 [31] Hazarath Munaga, J. V . R. Murthy and N. B. V enkateswarlu, A Nov el T rajectory clustering tech- nique for selecting cluster heads in W ireless sensor net- works, in the International Journal on Recent T rends in Engineering,1(1), pp. 357–361, 2009 [32] E. W . Dijkstra. A note on two problems in connexion with graphs. Numer . Math. (1), pp.269–271, 1959 [33] R. Agrawal, C. F aloutsos, and A. N. Swami. Ef ficient Similarity Search In Sequence Databases, in FODO: Proc. of the 4th International Conference of Founda- tions of Data Organization and Algorithms, D. Lomet, Ed. Chicago, Illinois: Springer V erlag, pp.69–84, 1993 [34] C. F aloutsos, M. Ranganathan, and Y . Manolopoulos. Fast subsequence matching in time-series databases, in SIGMOD 94: Proc. of the 1994 A CM SIGMOD inter- national conference on Management of data. New Y ork, NY , USA: A CM, pp. 419–429, 1994 [35] Z. R. Struzik and A. Siebes. Measuring time series similarity through large singular features revealed with wa velet transformation, in DEXA 99: Proc. of the 10th International W orkshop on Database and Expert Sys- tems Applications. W ashington, DC, USA: IEEE Com- puter Society , p. 162, 1999 [36] M. Gavrilo v , D. Anguelov , P . Indyk, and R. Motw ani. Mining the stock market (extended abstract): which measure is best? in KDD 00: Proc. of the sixth A CM SIGKDD international conference on Kno wledge dis- cov ery and data mining. Ne w Y ork, NY , USA: A CM, pp. 487–496, 2000 [37] H. Sakoe and S. Chiba, Dynamic programming algo- rithm optimization for spoken word recognition. San Francisco, CA, USA: Morgan Kaufmann Publishers Inc., 1990 [38] B. Bollobas, G. Das, D. Gunopulos, and H. Mannila. T ime-series similarity problems and well-separated ge- ometric sets, in Symposium on Computational Geome- try , pp. 454–456, 1997 [39] D. Berndt and J. Clif ford. Using dynamic time w arping to find patterns in time series, AAAI94 W orkshop on KDD, 1994 [40] D. P . Huttenlocher and K. Kedem. Computing the min- imum Hausdorff distance for point sets under trans- lation, in SCG 90: Proc. of the sixth annual sym- posium on Computational geometry . New Y ork, NY , USA: A CM, pp. 340–349, 1990