An efficient cluster-based power saving scheme for wireless sensor networks

Transcription

1 RESEARCH Open Access An effcent cluster-based power savng scheme for wreless sensor networks Jau-Yang Chang * and Pe-Hao Ju Abstract In ths artcle, effcent power savng scheme and correspondng algorthm must be developed and desgned n order to provde reasonable energy consumpton and to mprove the network lfetme for wreless sensor network systems. The cluster-based technque s one of the approaches to reduce energy consumpton n wreless sensor networks. In ths artcle, we propose a savng energy clusterng algorthm to provde effcent energy consumpton n such networks. The man dea of ths artcle s to reduce data transmsson dstance of sensor nodes n wreless sensor networks by usng the unform cluster concepts. In order to make an deal dstrbuton for sensor node clusters, we calculate the average dstance between the sensor nodes and take nto account the resdual energy for selectng the approprate cluster head nodes. The lfetme of wreless sensor networks s extended by usng the unform cluster locaton and balancng the network loadng among the clusters. Smulaton results ndcate the superor performance of our proposed algorthm to strke the approprate performance n the energy consumpton and network lfetme for the wreless sensor networks. Keywords: cluster, dstance, power consumpton, lfetme, sensor networks 1. Introducton Recently, there has been a rapd growth n the wreless communcaton technque. Inexpensve and low-power wreless mcro-sensors are desgned and wdely used n wreless and moble envronments [1-3]. A wreless sensor network conssts of a large number of sensor nodes. Each sensor node has sensng, computng, and wreless communcaton capablty. All sensor nodes play the role of an event detector and the data router. Sensor nodes are deployed n the sensng area to montor specfc targets and collect data. Then, the sensor nodes send the data to snk or base staton (BS) by usng the wreless transmsson technque. Wreless sensor networks have been pervasve n varous applcatons ncludng health care system, battlefeld survellance system, envronment montorng system, and so on. Fgure 1 shows an nfrastructure of wreless sensor networks. Power savng s one of the most mportant features for the sensor nodes to extend ther lfetme n wreless sensor networks. A sensor node consumes mostly ts energy n transmttng and recevng packets. In wreless sensor * Correspondence: jychang@nfu.edu.tw Department of Computer Scence and Informaton Engneerng, Natonal Formosa Unversty, Hu-We, Yun-Ln, Tawan networks, the man power supply of the sensor node s battery. However, n most applcaton scenaros, users are usually dffcult to reach the locaton of sensor nodes. Due to a large number of sensor nodes, the replacement of batteres mght be mpossble. However, the battery energy s fnte n a sensor node and a sensor node dranng of ts battery may make sensng area uncovered. Hence, the energy conservaton becomes a crtcal concern n wreless sensor networks. In order to ncrease energy effcency and extend the network lfetme, new and effcent power savng algorthms must be developed [4-6]. Low Energy Adaptve Clusterng Herarchy (LEACH) s a typcal cluster-based protocol usng a dstrbuted clusterng formaton algorthm [5]. In LEACH, the large number of sensor nodes wll be dvded nto several clusters. For each cluster, a sensor node s selected as a cluster head. The selecton of cluster head nodes s based on a predetermned probablty. Other non-cluster head nodes choose the nearest cluster to jon by recevng the strength of the advertsement message from the cluster head nodes. A non-cluster head node can only montor the envronment and send data to ts cluster head node. The cluster head node s responsble for 2012 Chang and Ju; lcensee Sprnger. Ths s an Open Access artcle dstrbuted under the terms of the Creatve Commons Attrbuton Lcense ( whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly cted.

2 Page 2 of 10 Fgure 1 Infrastructure of wreless sensor networks. collectng the nformaton of non-cluster head nodes n the cluster. Then, t processes data and sends data to the BS. As a non-cluster head node cannot send data drectly to the BS, the data transmsson dstance of the sensor node s shrunk. Therefore, the energy consumpton s reduced n the wreless sensor networks. However, the random selecton of the cluster head node may obtan a poor clusterng setup, and cluster head nodes may be redundant for some rounds of operaton. The dstrbuton of cluster head nodes s not unform, thus some sensor nodes have to transfer data through a longer dstance and the reasonable energy savng s not obtaned n wreless sensor networks. LEACH-centralzed (LEACH-C) s proposed as an mprovement of LEACH whch uses a centralzed clusterng algorthm to create the clusters [6]. In LEACH-C, the BS collects the nformaton of the poston and energy level from all sensor nodes n the networks. Based on ths nformaton, the BS calculates the number of cluster head nodes and confgures the network nto clusters. In [7], the authors propose the hybrd, energy-effcent, dstrbuted (HEED) clusterng protocol to prolong the network lfetme and support scalable data aggregaton. In ths protocol, the cluster heads are probablstcally selected based on ther resdual energy and the sensor nodes jon the clusters accordng to ther power level. The clusterbased power savng methods have been proposed n [8-11], whch have been valdated to some extent through smulaton. However, the crtcal problem s that the cluster sze s also not unform n these schemes. The energy consumpton cannot mprove effectvely. Fgures 2 and 3 show the extreme cases of clusterng structure for a certan round n the smulaton by usng LEACH. In ths artcle, we propose a savng energy clusterng algorthm (SECA) to provde effcent energy consumpton n wreless sensor networks. In order to make an deal dstrbuton for sensor node clusters, we calculate the average dstance between the sensor nodes and take nto account the resdual energy for selectng the approprate cluster head nodes. The lfetme of wreless sensor networks s extended by usng the unform cluster locaton and balancng the network loadng among the clusters. The man benefts of proposed scheme are that the energy consumpton s reduced and better network lfetme can be carred out. The rest of ths artcle s organzed as follow. In Secton 2, we present the system model of wreless sensor networks. In Secton 3, we llustrate the proposed scheme n detal. In Secton 4, we present our smulaton model and analyze the comparatve evaluaton results of the proposed scheme through smulatons. Fnally, some conclusons are gven n Secton System model The system nfrastructure s composed of a BS and some sensor nodes. We classfy all sensor nodes nto non-cluster head nodes and cluster head nodes. The non-cluster head nodes operate n sensng mode to montor the envronment nformaton and transmt data to the cluster head node. Also, the sensor node becomes a cluster head to gather data, compresses t and forwards to the BS n cluster head mode. The system framework of ths artcle s shown n Fgure 1.

3 Page 3 of 10 Fgure 2 Clusterng structure of LEACH (cluster heads = 5 and sensor nodes = 100). In wreless sensor networks, data communcatons consume a large amount of energy. The total energy consumpton conssts of the average energy dsspated by data transmsson of the non-cluster head nodes and the cluster head nodes. In addton, the energy consumpton for data collecton and aggregaton of cluster head nodes s consdered. Fgure 4 llustrates the rado energy dsspaton model n wreless sensor networks [5-7]. In ths model, to exchange an L-bt message between the two sensor nodes, the energy consumpton can be calculated by. E Tx (L, d) =E elec L + ε amp L, (1) E Rx (L) =E elec L, (2) where d s the dstance between the two sensor nodes, E Tx (L, d) s the transmtter energy consumpton, and E Rx (L) s the recever energy consumpton. E elec s the electroncs energy consumpton per bt n the transmtter and recever sensor nodes. ε amp s the amplfer energy consumpton n transmtter sensor nodes, whch can be calculated by ε amp = { εfs d 2, when d d 0 ε mp d 4, when d > d 0, (3) where d 0 s a threshold value. If the dstance d s less than d 0, the free-space propagaton model s used. Otherwse, the multpath fadng channel model s used. ε fs and ε mp are communcaton energy parameters. Usng theprevouslydescrbednthe lterature [5,6], the ε fs s set as 10 pj/bt/m 2 and ε mp s set as pj/bt/m 4. Also, the energy for data aggregaton of a cluster head node s set as E DA = 5 nj/bt/sgnal and the ntal energy of a sensor node s set as E nt =2J.Suppose that a non-cluster head node N transmts L N bts to the BS. Let d N, CH bethedstancebetweenthenon-cluster head node N and ts cluster head node CH. Letd CH, BS be the dstance between the cluster head node CH and the BS. Due to the mult-hop communcaton, a noncluster head node only sends data to ts cluster head node. The resdual energy of the non-cluster head node N s equal to E nt - E Tx (L N, d N, CH ). In addton, the resdual energy of the cluster head node CH s equal to E nt - E Rx (L N )-E DA - E Tx (L N,d CH, BS ), because the cluster

4 Chang and Ju EURASIP Journal on Wreless Communcatons and Networkng 2012, 2012:172 Page 4 of 10 Fgure 3 Clusterng structure of LEACH (cluster heads = 5 and sensor nodes = 150). head node must collect and process the nformaton of non-cluster head nodes n the cluster, and then send data to the BS. Fgure 4 Rado energy dsspaton model. It s obvous that the data transmsson between sensor nodes takes most of the energy consumpton n the wreless sensor networks. Takng nto account the energy consumpton of sensor nodes, the data transmsson dstance must be reduced and the packets delay should be avoded. Hence, the energy consumpton and routng desgn become an mportant ssue n the wreless sensor networks. 3. Proposed methods: SECA In order to ncrease energy effcency and extend the lfetme of the sensor nodes n wreless sensor networks, effcent power savng algorthm must be developed and desgned. Based on the centralzed clusterng archtecture, we propose a SECA to provde effcent energy consumpton and better network lfetme n the wreless sensor networks. In the proposed scheme, we assume that the BS receves the nformaton of locaton and resdual energy for each sensor node and the average resdual energy can be calculated. When the resdual energy of sensor node s hgher than the average resdual energy, the sensor node becomes a canddate of

5 Page 5 of 10 cluster head. We modfy k-means algorthm to make an deal dstrbuton for sensor node clusters by usng the nformaton of locaton and resdual energy for all sensor nodes [12,13]. In ths algorthm, the operaton ncludes two phases: set-up and steady-state phases Set-up phase The man goal of ths phase s to create clusters and fnd cluster head nodes. Durng the set-up phase, the BS collects the nformaton of the poston and energy level from all sensor nodes n the networks. Based on the characterstcs of statonary sensor nodes, the sutable ntal means of ponts for clusters can be obtaned. Let C be the center locaton for all sensor nodes. If there are n sensor nodes n the wreless sensor networks, C can be calculated by n =1 C = X, (4) n where X s the coordnate of sensor node. Let R be the average dstance between C and all sensor nodes, whch can be calculated by n =1 R = X C. (5) n Accordng to C and R, the locatons of ntal mean of pont m (m x, m y ) for the cluster s calculated by m x = R cos( 360 ( 1) π k m y = R sn( 360 k 180 )+C x ( 1) π 180 )+C y, (6) where k s the number of clusters and = 1, 2,..., k. Fgure 5 shows the example of the ntal means of ponts, where k s equal to 3. The ntal value k must be decded n the ntal set-up phase. Accordng to the defnton of optmum number of clusters n LEACH-C [6], k can be calculated by n εfs M k =, (7) 2π ε mp d 2 to BS where M sthesdeofthegvensquarefeld.thed to BS s the average dstance from the cluster head nodes to the BS whch s defned n LEACH-C. However, the cluster head nodes are selected by creatng some clusters n our proposed algorthm. Hence, we re-defne d to BS whch s the average dstance from the all sensor nodes to the BS. The settng of ntal means of ponts s very mportant. It can reduce the teraton tme for creatng clusters sgnfcantly. After the ntal means of ponts are Fgure 5 Example of the ntal means of ponts. set, based on the locaton of all sensor nodes, the BS creates some clusters. We use the k-means algorthm to partton the n sensor nodes nto k clusters n whch each sensor node belongs to the cluster wth the nearest mean of pont. If there are k clusters n the system, the k-means functon can be expressed by avg S mn k X j m 2, (8) X j S =1 where S s the cluster, X j s coordnate of sensor node j and m sthecoordnateofmeanofpont.the man reason for ths expresson s to obtan the mnmum average dstance between the means of ponts and the sensor nodes for all clusters. In order to create unform dstrbuted clusters, the mnmal dstance between the means of ponts and all sensor nodes s calculated. Then, the sensor nodes are classfed nto the cluster accordng to the mnmal dstance. If the X j s the closest to the m n the tth executon, the sensor node j wll jon the cluster, whch can be expressed by { S (t) = X j : X j m (t) 2 X j m (t) } 2 for all = 1,..., k, (9) where each sensor node jons exactly one cluster. The man goal of ths expresson s to decde whch cluster the sensor node j belongs to n the tth executon. When the classfcaton of all nodes s done, the new mean of pont s created whch s calculated by m (t+1) = where S (t) 1 S (t) X j S (t) X j, (10) sthenumberofsensornodesnthe cluster. Fgure 6 shows the example of the new means of ponts.

6 Page 6 of 10 Fgure 6 Example of the new means of ponts. Because the means of ponts are changed, all sensor nodes are re-classfed by executng Equatons (9) and (10) teratvely to obtan the mnmum average dstance between the means of ponts and the sensor nodes for all clusters. The fnal clusters are formed when each sensor node s fxed n the cluster. Fgure 7 shows the flowchart of the ntal cluster processng for our proposed scheme. The cluster head s a sensor node whch s closer to the fnal mean of pont and the resdual energy of the sensor node s hgher than the average resdual energy n each cluster. Fnally, the cluster archtecture s created. The BS broadcasts the routng nformaton of the clusters to all sensor nodes. Hence, each sensor node has ts own routng table and knows ts task (e.g., cluster head or non-cluster head). Also, each sensor node knows the dstances from any other sensor node n ts cluster and thereby calculates the transmsson power. Based on the number of the sensor nodes wthn the cluster, the cluster head node creates a schedule based on Tme Dvson Multple Access (TDMA) to allocate the tme for the cluster members Steady-state phase Once the clusters are created and the TDMA schedule s fxed, data transmsson can begn. The non-cluster head nodes send data to cluster head node durng ther allocated transmsson tme. When all the data have been receved, the cluster head node performs sgnal processng to compress the data nto a sngle sgnal. Then, ths sgnal s sent to the BS. The amount of nformaton s reduced due to the data aggregaton done at the cluster head node. Ths round s done and the next round begns wth set-up and steady-state phases repeatedly. To avod unnecessary nodes control messages transmsson and control overhead of the BS, the clusters are Fgure 7 Flowchart of the ntal cluster processng. re-created only when the sensor node cannot work n a certan round. So, the calculatng overhead s only cluster head selectng n the most set-up phase. 4. Performance analyss In ths secton, we evaluate the performance of our proposed SECA usng a smulaton model. We descrbe our smulaton model and llustrate the smulaton results, and compare our scheme wth the LEACH, HEED, and LEACH-C. We desgn a smulaton envronment by usng C#. The assumptons for our smulaton study are as follows. Table 1 Parameters used n smulaton mode Parameter Value Electroncs energy (E elec ) 50 nj/bt Energy for data aggregaton (E DA ) 5 nj/bt/sgnal Intal energy of node (E nt ) 2 J Packet sze 2000 bts Number of nodes (n) 50, 100, 150 Poston of BS (X, Y) (50, 175) Sensng area (M M) ,

7 Page 7 of 10 The smulaton envronment s composed of a BS and some sensor nodes. The BS s fxed and located far from the sensor nodes. The locaton of each sensor node s randomly dstrbuted n the sensng area. The non-cluster head node can montor the envronment and send data to the cluster head node. The cluster head node can gather data, compress t, and forward to the BS. All sensor nodes are statonary and the ntal energy s the same for each sensor node. All the parameters used n our smulaton are lsted n Table 1[5,6]. These performance measures obtaned on the bass of ten smulaton runs are plotted as a functon of the rounds and total network energy. A round s defned as the recevng data form all sensor nodes to the BS. The total network energy s defned as the sum of resdual energy at all sensor nodes. For far comparson, we set the number of clusters s equal to 5 whch s defned n LEACH-C. Fgures 8, 9, and 10 show the clusterng structure for a certan round n the smulaton by usng LEACH-C, HEED, and our proposed SECA scheme. Accordng to SECA features, t s ntutve that each cluster sze s almost the same and cluster heads locate more closely to the cluster centers. Hence, our proposed algorthm reduces the data transmsson dstances of sensor nodes and results n the lower energy consumpton n the wreless sensor networks. Fgure 11 shows the total network energy when the number of sensor nodes s 50 and the sensng area s 100 m 100 m. It s evdent that the resdual energy of our proposed SECA scheme s hgher than that of LEACH, HEED, and LEACH-C schemes after 900 runs. Ths s because our proposed SECA scheme provdes the unform cluster locaton and better cluster formaton. The data transmsson dstance from each sensor node to ts cluster head node s mnmzed. Thus, the energy consumpton s saved. Fgures 12 and 13 show the frst sensor node dead andhalfofthesensornodesalveforfourmethods when the number of sensor nodes s 50 and the Fgure 8 Clusterng structure of LEACH-C (sensor nodes = 100).

8 Page 8 of 10 Fgure 9 Clusterng structure of HEED (sensor nodes = 100). sensng area s 100 m 100 m. Accordng to Fgures 12 and 13, the lfetme of the sensor node of our proposed SECA scheme s better than that of LEACH, HEED, and LEACH-C schemes. The reason for ths behavor s that the data communcatons consume a large amount of energy n wreless sensor networks. However, the transmsson dstance between non-cluster head node and cluster head node s sutable by usng our scheme. The transmsson power of noncluster head nodes s reduced. Also, n order to prove our proposed scheme s well desgned, we ncrease the number of sensor nodes n the smulaton envronment. Fgure 14 shows the total network energy when the number of sensor nodes s 100 and the sensng area s 100 m 100 m. Due to the better energy savng approach n the proposed scheme, accordng to SECA features, t s ntutve that the proposed scheme results n hgher resdual energy than LEACH, HEED, and LEACH-C schemes when the sensor node ncreases n the same sensng area. Furthermore, we extend the sensng area to 200 m 200 m n the smulaton envronment. Fgure 15 shows the total network energy when the number of sensor nodes s 150. The dstrbutons of clusters n LEACH and HEED are not unform and some clusters consst of huge number of sensor node n a large area. Hence, the cluster head of the cluster wth huge number of sensor nodes wll suffer from heavy traffc load and result n sgnfcant energy consumpton. Accordng to Fgure 15, the curves ndcate that our scheme mproves the energy utltymoresgnfcantlythanleach,heed,and LEACH-C schemes when the sensng area s extended. From the smulaton results, t s clear that our proposed scheme strkes the approprate performance n the energy consumpton and network lfetme for the wreless sensor networks. 5. Conclusons The energy savng s a challengng ssue n the wreless sensor networks. To ncrease energy effcency and

9 Chang and Ju EURASIP Journal on Wreless Communcatons and Networkng 2012, 2012:172 Page 9 of 10 Fgure 10 Clusterng structure of SECA (sensor nodes = 100). extend the lfetme of sensor node, new and effcent energy savng schemes must be developed. In the proposed scheme, we calculate the average dstance between the sensor nodes and take nto account the Fgure 11 Total network energy (sensor nodes = 50). resdual energy for selectng the approprate cluster head nodes. The lfetme of wreless sensor networks s extended by usng the unform cluster locaton and balancng the network loadng among the clusters. Smulaton results ndcate our proposed algorthm acheves Fgure 12 Frst node dead (sensor nodes = 50).

10 Page 10 of 10 the low energy consumpton and better network lfetme n the wreless sensor networks. Competng nterests The authors declare that they have no competng nterests. Receved: 13 February 2012 Accepted: 16 May 2012 Publshed: 16 May 2012 Fgure 13 Half of nodes alve (sensor nodes = 50). Fgure 14 Total network energy (sensor nodes = 100). References 1. IF Akyldz, W Su, Y Sankarasubramanam, E Cayrc, A survey on sensor networks. IEEE Commun Mag. 40(8), (2002). do: / MCOM M Tubashat, S Madra, Sensor networks: an overvew. IEEE Potentals. 22(2), (2003) 3. JN Al-Karak, AE Kamal, Routng technques n wreless sensor networks: a survey. IEEE Wrel Commun. 11(6), 6 28 (2004). do: / MWC A Chamam, S Perre, On the plannng of wreless sensor networks: energyeffcent clusterng under the jont routng and coverage constrant. IEEE Trans Mob Comput. 8(8), (2009) 5. WR Henzelman, A Chandrakasan, H Balakrshnan, Energy-effcent communcaton protocol for wreless mcrosensor networks, n Proc 33rd Hawa Internatonal Conference on System Scences, pp (January 2000) 6. WB Henzelman, P Chandrakasan, H Balakrshnan, An applcaton-specfc protocol archtecture for wreless mcrosensor networks. IEEE Trans Wrel Commun. 1(4), (2002). do: /twc O Youns, S Fahmy, HEED: a hybrd, energy-effcent, dstrbuted clusterng approach for ad hoc sensor networks. IEEE Trans Mob Comput. 3(4), (2004). do: /tmc S Babae, AK Zadeh, MG Amr, The new clusterng algorthm wth cluster members bounds for energy dsspaton avodance n wreless sensor network, n Proc Computer Desgn and Applcatons (ICCDA), pp (June 2010) 9. Q Xuegong, C Yan, A control algorthm based on double cluster-head for heterogeneous wreless sensor network, n Proc Industral and Informaton Systems (IIS), pp (July 2010) 10. Y-U Yun, J-K Cho, N Hao, S-J Yoo, Locaton-based spral clusterng for transmsson schedulng n wreless sensor networks, n Proc Advanced Communcaton Technology (ICACT), pp (February 2010) 11. HD Targh, M Sabae, A new clusterng method to prolong the lfetme of WSN, n Proc Computer Research and Development (ICCRD), pp (March 2011) 12. T Kanungo, DM Mount, NS Netanyahu, CD Patko, R Slverman, AY Wu, An effcent k-means clusterng algorthm: analyss and mplementaton. IEEE Trans Pattern Anal Mach Intell. 24(7), (2002). do: / TPAMI J Zhu, H Wang, An mproved K-means clusterng algorthm, n Proc 2nd IEEE Internatonal Conference on Informaton Management and Engneerng (ICIME), pp (2010) do: / Cte ths artcle as: Chang and Ju: An effcent cluster-based power savng scheme for wreless sensor networks. EURASIP Journal on Wreless Communcatons and Networkng :172. Fgure 15 Total network energy (sensor nodes = 150).