A New Energy Efficient Approach by Separating Data

A New Energy Efficient Approach by Separating Data Collection and Data Report in Wireless Sensor NetworksYuning HeGraduate School of Systems and Information Engineering,University of Tsukuba,Tsukuba, Ibaraki305-8573,JapanY ongbing ZhangGraduate School of Systems and Information Engineering,University of Tsukuba,Tsukuba, Ibaraki305-8573,JapanYusheng JiNational Institute of Informatics,Chiyoda-ku, Tokyo101-8430,JapanXuemin(Sherman)Shen Department of Electrical and Computer Engineering,University of Waterloo,Ontario,CanadaABSTRACTA sensor network consists of a large number of distributed wireless sensors which are equipped with low power wireless transceivers. Network lifetime,scalability,and load balancing are important re-quirements for many data gathering sensor network applications. Network clustering is an effective approach for achieving these goals wherein sensors are grouped into multiple clusters.Multi-hop data transmission may provide efficient energy conservation further.A sensor in a cluster may work as a cluster head for data gathering,aggregation,and report to the base station(BS).These tasks can also be performed by distinct sensors in the same clus-ter.In this paper,we clarify some characteristics of multi-hop data transmission.Furthermore,we propose a new data gathering ap-proach for single-hop transmission wherein both the data gathering and the aggregation are performed by the same sensor in a cluster but the report to the BS may be done by a different sensor.The simulation results show that the proposed algorithm achieves better performance than other existing algorithms.Keywordssensor networks,energy efficiency,network lifetime,network clus-tering.1.INTRODUCTIONAdvantages in sensor technology,low-power electronics,and low-power radio frequency(RF)design have enabled the develop-ment of small,relatively inexpensive and low-power sensors,called microsensors,which can be connected via a wireless network[1, 2,3].These sensor nodes(or simply nodes)are usually deployed randomly and densely in hostile environment.They collaborate to observe the surroundings and send the information back to the net-Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on thefirst page.To copy otherwise,to republish,to post on servers or to redistribute to lists,requires prior specific permission and/or a fee.IWCMC’06,July3–6,2006,Vancouver,British Columbia,Canada. Copyright2006ACM1-59593-306-9/06/0007...$5.00.work manager(or base station)when abnormal events occur.For example,sensor networks can play an essential role in emergency situation such asfires,building collapses or extreme weather phe-nomena.Since battery replacement is not an option for networks with thousands of physically embedded nodes,an efficient energy saving protocol is required to prolong the sensor network lifetime [4,5].Network lifetime can be defined as the time elapsed from the network operation starts until thefirst node(or the last node)in the network depletes its energy(dies).Energy consumption in a node can be due to either”useful”or”wasteful”operations.The useful energy consumption includes transmitting or receiving data messages,and processing query requests.On the other hand,the wasteful consumption can be due to overhearing,retransmitting be-cause of harsh environment,dealing with the redundant broadcast overhead messages,as well as idle listening to the media.In order to save the transmission power,clustering[6,7],and multi-hop tranmission techniques[1]can be used.Adjacent sensors may sense the same data and therefore the data gathering can re-duce the redundant data collection.Sensors close to each other in the network can be grouped into clusters and data obtained from sensors in the same cluster are aggregated and then reported to the base station(BS).Data report to the BS can be performed by single-hop or multi-hop transmission.The collected data is directly trans-mitted to the BS in the former approach while via multiple inter-mediate sensors in the latter.Even though the multi-hop transmis-sion may lead to more power saving,it is more difficult to imple-ment.To the best of our knowledge,there is no efficient multi-hop transmission method until now.In this paper,we explore some characteristics of multi-hop transmission and propose a more ef-ficient single-hop transmission approach than existing ones.The data gathering,aggregation,and the report to the basestation(BS) can all be performed by a single node and this node is called the cluster head.Many clustering algorithms in this type have been proposed for wireless sensor networks[8,9,10,11].Heinzelman et al.[8]proposed a two-phase mechanism,called LEACH,based on single-hop transmission.Each node becomes a cluster head ac-cording to a certain probability.Non-head nodes choose the closest cluster to join.Any node is required to be able to communicate with any other node directly.Younis and Fahmy[9]proposed a distributed energy-based clustering algorithm,called HEED,which has been extended from LEACH by using cluster range limits andcost information.It is assured in HEED that a node with more en-ergy has more opportunity to be chosen as a cluster head,as well as cluster heads are distributed well over the area.Other clustering-based algorithms such as VCA[10]and EECS[11]can also be found in the literature.Clustering algorithms require re-clustering after a period of time (called round)and there are mainly two tasks during each re-clusteringprocess.One task is to receive and aggregate the messages from the non-head nodes,and the other is to transmit the aggregated data di-rectly to the BS.In the above algorithms,the cluster head plays an important role and the two tasks are all performed by the cluster head.The main disadvantage of those algorithms is that the cluster head may be the”best”node to collect data in a cluster but maynot be the”best”node to send the collected data to the BS.There-fore,it may be preferable to assign the two tasks to distinct nodes in order to reduce the energy consumption.In this paper,we pro-pose a new energy efficient clustering algorithm wherein the data gathering and aggregation are performed by one node,called the collector,while the report to the BS may be performed by another node,called the forwarder.The node for data aggregation and for report to the BS are determined based on the energy consumption and the data transmission distance to the BS.The paper is organized as follows.In Section2,we introduce the network model used in the paper.Multi-hop data transmission from a sensor to the BS is discussed here.Details of our proposedalgorithm are described in Section3.We evaluate the performance of our approach by simulation in Section4.In Section5,we sum-marize the concluding remarks of the paper.2.PROBLEM OUTLINEData gathering is a typical application in wireless sensor net-works.Sensor nodes monitor the environment and periodically transmit the sensed data to the BS.The BS gathers the messages from the sensors to draw some conclusions about the conditions of the service area,and broadcasts control messages to sensors in the network.In this section,we introduce the sensor network model under consideration.2.1Network ModelThe network model is basically similar to those in LEACH[8] and HEED[9].The nodes are grouped into small clusters.The sensed data in a cluster are gathered at a pariticular node and then sent to the BS directly or via multiple intermediate nodes.Unlike existing algorithms,the data gathering and aggregation in the pro-posed algorithm are performed by the same node,called the clustercollector(or simply collector),while the report to the BS can beperformed by another node,called the cluster forwarder(or sim-ply forwarder).We assume that nodes are uniformly dispersed in the service area of the network,and are quasi-stationary in the sense that they do not change their locations once deployed.Furthermore,the nodes are location-unaware and have the same characteristics. We also assume that data aggregation technique is available,and data collected from sensors in a cluster is packed into one packet. Table1lists the parameters and their definitions used in this pa-per.We use a simple model for the radio hardware energy dissi-pation.The energy consumptions for sending and for receiving anl-bit message are denoted by E t(l)and E r(l),respectively,and they are assumed to be equal and determined by the hardware factors, that is,E t(l)=E r(l)=lY.The energy consumption required to amplify the radio signal in order to send an l-bit message to a loca-tion with distance d away is denoted by E a(l,d).It is well-known that the transmission power depends on the environment and thetransmission distance.In this paper,two parametersε1andε2are used to indicate the amplification factors for different transmission distance.When the transmission distance d is less than a given threshold d0,that is,if d<d0,ε1is used,otherwiseε2is used.Ad-ditionally,the energy consumption for data aggregation is denoted by E g.Furthermore,energy needed to transmit an l-bit message to a location with a distance d away is given by:E t(l,d)=E t(l)+E a(l,d)=l(Y+ε1d2)d<d0,l(Y+ε2d4)d≥d0.(1)The energy consumption for receiving an l-bit message is given as follows.E r(l)=lY.(2)Table1:Meanings of the parametersTerm Definition UnitE t Energy dissipation rate to send the signals J/bitE r Energy dissipation rate to receive the signals J/bitd0Threshold distance for E a mE a Energy dissipation rate at the transmit amplifier J/bit/m2J/bit/m4E g Energy dissipation rate for data aggregation J/bit/signal N Total number of sensor nodesv i The i th sensor node(1≤i≤N)E i Energy dissipation rate from v i to the BS J/bite i Residual energy of sensor v i JR c Cluster radius mR t Inter-cluster transmission radius m2.2Problem StatementA typical clustering algorithm consists of four components:net-work clustering,channel assignment,data collection,and data re-port.The network clustering component divides nodes in the net-work into clusters.The channel assignment component determines the channel used for data transmission in each cluster.The data collection component is used for gathering and aggregating data in a cluster.Finally,the data report component is used for sending the compressive data to the BS.Most of the existing algorithms fo-cus on thefirst component.Specifically,they consider the way to choose cluster heads as well as achieve the clusters well-balanced over the network.The differences among the clustering algorithms (such as LEACH[8],HEED[9],and VCA[10])are in the cluster forming period.After the formation of clusters,all the algorithms execute the same following components.In this paper,the cluster forming method is the same as the existing approaches,but a dif-ferent method for data reporting to the BS is developed to prolong the network lifetime.In most distributed clustering algorithms,a cluster head node does not only receive and aggregate data from member nodes,but also transmits the compressive data to the BS.Since the energy consumption is quite concentrated on the head node,it causes tra-ditional cluster algorithms unsatisfied in some conditions.For ex-ample,as shown in Figure1,four sensors have the same cluster radius.Both A and C have1J energy,while B and D have2J and 3J energy,respectively.Nodes A,C and D are within the commu-nication range of B.However,A,C and D can not communicateFigure 1:The traditional clustering algorithm.with each other directly.In this case,node B is the most suitable to be the head,because data fusion can be performed by B and only one piece of message needs to be delivered to the distant BS.How-ever,compared to node D,node B is not suitable for transmitting the compressive message,because D has more residual energy and is nearer to the BS.2.3Data Transmission to the Base Station(a)Single-hop method.(b)Multi-hop method.Figure 2:Single-hop and multi-hop data transmission.We can consider two distinct approaches to transmit the collected data from a cluster to the BS as shown in Figure 2.In the figure,the collectors and forwarders of clusters are denoted by the gray and the black nodes,respectively,while the other nodes are de-noted by the white nodes.One of the communication methods is single-hop transmission,that is,each cluster transmits the collected data directly to the BS.This approach is simple to implement but clusters far from the BS have to consume much energy.The other is multi-hop transmission,that is,a cluster can transmit the collected data to the BS via multiple intermediate forwarders.In this section,we investigate the multi-hop transmission in details.Younis et al.[9]considered a sensor network with N nodes.They divided the network into n square cells each with a cell side length of R c /√2,where R c denotes the transmission range of a cluster.They showed that each cell contains at least one node when N and n become infinite.In this paper,we consider the same assumptions as in [9].Furthermore,we assume that no collector is located in the coverage of another collector,even though the sevice areas of two collectors may overlap.In the following,we first introduce two Lemmas and then derive two theorems.L EMMA 1.If the inter-cluster transmission range satisfies the condition R t ≥2.8R c ,a cluster collector can find at least one other collector within the transmission radius R t .Proof.According to the assumptions in [12,9],if the the node density of the network area is large enough,there exists at least onenode in any square cell with side length of R c /√2.Let us consider a situation shown in Figure 3where there are two cluster collectors,A and B .Let a square cell tangent to the cluster area centered at A with a side length of R c /√2as shown in the figure.According to [12],there must exist a node in the square cell.Assume that node v belongs to the cluster centered at B and is located at the boundary of cluster B .Let us consider the case where node v is located at the right bottom of the cell,which is the farthest point in the cell from collector A .Then,the Euclidean distances among A ,B ,and v satisfy the following relation.A −B ≤ A −v + v −B=(1+1√2)2+(12√2)2×R c +R c ≈2.8R c.Figure 3:Minimum transmission range for inter-cluster com-munication.We see that the distance of 2.8R c indicates the farthest distance between the neighboring cluster collectors,A and B.Therefore,if R t ≥2.8R c ,there is no isolated cluster collector in the network.L EMMA 2.The network graph of the cluster collectors is con-nected if R t ≥2.8R c .Figure 4:Minimum transmission range for inter-cluster com-munication.Proof.If we can prove that any two collectors are connected when the transmission radius between two cluster collectors R t is just equal to 2.8R c ,then we can prove the lemma.Now,we choose two collecters arbitrarily,H 1and H 2.If H 1−H 2 ≤2.8R c ,H 1and H 2are connected directly,since R t =2.8R c .Let us consider the case when H 1−H 2 >2.8R c as shown in Figure 4.We can draw a square area with side length of R c /√2,which is not tangent to but very close to the area centered at H 1.Therefore,there must exist a node v in the square area and it does not belong to H 1or H 2.Assuming that node v belongs to the cluster centered at H 3,we have H 1−H 3 ≤2.8R c .Therefore,H 1is connected with H 3.Since no two collectors are within each other’s range,we have H 3as shown in Figure 4that indicates the farthest point from H 2.We see that H 2−H 3 ≈2.7R c ,when H 1−H 2 =2.8R c .Therefore,we can find a collector H 3for H 1that is connected to H 1and is nearer to H 2.By using the similary way,we can find another collector that is connected to H 3and is closer to H 2if H 2−H 3 >2.8R c .Since the distance between H 1and H 2is limited,we can finally find a collector that is connected with H 1,and directly connected to H 2.Therefore,H 1and H 2is connected.In order to realize the multi-hop transmission from a cluster to the BS,the network graph of the cluster forwarders should be con-nected.We define R tas the transmission radius between two for-warders.In the following theorem,we give the condition under which the forwarders in the network are connected.T HEOREM 1.The network graph of the cluster forwarders is connected if the transmission range between forwarders satisfiesthe condition R t ≥4.8R c .Proof.In Lemma 2we see that all the collectors are connected ifR t ≥2.8R c .Therefore,the forwarders are connected ifR t≥R t +2R c ≥2.8R c +2R c =4.8R c .Let m denote the shortest distance from the BS to the nearest point of the sensor service area.In the following,we derive the conditions under which sensors are able to reach to the BS.T HEOREM 2.When the maximum transmission power level of sensor nodes is stronger than max {4.8R c ,m +2.8R c },all the nodes can deliver the data to the BS using multi-hop communication.Figure 5:The maximum transmission range to the BS.Proof.In Figure 5,we see that there is at least one node in the semicircle area,or else a blank (R c /√2×R c /√2)size of cell ex-ists.The maximum distance from the BS to the nearest forwarder is given by:BS −f orwarder ≤m +√104R c +2R c≈m +2.8R c .(3)That is,if the transmission range is longer than m +2.8R c ,there is at least a forwarder w which can communicate with the BS.Moreover,according to Theorem 1,if R t ≥4.8R c,any forwarder can reach forwarder w .Therefore,the transmission range must be longer than both m +2.8R c and 4.8R c .3.PROPOSED APPROACHIn this section,we describe the proposed algorithm in details.Because of the specific characteristics of sensor network,it is hard to determine an efficient way to transmit the collected information via multi-hop transmission to the BS.We showed the necessary conditions for multi-hop,and we will make further research in fu-ture works.In this paper we only focus on the single-hop transmis-sion.The determination of the collector in a cluster is similar to the method to determine the cluster head in existing algorithms [9].However,the forwarder is chosen based on the residual energy of each node in a cluster and the distance from the node to the BS.It is assumed that the BS broadcasts a ”hello”message periodi-cally with a certain given power level.Then,each node computes the approximate distance to the BS based on the strength of signal received.For example,node i can determine the value of parameter E i shown in Table 1using the distance from node i to the BS.3.1Forwarder DeterminationThe forwarder of each cluster is selected according to the report cost of nodes.The node report cost is determined by the current re-maining energy and the distance to the BS.The remaining energy of node i in a cluster is denoted by e i .Furthermore,the energy re-quired for transmission from node i to the BS is denoted by E i .We consider the report cost that is proportional to the energy consump-tion from node i to the BS,E i ,and is inversely proportional to the current remaining energy of node i ,e i .That is,the report cost of node i in a cluster,denoted by c i ,is given by:c i =E i e i.(4)3.2Implementation of the Proposed AlgorithmThe mechanism for determining the cluster collector is similar to the method of determining the cluster head in HEED [9].In order to determine the cluster forwarder,each member of a cluster com-putes its report cost,adds the cost information to the cluster-joining broadcast packet,and then sends it to the collector.Therefore,no much overhead will be appended to the proposed algorithm com-pared with the existing algorithms.The collector selects a node with the minimum report cost as the forwarder.Note that the col-lector can also be the forwarder in its cluster if its report cost is the minimum.Then,the collector determines the TDMA slot as-signment for the cluster members.The collector gathers the data from the members except the forwarder,aggregates the collected data into one packet,and then transmits it to the forwarder.The forwarder aggregates the compressive data with its own data,and then transmits to the BS directly (see Figure 6).The proposed algorithm consists of four procedures:initializa-tion ,cluster construction ,TDMA slot allotment ,and data trans-mission .The pseudo-code of the proposed algorithm is shown inFigure6:Data aggregation and report to the BS in the pro-posed algorithm.Figure7.In the proposed algorithm,the variable v i.status indi-cates the status of node i,which may be a collector,a non-collector, or a forwarder(a forwarder is also a non-collector).The variable v i.cluster indicates the set of member nodes of a cluster cen-tered at collector v i.v i.collector and v i.forwarder denote the collector and the forwarder of the cluster which contains node v i.4.EXPERIMENTSIn this section,we evaluate the performance of the proposed approach through the simulations.We compare the proposed ap-proach with HEED.In our experiments,we consider a network topology where nodes are randomly distributed between(x=0, y=0)and(x=100,y=100)with the BS at location(x=50, y=200).We repeat the simulation100times.The parameters used in the simulation are given in Table2.We consider two scenarios in simulation,each of which has a dis-tinct ratio of data packet size to the broadcast packet size.In Sce-nario1,the data packet size is500bytes,and the broadcast packet size is25bytes.On the other hand,in Scenario2,the data packet size is150bytes,and the broadcast packet size is50bytes.Here, we define the network lifetime as the time elapsed from the net-work operation starts until thefirst node dies,since if a node dies, it causes a vacancy area which can not be covered by a sensor node.Table2:Parameters used in experiments.Parameter ValueNetwork grid from(0,0)to(100,100)Base station At(50,200)Threshold distance(d0)100mCluster radius25mE t,E r50nJ/bitε110pJ/bit/m4ε20.0013pJ/bit/signalE g5nJ/bit/signalRound5TDMA framesInitial energy0.5J/batteryWe consider the number of nodes is100,and we compare the proposed approach with HEED and LEACH respectively in the Scenario1.In Figure8,we can see that the our approach is su-perior to both HEED and LEACH on network lifetime.In the following experiments,we compare our approach with HEED to observe the impact on performance by various parame-ters.Figure9and Figure10show the network lifetime by chang-•Initialization1.Receive”hello”message from BSpute E i•Cluster construction1.Construct clusters using HEED algorithm2.If(v i.status=collector)3.Wait for the joining messages from members4.If(v i.status=non-collector)5.Join a cluster with E i information•TDMA allotment1.If(v i.status=collector)2.v i.forwarder←least cost(v i.cluster)3.Assign TDMA slots to all the members4.Else5.Receive TDMA information from v i.collector•Data transmission1.If(v i.status=collector)2.If(v i.forwarder=v i)3.Receive all the data from members4.Aggregate and transmit to the BS5.Else6.Receive all the data from members7.Aggregate and transmit to v i.forwarder8.If(v i.status=non-collecter)9.If(v i.status=forwarder)10.Receive data from v i.collector11.Aggregate and transmit to the BS12.Else13.Transmit to v i.collectorFigure7:Pseudo-code of the proposed algorithm.ing the number of nodes and cluster radius respectively.From the figures,we can see that many conditions,such as density of nodes, cluster radius have big impacts on network performance.Due to the complexities,it is hard to determine the optimal values for all the parameters.In Figure11,we consider the case that there are100nodes in the network area.We keep the x-coordinate of the BS as50,which is introduced in Table2,and vary the y-coordinate of the BS.We change the y-coordinate of the BS from150to300.We can see that network lifetime severely deteriorates when using single-hop trans-mission as the distance increases,which emphasizes the advantages of network clustering.5.CONCLUSIONSIn this paper,we proposed a new clustering approach for sen-sor networks wherein the data gathering and data aggregation are separated from the data report to the BS,and these tasks can be assigned to distinct nodes in a cluster.The proposed algorithm pro-20 40 60 80 100 120 0 20 40 6080 100 120 140 160 180 200N u m b e r o f a l i v e s e n s o r n o d e sRoundsHEED Proposed(a)HEED vs.Proposedapproach.20 40 60 80 100 120 0 20 4060 80 100120N u m b e r o f a l i v e s e n s o r n o d e sRoundsLEACH Proposed(b)LEACH vs.Proposed approach.Figure 8:Network lifetime in HEED and LEACH.0 20 40 60 80 100 120 50150250 350 450 550R o u n d s u n t i l t h e f i r s t n o d e d i e sNumber of nodesHEED Proposed(a)Scenario 1. 80 100 120 140 160 180 200 50150250 350 450 550R o u n d s u n t i l t h e f i r s t n o d e d i e sNumber of nodesHEED Proposed(b)Scenario 2.Figure 9:Network lifetime with various number of nodes.0 20 40 60 80 100 120 140 152535 45 55 65R o u n d s u n t i l t h e f i r s t n o d e d i e sCluster transmission rangeHEED Proposed(a)Scenario 1. 60 80 100 120 140 160 180 200 152535 45 55 65R o u n d s u n t i l t h e f i r s t n o d e d i e sCluster transmission rangeHEED Proposed(b)Scenario 2.Figure 10:Network lifetime with various cluster range.0 20 40 60 80 100 120 140 150175200 225 250 275 300R o u n d s u n t i l t h e f i r s t n o d e d i e sVertical coordinates of the BSHEED Proposed(a)Scenario 1. 0 50 100 150 200 250 150175200 225 250 275 300R o u n d s u n t i l t h e f i r s t n o d e d i e sVertical coordinates of the BSHEED Proposed(b)Scenario 2.Figure 11:Network lifetime with various position of the BS.vides much better performance than previous algorithms.Simula-tion results show that our proposed algorithm prolongs the network lifetime compared with the traditional approaches.In large scale sensor networks,multi-hop communication be-comes important for energy conservation.In this paper,we showed the necessary conditions for multi-hop data transmission.We are now working on designing multi-hop data transmission algorithms to achieve better performance.Since it is generally difficult to have the optimal solution,we are trying to develope heuristic algorithms for multi-hop data transmission.6.REFERENCES[1]I.Akyildiz,W.Su,Y .Sankarasubramaniam,and E.Cayirci,“A survey on sensor networks,”IEEE Commun.Mag.,pp.102–114,August 2002.[2]G.Pottie and W.Kaiser,“Wireless intergrated networksensors,”Communications of the ACM ,vol.43,pp.51–58,May 2000.[3]J.Agre and L.Clare,“An integrated architecture forcooperative sensing networks,”IEEE Computer ,vol.33,pp.106–108,May 2000.[4]J.Chang and L.Tassiulas,“Energy conserving routing inwireless ad-hoc networks,”in Proc.of IEEE INFOCOM ,March 2000.[5]C.Chiasserini,I.Chlamtac,P.Monti,and A.Nucci,“Energyefficient design of wireless ad hoc networks,”in Proc.of European Wireless ,February 2002.[6]T.Kwon and M.Gerla,“Clustering with power control,”COM ,November 1999.[7]V .Kawadia and P.Kumar,“Power control and clustering inad hoc networks,”in Proc.of IEEE INFOCOM ,April 2003.[8]W.Heinzelman,A.Chandrakasan,and H.Balakrishnan,“Anapplication-specific protocol architecture for wireless microsensor networks,”IEEE Trans.Wireless Commun.,vol.1,pp.660–669,October 2002.[9]O.Younis and S.Fahmy,“Distributed clustering in ad-hocsensor networks:A hybrid,energy-efficient approach,”in Proc.IEEE INFOCOM ,March 2004.[10]M.Qin and R.Zimmermann,“An energy-efficientvoting-based clustering algorithm for sensor networks,”in ACIS Int.Workshop Self-Assembling Wireless Networks(SAWN 2005),May 2005.[11]M.Ye,C.Li,G.Chen,and J.Wu,“EECS:An energyefficient clustering scheme in wireless sensor networks,”in IEEE Int.Workshop on Strategies for Energy Efficiency in Ad Hoc and Sensor Networks (IWSEEASN’05),April 2005.[12]D.Blough and P.Santi,“Investigating upper bounds onnetwork lifetime extension for cell-based energyconservation techniques in stationary ad hoc networks,”in Proc.8th Int.Conf.Mobile Computing and Networking (MobiCom 2002),pp.183–192,September 2002.。