An Energy-aware Awakening Routing Algorithm in Heterogeneous Sensor Networks

Transcription

1 An Energy-aware Awakenng Routng Algorthm n Heterogeneous Sensor Networks TAO Dan 1, CHEN Houjn 1, SUN Yan 2, CEN Ygang 3 1. School of Electronc and Informaton Engneerng, Bejng Jaotong Unversty, Bejng, Chna 2. School of Computer Scence & Technology, Bejng Unversty of Posts and Telecommuncatons, Bejng, Chna 3. School of Computer and Informaton Technology, Bejng Jaotong Unversty, Bejng, Chna e-mal: dtao@bjtu.edu.cn, hjchen@bjtu.edu.cn, sunyan@bupt.edu.cn, ygcen@bjtu.edu.cn Abstract: Energy conservaton s of prme mportance for wreless sensor networks. In ths paper, we advocate explotng heterogenety for energy-aware routng by takng advantage of the dfferent capabltes of sensors n a heterogeneous sensor networks. Frst, we descrbe a knd of energy-aware awakenng mechansm, whch allows hgh-cost nodes (eg: mage/vdeo nodes) to go to sleep to conserve energy. Once a low-cost node (eg: temperature/humdty/lght nodes) senses and judges an exceptonal event happens, t wll wake up neghborng sleepng hgh-cost nodes to cooperatvely perform montorng. Moreover, we propose an energy-effcent routng algorthm to determne whch hgh-cost node s the optmal one to be woken. Extensve smulatons show that our soluton can save energy consumpton well and make quck response to the exceptonal events. Keywords: heterogeneous sensor network; awakenng routng; energy-aware routng; floodng 1. Introducton Early research n sensor networks focused on network n whch all sensor nodes possess dentcal capacty n terms of sensng, computaton, communcaton, and battery energy. Ths homogeneous archtecture s attractve because t s reslent to ndvdual falures. More recently, however, heterogeneous sensor networks have become popular due to ther potental to ncrease network lfetme and relablty wthout sgnfcantly ncreasng the whole cost [1]. The smplest operatonal scenaro n a sensor network could be to run all the sensors at the maxmum possble frequency, capture all sensory data, and transmt all raw data to a central center. However, lmted power, storage and bandwdth along wth mnmum desred operatonal lfetme makes such a smplstc soluton nfeasble for autonomous montorng of hghly dynamc, complex envronment. In heterogeneous sensor networks, typcally, a large number of nexpensve nodes perform sensng, whle a few expensve nodes (perhaps embedded PCs) provde data flterng, fuson and transmsson. Ths parttonng of tasks ensures a cost effectve desgn as well as a more effcent mplementaton of the overall montorng applcaton. In general, a sensor has four energy levels accordng to four dfferent states: transmttng, recevng, dle lstenng, and sleepng [2]. Whle ther power consumpton rato for a typcal sensor (eg. Mca2 Mote sensors) s 81: 30: 30: We can conclude that an effectve way of The work s supported by the Natonal Natural Scence Foundaton of Chna (No and No ); Bejng Natural Scence Foundaton of Chna (No ); the Talent Fund of Bejng Jaotong Unversty (No.W07J0260); Honhguoyuan Talent Tranng Program of Bejng Jaotong Unversty. energy conservaton n sensor networks s sleepng as much as possble nodes when they are not engaged n workng. The energy-aware awakenng mechansm proposed here s based on the sleep-awake protocol (SWP) that frst appeared n paper [3]. Accordng to the SWP, each sensor goes though alternatng states of sleepng and workng. When a sensor s n sleepng state, ts processor, data rado and other components are turned off, thus the energy consumpton s mnmal. Authors of paper [4,5] ntroduced the awakenng mechansm nto homogeneous sensor networks. To extend network lfetme, they deploy far more than the expected amount of sensors n a regon. These sensors wll be dvded nto some groups; and each group s alternatvely actvated n turn. In such a case, energy conservaton comes at the expense of hgh deployment cost. Some poneer papers have addressed awakenng mechansm n heterogeneous sensor network. The works of [6,7] focus on the rather smple heterogeneous scenaros-un whch sensors are equpped wth dfferent battery power. Cluster head wake up the basc sensor nodes n ts cluster to swtch the state of the entre cluster from sleepng to workng. Ther man goal s to save energy and prolong network lfetme. Our soluton, however, concentrates on sleep-workng collaboraton by takng advantage of the dfferent capabltes of sensors n a heterogeneous system, thus conserve the entre energy and make quck response to exceptonal events. In other words, we regard the exceptonal events detected as state varables to ensure that varous knds of sensory nformaton wll be feedback to control center. The remnder of ths paper s organzed as follows. Secton 2 descrbes the energy-aware awakenng mechansm. In Secton 3, we propose a novel awakenng routng algorthm to determne whch hgh-cost node s the ScRes.

2 optmal one to be woken. A seres of smulatons have been performed to demonstrate the effectveness of our soluton n Secton 4. Fnally, we make a concluson n Secton 5. 2 Energy-aware Awakenng Mechansm Fundamentally dfferent from conventonal sensor networks, heterogeneous sensor networks are characterzed by ts collaboraton among heterogeneous sensors. By ther dfferent capabltes, we classfy the heterogeneous nodes nto two categores: Low-cost node (L): t has lmted resource, and wth the capacty of sensng, calculatng, transmttng smple data wth low power (eg. temperature, humdty, lght node etc.). Hgh-cost node(h): compared to low-cost one, t has relatvely rch resource, and wth hgh capacty of sensng, calculatng, transmttng complex data, however, t consumes greater power (eg. audo, mage, vdeo node etc.). Fgure 1. Flow chart of the awakenng mechansm The flow chart of the awakenng mechansm s llustrated n Fgure 1. The functon modules n the left (rght) dashed frame are performed by low-cost (hgh-cost) nodes. Durng the process of montorng, low-cost nodes are n a long-term workng state wth low power, and perodcally send sensory data to control center. Each low-cost node contans a smple classfer and rule-based knowledge proposed n our pror work [8], whch can analyze and judge whether an exceptonal event happens or not. If yes, the awakenng mechansm wll be actvated. The workng low-cost nodes wake up the sleepng hgh-cost nodes to cooperatvely acheve montorng actvty. Varous types of sensory data wll be transmtted back to control center. Fnally, control center fuses these sensory data to make a comprehensve, accurate decson and make a quck response. We can fnd that low-cost nodes not only perform perodc envronmental data gatherng, but also work as an alarm clock to wake up hgh-cost nodes once an exceptonal event happens. Seen from Fgure 1, the realzaton of awakenng mechansm contans three man procedures: wake-up whom, how to wake-up, what to do after wake-up. In ths paper, the procedure of wake-up whom s our prmary concern. Whereas, the two procedures of how to wake-up, what to do after wake-up are related wth communcaton model and task logc, respectvely. Here we don t address more. In heterogeneous sensor networks, the transferrng data can be classfed nto two categores: message and sensory data. Our awakenng mechansm determnes a wake-up message s always transmtted from low-cost node to hgh-cost one. After recevng a wake-up message, hgh-cost nodes wll return a confrm message. Assume that each type of sensor networks satsfes wth ndvdual network communcaton connectvty, and each type of sensory data can only be transmtted va ts ndvdual sensor network. 3 Awakenng Routng Algorthm Generally, we utlze tradtonal floodng as routng protocol to transmt data wthn a small range. Snce floodng s a reactve technque, and t doesn t requre costly topology mantenance and complex route dscovery algorthms. Once a low-cost node senses and judges an exceptonal event, t wll turn nto a wake-up source node denoted as L S and send a wake-up message to neghborng hgh-cost nodes. Here, we adopt n-hop floodng. The value of n should be moderate. Too small value wll cause the dsablty of the whole awakenng mechansm. Too large value, however, wll consume over much energy. Hence, how to determne the value of n s the kernel ssue, the soluton can be referred n our pror work [9]. Once the exceptonal event s sensed, the wake-up source node L S broadcasts a wake-up message by Mult- Hopfloodng algorthm whch s depcted as Algorthm 1. When a hgh-cost node receves the wake-up message, t wll return a packet, whch contans a confrm message and ts ndvdual resource data (ncludng locaton, remaned energy, route etc.). The packet data wll be coped onto the low-cost transferrng nodes along each path. Then L S wll determne and select the lowest-cost path to wake up a sutable hgh-cost node. In ths part, we dscuss strategy formulaton to evaluate whch hgh-cost node should be woken up. The total cost of transferrng path (C ) would be decded by some parameters: ncludng hop number n n-hop floodng (H), communcatng dstance (D) and the remaned energy of hgh-cost nodes (E remaned ). C (1 ) wh 1 w2d w3 Eremaned (1) ScRes. 624

3 Algorthm 1. MultHopfloodng Procedure MultHopfloodng(L S ) /*L S denotes the wake-up source node senses and judges an exceptonal event happens*/ Step1. N:=get_n_value();//get the value of n used n the current n-hop floodng route protocol; Step2. f (N>0) then { Step2.1 HNode[]=MultHopfloodng() Step2.2 hgh-cost nodes H return resource data to L S, and backup resource data onto all the low-cost transferrng nodes L j along each path. Step2.3 L S call Strategy() to determne whch hgh-cost node to wake up and along whch path to transmt. Step2.4 L S wake up a hgh-cost node H k (k ) whch meets wth the optmal strategy; Step2.5 the state of hgh-cost node H k s swtches from sleepng to workng, gathers and sends sensory data back to control center. Step2.6 N=N-1; Goto Step2.} 3 where w 0 and w. The smaller C becomes, the 1 1 greater the probablty of beng selected as transferrng path wll be. Notce that parameter D s proportonal to the error rate n wreless communcaton. Wth the decrease of the error rate, the probablty of beng selected as transmttng path wll ncrease. In order to untze the value felds of the three dfferent parameters, we defne H {1, 2, 3,...}, H Dj j 1,1 [ 0,1]. D H R CL [0,1] E remaned 4 Case Study & Performance Analyss To perform emprcal evaluaton of our soluton, we mplemented a stand alone VC++ smulaton platform. 4.1 Case Study In ths secton, we utlze two examples to llustrate the realzaton of our awakenng mechansm; especally the strategy to determne whch hgh-cost node s the optmal one to be woken. Frst, we show an example of awakenng routng strategy n Fgure 2. Fgure 2. The example of awakenng routng strategy (n=1) When L S senses an exceptonal event, t broadcasts a wake-up message by sngle-hop floodng. The wake-up message reaches the 4 hgh-cost nodes H 1, H 2, H 3, H 4. Each of them returns a correspondng packet, whch contans a confrm message and ts ndvdual nformaton ncludng locaton, remaned energy. By formula (1), L S selects the lowest cost path P 3 : L S H 3 to wake up the hgh-cost node H 3. The expermental data s lsted n Table 1. Table 1. The example for n=1 Path D(w 1 =0.3) 1-E remaned (w 2 =0.7) P 1 : L S H 1 88/ % P 2 : L S H 2 92/ % P 3 : L S H 3 65/ % P 4 : L S H 4 58/ % Among the two parameters D and 1- E remaned, The effect of 1-E remaned s greater than D, hence, the weght of 1-E remaned s larger than that of D, accordngly. Seen from Table 1, although the remaned energy of H 3 s 80%, t sn t the hghest among the four (the remaned energy of H 2 s 90%), regardng of the nfluence upon parameter D, the total cost of P 3 s less than the total cost of P 2 n ths example. When we evaluate the total cost of path, we should balance between parameter D and 1- E remaned. Fgure 3. The example of awakenng routng strategy (n>1) Next, we wll explore the example of awakenng routng strategy when n s greater than 1. For the case llustrated n Fgure 3, when L S senses an exceptonal event, t broadcasts a wake-up message by 2-hop floodng. Wthn 2 hops, the wake-up message can arrve at the two hgh-cost nodes H 1, H 2. Each of them returns a packet, whch contans a confrm message and ts ndvdual nformaton, such as locaton, remaned energy, route. The packet data wll be coped onto the low-cost transferrng nodes along each path. For example P 1 : L S L 1 H 1, where L 1 s the low-cost transferrng node. Once L 1 senses the exceptonal event happens durng the mmedate perod, t drectly wakes up hgh-cost nodes accordng to route data on t nstead of re-broadcastng a wake-up message by 2-hop floodng. Accordng to formula (1), L S choose the lowest cost path P 2 : L S H 1 to wake up the hgh-cost node H 1. The expermental data s lsted n Table 2. C ScRes.

4 Table 2. The example for n>1 Path H(w 1 =0.4) D(w 2 =0.1) 1-E remaned (w 3 =0.5) P 1 : L S L 1 H 1 2 (52+48)/(2*100) 1-80% 0.95 P 2 : L S H /(1*100) 1-80% P 3 : L S L 2 H 1 2 (63+45)/(2*100) 1-80% P 4 : L S L 3 H 2 2 (50+90)/(2*100) 1-85% C 4.2 Performance Analyss Through a seres of smulatve data (Expermental parameters are set as S=500*500, p 0 =85%, R SH =140m, R SL =50m 1 ), we dscuss how the optmal hop number n nfluences the two key parameters n our energy-aware awakenng routng algorthm, they are energy consumpton and the successful awakenng probablty. We generated 100 random topologes per scenaro. Fgure 4 shows the relaton between the optmal hop number and energy consumpton. From the curve, we can educe the optmal hop number s proportonal to energy consumpton. Suppose that the average energy consumpton of sngle-hop floodng s 1E. The energy consumpton of floodng wll sharply ncrease wth the ncrease of hop number. When hop number reaches 2 and 3, the average energy consumpton of each approaches 8.05E and 23.9E, respectvely. Fgure 4. The relaton between the optmal hop number and energy consumpton Fgure 5 shows the relaton between the optmal hop number and the successful awakenng probablty. We fnd the average probablty value s close to 66.5% n the case of sngle-hop floodng, whch s comparatvely low. However, n the case of 2-hop or 3-hop floodng, the dfference of ther average probablty value s slght, and approaches 97.7%, 99.2% respectvely. 1 The area of detected regon s denoted as S, and the ntal coverage rate of network s denoted as p 0. The sensng range of hgh-cost (low-cost) node s denoted as R SH (R SL ). Smlarly, the communcatng range of hgh-cost (low-cost) node s denoted as R CH (R CL ). We assume R C =2R S n ths paper. Fgure 5. The relaton between the optmal hop number and successful awakenng probablty Hence, we adopt 2-hop floodng. From the vew of energy consumpton, by 2-hop floodng we can economze at least 15E by 2-hop floodng compared to the case of 3-hop floodng. In Fgure 3, from the other vew of the successful awakenng probablty, we get an ncrease of 30 percent n the successful awakenng probablty compared to usng sngle-hop floodng. And the probablty of 2-hop s close to that of 3-hop floodng, they almost reach 100%. We can conclude that, our soluton can effectvely preserve energy and prolong an extended lfetme for heterogeneous sensor networks. 5 Conclusons A crtcal ssue n sensor networks s represented by the lmted energy wthn network nodes; therefore effcent utlzaton of energy s a must. Motvated by ths, we ntroduce an awakenng mechansm. In a usual montorng perod, only low-cost nodes are n workng state and hgh-cost nodes are n sleepng state. Once a low-cost node senses an exceptonal event happens, t broadcast a message to wake up neghborng hgh-cost nodes to cooperatvely perform montorng. In partcular, we present an energy-aware awakenng routng algorthm to wake up the optmal hgh-cost nodes. Fnally, the smulaton verfes the effectveness of our soluton. 6 Acknowledgements The work reported n ths paper s partally supported by the NSFC and the Talent Fund of Bejng Jaotong Unversty, Chna. References [1] Yarvs, M.; Kushalnagar, N.; Sngh, H.; Rangarajan, A.; Lu, Y.; Sngh, S. Explotng Heterogenety n Sensor Networks. INFO- COM March. pp: [2] Matthew J. Mller, Ntn H. Vadya, Power Save Mechansms for Mult-HopWreless Networks, n Proceedngs of the Frst Internatonal Conference on Broadband Networks, pp: [3] Chatzgannaks, S. Nkoletseas. A Sleep-awake Protocol for Informaton Propagaton n Smart Dust Networks, In 3rd Internatonal Workshop on Moble, Ad-hoc and Sensor Networks, pp. 225a, ScRes. 626

5 [4] Huadong Ma, Yonghe Lu. On Coverage Problems of Drectonal Sensor Networks, Internatonal Conference on Moble Adhoc and Sensor Networks (MSN 05), Dec, pp: [5] Athanasos Stathopoulos. Explotng heterogenety for routng n wreless sensor networks. Doctoral Thess. Unversty of Calforna, Los Angeles. CA, USA [6] I. Chatzgannaks, AKnals and S. Nkoletseas, Power Conservaton Schemes for Energy Effcent Data Propagaton n Heterogeneous Wreless Sensor Networks, n Proceedngs of the 38th Annual Smulaton Symposum, pp: [7] Z. Zhang, M. Ma and Y. Yang, Energy Effcent Mult-hop pollng n Clusters of Two-layered Heterogeneous Sensor Networks, n Proceedngs of the 19th IEEE Internatonal Parallel and Dstrbuted Processng Symposum, pp: 81b. [8] D. M. Zhang, H. D. Ma, L. Lu, D. Tao. EAAR: An Approach to Envronment Adaptve Applcaton Reconfguraton n Sensor Network. The 1st Internatonal Conference on Moble Ad-hoc and Sensor Networks, pp: [9] LIU Wenhong, Zhang Hongke, TAO Dan. A Novel Cooperatve Wake-up Routng Algorthm for Wake-up Mechansm n Heterogeneous Sensor Networks. Acta Electronca Snca. 2007, 35 (7): ScRes.