Intelligent Wakening Scheme for Wireless Sensor Networks Surveillance

Transcription

1 The Frst Internatonal Workshop on Cyber-Physcal Networkng Systems Intellgent Wakenng Scheme for Wreless Sensor Networks Survellance Ru Wang, Le Zhang, L Cu Insttute of Computng Technology of the Chnese Academy of Scences, Beng, Chna {wangru,zhangle82,lcu}@ct.ac.cn Abstract The effectve energy control whle mantanng relable detecton performance s a key problem n wreless sensor networks survellance, such as blue-green algae survellance. An ntellgent wakenng scheme (IWS) s proposed n ths paper whch consders dfferent mportance degrees of the grds n the survellance zone. It uses a vorono dagram to determne the effectve scope of each sensor node and calculates the node wakenng probablty wth the mportance degree n the effectve scope. The nodes are then turned on stochastcally accordng to the node wakenng probablty. Smulaton results llustrate that ths scheme greatly reduces the number of wakenng nodes whle mantanng hgh relablty n survellance. Keywords-wreless sensor networks; ntellgent wakenng scheme; energy control; blue-green algae survellance I. INTRODUCTION The maturng of ntegrated crcutry, dgtal sgnal processng, mcro electromechancal systems (MEMS) and low-range rado electroncs on a sgnal node have led to the desgn of wreless sensor networks (WSN), whch have attracted many researchers due to ts wde range applcaton potentals. A WSN provdes a new class of computer system and epands human ablty to remotely nteract wth the physcal world. Applcatons of WSN nclude battlefeld survellance, bologcal detecton, home applance, smart spaces and nventory trackng. By applyng WSN n the montorng of Blue-green Algae Bloom on Lake Ta[1], we can obtan the physcal nformaton, lke water temperature and water color, and t s possble to predct bologcal growth and potental ecologcal dsaster. WSN has some unque characterstcs, of whch the most mportance one s lmted energy supply. A sensor node has a fnte energy reserve suppled by a battery. It s often unfeasble to recharge the node s battery. Thus mnmzng energy consumpton whle mamzng the system lfetme s a major objectve n desgnng wreless sensor networks. Generally, the energy consumpton of sensor nodes conssts of three parts: the mcroprocessor, the rado communcaton module and the senor module. When the sensor node s put nto workng status all the tme, sensor module s often swtched on and the rado module s turned on when necessary. Although the rado communcaton s the largest energy consumer, we can also save much energy by reduce the workng tme of sensor module. In wreless sensor networks, adjacent nodes share common sensng tasks, whch mples that not all sensors are requred to perform the sensng task durng the whole system lfetme. That s to say, the functon of whole system wll not be affected by some sleepng nodes as long as there are enough workng nodes. Therefore, f the sensors can be well scheduled, the system lfetme can be prolonged correspondngly;.e. the system lfetme s prolonged by eplotng redundancy. In ths work, IWS (Intellgent Wakenng Scheme) s presented. In IWS, every node has a local decson on whether t needs to be turned on or off by dynamc calculaton of ts mportance degree n WSN. Our desgn has been drven by the followng requrements: frstly, the self confguraton s mandated because t s nconvenent or mpossble to manually confgure sensors after they have been deployed n hostle or remote workng envronments. Secondly, the desgn has to be fully dstrbuted and localzed. Fnally, the algorthm should provde a dfferentated survellance servce for dfferent mportance degree areas whle mantanng hgh enough detecton probablty to targets. II. RELATED WORKS Meguerdchan et al. [2] addressed one of the fundamental problems, namely, coverage, whch n general answers the questons about the qualty of servce that can be provded by a partcular sensor network. A node-schedulng scheme by [3] turns some nodes on or off whle guarantees certan redundancy. A node wll turn off f t dscovers that ts neghbors can substtute t to montor ts sensng area. The soluton does not need the global knowledge of the network and thus can be mplemented locally at each node. The proposed scheme ncreases communcaton cost, requres synchronzaton and nvolves n the calculaton of a geometrc representaton s ntersecton. In [4], Ye et al. developed a dstrbuted densty control algorthm named PEAS, whch s probng based. In PEAS, each node sleeps for an eponentally dstrbuted duraton. A sleepng node wakes up and broadcasts a probng message wthn a certan range after ts sleepng perod; f no reply s receved after a tmeout, t wll turn on to work untl t depletes ts energy. Prevous research [2-7] focused on how to provde full or partal sensng coverage n the cost of energy consumpton. In such an approach, a node wll be set to be asleep as long as ts neghbors can cover ts sensng area. These solutons treat the /11/$ IEEE 755

2 dfferent survellance zones wth the same mportance. However, n most scenaros such as battlefelds, there est some geographc sectons such as the general command center that s much more securty-senstve. Based on the fact that ndvdual sensor nodes are not relable and subject to falure and sngle sensng readngs can be easly dstorted by background nose and cause false alarms, t s smply not suffcent to rely on a sngle sensor to safeguard a crtcal area. In ths case, t s desred to provde hgher degree of coverage n whch multple sensors montor the same locaton at the same tme n order to obtan hgh confdence n detecton. However, t wll be too energy consumng f the same hgh degrees of coverage are appled n some non-crtcal areas. The rest of the paper s organzed as follows: The detals of IWS algorthm are dscussed n Secton 3. Smulaton results are presented and dscussed n Secton 4. Concludng remarks are provded n Secton 5. III. INTELLIGENT WAKENING SCHEME In ths paper, "Importance Degree" and "Effectve Scope" are ntroduced nto IWS. By ths way, the mportance of ndvdual sensng scope s quantfed. The waken up probablty s decded by the quantfed value. More mportant degree wll be assgned a larger wake up probablty. In ths way, we reduce the number of workng nodes whle keeps Survellance performance and save more energy. A. Importance Degree Accordng to the eperence of prevous blue-green algae survellance, the green algae outbreak possblty dffers n dfferent water areas, t s nfluenced by water temperature weather condtons even geologcal terran. Correspondngly, water area whch has a hgh possblty of green algae outbreak, should be allocated more sensng resource to assure the tmely outbreak predcton. In our desgn, the survellance zone s dvded nto a lot of dscrete grds. Defne the mportance degree of a grd as the frequency of targets appearng n the grd. The smlar dea Certanty Grds has been used successfully n several moble robot control programs [8]. It can be obtaned through a pror nformaton. The dfferent mportance of the grds n the survellance zone can be drectly dsplayed by ther dfferent mportance degrees. The hgher the mportance degree s, the hgher the frequency targets appear, and the more mportance the grd s. Consder the case where the survellance zone D s a 2- Dmenson regon, then t can be dvded nto m n grds D g and an mportance degree matr s defned as mn follows, A degree of grd g. a, wth mn a representng the mportance In ths paper, each grd s be sensed wth an eponentally dstrbuted duraton generated accordng to a probablty at densty functon f () t ae, where a s the mportance degree of the grd, t denotes the tme duraton and s set by msson requrements. The more mportant the grd s, the hgher frequency t should be sensed. Obvously, E t 1/ a s actually the mean of ntervals to be sensed. Then the probablty dstrbuton functon of grd g at tme t s as followed: at F ( t) P( t) 1 e (1) At tme t, the larger a s, the larger the value of F () t s. B. Effectve Scope of Node In ths paper, the Vorono dagram s used to defne the effectve scope of the nodes. Let S { p1, p2,, p n } be an aggregaton of ponts n a two-dmensonal Eucldean plane. These ponts are called stes. A Vorono dagram decomposes the space nto regons around each ste, so that the ponts n the grds around p are closer to p than any other pont n S. Usng the defnton n [9], the Vorono regon V( p ) for each p s epressed as: V( p ) { : d( p, ) d( p, ), j } (2) j V( p ) conssts of all ponts that are closer to p than any other ste. The aggregaton of all stes form the Vorono Dagram V() s. If there are denser nodes n survellance zone, the vorono regon acreage wll be smaller because of large overlapped proporton. C. The Algorthm Here we wll propose the IWS (Importance Degree based Wakenng Scheme) that calculates node wakenng probablty accordng to the dfferent mportance degrees of survellance grds. To smplfy the problem, we assume that all nodes have the same sensng range r and communcaton range R. We consder the S-MAC protocol [10] for the wreless sensor networks communcaton, where n each operaton perod, also called Round, the nodes operate n two phases. In the frst phase, the node determnes whether t should be awake or not stochastcally accordng to the wakenng probablty. If t chooses wakenng state, t wll mplement detecton and may further do some smple computaton. Otherwse t wll just turn off so that the energy can be saved. In the second phase, the node communcates wth ts neghbors through recevng or sendng nformaton. The operaton flow can be represented as Fg.1. Round Wakng/Sleepng R/S Wakng/Sleepng R/S... Fgure 1. Operaton flow of ant nodes operaton. c Tme The sensor detecton model s consdered as followed: 756

3 1 : d(, ) r, R C () (3) 0 : d(, ) r, R Where R s the zone of survellance, d(, ) denotes the dstance between node and geographcal locaton pont. Gven sensor nodes S { p1, p2,, p n }, the Vorono Dagram s: V( s) V ( p1), V( p2),, V( p n ) where V( p ) s the Vorono regon satsfyng (2). Then, we defne a mappng from V( p ) to U( p ), as followed: U( p ) { g : g V( p )} (4) Defne the weght of the Vorono regon V( p ) at tme t I( p, t ) as the sum of F () t n the regon. Gven F () t and U( p ), I( p, t ) can be calculated as follows: I( p, t) F ( t) (5) g U ( p ) I( p, t ) eplcates the total amount of probablty of the grds to be sensed. As we can deduce from (5) that numerous grds and hgh probablty F result n the large weght I. Obvously, the large I( p, t ) may ndcate an urgent detecton task on the node p. Then the node wakenng probablty may be calculated from I( p, t ). In ths paper, we construct the calculaton as follows: I( p, t) N W (, t) mn C,1 (6) M Where N s the amount of sensor nodes n the survellance zone, C s the parameter chosen by epermentaton, and M s the total amount of grds contaned n survellance zone D that can be calculated as follows: D (7) After the node wakenng probablty W(, t ) s calculated, the sensor node can make a stochastc decson on swtchng from sleepng to actvty accordngly. If a target s detected by an actve node, the nodes around t wll be aroused. Otherwse they wll return to sleep and set t to 0. the locaton nformaton s avalable to the sensor node ether through hardware such as embedded GPS or through locaton dscovery algorthms [11]. Furthermore, we assume that the rado communcaton radus of sensor node always meets the crtcal densty[12] condtons, whch means the network s connected all the tme. Hence the desred area to montor of a sensor node s the polygon that s defned by the Vorono dagram. As fgure 2 shows, the nodes do not dstrbute evenly due to random confguraton n real applcatons. Here the sensng range r n (3) s set to be 20. Suppose that before WSN deployment, we have assgned proper mportance degrees to grds accordng to temperature, weather condtons and geologcal terran (Hgh outbreak possblty wll be assgned hgh mportant degree) Fgure 2. The sensor nodes and ts Vorono dagram The dfferent mportance degrees n the survellance zone are showed n fg 3. Defne the whole survellance zone as D, where the mportance locaton s fgured as the green pentagram named G. G may be a road or a battlefeld etc. The sensng msson requres the grds n G to be detected n t 1 =2.5 second nterval whle the other grds to be detected n t 2 =5 second nterval. Then a can be calculated as follows: a 1/ t1 g G 1/ t2 g D G (8) IV. EXPERIMENTAL ANALYSIS In ths secton, we present detaled smulaton results to verfy the effectveness of our algorthm. To evaluate the algorthm performance and energy savng result, and to evaluate the parameter choce of "C" n algorthm performance, we use algorthm wth fed wakenng probablty as baselne algorthm. We also adopt dfferent value of C to conduct the smulaton. A. Smulaton Envronment Suppose that the blue-green algae survellance range s , and s dvded nto a lot of dscrete grds wth There are 100 sensor nodes randomly scattered n the zone, and Fgure 3. The mportance locaton fgured as the area wth green pentagram 757

4 Gven sensor node and grd unt g, the correspondng grds set U( p ) can be calculated accordng to (4). F () t and I( p, t ) can be calculated accordng to (1) and (5) respectvely. At last, the nodes wakenng probablty can be calculated accordng to (6) and (7). In order to evaluate IWS n a dense targets envronment, we generate the targets (green algae outbreak event) stochastcally wth the number K 50,100,150, 200 respectvely. And the probablty targets appearng n G (the mportance locaton) s twce of that n the other locaton. Fg 4 s a snapshot of 200 targets generated n our smulaton. Fgure 5. Smulaton tme wth dfferent target numbers of IWS Fgure 4. Targets appearng n the survellance zone B. Results Analyss Scenaro. Before each smulaton, we ntalze the system by settng the tme T 0 =0. Every grd s set to actvty accordng to the calculated node wakenng probablty. At tme T 0 +1, K random targets occur n the survellance area. When all the targets are detected the smulaton s demonstrated successful and the smulaton tme T s recorded. If there are stll targets fal to be detected by the tme T 0 +5, then the smulaton s demonstrated a falure. Evaluaton. In most wreless sensor networks, energy s a key performance crteron. In our eperment, energy E s defned as the accumulated tmes of wakenng nodes durng one smulaton. The smaller E s, the less energy the system consumes. Smulaton tme T demonstrates the aglty of the system to targets. The number of faled smulatons n the eperment s accumulated to reflect the robustness of the algorthm. Comparson. To see whether IWS s vald or not, we compare t wth the algorthm (named p-algorthm) wth fed wakenng probablty p evaluated n the nterval [0, 1]. Whle the parameter C n IWS s evaluated n the nterval [0, 10]. In the smulaton, the number of smulatons s set to 30. Let T and E be the mean value of 30 smulatons respectvely and F be the tmes of faled smulatons n the eperment. The results are as follows: Fgure 6. Smulaton tme wth dfferent target numbers of p-algorthm Fgure 5 and 6 show that smulaton tme s attenuatng wth the ncrease of parameter C and probablty p. When C>2 and p>0.9, smulaton tme remans stable at the value of 6, whch means the targets can be detected as soon as they are generated. Fgure 7. Energy cost wth dfferent target numbers of IWS 758

5 Fgure 8. Energy cost wth dfferent target numbers of p-algorthm Fgure 7 and 8 llustrate that energy cost ncreases monotonously wth parameter C and probablty p. Compared to p- algorthm, the energy cost of IWS s robust to dfferent targets number K. From fgure 5-8, we can also observe that p- algorthm usually costs more energy than IWS wth the same smulaton tme T. Fgure 9 and 10 show the relatonshp of the falure tmes wth parameter C and p when dfferent targets numbers are appled. These fgures ndcate that falure tme ncreases when ether of parameter C and p decreases. Meanwhle, t s clear that wth respect to the p-algorthm, falure tmes n IWS are less affected by the targets number K, whch mples the superor robustness of IWS. From the above smulaton results we can draw the followng conclusons: The energy cost E s reduced by the decrease of C and p. Besdes, the falure tmes wll ncrease f the parameter C or p decreases. T and E wll ncrease through addng targets number K. Thereby, t s vtal to reasonably choose the parameters C accordng to dfferent applcatons n order to ensure the optmzng performance of the system. IWS s superor to p-algorthm n performance. The man reason s that IWS not only calculates the geographcal locatons of the sensor nodes but also utlzes mportance degree based schedulng scheme to rase effcency. Fgure 9. Falure tme wth dfferent target numbers of IWS Fgure 10. Falure tme wth dfferent target numbers of p-algorthm V. CONCLUSIONS We propose the IWS to save network resources through consderng dfferent mportance degrees of blue-green algae survellance locatons. In the algorthm, mportance degree s ntroduced to reflect the dfferent mportance of the grds n the survellance zone. It s used to calculate the wakenng probablty whch determnes the modes of the sensor nodes. Our conducted green algae survellance smulaton results show that our approach s robust and energy effcent. Moreover, a compromse s acheved between energy cost and system aglty. In the future, we wll focus on the wake up mechansm based on comprehensve factor combnaton of regon mportance degree, sensor node energy and network connectvty. ACKNOWLEDGMENT Ths work was supported n part by the Natonal Basc Research Program of Chna (973 Program) under Grant No.2011CB302803, Natonal Natural Scence Foundaton of Chna under Grant No , Beng Natural Scence Foundaton under Grant No and the Natonal S&T Major Project of Chna under Grant No. 2010ZX

6 REFERENCES [1] Dong L, Ze Zhao, L Cu, He Zhu, Le Zhang, Zhaolang Zhang, and Y Wang. The Desgn and Implementaton of a Survellance and Self- Drven Cleanup System for Blue-green Algae Bloom on Lake Ta. The 7th IEEE Internatonal Conference on Moble Ad-hoc and Sensor Systems (MASS 2010), Demo Paper, pages , San Francsco, CA, USA, Nov., [2] S. Meguerdchan, F. Koushanfar, M. Potkonjak, and M.Srvastava. Coverage Problems n Wreless Ad-Hoc Sensor Network, Proc. IEEE INFOCOM 01, pp , [3] D. Tan and N. D. Georganas. A coverage-preservng node schedulng scheme for large wreless sensor networks. In Frst ACM Internatonal Workshop on Wreless Sensor Networks and Applcatons, pages 32 41, [4] F. Ye, G. Zhong, S. Lu, and L. Zhang. PEAS: A Robust Energy Conservng Protocol for Long-lved Sensor Networks. In Proc. of Internatonal Conference on Dstrbuted Computng Systems (ICDCS), May [5] Lu, Changle; Cao, Guohong;, "Spatal-Temporal Coverage Optmzaton n Wreless Sensor Networks," Moble Computng, IEEE Transactons on, vol.10, no.4, pp , Aprl 2011 [6] Dezun Dong; Yunhao Lu; Kebn Lu; Xangke Lao;, "Dstrbuted Coverage n Wreless Ad Hoc and Sensor Networks by Topologcal Graph Approaches," Dstrbuted Computng Systems (ICDCS), 2010 IEEE 30th Internatonal Conference on, vol., no., pp , June 2010 [7] Yang Xao; Hu Chen; Ku Wu; Bo Sun; Yng Zhang; Xnyu Sun; Chong Lu;, "Coverage and Detecton of a Randomzed Schedulng Algorthm n Wreless Sensor Networks," Computers, IEEE Transactons on, vol.59, no.4, pp , Aprl 2010 [8] H. P. Moravec, Certanty grds for moble robots, Techncal Report Preprnt, The Robotcs Inst., Carnege-Mellon Unv., 1986 [9] Vera M A M, et al. Schedulng nodes n wreless sensor networks: a Vorono approach[a]. Local Computer Networks[C], LCN '03. Proceedngs. 28th Annual IEEE Internatonal Conference on Oct Page(s): [10] We Ye, Hedemann, J, and Estrn, D, An energy-effcent MAC protocol for wreless sensor networks, INFOCOM Twenty-Frst Annual Jont Conference of the IEEE Computer and Communcatons Socetes. Proceedngs. IEEE Vol 3, pp June 2002 [11] He T, Huang CD, Blum BM, Stankovc JA, Abdelzaher T. Range-Free localzaton schemes n large scale sensor networks, In: Proc. of the 9th Annual Int'1 Conf. on Moble Computng and Networkng. San Dego: ACM Press, [12] Adlakha, S.; Srvastava, M. Crtcal densty thresholds for coverage n wreless sensor networks, Wreless Communcatons and Networkng, WCNC IEEE, vol.3, no., pp vol.3, March