Optimized Forwarding for Wireless Sensor Networks by Fuzzy Inference System

Transcription

1 Optmzed Forwardng for Wreess Sensor Networs by Fuzzy Inference System Mohammad Abdu Azm and Abbas Jamapour Schoo of Eectrca and Informaton Engneerng The Unversty of Sydney, NSW 6, Austraa {azm, Abstract For the fat wreess sensor networs, we nvestgate the optmzaton probem of the routng path based on the metrcs: tance, and n usage to maxmze the fetme of the sensor networs. We empoy the we nown fuzzy nference systems (FIS) for the seecton of the best node, from the canddate nodes, n order to forward pacet to the sn. Smuaton resuts show that networ fetme can be mproved by empoyng the optmzed routng protoco.. Introducton Wreess sensor networ (WSN) s composed of cheap and tny unreabe sensors wth mted resources, where the sensors possess sensng, computng and communcatng capabtes []. Due to the advancement of the mcro eectro mechanca system (MEMS) technoogy and sensor networs prospectve dversfed appcatons (such as home automaton, ndustra montorng, mtary, envronmenta and many more) WSN s expectng a huge growth n near future and aso experencng an ntense research nterest. Sensor networs are generay consdered as composed of randomy and densey depoyed arge number of nodes. Based on the underyng networ structure WSN can be fat or herarchca. In the fat networs a the sensor nodes perform the same functons, on the other hand n the herarchca networs the hgher energy nodes nown as custer hea mantan the custer, aggregate data from the non-custer head sensor node and transmt the congomerated data to the sn. Dependng on how source fn a route to the destnaton, routng protocos n sensor networs coud be ether proactve or reactve. In proactve routng, routes are computed before they are needed; on the other hand a reactve routng cacuates the route ony when t s needed. The desgn consttuent of the routng protoco depen many on the appcaton because of the appcaton s traffc demand and pattern may vary enormousy. Power consumpton, mobty, scaabty and QoS are the other most sgnfcant ssues n desgnng routng protocos n WSN. Today's man chaenge for the desgners and deveopers of protocos and appcatons for WSN s the resource scarcty of nodes, most mportanty ts avaabty, snce n sensor networs the battery fe s consdered as the networ fe. To extend the sensor networ fetme, we utze the fuzzy nference system (FIS) that optmzes the routng path (dependng on the metrcs: tance, remanng battery and n usage) n a trbuted fashon. The remanng paper s organzed as foows. Secton II, descrbes some reated wors consders the metrcs: tance, energy and oad trbuton. Secton III states the probem statement. In secton IV, we present our protoco, ts advantages and drawbacs. In Secton V, we present our smuaton resuts and fnay n Secton VI, the ey concusons and the future wors are stated.. Bacground A number of protocos have been proposed n the area of sensor routng. Reference [] proposes ow energy adaptve custerng herarchy (LEACH) a custer based herarchca networ routng protoco, where the sensor nodes transmt to the custer head drecty. Custer hea then transmt the data to the sn. In [], authors propose a varant of herarchca agorthm where, the sensor nodes forward the data by severa hops, optmzed by the Djstra s [4] agorthm. Reference [5] stochastcay trbutes the oad by choosng a random node from the forwardng path. Aternatvey, [6] proposes the agorthm where a probabty s assgn to each node for oad trbuton. Here, the probabty s nversey proportona to the cost functon of the partcuar path and the data s forwarded based on the desgnated probabty. Reference [7] proposes a protoco that combnes both stochastc and cost based schemes ntroduces n [5] and [6]. Reference [8] proposes the routng protoco that seects the hghest energy node from the forwardng tabe to forward data. In an mproved verson of LEACH, a recent paper proposes, herarchca battery aware routng (H-BAR) [9]. Protoco seects the hghest battery ed node as the CH. H-BAR shows a favorabe mprovement n the

2 performance. Reference [] proposes geographca mutpath routng protoco (GMR) based on the ocaton nformaton. In [], energy and mobty s consdered n addton to GMR. They optmze a the metrcs by FIS for ther mproved routng protoco, energy and mobty aware geographc routng protoco (EM-GMR).. Probem Statement From the aforementoned teratures we fnd some very smpe crteron to engthen the fetme of the sensor networs. These ncude: Sma mutpe hops: As the energy consumed for the transmsson s proportona to the square of the tance from sender to recever, mutpe short hops s preferabe nstead of a snge arge hop. Shortest path: Shortest path from the sender to recever s the straght ne connectng the nodes. Forwardng pacets aong ths ne s more effcent than a detour. Load trbuton: In case, concentraton of events n some partcuar areas s more than that of other areas, usng shortest path w cause mposon aong the path. So unform trbuton of traffc s needed. Hghest remanng energy: Nodes havng greater remanng energy partcpates more than the nodes havng sma amount of can extent the networ fetme. Ths paper presents a souton that optmzes the routng path accordng to a the abovementoned crtera by a snge trbuted agorthm. 4. Proposed Routng Protoco 4.. Assumptons The proposed protoco assumes that the nodes can access ther own battery eve and transmt can be adjusted dependng on the tance of the destnaton. Protoco aso assumes that the sensors now ther ocaton nformaton. Sensors shpped wth the GPS recevers, can ready sense ts ocaton nformaton. Aternatvey, ocaton nformaton can aso be acqured through a ocazaton agorthm. In fact ocaton nformaton s mportant when an event occurs. Most of the appcatons w probaby need the nformaton to montor an nterested area, at east n a course gran. So, ths s very much justfabe to nfer that ocaton nformaton s avaabe to the nodes. Locazaton tsef s an ongong area of research and s not wthn the scope of our research. 4.. Goas Our man objectve of desgnng the protoco s to fnd an optma path from the avaabe metrcs; shortest path, mnmum tance, battery usage and number of pacets forwarded prevousy by the same n. Optmzng the path w resut n maxmzng the fe of the networ. 4.. Protoco Operaton Nodes coect the routng metrcs through the ocazaton agorthms, accessng ther own battery eve and eepng trac of the n usage. The protoco has the potenta to be mpemented n both the reactve and proactve manner. In reactve routng, when a node nee to transfer data t generates routng query and ass for ts snge hop neghbor s nformaton, n order to cacuate the routng path. On the other hand, proactve routng, updates the neghborng nodes by perodca broadcastng. When a data s needed to be sent the protoco seects the optma path through the FIS. Fnay, t adjust the transmt accordng to the tance of the recever node and forward the data. By usng the FIS [] we can ntegrate the dfferent types of metrcs (tance, battery and n usages n our case) even when the correaton between the metrcs s dffcut to mode mathematcay. Each node can mae trbuted forwardng decsons. Ths emnates the necessty of herarchca networs. Fg. shows an exampe networ where a source node nee to send a data pacet to the destnaton sn. The shortest path and the rado ranges are shown n the fgure too. To emnate the burden on the FIS agorthm t smpy car some of the nodes as a potenta canddate. Lght shaded nodes are carded, as they are not n the forward drecton. In ths case n,n and n Source n n Fg.. Sensor Networs. Sn

3 n are the potenta canddates Routng Matrces The routng metrcs are shown n the tabe above. Here, d s the tance of canddate nodes from the source, s the tance of the canddate nodes from the shortest path, whe p and denote the and the n usage respectvey. Here, a the metrcs are assumed to be normazed n order to mpement the fuzzy rues. Tabe : Routng Tabe. equaton () the heghts are gven as for d =.... h = t. C. Battery Used: For the battery usage the MF s set n such a way that, up to % there s tte effect of the usage. When the usage goes hgher %-7% t shows moderate resstance to forwardng. But when t s at 7%- % t shows the hghest resstance to forwardng a pacet. Let, p < p < p. Therefore, Node Dstance (d) Dstance () Power (p) Ln Usage () n d p n d p hp.. = p. p < p <=.. p f.< p <=.7.7< p <= n d p 4... Optmum Seecton by Fuzzy Logc n our case The frst step of desgnng fuzzy optmzaton requres characterzng the membershp functon (MF), whch gves the nput output reatons. MFs are dfferent for the dfferent metrcs. The nput parameters are the routng metrcs (x-axs) wth respect to the correspondng cost (yaxs) of the MF and the outputs are projected to form the trapezo as shown n fgure (a-d). For a partcuar node a dfferent trapezo are added up and fnay fndng the centrod maes decson. We w see the agorthm step by step for the case of WSN routng. 4...Membershp functons (s) A. Dstance from the node: As the s proportona to the square of the tance, n case of the frst order rado mode, the MF of the tance (from the node) s the curve as shown n Fg. (a). The tances d,, d and d are the nputs of the MF (Fg. (a)). Let, d < d <. Outputs, the projected trapezo are the d weghts for the correspondng nodes. The heght of the th trapezo for the node are defned as hd = t. d. () where t a vaues of d =.... s a constant for B. Dstance from the shortest path: The MF, n ths case, s the same as the prevous one because t s aso a tance. Inputs < <. The outputs are the correspondng trapezo (Fg. (b)). Smary as where hp and are the heght of the th node and the co-effcent respectvey (derved from the Fg (c)). D. Ln Usage: We use a near functon as a n usage MF. The more the path s used the more t becomes reuctant to forward the pacet. Here we assume < <. The heghts become h = n. where s the correspondng coeffcent. n Decson (based on A, B, C & D): A the four types of outputs are added and the weghted average s taen (bac crces n Fg. (e)). The area of the trapezo, are cacuated by the foowng expressons. A, j ( ( h = where A denotes the area of the trapezod and j s the th j membershp functon. In ths case j =,,, 4, j ) ). As the tota number of parameter consdered s 4. Therefore, the weghted averages are cacuated as. Av 4 = j= Let Av, Av and Av be the respectve weghted average of the nodes n, n and n. As, Av > Av > Av, n s the optma node that the source w forward to due to ts mnmum cost. A 4, j

4 d d d Dstance hd hd hd Ad Ad Ad n n n n n Fg. (a) Membershp Functon (Dstance from the Node) Dstance h h A, A, h A, n n n Fg. (b) Membershp Functon (Dstance form the Shortest Path) p p p Power Used hp A, Fg. (c) Membershp Functon (Power) Ln Usage 4.4. Dscusson h A,4 Fg. (d) Membershp Functon (Ln) Protoco Advantages hp A, hp A, n n n h A,4 h A,4 n n n As sensor networs coud be composed of a arge number of sensors, t s not desrabe that sensor networs w have a goba addressng scheme n Fg. (e) Weghted Average. because of ts huge mantenance overhead. Our protoco does not need to mantan the ID, hence the cost of goba addressng mechansm s saved and maes the networ scaabe. Nodes need to mantan a sma tabe because t nee ony to mantan the cost metrcs for the neghborng nodes. Thereby, the protoco saves the storage cost to store the routng tabe. It aso saves communcaton cost such as transmt-receve energy and bandwdth. In case of reactve mpementaton the protoco s fast n respondng to the networ dynamcs because of ts mnmum covery overhead. Optma seecton of the node saves data transfer cost. Transmsson energy s consdered the prmary consumer of the energy usage for wreess sensor networs. Nodes havng more remanng energy contrbute more to the forwardng of pacets. The protoco s far as t trbutes the woroad of forwardng data eveny. For the custer based sensor networs, the faure of a custer head may cause the whoe custer to become non- operatona. Moreover, for mantanng the custers, (seecton or eecton of the custer hea, and nodes jonng to the custers) requres contro message exchange. Ths overhead may be consdered as an extra burden to the resource crtca sensor nodes. By usng a fat archtecture ths protoco emnates both the aforementoned ssues Lmtatons Faure reachng the sn: The protoco w fa to converge n the presence of vo or dead en even when there exsts a routng path through farther nodes. The souton to the faure s to ocay food the networ to fnd a path. In the

5 worst case scenaro, when the oca foodng aso fas then foodng the whoe networ becomes an opton. Permeter routng [] can aso be used where message traverse through the face of ntersectng ne between source and destnaton thereby gudng the pacets out of the oca mnma. Processng cost: To run the fuzzy agorthm nstructons, nodes requre some amount of battery because of the agorthm compexty. However, the processor wthn the sensor node consumes sgnfcanty ess energy than the transmtter. The amount of requres to transmt -bt to m tance s equvaent to the amount of requres to run mons of nstructons []. The protoco expots the reaton and uses t favoraby.e. uses cacuaton to optmze transmsson Normazed tance (meters) Normazed tance (meters) Fg.. Randomy depoyed Sensor Nodes. 5. Performance Evauaton To evauate the performance of the protoco, we smuate the protoco n MATLAB. We appy the same rado mode ntroduced n [] and used by severa papers [9]. In ths rado mode, the transmsson and receve cost s defned as EnT ( ) = Eec. + Eamp.. d and EnR ( ) = Eec. respectvey, where, s the number of bt per pacet, d s the tance, Eec and Eamp are 5nj/bt and pj/bt/m^ respectvey. For the smuaton, we randomy depoy statc sensor nodes n m x m fed, wth the sensors transmttng over a radus of m (Fg.). The sn s movng randomy as shown n Fg. 4. Each sensor generates pacet randomy, checs whether the sn s wthn ts drect rado range. If yes, the node drecty transmts the data to the sn otherwse va the ntermedate nodes accordng to the proposed protoco. The networ w become parttoned and communcatons w degrade drastcay when too many nodes de. For the reason we ony evauate the frst deaths. We observe the performance of the optmzatons usng dfferent combnatons of the metrc parameters. As shown n Fg. 5 the hghest performance s found when a the four metrcs are consdered and optmzed. In rea word, the generatng pacets coud be nonunform. We expect that the protoco w perform even better s such a case. 6. Concusons Motvated by the sensor fetme eongaton probem, we optmzed the tances, energy and n usage to semnate data for a statc sensor bed where the ony Normazed tance (meters) Number of transmsson Normazed tance (meters) Fg. 4. Sn Mobty Mode. d dp dp Number of dead nodes Fg. 5. Performance (Number of Transmsson vs. dead nodes). mobe entty s the sn. Smuaton resuts show that the networs fetme coud be extended by the scheme. In ths wor we ony nvestgated the eveny trbuted traffc pattern. We w extend our wor to ncude hot spots n the networs. We aso ntend to ntegrate mobty as an addtona metrc n the routng protoco for mobe sensors.

6 7. References [] J. N. A-Kara, A. E. Kama, Routng Technques n Wreess Sensor Networs: A Survey, IEEE Wreess Communcatons, December 5, pp [] W. Henzeman, A. Chandraasan, and H. Baarshanan, An Appcaton-Specfc Protoco Archtecture for Wreess Mcrosensor Networs, IEEE Transactons on Wreess Comuncatons, Vo, No.4, October, pp [] M. Youns, M. Youssef and K. Arsha, Energy-Aware Routng n Custer-Based Sensor Networs, Modeng, Anayss and Smuaton of Computer and Teecommuncaton Systems, October, pp [4] W. Stangs, Data and Computer Communcatons, Macman Pubshng Company, rd edton, 99. [5] C. Schurgers and M. Srvastava, Energy Effcent Routng n Wreess Sensor Networs, Mtary Communcatons Conference, v., October, pp [6] R. C. Shah and J. M. Rabaey, Energy Aware Routng for Low Energy Ad hoc Sensor Networs, Wreess Communcatons and Networng Conference, v., March, pp [7] L. Gan, J. Lu, and X. Jn, Agent-Based, Energy Effcent Routng n Sensor Networs, IEEE Autonomous Agent and Mutagent Systems, 9- Juy 4, pp [8] X. Hong, M. Gera, H. Wang and L. Care, Load Baanced, energy-aware communcatons for mars sensor networs, IEEE Aerospace Conference,, pp [9] R. Musunur and J.A. Cobb, Herarchca-Battery Aware Routng n Wreess Sensor Networs, Vehcuar Technoogy Conference, 5. [] R. Jan, A. Pur and R. Sengupta, Geographca Routng Usng Parta Informaton for Wress Ad hoc Networs, IEEE Persona communcatons, February, pp [] Q. Lang and Q. Ren, Energy and Mobty Aware Geographc Mutpath Routng for Wreess Sensor Networs, Wreess Communcatons and Networng Conference, 5, pp [] J. M. Mende, Fuzzy Logc Systems for Engneerng: A Proceedngs of the IEEE vo. 8 no. Tutora, March 995, pp [] B. Kerp and H. T. Kung, GPSR: Greedy Permeter Stateess Routng for Wreess Networs, ACM MOBICOM.