Energy Aware Distributed Estimator System over Wireless Sensor Networks with Ad-hoc On-Demand Distance Vector (AODV) Routing Algorithm

Transcription

1 Vol.8/o.1 (016) ITERETWORKIG IDOESIA JOURAL 3 Energy Aware Dstrbuted Estmator System over Wreless etwors wth Ad-hoc On-Demand Dstance Vector () Routng Algorthm Husnul Abady and Endra Joelanto, Member, IEEE Abstract Wreless sensor networs (WSs) have many applcatons such as envronmental montorng, ndustral montorng and control, structural health montorng. WSs have receved consderable attenton n recent years. The paper presents comparson of the performance of the combned centralzed-decentralzed Kalman flter wth the centralzed Kalman flter and the decentralzed Kalman flter for state estmaton over wreless sensor networ. The wreless sensor networ uses energy aware routng algorthm to reduce energy consumpton of sensors. In addton, the paper compares the energy consumpton of each archtecture and the dfference between the energy aware and the regular algorthm. Index Terms routng algorthms, Centralzed, Decentralzed, Estmator, Kalman Flter, Wreless etwor (WS). A I. ITRODUCTIO Wreless etwors (WS) s a networ that conssts of many sensors connected wrelessly wth lmted memory unt, lmted calculatons and communcaton capabltes, and lmted energy resources [1]. The progress of WS has promoted multple sensor fuson and state estmaton as an mportant and emergng topc cover many applcaton areas. A wreless sensor networ (WS) [] s spatally dstrbuted autonomous sensors to montor (physcal or envronmental condtons, such as temperature, flow, sound, pressure, etc.) and to cooperatvely pass ther data through the networ to a man locaton. Modern networs are bdrectonal, also have enablng control of sensor actvtes. The WS conssts of nodes, from one or two to several hundreds or thousands. Every node s lned to one sensor or a number of sensors. Each sensor networ node comprses some parts such as a rado transcever ncludng an nternal antenna or ln to an external antenna, a mcrocontroller, an electronc Manuscrpt receved July 6, 015. Ths wor was supported n part by the Decentralzed Research rogram, DIKTI, Indonesa, 01. H. Abady was wth the Engneerng hyscs Master rogram, Faculty of Industral Technology, Bandung Insttute of Technology, Bandung 4013, Indonesa. E. Joelanto s wth the Instrumentaton and Control Research Group, Faculty of Industral Technology, Bandung Insttute of Technology, Bandung 4013, Indonesa (correspondng author e-mal: ejoel@ tf.tb.ac.d). devce for nterfacng wth the sensors and an energy source from a battery or an energy harvestng form. The sze of a sensor node mght alter from that of a small box down to a gran. The cost of sensor nodes s also vared from a few to hundreds of dollars determned by the complexty of the pecular sensor nodes. Constrants on sze and cost of sensor nodes lead to lmtatons on propertes of the nodes, for examples: energy supply, memory capacty, computatonal speed and bandwdth n communcatons. In the WSs, the topology can be desgned from a smple star networ to a complex mult-hop wreless mesh networ. Routng or floodng s the transmsson technque among the hops of the networ [3][4]. A WS mght be a networ that has flexble topology governed by a routng algorthm. One of the most popular routng algorthms s nown as Ad-hoc On- Demand Dstance Vector (). Unfortunately, the algorthm does not account for energy effcency. As energy s an essental part n wreless networs for the networ lfetme, t s of nterest to have an archtecture and a networ topology based on energy effcency consderaton [1][5][6]. The (Ad-hoc On-Demand Dstance Vector) routng protocol [7] s a reactve routng protocol based on some characterstcs of practcal routng protocols. In ths protocol, routes are establshed on-demand, once establshed, a route s retaned as long as t s requred. On-demand (reactve) routng protocols locate a path between the source and the destnaton only when the path s requred for data exchange. The routng method greatly reduces the routng expenses, but t may produce large delay from the moment the route s requred to send a pacet untl the tme the route s actually obtaned. In, f there s no requrement of connectons, the networ remans slent. At that pont n tme, the networ node that requres a connecton announces a connecton request. Other nodes pass on ths message and record the node, then mang a sudden ncrease of temporary routes bac to the needy node. When a node pcs up such a message and already establshes a route to the desred node, t dspatches a message bacwards through a provsonal route to the requestng node. The needy node then begns usng the route that has the least number of hops through other nodes. Vacant entres n the routng tables are reused afterwards. When a ln breas down, a routng error s returned to a transmttng node, and the ISS: / 016 IIJ

2 4 ITERETWORKIG IDOESIA JOURAL ABADY & JOELIATO process s repeated agan. The has advantages as t creates no extra traffc for communcaton along exstng lns, and the dstance vector routng s smple, less memory requrement or calculaton. Conversely, the needs more tme to begn a connecton and has heaver ntal communcaton to set up a route than other methods. The estmators n WS mght have several archtectures ncludng centralzed and decentralzed archtectures [8][9][10][11]. Estmator uses Kalman flter algorthm to combne sensor data and estmate the state of process. In ths paper, the centralzed and the decentralzed archtectures are combned n a WS and use the energy-aware routng algorthm n order to produce better performance WS. II. ROBLEM FORMULATIO In ths paper, t s consdered a closed loop control system wth measurement carred out va a sensor networ system as llustrated n Fg. 1. The control system conssts of the plant and sensors. The paper only consders the problem of state estmaton over the sensor networ system of the closed loop control system. The process dynamcs of the plant and the measurement equaton of a sensor are gven by a dscrete tme state space equaton where x x Ax w (1) y n s the state, a sensor (=1,, ). 1 1 C x v () w q y s measurement output of n and v q are the process nose and the measurement nose for tme step. The noses are whte Gaussan nose wth zero mean and covarance matrx Q 0 and R 0 respectvely. Fg. 1. A closed loop control system wth a sensor networ system Estmators have several archtectures such as centralzaton and decentralzaton. Both archtectures have ther advantages and dsadvantages respectvely. As t s possble to defne the archtecture easly over a large sensor networ, n ths paper, t s consdered to combne the two archtectures n the networ. The performance of the combned archtecture s then compared wth the centralzaton and the decentralzaton archtectures. In the control system, sensors transmt data measurement to an estmator or a fuson center. The estmator combnes the data and computes the optmal state estmaton. Flter Kalman and ts varants are the well nown optmal state estmaton. Data transmsson route n the networ s governed by a routng algorthm. The wdely used routng algorthm s routng algorthm [7]. routng algorthm does not use energy parameter n the route dscovery. Therefore, t s of nterest to consder energy threshold n the routng algorthm. In ths paper, the energy threshold s calculated by usng Current Average Energy of The etwor (CAE) [1]. Ths consders other factors that may affect the performances of wreless communcaton nstead of the excessve roamng actvtes consdered n [13], acton latency [14][15], stablzaton and control networ [15][16]. The dfference n energy consumpton between the energy aware and the regular algorthm wll be shown n the paper. The Flter Kalman estmator over WS usng the centralzed-decentralzed archtecture wth energy aware routng algorthm s then smulated n order to show the effectveness of the consdered archtectures over other archtectures and routng algorthm. Kalman flter can be combned wth other algorthms n order to mprove ts performance, such as dscussed n [17]. III. EERGY AWARE ROUTIG ALGORITHM An energy aware algorthm s an extenson of the algorthm wth the addton of energy consderaton n the route dscovery. The algorthm s modfed by usng the followng steps. A. Revsed acets Hello pacet: Hello pacet s used to exchange nformaton of remanng energy wth neghborng sensors. The new data Remanng Energy (RE) wll be added n order to store energy nformaton. The new nformaton s also added to local Remanng Energy Table (LRET) whch records the remanng energy of all one hop neghbors. RREQ pacet: RREQ s also modfed n order to obtan accumulaton of local average remanng energy nformaton. The nformaton s denoted as E sum and each ntermedate sensor wll update E sum by cumulatng ts local average energy to t. B. Route Dscovery Modfcaton Route dscovery process occurs when a source sensor needs to send data to a destnaton sensor. Ths process ncludes the operatons at the source sensor and ntermedate sensors.

3 Vol.8/o.1 (016) ITERETWORKIG IDOESIA JOURAL 5 Source : When a source sensor would le to communcate wth a destnaton sensor and no route s avalable, the source sensor then broadcasts RREQ. As the frst sensor on the route, the source sensor ntates E sum wth ts local average energy defned as [1]: s n s Er 1 Er Esum El (3) n 1 s where Er denotes the remanng energy of the source sensor, Er denotes the remanng energy of ts 1-hop neghbor sensors and n denotes the number of neghbors. Intermedate : Once sensor receves RREQ, an ntermedate sensor decdes ts own response to the RREQ receved by comparng ts remanng energy wth energy threshold whch s descrbed as follows. Followng the same procedures as that n the source sensor, the ntermedate sensor calculates ts local average energy usng the followng equaton [1]: Et El (4) n 1 Where Et s the total remanng energy of sensor and ts neghbors, the number of whch s defned as n. There s a new parameter called Current Average Energy of The etwor (CAE), whch can be calculated as:[1] Esum El Ec (5) Where s the number of nodes along the bacward path, and E sum s accumulaton of local average energy on route. Gven CAE, the energy threshold can be calculated:[1] E thres E (6) where α s system parameter, 0 < α < 1. When the remanng energy of ntermedate sensors s greater than the energy threshold, t s consdered that the sensors have enough energy. These sensors then accumulate ther local average energy to E sum and contnue to broadcast RREQ. Otherwse, the sensors should wat for a whle and determne whether to forward RREQ or drop RREQ based on the number of dentcal RREQ receved durng the delay perod. c w v lant x +1 = A x + w y = C x +v Fg.. Centralzed Kalman Flter y 1 y 1 JS Fuson Estmas x dan ye, x The advantage of centralzed archtecture s each sensor may not have computng capabltes because sensors only measure the varable and then transmt data to the fuson center. The energy consumpton and bandwdth are also relatvely small compared to the decentralzed archtecture. However, the centralzed Kalman flter calculates all of the process n the fuson center. Ths s the man weaness n case of falure n the centralzed Kalman flter. Kalman flter algorthm for centralzed can be formulated as follows: [9] x Ax (7) 1 K T A 1 A Q (8) T T 1 ( C ) ( C ( C ) R) (9) e y C x (10) 1 x x ( 1 Ke ) (11) 1 [ I ( 1 KC )] (1) The varable ndcates the -th sensor. B. Decentralzed Kalman Flter In ths archtecture (Fg. 3), the flter has two nds of flter whch are local flter and global flter. The local flter s n each sensor whle the global flter can occur n each sensor or n the fuson center. Because sensors have the local flter and the global flter separately, there s no pont of falure. IV. ESTIMATOR In ths paper, we consder Kalman flter algorthm to estmate the state varables and to combne data measurement from sensors. Kalman flter s desgned to have multple archtectures such as centralzed and decentralzed. A. Centralzed Kalman Flter In the centralzed Kalman flter (Fg. ), the sensor measurement s sent drectly to the fuson center through communcaton lnes. The fuson center s used to combne the data measurement from sensors and to estmate the states. The transmtted data are the measurement receved from sensors. Fg. 3 Decentralzed Kalman Flter ISS: / 016 IIJ

4 6 ITERETWORKIG IDOESIA JOURAL ABADY & JOELIATO The local flter at each sensor uses ts measurement to obtan estmaton of the local state, x and the error covarance,. Both of these data are then transmtted to a global flter n surroundng sensors. The global flter combnes the data. The data receved n the global flter are State Error Informaton (SEI) and Varance Error Informaton (VEI). Once these data receved by each sensor, the global flter estmates the new state and the error covarance. The results are then used to update the data n the local flter on each sensor. At the local flter, Kalman flter algorthm s added wth SEI and VEI as follows: [9] x Ax (13) 1 K T A 1 A Q (14) T T 1 C ( C C R) (15) x x K ( y Cx ) (16) ( I K C) (17) SEI x x (18) VEI (19) The algorthm n the global flter s gven by: [9] ( 1VEI ) (0) 1 1 x ( x 1 ( 1 SEI )) (1) T 1 A A Q () x1 Ax (3) a. If sensors have a route to the fuson center, the measurement data wll be sent. If sensors do not have a route, then sensors wll broadcast RREQ. b. If sensors are not a fuson center, sensors that receved RREQ wll calculate the energy threshold n Equaton (3)-(6). c. If the energy threshold s less than the energy battery, sensors wll broadcast RREQ to other sensors. Otherwse, sensors wll wat whether to contnue or to drop the RREQ whch s based on the number of dentcal RREQ receved by sensor. d. If sensors are fuson center, sensors wll send an RRE to the source sensor. e. When the source sensor receved RRE, route has been obtaned. 3. s wll transmt measurement data after the route s obtaned. 4. Fuson center receves measurement data from the sensor and processes the data n the local flter. 5. The results of the local flter (SEI and VEI) are sent to another fuson center. 6. Fuson center receves SEI and VEI from another fuson center and then processes the data on the global flter. 7. The results of global flter then wll be used to update the data n the local flter at the fuson center. C. Centralzed-Decentralzed Kalman Flter In the centralzed-decentralzed archtecture (Fg. 4), the sensors are dvded nto several areas. Each area has ts own fuson center. The local flter n fuson center has the same algorthm wth the regular Kalman flter algorthm wth addton SEI and VEI calculaton. The result of ths local flter s then transmtted to another fuson center and processed n a global flter. The result of the global flter (state estmaton and error covarance) s used to update the data n the local flter at each fuson center. The Kalman flter algorthm s the same as the local flter algorthm n Equaton (7)-(1), and Equaton (18)-(19). The algorthm n the global flter s same wth Equaton (0)-(3). V. ESTIMATOR DESIG I WS In the paper, the man consdered archtecture s descrbed by the centralzed-decentralzed archtecture usng energy aware routng algorthm to organze the topology of WS. The algorthm s descrbed as follow: 1. s receve measurement data.. Energy aware routng algorthm ntates accordng to the logc as follow: Fg. 4. Centralzed-Decentralzed Kalman Flter VI. SIMULATIO As an example, the consdered system s a heat transfer system n a closed room wth the followng partal dfferental equaton: u ( u u ) (4) t x y wth boundary condtons descrbed as follow:

5 Vol.8/o.1 (016) ITERETWORKIG IDOESIA JOURAL 7 u x t,0, y u x t, l, y u y t, x,0 u y t, x, l 0 (5) where x and y are the coordnates of the area, u(t,x,y) ndcates the temperature at tme t at locaton (x,y), and α s the speed of dffuson process. Equaton (4) s converted nto dscrete form usng a fnte dfference method. The area s assumed to be a rectangular wth length l meters and dscretzed wth grd wth grd sze h = l/(-1). By use of the fnte dfference method, the equaton (4) becomes: that the state estmaton approaches the true value of measurement data. The MSE comparson between the archtecture s summarzed n Table 1. The smulaton result shows that the centralzeddecentralzed yelds the best MSE value than the other archtectures. u( 1,, u(,, / h [ u(, 1, u(,, j 1) u(, 1, u(,, j 1) 4u(,, ] (6) where u(,, denotes the temperature at tme at locaton (h,. If all temperatures at are combned nto a vector U [ u(,0,0),... u(, 1, 1)], we wll obtan the dscrete equaton where the matrx A can be calculated from Equaton (6). By ntroducng process nose, w, the equaton wll be same as Equaton (1). For the smulaton, the followng parameters are selected: α = 10 m /s. l = 40 m and = 4 therefore h = 140 m. We use 1 sensors: 9 sensors for measurement and 3 sensors for fuson center. The placement of sensors s shown n Fg. 5. The consdered archtectures are smulated by usng the defned system and the energy-aware routng algorthm. In the smulaton, only the centralzed-decentralzed archtecture wll use two nds of routng algorthm (.e. energy-aware and regular ). Results from the smulaton are then compared between the routng algorthm and the energy-aware routng algorthm. Fg. 5 Topology of centralzed-decentralzed The smulaton result for the centralzed-decentralzed archtecture n sensor 4 s shown n Fg. 6. Ths fgure shows Fg. 6 Centralzed-Decentralzed 4 MSE Before flterng After Flterng TABLE 1. MSE Comparson between Archtecture Centralze d Decentralz ed Centralzed- Decentralzed Energy Aware Regula r The centralzed archtecture has better performance than the decentralzed archtecture. Ths s caused by the measurement data from all the sensors can be combned nto one n order to provde better accuracy. In the decentralzed archtecture, the data are only sent to neghborng sensors. Ths causes the MSE value s hgher than the centralzed. Comparson on the energy consumpton of the archtectures s gven n Table. Ths table shows that the largest energy consumpton s taen by the decentralzed archtecture. Ths archtecture dd not use routng algorthm to determne the networ topology. Topology n ths archtecture s determned by neghborng sensors. The decentralzed archtecture wll broadcast ts data to all sensors wthn ts range. If there are a lot of sensors wthn ts range, the energy consumpton wll be large. The bggest energy consumpton n WS s due to data transmsson. Table also shows the dfference between the energyaware routng algorthm and the regular routng algorthm. The energy consumpton n the energy-aware routng algorthm s smaller than the regular. Ths s because of n the regular, sensor 3 wll not be ncluded n the route dscovery f there s another route. ISS: / 016 IIJ

6 8 ITERETWORKIG IDOESIA JOURAL ABADY & JOELIATO However, f the sensor 3 s the only route then t wll be ncluded nto the route dscovery. S e n s o r Centralzed TABLE. Comparson of Energy Consumpton Decentralzed Centralzed- Decentralzed Energy Aware Centralzed- Decentralzed Regular VII. COCLUSIO The paper consdered a centralzed-decentralzed archtecture n a wreless sensor networ. The smulaton results showed that the combned archtecture yelded the best performance. The worst performance was produced by the decentralzed. The centralzed archtecture had smlar performance wth the centralzed-decentralzed but t had the pont of falure n the fuson center. A Wreless etwor s a networ that promotes energy effcency. The smulaton results showed that the smallest energy consumpton was obtaned by usng the centralzeddecentralzed archtecture wth the energy-aware routng algorthm. The energy-aware routng algorthm can be used as a way to reduce energy consumpton n the wreless sensor networs. REFERECES [1] L. Sh, K.H. Johansson, and R.M. Murray, "Estmaton Over Wreless etwors: Tradeoff Between Communcaton, Computaton and Estmaton Qualtes," roc. of The 17th World Congress The Internatonal Federaton of Automatc Control, Seoul, Korea, 008. [] F. Ayldz and I.H. Kasmoglu, "Wreless and Actor etwors: Research Challenges," Ad Hoc etwors, vol., no. 4, 004, pp [3] W. Darge and C. oellabauer, Fundamentals of wreless sensor networs: theory and practce, John Wley and Sons, 010, pp , [4] K. Sohraby, D. Mnol, and T. Znat, Wreless sensor networs: technology, protocols, and applcatons, John Wley and Sons, 007, pp [5] Y.L. Mo, R. Ambrosno, and B. Snopol, "etwor Energy Mnmzaton va Selecton and Topology Control," roc. of The Frst IFAC Worshop on Estmaton and Control of etwored Systems, Vence, Italy, 009, pp [6] L. Sh, K.H. Johansson, and R.M. Murray, "Change Topology When eeded : How to Effcently Use System Resources n Control and Estmaton Over Wreless etwors," roc. of The 46th IEEE Conference on Decson and Control, ew Orleans, LA, USA, 007. [7] C. erns, E. Beldng-Royer, and S. Das, Ad hoc On-Demand Dstance Vector () Routng, IETF. RFC Retreved , July 003. [8] L. Baramov, D. achner, and V. Havlena, "Kalman Flter for Systems Wth Communcaton Delay," roc. of The Frst IFAC Worshop on Estmaton and Control of etwored Systems, Vence, Italy, 009, pp [9] M.S.M. azar, "A Comparatve Study of Dfferent Kalman Flterng Methods n Mult Data Fuson," roc. of The Internatonal MultConference of Engneers and Computers Scentst (IMECS), Hong Kong, 009. [10] R. Olfat-Saber, "Dstrbuted Kalman Flterng for etwors," roc. of IEEE Conference Decson and Control, pp , 007. [11] S. Oruc, J. Sjs, and..j. Van den Bosch, "Optmal Decentralzed Kalman Flter," roc. 17th Medterranean Conference on Control and Automaton, Thessalon, Yunan, 009, pp [1]. Ren, J. Feng,. Hu, and J. Ca, "Energy Savng Ad-hoc On-Demand Dstance Vector Routng for Moble Ad-hoc etwors," IEEE Internatonal Conference on Communcaton, 009, pp [13] R. Wjaya, J.M. Kerta, and R. Hedwg, "Influences of Wreless Traffc toward WF erformance at BIUS Square-Jaarta," Internetworng Indonesa Journal, vol. 5, no. 1B, 013, pp [14] E. Joelanto, and Hosana, Loop-bac Acton Latency erformance of an Industral Data Communcaton rotocol on a LC Ethernet etwor, Internetworng Indonesa Journal, vol. I, o.1, 009, pp [15] E. Joelanto, etwored Control Systems: Tme Delays and Robust Control Desgn Issues, 011 nd Inter. Confer. on Instrumentaton, Control and Automaton, Bandung, Indonesa, 011, pp [16] F.. Jusuf and E.Joelanto, Stablzaton of etwored Control System wth Tme Delay Induced by etwor Imperfectons, 01 IEEE Conference on Control, System and Industral Informatcs (ICCSII), Bandung-Indonesa, 3-6 September 01, pp [17] E. Joelanto and I. Wranto, An Applcaton of Ant Colony Optmzaton, Kalman Flter and Artfcal eural etwor for Multple Target Tracng roblems, Int. J. of Artfcal Intellgence (IJAI), vol. 7, no. A11, , 011. Husnul Abady receved hs M.Eng degree n Engneerng hyscs from Bandung Insttute of Technology, Bandung, Indonesa n 013. Endra Joelanto (M 01) receved hs B.Eng. degree n Engneerng hyscs from Bandung Insttute of Technology, Indonesa n 1990, and hs h.d. degree n Engneerng from The Australan atonal Unversty (AU), Australa n 00. He was a Research Assstant n the Instrumentaton and Control Laboratory, Department of Engneerng hyscs, Bandung Insttute Technology, Indonesa from Snce 1999, he has been wth the Department of Engneerng hyscs, Bandung Insttute of Technology, Bandung, Indonesa, where he s currently a Senor Lecturer. He has been a Senor Research Fellow n the Centre for Unmanned System Studes (CETRUMS), Bandung Insttute of Technology, Bandung, Indonesa snce 007. He was a Vstng Scholar n the Telecommuncaton and Informaton Technology Insttute (TITR), Unversty of Wollongong, SW, Australa n February 00. He was a Vstng Scholar n the Smart Robot Center, Department of Aerospace and Informaton Engneerng, Konu Unversty, Seoul, Korea n October 010. Hs research nterest ncludes hybrd control systems, dscrete event systems, computatonal ntellgence, robust control, unmanned systems and ntellgent automaton. He has edted one boo on ntellgent unmanned systems publshed by Sprnger, 009 and publshed more than 100 research papers. Dr. Joelanto s an IEEE Member and IEEE Indonesa Control Systems/Robotcs Automaton Jont Chapter Char. He s the Charman of Socety of Automaton, Control & Instrumentaton, Indonesa. He s an Edtor of the Internatonal Journal of Artfcal Intellgence (IJAI) and the Internatonal Journal of Intellgent Unmanned Systems (IJIUS). He was the Guest Edtor at the Internatonal Journal of Artfcal Intellgence (IJAI), the Internatonal Journal of Imagng and Robotcs (IJIR) and the Internatonal Journal of Appled Mathematcs and Statstcs (IJAMAS). He was the General Char of the 1st Internatonal Conference on Instrumentaton, Control and Automaton (ICA), Bandung, Indonesa n 009, the 01 IEEE Conference on Control, Systems and Industral Informatcs (ICCSII), Bandung, Indonesa, the 014 Internatonal Conference on Intellgent Autonomous Agents, etwors and Systems (IAGETSYS), and the 3rd 015 Internatonal Conference on Technology, Informatcs, Management, Engneerng and Envronment (TIME-E).