Experimental Assessment of the Battery Lifetime in WSN Based on the Duty-Cycle Current Average Method

Transcription

1 Wirele Senor Network, 2014, 6, Publihed Online Otober 2014 in SiRe. Experimental Aement of the Battery Lifetime in WSN Baed on the Duty-Cyle Current Average Method Leonardo Barboni 1, Maurizio Valle 2 1 Faultad de Ingenieria, Univeridad de la Republia, Montevideo, Uruguay 2 Department of Naval, Eletrial, Eletroni and Teleommuniation Engineering, Univerity of Genova, Genova, Italy lbarboni@fing.edu.uy, maurizio.valle@unige.it Reeived 25 Augut 2014; revied 25 September 2014; aepted 24 Otober 2014 Copyright 2014 by author and Sientifi Reearh Publihing In. Thi work i liened under the Creative Common Attribution International Liene (CC BY). Abtrat A great amount of work addreed method for prediting the battery lifetime in wirele enor ytem. In pite of thee effort, the reported experimental reult demontrate that the duty-yle urrent average method, whih i widely ued to thi aim, fail in aurately etimating the battery life time of mot of the preented wirele enor ytem appliation. The aim of thi paper i to experimentally ae the duty-yle urrent average method in order to give more effetive inight on the effetivene of the method. An eletroni metering ytem, baed on a dediated PCB, ha been deigned and developed to experimentally meaure node urrent onumption profile and harge extrated from the battery in two eleted ae tudie. A battery lifetime meaurement (during 30 day) ha been arried out. Experimental reult have been aeed and ompared with etimation given by uing the duty-yle urrent average method. Baed on the meaurement reult, we how that the aumption on whih the method i baed do not hold in real operating ae. The rationality of the duty-yle urrent average method need reonidering. Keyword Battery Lifetime, Current Conumption, TinyOS, Meaurement, Duty-Cyle Current Average Method 1. Introdution Nowaday, WSN are ued for many different appliation, but ytem are deigned in a rather ad-ho manner without unified and ytemati battery lifetime analyi. Sine the quantity of battery harge i limited, it i deirably the longet poible lifetime. How to ite thi paper: Barboni, L. and Valle, M. (2014) Experimental Aement of the Battery Lifetime in WSN Baed on the Duty-Cyle Current Average Method. Wirele Senor Network, 6,

2 For intane, in [1] author preent an ad-ho WSN for motor energy monitoring but the node lifetime etimation method i not diued. In [2] author diu WSN deign for indutrial automation appliation; tehnial apet and deign approah are remarked, but iue related to node lifetime etimation are not addreed. In [3] author introdue a olar energy elf-powered wirele enor network. Iue regarding the devie for energy toring are diued but the node urrent onumption i not modeled. In [4], author highlight the need of ahieving redued urrent onumption in order to maximize battery lifetime but a ytemati methodology for ahieving i not propoed. In [5] a WSN appliation for pedetrian traking and loalization i preented, but the node lifetime etimation i not addreed. On the other hand, nowaday WSN are alo ued a part of preiion agriulture tool [6] [7], but method related to node lifetime etimation till remain omitted. To the extent of our knowledge, after reviewing the tate of the art (by mean of well-known bibliographi and itation databae of peer-reviewed ientifi literature like SCOPUS, EBSCO, JSTOR, IOP iene, and IEEE Xplore), we argue that WSN deigner avoided arrying out long-term meaurement in order to validate battery depletion etimate model for WSN appliation. Deigner ue the duty-yle urrent average method, whih i nowaday the mot aepted method. However, deigner often report that batterie are depleted prematurely with repet to the etimation given by uh method (e.g. [8]-[12]). The aim of thi paper i to experimentally ae the duty-yle urrent average method in order to give more effetive inight on the effetivene of the method. We preent a ounterexample baed on meaurement in order to how how the duty-yle urrent average method an fail the battery lifetime etimation. A we need a olid undertanding of the battery urrent onumption profile, we have deigned an eletroni metering ytem baed on a dediated PCB, whih viualize the node urrent onumption profile and harge extrated from the battery during the wirele enor node operation (e.g. tranmiion, ignal ampling). The node ued in the experiment i the MICAz [13] running appliation developed by uing nec language [14]. The TinyOS [15] verion i provided by the oftware platform MOTEWORK [16] and the radio traneiver i the Chipon CC2420 [17]. The node i powered by two non-rehargeable alkaline IEC-LR6 ell, ize AA. By uing the propoed eletroni metering ytem, meaurement of two ae tudie have been performed; the experimental reult provide evidene to explain why the duty-yle urrent average method i not reliable for the etimation of the battery lifetime. An experimental battery full-depletion time meaurement ha been onduted to ompare the reult againt the etimation given by the duty-yle urrent average method. 2. Wirele Senor Tak and Duty-Cyle Definition The TinyO operating ytem manage the node hardware reoure; it i baed on the hardware abtration layer oftware model (TEP 2 [15]), where eah level give ervie to upper layer through interfae. The bottom layer i named Hardware Preentation Layer (HPL) and it i reponible for aeing hardware reoure. Thi i explained in the doumented TinyOS Enhanement Propoal alo named TEP (in thi ae ee TEP 2 from [15]). By uing thi oftware model, deigner an develop program by uing utomized interfae through whih the lower hardware abtration layer are inviible. The WSN appliation i implemented at the top level (Hardware Interfae Layer or HIL). In thi framework, the tak repreent appliation performed by the wirele enor node: e.g. enor ignal ampling, radio tranmiion and ignal proeing. The node wake up (from power down tate) in order to exeute the tak and afterward goe into leep mode (power down). In other word, a tak i the ativity that the node perform between two power down tate. Eah tak i exeuted at the rate R (number of tak per eond) and the tak time duration i defined a the ative period T A (eah tak ha it own T A ). The node duty-yle time interval D i defined a D = RT A. During the exeution of a T A Q = i t dt whih i named tak harge ot hene- tak, the node iruit onume the battery harge ( ) forth. Figure 1 how the urrent onumption profile for a tranmiion tak, where the marked time interval are: interval A ( T 1 ) node wake up; interval B ( T 2 ), radio traneiver tart-up; interval C ( T 3 ), MAC layer funtion uh a hannel aement and bakoff-time alulation and reeiver ytem alibration; interval D ( T 5 ), tranmiion; interval E ( T 7 ), reeption mode aording to the MAC protool [17], waiting for ACK paket. At the end, the node get power down tate (P.D.), waiting for the next timer interruption for next tak exeution. 3. Battery Lifetime Etimation Approah A Figure 2 how, we need to etimate the time 0 L S (node lifetime or ervie hour) at whih the battery vol- 213

3 Figure 1. Battery urrent profile at 25 dbm output power. Time value are: T 1 = 0.84 m, T 2 = 0.58 m, T 3 = 2.00 m, T 4 = 0.49 m, T 5 = 1.1 m, T 6 = 0.36 m, T 7 = 1.07 m. The um i the ativity time (or tak time) T = T1+ T2 + T3 + T4 + T5 + T6 + T7 = 6.44 m. A Figure 2. Battery voltage profile; B ( ( )) battery voltage, L i the node lifetime (ervie hour) and ( ) S V it i the battery voltage at time t ; V Bo i the initial it i the battery urrent. tage V B drop below the ut-off voltage V utt-off. To etimate the L S value, the node urrent onumption profile i( t ) and the equation that relate i( t ) to the battery voltage, i.e. the equation VB ( i( t )) are needed. The urrent onumption i( t ) i determined by the hardware platform and the operating ytem feature. There ha been a lot of effort for building model apable of prediting the urrent onumption profile i( t ), ee for intane [18]-[21], but the problem i diffiult and not eaily tratable. On the other hand, ee for intane [22]-[25], there i no onenu about the deription of the battery voltage VB ( t ) a funtion of i( t ). Approahe have onverged toward the duty-yle urrent average method, whih i the mot widely ued to etimate the L S value. The method approximate the node urrent onumption with tep-haped waveform (named i ( t ) from herein) that bet approximate i( t ). The node duty-yle i aumed to be determiniti and known beforehand. The i ( t ) integration over the time T A etimate the node harge onumption value. Thi harge value i then ued to etimate the average ontant urrent onumption i d for uh duty-yle; it i ued to etimate the lifetime T value by mean of the graph ontant urrent diharge v. hour ervie, a Figure 3 how, whih i provided by the battery dataheet. From the battery ervie hour graph it i poible to derive the log-linear relation between the ontant urrent i d and the ervie hour T for a given utoff voltage (e.g. the marked one with 0.8 V, 1 V and 1.2 V in Figure 3). For example, for the utoff voltage value 1.2 V, the traight line linking the point (20 ma, 100 h) and (80 ma, 20 h) i a follow: The oeffiient m i given by: ( T) ( ) m( ( id ) ( )) log log 20 = log log 80 (1) m ( ) ( ) ( ) ( ) log 100 log = = log10 20 log (2) 214

4 Figure 3. Battery ervie hour for different ontant urrent onumption value (figure from [29]). Then: ( T ) m ( i ) m ( ) ( ) log = log log 80 + log 20 (3) d ( T ) ( i ) log = 1.2log (4) d Some node platform vendor provide eletroni heet baed on thi method; ee for example the Mote Battery Life Calulator from Crobow Tehnology [26] and other example given by [27] [28]. For example, in Equation (4) if: 1) i = 2.04 ma and 2) i = 3.02 ma then repetively: a) T = h (i.e day) d d and b) T = h (i.e day). To ummarize, the hypothee for uing the duty-yle urrent average method are the following: H 1 ) the node urrent onumption profile a funtion of the given appliation i known. It i a determiniti waveform that an be alway approximated by a tep-haped waveform. The node duty-yle will be ontant during the node lifetime. H 2 ) uperpoition priniple: let aume that a node perform tak 1 with harge ot Q 1 during lifetime L 1. On another hand, let aume that the ame node perform tak 2 with harge ot Q 2 during lifetime L 2. If tak 3 i defined a the tak 1 plu tak 2 (e.g. tak 1 perform tranmiion and tak 2 perform ignal ampling and data torage, then, tak 3 implement both ativitie, ignal ampling, data torage and tranmiion), the uperpoition priniple tate that Q3 = Q 1+ Q2. We aim to demontrate that H 1 ) and H 2 ) hypothee do not alway hold. In the next etion, we preent the metering ytem that ha been ued to experimentally ae the two hypothee. 4. Eletroni Metering Sytem The metering ytem i baed on a dediated PCB: it viualize the node urrent onumption and the tak harge ot Q. The prototype PCB i baed on the following devie: the High Side Current Shunt Monitor INA139 [30] and the Coulomb Counter LTC4150 [31]. Figure 4 how the metering ytem blok diagram. The INA139 amplifie the voltage aro a hunt reitane between battery and load. The meaured voltage i proportional to the battery urrent onumption. The output voltage ignal i in input to an amplifiation tage and ubequently to an oilloope where the urrent waveform aoiated to the node tak an be analyzed. The hunt reitane value ha been eleted o that the INA139 gain provide the maximum output voltage range at the maximum node urrent onumption. The board i powered by a 3 V voltage. The harge ounter haraterize the node in term of onumed harge; value are expreed in harge unit (mah or μah). The LTC4150 devie i a voltage integrator with a voltage-to-frequeny onverter. The oulomb ounter devie fore to zero the output pin voltage when a fixed quantity of harge Q M ha been meaured. The value of Q M = 5.2 μah ha been obtained by alibration a it i uggeted in the devie dataheet. The time period between oneutive output pule i proportional to the value of Q M. Figure 4 how the blok diagram of the experimental etup. If the number of ounted pule within a given time obervation window T i equal to K, then the total harge onumed by the node i QT = K QM. Under the aumption that the node 215

5 Figure 4. Blok diagram of the experimental etup. exeute N time the ame tak during the time T, we etimate the average tak harge ot value Q a Equation (5), where N i uh that N= RT (the meaurement end when time T i reahed and the tak ha been exeuted N time). The error i given by Equation (6) where K = 1 Q Q N KQ N = T = M (5) K Q N M Q = (6) We ue the following example to illutrate the methodology. If the node perform paket tranmiion at the rate R = 10 paket/, for a given time, e.g. T = 3600, the number of performed tak (equal to the number of paket tranmiion) i N = If we ount, for intane, K = 1000 pule, from Equation (5) the harge ot aoiated to the tranmiion tak and the related error reult a follow: A a final remark, we oberve that the bet et of value {, } 5.2 μah Q = 1000 = uah (7) μah Q = = pah 0.15 nah (8) T R to perform the experiment i uh that N 1 value (e.g or more), in order to meet the ondition Q Q. The propoed method ha the diadvantage that it require long time meaurement. A uitable value of T (or N) i not known a priori, beaue the value Q i obviouly not known, a well a the K value. Thi problem an be overome with an iterative proedure, by performing everal meaurement. In general from our experiene, we have not onduted meaurement of more than 4 hour. In thi artile, the node tak under evaluation are tranmiion (tak 1) and temperature enor ampling (tak 2). The meaured amount of harge Q for eah tak i hown in Table Battery Lifetime Time Meaurement The MICAz node ha been programmed to tranmit one paket with a payload of 25 byte, at the output power 1 level of 25 db m. The tranmiion rate i R = 20 (i.e. 50 m time interval between paket tranmiion). Pleae note that the given tranmiion rate value i high for uual real wirele enor network appliation. 216

6 Table 1. Meaured harge Q. Tak Charge ot Q [μah] Tak Tak Power voltage upply: 3 V; Output power: 25 db m $, payload ize 25 byte. The tranmiion rate value ha been eleted to perform the experiment in a reaonable time (i.e. ome day, no more than one month). The node ould be onidered not working when the voltage power upply drop value below 2.4 V (i.e. individual ell voltage value of 1.2 V). We daily meaured the battery voltage and the reult i hown in Figure 5; the voltage value 2.4 V ha been ahieved in 30 day. The urrent onumption profile ha been meaured during tranmiion; it i hown in Figure 1. The reulting duty-yle i D = 6.44 m 50 m = In aordane with the duty-yle urrent average method, it i aumed that thi urrent profile i repeated eah time that the radio perform a tranmiion. By numerial integration, the average diharge urrent inide the duty-yle time interval i given by i d = 2.04 ma ; the etimated battery lifetime (Equation (4)) i T = h, i.e day, with an error of 27 day with repet to the meaured one. At thi point, it i neeary to undertand why there i uh big differene between etimated and meaured reult. To thi aim, we ued the metering ytem for viualizing the node urrent onumption profile. 6. Aement of H 1 Hypothei Differently to the previouly preented lifetime etimate, we ue Q = μah (ee Table 1). The average urrent onumption i id = Q 50 m = 3.02 ma and by mean of Equation (4), the etimated lifetime i T = h (i.e day), whih i a better approximation with repet to the previou one. We onlude that the H 1 hypothei doe not hold. The metering ytem provide exellent auray to meaure Q for large N value. It i not neeary to aume any node urrent onumption profile beforehand and the overall urrent onumption i taken into aount to meaure Q (inluding node tart-up and power down tate). The redued urrent onumption level are frequently negleted in the duty-yle urrent average method, neverthele, they ould have umulative effet during the node lifetime. It i not enough to oberve the urrent onumption profile for a hort time to etimate the real required harge Q beaue our meaurement how non periodi urrent onumption profile. In fat, we oberved a time-varying duty yle (time-varying urrent onumption). Figure 6 how three different oberved ae, among other. The differene between duty-yle value of ae A and C i 48.9%. Then, it i not poible to have a determiniti known urrent profile during tranmiion and the H 1 hypothei whih tate that the duty-yle i ontant, doe not hold. Therefore, without a reliable value of the duty yle value, the duty-yle urrent average method an etimate the lifetime value very roughly. In fat, it i not poible to aume a priori node urrent onumption profile and the experimental haraterization hould be done for every given appliation. The node tak are ontrolled by the program that run inide the miroontroller under the uperviion of an operating ytem like TinyOS [14]. Mot of the time, the operating ytem aynhronouly manage hardware reoure in order to provide ervie to aynhronou event related to the wirele ommuniation. More peifially, the origin of thi non determiniti behavior i given by the method ued for the lear hannel aement in mot of the MAC layer protool (ee Chapter 5 MAC Protool in [20]). Therefore, generally peaking, the node awake and leep in a non-determiniti way and the battery i fored to deliver urrent pattern whih are not eaily preditable, neither the duty-yle value. The duty-yle urrent average method require the knowledge of the determiniti node urrent onumption, but we aeed that it i not poible. 7. Aement of H 2 Hypothei Table 1 how harge ot for the tak 1 and 2. Tak 3 i defined a the ampling plu the tranmiion tak, i.e. the node perform ignal ampling and then paket tranmiion. The H 2 hypothei tate that the expeted battery harge ot Q 3 for tak 3 hould be equal to the um of Q 1 (tak 1) and Q 2 (tak 2). To verify the H 2 hypothei, we implemented the tak 3 and experimental meaurement have been performed. The meaurement how that H 2 hypothei doe not hold. The reult are preented in Table 2; they are ig- 217

7 Figure 5. Meaured battery voltage during node tranmiion. Figure 6. Three different meaurement that how the time-varying node urrent profile behavior beaue of the random nature of bakoff-time alulation for the MAC layer. Pleae, note how the T A value varie. Cae A: D 9.2 m 50 m or 18.4%; Cae B: D 7.5 m 50 m or 15.0%; Cae C: D 4.5 m 50 m or 9.0%. Table 2. Meaured harge Q. Tak Charge ot Q [μah] Tak 1 + Tak = Tak nifiantly different with repet to the expeted one. The reaon behind thi reult i due to the effet of node programming method and ompiler feature whih have a trong influene on the node urrent onumption profile. In fat, the operating ytem behavior and how hardware reoure are managed, make unreliable the appliation of the uperpoition priniple. 218

8 No withtanding the fat that program are written in nec, i.e. they are baed on utomized interfae and module whih provide modularity, enapulation and re-uability feature; uh benefit are ounterbalaned by the impoibility of uing the duty-yle urrent average method for battery lifetime etimation. 8. Conluion The WSN are ompried of battery-powered embedded eletroni devie. A great amount of reearh ha been devoted to the development of method apable of prediting battery lifetime. Regardle of uh effort, mot of the wirele enor appliation have prematurely depleted batterie, ontrary to the etimation baed on the widely adopted duty-yle urrent average method. Our work experimentally aee uh direpany. An eletroni metering ytem, baed on a dediated PCB to experimentally meaure node urrent onumption profile and harge extrated from the battery for two eleted ae tudie, ha been deigned and implemented. Baed on the meaurement reult, we demontrate that the hypothee on whih the method i baed are not reliable. The onluding remark to be taken into aount for the development of WSN appliation are: 1) the oftware arhiteture and how to develop the appliation program have trong influene on the node urrent onumption profile and hene, on the battery lifetime; 2) the node urrent onumption profile a funtion of the given appliation a well a the duty-yle i not determiniti. The duty-yle urrent average method ha another hypothei whih ha not been onidered: the battery operating temperature, whih would have ignifiant effet on the graph ontant urrent diharge v. hour ervie. It hould be addreed a future reearh trend. Referene [1] Lu, B. and Gungor, V.C. (2009) Online and Remote Motor Energy Monitoring and Fault Diagnoti Uing Wirele Senor Network. IEEE Tranation on Indutrial Eletroni, 56, [2] Gungor, V.C. and Gerhard, P.H. (2009) Indutrial Wirele Senor Network: Challenge, Deign Priniple, and Tehnial Approahe. IEEE Tranation on Indutrial Eletroni, 56, [3] Dondi, D., et al. (2008) Modeling and Optimization of a Solar Energy Harveter Sytem for Self-Powered Wirele Senor Network. IEEE Tranation on Indutrial Eletroni, 55, [4] Agha, K.A., et al. (2009) Whih Wirele Tehnology for Indutrial Wirele Senor Network? The Development of OCARI Tehnology. IEEE Tranation on Indutrial Eletroni, 56, [5] Ahn, H.-S. and Ko, K.H. (2009) Simple Pedetrian Loalization Algorithm Baed on Ditributed Wirele Senor Network. IEEE Tranation on Indutrial Eletroni, 56, [6] Piere, F.J. and Elliott, T.V. (2008) Regional and On-Farm Wirele Senor Network for Agriultural Sytem in Eatern Wahington. Computer and Eletroni in Agriulture, 61, [7] Lopez, J.A., et al. (2009) Development of a Senor Node for Preiion Hortiulture. Senor, 9, [8] Torvmark, K.H. Low Power Sytem Uing the CC1010. Chipon Appliation Note AN017. [9] Baleri, G. Guideline for WSN Deign and Deployment. obow+tehnology%2c+in.+guideline+for+wsn+deign+and+deployment [10] The Battery Life Etimator. A Web-Baed Calulator to Help Optimize Low-Power Appliation. [11] Yang, O. and Heinzelman, W.B. (2012) Modeling and Performane Analyi for Duty-Cyled MAC Protool with Appliation to S-MAC and X-MAC. IEEE Tranation on Mobile Computing, 11, [12] Jiang, X., Taneja, J., Ortiz, J., Tavakoli, A., Dutta, P., Jeong, J., et al. (2007) An Arhiteture for Energy Management in Wirele Senor Network. Speial Iue on the Workhop on Wirele Senor Network Arhiteture, 4, [13] Tehnial Data MPR-MIB Uer Manual Reviion B, June 2006 PN: Tehnial Data MTS/MDA Senor Board Uer Manual Reviion A, June 2007 PN:

9 [14] Gay, P., Welh, M., Levi, P., Brewer, E., Von Behren, R. and Culler, D. (2003) The nec Language: A Holiti Approah to Networked Embedded Sytem. Proeeding of the ACM SIGPLAN 2003 Conferene on Programming Language Deign and Implementation, San Diego, 9-11 June [15] [16] MoteWork Crobow Platform for the Development of Wirele Senor Network from Crobow. [17] CC GHz IEEE ZigBee-Ready RF Traneiver Chipon, Produt from Texa Intrument. [18] Kan, B., Cai, L., Zhao, L. and Xu, Y. (2007) Energy Effiient Deign of WSN Baed on an Aurate Power Conumption Model. Proeeding of the Wirele Communiation, Networking and Mobile Computing (WiCom 2007), Shanghai, September 2007, [19] Barberi, A., Barboni, L. and Valle, M. (2007) Evaluating Energy Conumption in Wirele Senor Network Appliation. Proeeding of the 10th Euromiro Conferene on Digital Sytem Deign Arhiteture, Method and Tool, Lubek, Augut 2007, [20] Holger, K. and Willig, A. (2005) Protool and Arhiteture for Wirele Senor Network. John Wiley & Son, Hoboken. [21] Bougard, B., Daly, D.C., Chandrakaan, A. and Dehaene, W. (2005) Energy Effiieny of the IEEE Standard in Dene Wirele Miroenor Network: Modeling and Improvement Perpetive. Proeeding of the Conferene on Deign, Automation and Tet, 1, [22] Rao, R., Vrudhula, S. and Rakhmatov, D. (2003) Battery Modeling for Energy-Aware Sytem Deign. IEEE Computer Soiety, 36, [23] Rakhmatov, D. and Vrudhula, S. (2003) Energy Management for Battery-Powered Embedded Sytem. ACM Tranation on Embedded Computing Sytem, 2, [24] Rao, R., Vrudhula, S. and Chang, N. (2005) Battery Optimization v Energy Optimization: Whih to Chooe and When? Proeeding of Conferene on Computer-Aided Deign, ICCAD IEEE/ACM, San Joe, 6-10 November 2005, [25] Panigrahi, D., Chiaerini, C., Dey, S., Rao, R., Raghunathan, A. and Lahiri, K. (2001) Battery Life Etimation of Mobile Embedded Sytem. Proeeding of 14th International Conferene on VLSI Deign (VLSID 01), Bangalore, 3-7 January 2001, [26] Expeted Battery Life v. Sytem Current Uage and Duty Cyle, PowerManagement.xl. Exel Workheet from Crobow Tehnology. [27] Kyaw, Z.T. and Sen, C. (2008) Uing the CC2430 and TIMAC for Low-Power Wirele Senor Appliation: A Power Conumption Study. [28] Appliation Note AN053 Meauring Power Conumption with CC2430 and Z-Stak. [29] Produt Dataheet Energizer E91, Alkaline Battery AA. Energizer Holding, In. Form No. EBC-1202M. [30] High-Side Meaurement Current Shut Monitor INA 139, Dataheet Burr-Brown Produt from Texa Intrument. [31] Coulomb Counter Battery Ga Gauge LTC4150. Linear Tehnologie IC Produt

10