Research Article elighthouse: Enhance Solar Power Coverage in Renewable Sensor Networks

Transcription

1 Hndaw Publshng Corporaton Internatonal Journal of Dstrbuted Sensor Networks Volume 213, Artcle ID , 16 pages Research Artcle elghthouse: Enhance Solar Power Coverage n Renewable Sensor Networks Peng Lu, 1 Yfan Wu, 1 Jan Qu, 1 Guojun Da, 1 and Tngtng Fu 2 1 Insttute of Computer Applcaton Technology, Hangzhou Danz Unversty, Hangzhou 3118, Chna 2 Insttute of Graphcs and Image, Hangzhou Danz Unversty, Hangzhou 3118, Chna Correspondence should be addressed to Tngtng Fu; ftt@hdu.edu.cn Receved 16 May 213; Revsed 15 August 213; Accepted 7 September 213 Academc Edtor: Aravnd Kalas Copyrght 213 Peng Lu et al. Ths s an open access artcle dstrbuted under the Creatve Commons Attrbuton Lcense, whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly cted. Energy harvestng from ambent resource such as solar power mproves the sustanablty and contnuous montorng capablty of sensor nodes. However, energy conservaton and harvestng can hardly provde a complete soluton due to two man reasons: the energy harvestng capabltes/opportuntes are serously affected by spatal and temporal facts regardng the locaton of sensor deployment, and dstrbuted processng of sensng and communcaton are also uneven throughout the network whch leads to unbalanced energy consumpton. In ths paper, we propose an energy transfer system called elghthouse to level the energy harvestng and consumpton of solar-powered WSN. An energy-hub equpment has been brought forward utlzng a controlled retroreflector to forward sunlght to the approprate nodes whch urgently requre rechargng. The schedulng algorthm among multple nodes s also ncluded n ths paper whch helps to mantan the maxmum number of nodes n actve state; the localzaton algorthm whch enables the controlled retroreflector for pontng correctly s also a key contrbuton of ths paper. The expermental and smulaton results both demonstrated that the proposed system can maxmze the utlzaton of the solar power chargng technque effectvely to prolong the network lfetme and buld up a sustanable WSN. 1. Introducton Lmted energy supply has been the man constrant of battery-powered system such as wreless sensor networks whch provde contnuous montorng n ambent envronment. It greatly affects the sustanablty of WSNs and has dsproportonate mpact on the fdelty. In ths area, the recent research topcs are manly focused on two categores: energy savng and energy harvestng. A natural mechansm of energy savng s enhancng network protocols, such as energy-effcent MAC protocols, routng protocols, and data aggregaton as shown n [1 3]. The goals of these researches are not only to reduce the energy consumpton rate but also to balance t. However, these schemes can only prolong the network lfetme to a lmted extend but fal to make t sustanable. A promsng soluton to ths problem s to explore energy harvestng and acqurng technques. Solar power, wnd power, and vbraton power have all been ntroduced as perspectve energy resource of tny sensor nodes. Correspondng hardware desgn and experments have also been ntroduced n many papers although t stll suffers some lmtatons such as costlness, nflexblty, and low effcency. Furthermore, wthout suffcent energy acqurng approaches, these methods cannot work well. On one hand, as a network performng dstrbuted montorng tasks, data communcaton and collecton process are serously dependent on task requrement and network topology. Snk nodes and other nodes near hot spots or n a busy route tend to consume ther energy more rapdly whch means energy consumpton among the network that s unevenly dstrbuted. On the other hand, the energy harvestng capablty at each node may vary sgnfcantly due to envronmental geometrc. For example, the amount of harvested solar power n a sensor node vares due to factors of weather, locaton, and tme, n spte of Maxmum Power Pont Trackng functon beng used. In one word, the energy harvestng s unbalanced as well as energy consumpton. Researches have been carred out to balance the energy dstrbuton, such as [4, 5]. However, these approaches only

2 2 Internatonal Journal of Dstrbuted Sensor Networks consder the energy usage pattern correspondng to the energy harvestng rate, n whch the faster recharged nodes carry more operaton responses. It s not always feasble to leveldatacommuncatonbytheenergyharvestngcapablty. A heavy loaded sensor node may be deployed at energy-poor locatons and become the bottleneck of the entre network. To balance energy between nodes, energy share system could be a soluton. Energy can be transferred to other nodes va wres, mcrowave, magnetc resonance, laser, and so on. Wred transfer does not ft for wreless sensor networks and Magnetc resonance has very low effcency. Mcrowave and Laser requre addtonal hgh-cost equpments whch have huge overhead under current technology. Despte the poor performance of current energy sharng system, the dea of energy transfer between sensor nodes stll has ts potental n energy balancng. The key desgn goal s to mprove the energy transfer effcency. In our opnon, transferrng the ntal state of the harvested energy may be a promsng way. Therefore, n ths paper, we propose a sunlght reflecton based resource energy transfer method for solar energy harvestng sensor networks. Durng daytme, t does not produce specal means of energy such as electrcal, laser, magnetc resonance, mcrowave, and so on but only relay the ambent energy n the ntal state. Sunlght s reflected by a retroreflector for rechargng capablty enhancement to specfed nodes n an napproprate place, for example, nodes n shaded area or wth heavy loads. To mprove effcency, spare energy accumulated durng daytme can also be transferred to nodes by wreless energy transfer means n thenghtordurngbadweather. The proposed system conssts of (1) asunlghtreflecton and solar power harvestng staton called energy hub (2) wreless sensor nodes equpped wth solar panel and lght sensor. The energy hub can adjust the onboard retroreflector andreflectsunlghttoenhancetherechargngcapabltyof specfed nodes and charge them lke a lghthouse. Fundamental experments have been carred out to prove the feasbltyoftheproposedmechansmaswellasevaluaton of the energy transfer effcency. A prototype system has been produced and the correspondng schedulng algorthm has also been ntroduced to ensure the network lfetme extenson, whch ams to buld up a sustanable sensor network system. However, we meet some desgn and mplementaton challenges, such as accurate localzaton of the solar board wth a maxmum tolerable error of centmeter level, precse sunlght reflecton postonng, and real-tme energy rechargng schedulng. Ths method fts well for nfrastructured envronment as t requres lne-of-sght (between the energy hub and the sensor node wth solar panel). As shown n Fgure 1, each node may have dfferent opportuntes to receve sunlght (denoted by percentage) due to the movement of the sun. By applyng the proposed method, t can effcently provde reflected sunlght or wreless energy aganst the shadow. For large networks n a more complex envronment, our method can also mprove energy effcency by deployng energy hubs locally and carefully. More specfcally, the man contrbutons of ths paper are concluded as follows. 1% 7% 8% 2% 3% 9% 8% 1% 1% Fgure 1: Energy harvestng ablty vares from node to node. () We desgned and mplemented a solar power transfer system for energy harvestng sensor networks, whch s a novel means of effcent ntal state energy transfer scheme. () We proposed a sunlght reflecton strategy whch can reflect sunlght to the solar panel of the nodes n need. The strategy ncludes solar panel coverng and retroreflector adjustng. () We have proposed an accurate node locatng algorthm for sensor node localzaton, whch combnes camera and RSSI locatng method. (v) We have developed a chargng schedulng algorthm whch guarantees the sustanablty of sensor nodes. (v) We conducted model smulatons and also bult a proof-of-concept prototype of the system and conducted extensve experments on the prototype to evaluate ts feasblty and performance. The results ndcate that our system can effcently enhance sunlght coverage and mprove network performance. The remander of the paper s organzed as follows. In Secton 1, the motvaton s ntroduced, whle, n Secton 2, related works are descrbed and evaluated. The system model and desgn are dscussed n Secton 3, and the node locatng mechansm s presented n Secton 4. Secton 5 dscusses the schedulng of the chargng sequence, followed by smulaton and expermental evaluatons n Secton 6. Fnally,Secton 7 concludes the paper. 2. Related Work To prolong lfetme of sensor networks, energy harvestng has been proved to be a promsng key. Solar power, wnd power, vbraton, and even human power [6] can all be the energy source. Desgn, modelng, and capacty plannng for energy harvestng sensor network are evaluated n [7]. However, the energy harvestng opportunty of each node s far from each other and vares wth respect to temporal and spatal facts. Energy share methods are ntroduced to solve ths bottleneck problem, whch mples energy transfer between nodes

3 Internatonal Journal of Dstrbuted Sensor Networks 3 S B A C D E F Energy hub G D movement and trajectory of moble energy staton (also be called ferry, chargng vehcle, moble charger, etc.). Watfa et al. [24] try to extend current one-hop wreless energy transfer to a multhop manner. They nvestgate both theoretcally and through smulatons whether the physcal phenomenon of long-lfetme resonant electromagnetc states wth localzed slowly evanescent feld patterns can be used to transfer energy effcently over multple hops. In [25], the placement of RF power transmtters s bounded wth mssons of sensor nodes. RF power transmtters wll be moved to Landmarks among sensor networks accordng to current mssons and maxmze theenergyusageandtheproft.thekeyproblemshowto choose Landmarks and the authors propose ILP to solve t. Fgure 2: Sensor network archtecture wth energy hubs. or from energy staton to nodes by means of wred or wreless electrcal power exchange technologes. Magnetc resonance, developed by Kurs et al. [8], has become a major technque for wreless energy transfer, based on whch some rechargng schedulng algorthms have been developed, [9, 1]. Other studes utlzng wred energy transfer are conducted as well. Zhu [11] developed an energy chargng and dschargng mechansm usng an array of ultracapactors as the man component of an energy router through wred connectons, whch s known as eshare. However, accordng to [1], wreless chargng effcency s serously affected by dstance. Chargng effcency s approxmately 1.5% when the energy recever s 1 cm away from the charger. The chargng effcency decreases to zero when the dstance approaches 1 cm. The controlled electrcal energy transfer between wred ultracapactors s also not as regular as mentoned n [11] due to physcal characterstcs of the component, whch has some nonlnear features. Another man dsadvantage s the low effcency caused by the wasted energy n transfer crcut. Above all, electrcal energy share schemes are stll n a desgn or laboratory experment level at the recent stage. Localzaton has been a hot topc for years. However, the accuracy depends on node densty and avalable facltes. Most of the range-based methods can be concluded nto four categores, whch are tme of arrval (TOA), tme dfference of arrval (TDOA) [12], angle of arrval (AOA) [13]andreceved sgnal strength ndcator (RSSI) [14]. Range-free methods are mentoned n [15 17]. References [18, 19], whch use Corner Cube Retroreflector (CCR) to transfer man-made laser and mrrors to transfer lamplght by optcal means, are the most smlar work compared wth our approach. In [2],theauthorsproposeusng external energy transmtters (ETs) to chargng sensor nodes by wreless means. ETs are assumed that they have endless power and can be deployed randomly or predetermned. The paper does not dscuss the placement of ETs but proves the feasblty of ths energy staton-based method and the routng and data transmsson ssues under the crcumstance. References [9, 21 23] both focus on moble energy staton whch can prolong network lfetme by travelng among nodes and chargng them wrelessly. The man concern s the 3. System Descrpton 3.1. System Overvew. As can be seen from Fgure 2, n elghthouse system, there are sensor nodes and energy hubs. Each energy hub can cover certan amount of sensor nodes. The coverage area can be ether dentcal or varant. The energy to power a sngle sensor node comes from energy harvestng from solar power. Sensor nodes mght run out ther batteres so that part of communcaton connectons between nodes becomes ntermttent shown as dashed lnes n the fgure. There are two major problems. Frst, because of the movement of the sun and shadow of the envronment, each node does not have an equal opportunty to receve sun radaton. The prncple of elghthouse system s based on sunlght reflecton by the retroreflector. By adjustng the angle of the retroreflector, the energy hub can redrect the sunlght to those nodes sheltered from trees or other obstacles. The energy hub tself s equpped wth a powerful solar panel so that ts energy supplement s not the key pont of ths paper. That s to say, even durng the nght, energy hubs can provde energy wth means of wreless energy transfer. Second, because the sunlght and wreless energy transfer can only generate feeble electrc current, we need to fnd ways to speedandeaslystoretheenergy.regardngthefrstproblem, we need to solve further two more problems. One s how we can fnd the node and compute the correspondng angle of retroreflector. The other s who and how long we should forward the sunlght to. elghthouse s sutable for plan area deployment where trees and bush are not too flourshng. The energy hub s often deployed n a relatvely open area or hgher locaton and wth large energy harvestng equpment so that ts own energy consumpton needs no consderatons. Regardng the second problem, we use super capactors as the energy storage Hardware Implementaton. The archtecture of elghthouse sensor network s shown as n Fgure 3, whchs composed of two parts, energy harvestng sensor nodes and the equpment called energy hub. The sensor nodes are composed of () the sensor board equpped wth multple sensors n whch lght sensor s a fundamental part whle other sensors are optonal, () a square 15 cm 15cm solar panel for power rechargng, () a 1 or 2 F super capactor for energy storage and suppler, (v) perpheral crcut for

4 4 Internatonal Journal of Dstrbuted Sensor Networks Solar panel Interface crcut Camera and mrror energy hubs and how to schedule chargng sequence wll be addressed n Secton5. Super capactor Sensor node and backup battery Rotary platform Fgure 3: Hardware components of elghthouse. energy harvestng and voltage convert; (v) rado transcever component and a set of backup batteres. L-on or other batteres normally cannot support fast charge and respond slowly to weak chargng current. Therefore, usng a sngle L-on battery s not enough to satsfy future development of wreless sensor networks. Also the resdual energy of L-on battery s hard to compute accurately and t does not have a lnear relaton between resdual energy and voltage. Furthermore, for a long-tme deployment, chargng cycle s essental to the sensor battery. To maxmze the usage of energy and schedule sensor nodes effcently, as well as routng protocol, the resdual and chargng energy of battery should be accurately calculated. Regardng super capactor, ts resdual energy s n accordance wth the formula E = (1/2)CV 2. Furthermore, super capactors almost have unlmted chargng cycles and can be charged very quckly. In ths paper, we desgn our system usng super capactors as man energy storage and a L-on battery as a backup battery. If there s redundant energy, t wll be stored n L-on batteres. However, energy leakage s a man concern of super capactors. It means the desgn of energy usng polcy, e-hub schedulng, and routng protocols should be carefully addressed aganst ths. The energy hub conssts of () a rotary platform that can pan and bend so as to adjust angle of the retroreflector, () a retroreflector whch s used to reflect sunlght, and () a camera for accurate localzaton of node as an assstant of RSSI method. The energy hub s connected to a TelsoB node whch wll receve resdual energy nformaton from other nodes and perform schedulng algorthm to control the rotary platform. In most occasons, the energy hub tself s equpped wth a set of powerful energy harvestng devces, for example, wth larger area of solar panel. The energy harvested durngdaylghtcanbeusedtochargenodesdurngthenght. However, energy has to be transmtted usng wreless means when there s no sunlght. Wthout losng generalty, f we regard energy hubs as all-weather energy supplers, problems that nclude how to cover sensor nodes wth mnmum 3.3. Software Implementaton. The software components are bultontnyos2.1andrunoneachsensornodenthe network. Besdes normal data collecton, the sensor node s runnng status report task perodcally. It also keeps measurng ts resdual energy status and wll requre the energy hub to transfer sunlght when the energy s below certan thresholds.thecontrolandschedulngsoftwarecomponent s runnng on the TelosB node attached wth the energy hub. It communcates wth other nodes through nterface. Normally, one energy hub provdes sunlght transfer to several nodes. We need to dentfy the followng thngs. Frst,whoshallberesponsbleforntalzngtheenergy transferrng process? Second, how to control sunlght reflecton between a par of sensor node and energy hub? Thrd, how to balance the collson of multple energy transferrng requrements? Tobegnanenergytransferrngprocess,therecouldbe two knds of methods. Namely, energy hub ntalzed and sensor node ntalzed schemes. In [11], the node whose energy s gong to be below crtcal level wll cast an energy sharng request to nodes lsted n descendng order of effcency. However, n a network scale vew, energy should be reserved as much as possble. In our desgn, energy s stored n the super capactor. Therefore, hgher energy reservaton wll lead to hgher energy leakage. Hence, energy parng s relatedtotwothngs,energylevelandusngpattern.when a node s battery s gong to be flat, t wll be an ntalzer as n [11].Also,whenndaytme,anenergyhubwllbecomean ntalzer. If an energy hub and node go nto sunlght reflecton procedure, the ntalzer sends out a REFLECTION REQUEST message. After recevng the message, the other one may answer wth message REFLECTION RESPOND. When a sunlght reflecton process starts, the energy hub wll generate a REFLECTION START packet and try to locate the node. The node wll drect the locatng process wth ts lght sensor. When the desred amount of sun radaton has been transferred, the energy hub sends out the REFLECTION STOP packet. However, the node can also send out REFLEC- TION STOP packet to termnate the sunlght reflecton process under certan crcumstances. The software flowchart of sensor nodes s shown n Fgure 4. After ntaton, sensor nodes always measure ther battery level and report to energy hubs. If energy level s below certan threshold, for example, E mn,thesensornodewll send request to ask for energy transferrng from the energy hub. Both energy hubs and sensor nodes can ntate energy transfer process so that an energy transfer parng procedure s brought forward to help balance energy schedulng and emergency battery chargng, and so forth. If energy parng s successful, then t wll start sunlght reflect process. The software flowchart of energy hubs s shown n Fgure 5. They montor battery status of appurtenant sensor nodes, analyze ther energy consumpton and harvestng model, and then perform chargng sequence schedulng to

5 Internatonal Journal of Dstrbuted Sensor Networks 5 Start Start System ntate (sensor test, communcaton setup and energy harvestng unt ntate) System ntate (rotary platform test, communcaton setup, energy harvestng unt ntate) Measure battery level Receve battery level reports No Resdual energy below threshold E mn? Yes No Nodes need mmedate energy transferrng? Yes Perform regular sensng tasks Perform low dutycycle sensng tasks Perform chargng sequence schedulng Perform energy transfer Report energy level to the energy hub Request energy transfer End No Energy transfer parng successful? Yes Fgure 5: Software flowchart of energy hub. Sunlght reflect (energy transferrng) process End Fgure 4: Software flowchart of sensor nodes. enable sensor nodes stay workng as long as possble. Besdes ntatng chargng n schedulng they need also to respond to mmedate energy transfer request. 4. Energy-Aware Node Locatng 4.1. Accurate Node Localzaton. In our proposed scheme, n order to reflect the sunlght accurately to the solar panel of each sensor node, an mportant step s the localzaton of all thesolarboards.therehavebeenmanymethodsntroduced for localzaton, whch are ether range-based or range-free [15].Sncerange-freeapproachesusehopcountandthe average dstance as the measurement standard, they are not able to be appled n our scenaro whch only consders onehop dstance stuaton. In our scenaro, as the localzaton precson s relatvely hgh whch requres the reflected lght drectly postonng the solar board, all the above schemes cannot satsfy the requrement. However, the localzaton n our scenaro s not a global network localzaton; t only covers approxmately one-hop dstance, or even shorter. A beneft rased from ths stuaton s that we can use vdeo camera on the platform to coordnate the accurate postonng. Lght sensor embedded nthesensornodealsocanhelptoncreasetheaccuracy by communcaton feedback. The detaled system procedure, the localzaton process, and the correspondng geometrcal model are shown as follows. We assume all the sensor nodes are deployed n a plane area, and the heght of the retroreflector n the energy hub s known as H. Thesnknodesconnectedtotheenergy hub and located exactly below the retroreflector whch s set to be the orgnal pont of the coordnate systems. In the sensor node deployment and network ntalzaton stage, the postons of the manually deployed sensor nodes are set to be known parameters. By convertng the rectangular coordnate system (x, y, z) to the polar coordnate system (γ, θ, φ), the horzontal angle θ and the vertcal angle φ of the poston from the orgnal pont to these nodes can be obtaned, whch wll be dscussed n the next subsecton. Durng the long-term montorng of the sensor network, newly added nodes may change the topology and the connectons of the network. Correspondngly, the localzaton nformaton requres updatng. In our method, fundamental RSSI s used as a frst step for rough localzaton. The newly added nodes that wll broadcast a short beacon message n a partcular transmt power strength, whch s power level 1 n CC242 for our experment. Those neghborhood nodes that receve ths nformaton wll use the receved sgnal strength tocalculatethedstancefromtheradosourceandreport t to the snk. The correspondence between dstance and the sgnal strength has been measured n advance. Ths snk node wll select two-dstance nformaton from the locatonknown nodes and that collected by tself to determne the (x, y) locaton of the new nodes. Wth the knowledge of D, D 1, D 2, (X 1,Y 1 ),and(x 2,Y 2 ),tseasytoobtanthevalueof

6 6 Internatonal Journal of Dstrbuted Sensor Networks C (X, Y) γ A (X 1,Y 1 ) D 1 D 2 H y α D β Hn D B (X 2,Y 2 ) Fgure 8: Solar power reflecton. O (Snk node) H n and coordnaton are known from node localzaton. H y s a constant. Hence, Fgure6:Exampleofnodelocatng. D e : maxmum tolerant error dstance γ=9 arctan ( H y H n x 2 +y 2 ), (1) D 15 Fgure 7: Camera asssted node locatng. (X, Y), whch only requres some fundamental geometrcal knowledge, as shown n Fgure 6. However, our experment results show that the error of ths RSSI measurement and locaton calculaton s up to a level of 1-2 meters wth an approxmately 2-meter D,whch s unacceptable for the applcaton. Therefore, to mprove the accuracy of localzaton, the onboard camera n the rotary platform of energy hub wll take the second run. The drectonal camera used n our experment has a vsual angle of 15, whch s a typcal value. In the case of 2-meter D, the maxmum tolerant error dstance for RSSI localzaton s shown n Fgure 7, whchs2 tan(7.5 ) = 2.63 m, larger than the experment results. After the camera captured the sensor node solar board, the rotary platform wll slghtly be rotateduntlthesolarboardmovestothecenterofthecamera vsual area. Then the exact drecton of the solar board can be determned, as shown n Fgure Sngle-NodeSolarPanelCoverage. It s also a challenge to enable reflected lght to fully cover the desgnated solar panel so that the utlty of solar power can be maxmzed. After accurate node localzaton, the rght angle of retroreflector should be calculated and the hardware component has to be correctly postoned by the rotaton of the rotary platform. Then, the mnor adjustment wll take place to make full coverage accordng to the response of the sensor node. Assume that the locaton of a node has been acqured usngourlocatngalgorthmsothatweknowthedstanceand heghtandsabletocomputetsrelatvelocatontotheenergy hub. The detaled process s descrbed as follows. In Fgure 8, α=arctan ( y x ). After obtanng γ and α we can control the rotary platform to reach the poston. Wth the knowledge of the accurate node locaton, t does not guarantee that the entre surface of the solar panel can be covered by the reflected sunlght. The solar panel may be located to any sde of the sensor node, whch generates a desgn challenge of the proposed scheme. In ths paper, we use slow movng capturng combned wth feedback from the nodes to solve ths problem. When the lght s reflected to the node, t wll result n ncreasng of lumnous flux of lght sensor.however,thelocatonrelatedtothesolarpowerhasa greatmpactonperformanceofthemethod.wesummarze four stuatons as shown n Fgure 9.InFgure 9(a), the lght sensor s covered by all the crcles whle none of the crcles provdes full coverage of the solar panel. In Fgure 9(b), the lght sensor s covered by none of the crcles whle each crcle provdes full coverage of the solar panel. In Fgure 9(c), t s desred stuaton that the solar panel and the lght sensor are both covered. Let us analyze the stuaton usng multple lght sensors. All possble placements of lght sensors around the solar panel are shown n Fgure 9(d).Whentwolghtsensors are deployed separately on the two sdes of the solar panel as two yellow dots, n most occasons, we can clarfy that the panel s fully covered upon both lght sensors beng covered. The choces of locaton are proved as follows. The placement of lght sensors s decded by the sze of the solar panel and the lght coverage crcle. For smplcty, let r stand for the radus of the lght crcle, let a denote the sde length of the solar panel, and let d be the dstance between the lght sensor and the center of the lght panel. Fgure 1(a) shows the specfed case that two sensor nodes are wthn maxmum dstance between each other. The dstance d can be calculated as follows: d 1 = r 2 ( r 2 ( a 2 )2 a 2 ) 2. (2)

7 Internatonal Journal of Dstrbuted Sensor Networks 7 (a) Sensed but not fully covered (b) Fully covered but not sensed (c) Fully covered and sensed (d) Multple lght sensors Fgure 9: On sngle node solar pad coverage. r d (a) a a r (b) Fgure 1: Computng of lght sensor placement. The best placement of lght sensor s shown n Fgure 1(b), and t can be computed as follows: d 2 = a 2 +(r r 2 ( a 2 )2 ). (3) As a result, the dstance of the lght sensor should not be greater than d 1 and s preferred to be d Schedulng of the Chargng Sequence We consder a wreless sensor network N consstng of N sensor nodes that are randomly dstrbuted n a two-dmensonal area of sze X Y. It s assumed that locatons of all sensor nodes are known once they are deployed. Each sensor node n, 1,...,N, s powered by rechargeable battery wth maxmum capacty E max, and a mnmum energy level E mn srequredtokeepthenodeoperatonal.theenergy consumpton rate of node n sassumedtobecr.lete (t) be the energy level of node n at tme t, and we say that a node n stays alve f t, e (t) E mn. AnumberofM energy hubs H are deployed n the same area to charge the sensor nodes, where each energy hub H j,j 1,...,M, s responsble for chargng a subset of d sensor nodes N j.wehave j 1,...,M N j = N, and j, k 1,...,M,j=k,N j N k =.Whenasensornoden s charged by energy hub H j,thats,n N j,tssadthatn s covered by H j. We assume that all energy hubs have unlmted energy capacty, a smlar chargng power of E c and effectve chargng range of R, whch means that only nodes wthn the chargng range can be charged by the energy hub. Denote by η,j the chargng effcency of n when n s charged by H j,andthe effcent chargng rate of n s then η,j E c.notcethatη,j s affected by several factors. For nstance, n wreless chargng, η,j s strongly related to Dst,j, the dstance between the node and the energy hub [1], whle, usng lght-reflecton chargng, η,j does not depend on Dst,j that much. In ths paper, we assume that η,j can be decded once the locatons of node n and energy hub H j are settled. Therefore, for the sake of clarty, we also denote η,j as η f n s mpled to be covered by a nonspecfc energy hub. Also notce that η,j = f Dst,j >R. To keep contnuous energy replenshment, each sensor node n s charged perodcally by ts correspondng energy hub H j. That s, for each chargng perod T c, n s charged by H j wth chargng tme t c. Theproblemthatwentendtosolvestofndthemnmum number of energy hubs, together wth ther deployment locatons and chargng sequences of ther covered sensor nodes, such that all nodes stay alve System Model. We consder a system consstng of N sensors nodes. Each node n hasabatterycapactyofe max and a mnmum energy level E mn at whch the node s operatonal. The energy consumpton rate of node n s a nonncreasng functon cr (t), and we do not restrct t to be lnear, as assumed n some related works [9]. Regardng the fact that most sensor nodes perform perodc actvtes n a certan

8 8 Internatonal Journal of Dstrbuted Sensor Networks scale of tme, t s reasonable to assume that cr (t) s a perodc functon, that s, cr (t+k T )=cr (t), k=1,2,3,...,wheret s the perod of cr (t). Thereby, the total power consumpton durng T s T E (T )= cr (t) dt. (4) The sensor nodes are recharged perodcally by the energy hub. Denote E c the chargng power of the charger and η the chargng effcency w.r.t. node n,andtheeffcentchargng rate for node n s then η E c.notcethatη equals zero f node n s not chargeable by the energy hub due to reasons lke beng outsde the effcent chargng range or beng blocked by obstacles. Context swtch cost whch ncludes reflector adjustment delay s mpled wth η. Let the length of the rechargng tme t c be nvarable n each perod, and the total energy recharged to node n durng one perod T s E c (T )=η E c t c. (5) Let e (t) be the energy level of node n at tme t, andwe say that a node n stays alve f t e (t) E mn.wentend to fnd that, gven the above system, does t exst a feasble schedule such that all nodes stay alve, and f so, how the feasble schedule s generated Problem Descrpton. We frst examne some condtons that must be satsfed before the problem s formally defned. Lemma 1. A node n stays alve f t satsfes e (t,k ) E (T )+E mn, k=1,2,3,..., (6) where t,k denotes the tme nstance at the begnnng of the kth perod. Proof. We prove the followng by contradcton. Assume the node goes dead durng the mth perod at tme t [t,m,t,m + T ],andthen e (t )=e (t,m ) E (t,m,t )+E c (t,m,t )<E mn, (7) where E (t 1,t 2 ) and E c (t 1,t 2 ) denote the consumed energy and recharged energy durng tmes t 1 and t 2,respectvely. Therefore, we have e (t,m )<Emn +E (t,m,t ) E c (t,m,t ) whch contradcts (6). E mn +E (t,m,t ) E mn +E (T ) Assume that e nt = e (t,1 ) s the energy level at the begnnng of the frst perod; accordng to Lemma1, the followng condton must be satsfed e nt (8) E mn +E (T ). (9) Lemma 2. A node n stays alve only f t satsfes E c (T ) E (T ). (1) Proof. We prove the followng by contradcton. If E c (T )< E (T ),then m,suchthatattmet [t,m,t,m +T ] e (t )=e nt m + k=1 (E c (T ) E (T )) +E c (t,m,t ) E (t,m,t ) < E mn (11) whch means the node goes dead at tme t, and therefore the lemma follows. From Lemma 2 and (5), we deduce that whch gves the rechargng tme E c (T ) =η E c t c E (T ) (12) t c E (T ) η E c. (13) However, t s not suffcent to clam that a node stays alve f the rechargng tme satsfes (13).Thssbecausef rechargng happens when the energy level of node n reaches E max, the rechargng effcency reduces to the equvalent value of the energy consumpton rate at that tme (thus keepng the energy level at E max ). Therefore, the actual recharged energy durng that perod s less than the desgnated E c (T ). On the other hand, such rechargng wastes energy and tme of the charger and accordngly the opportunty of mantanng other nodes operatonal. The followng lemma prevents such nsuffcent chargng tme. Lemma 3. A node n wth chargng effcency η E c s guaranteed to be recharged wth energy of amount E c (T ) evaluated as n (5) usng rechargng tme t c f e (t,k ) Emax E c (T ), k=1,2,3,... (14) Proof. The node s guaranteed to be recharged wth energy of amount E c (T ) usng rechargng tme t c f and only f durng chargng the energy level does not exceed E max. We then prove the followng by contradcton. Assume the energy level exceeds E max at tme t [t,m,t,m +T ],and then e (t )=e (t,m )+Ec (t,m,t ) E (t,m,t )>E max (15) whch gves e (t,m )>Emax E c (t,m,t )+E (t,m,t ) E max whch contradcts (14). E c (T ) (16)

9 Internatonal Journal of Dstrbuted Sensor Networks 9 Lemma 4. A node n wth chargng effcency η E c s guaranteed to be recharged wth energy of amount E c (T ) evaluated as n (5) usng rechargng tme t c f E c (T ) =E (T ). (17) Proof. Regardng (1), f E c (T )>E (T ),wehavee (t,k+1 )= e (t,k )+Ec (T ) E (T )>e (t,k ), k, whchmeansthatthe energy level at the begnnng of each perod keeps ncreasng. Therefore, t s clear that m,suchthat e (t,m )+Ec (T )>E max (18) whch volates (14) Lemma 3. By contradcton, the lemma follows. Equaton (17) means that durng each perod T,the consumed energy s equvalent to the recharged energy, andhencetheenergylevelatthebegnnngofeachperod remans the same; that s, e (t,k )=ent k. (19) We now formally defne the schedulablty problem mentoned at the end of Secton 5.1. Defnton 5. The system descrbed n Secton 5.1, subjected to the followng constrants: N e nt e nt E mn +E (T ), (2) E max E c (T ), (21) t c = E (T ) (22) η E c stays alve f each node n receves t c rechargng tme n each perod T. Notce that (22) can be drectly derved from (5) and(17) Schedulablty Analyss. To guarantee that each node n receves t c rechargng tme durng each T,tsuffcesto consder the equvalent problem of tasks schedulng on a unprocessor, whch s descrbed as follows. Defnton 6. The system descrbed n Defnton 5 stays alve f and only f N tasks, each of whch has perod T,constant executon tme t c, and relatve deadlne D = T, are schedulable on a unprocessor. Usng the well-establshed schedulng theory of real-tme systems [26], we have the followng. Lemma 7. The system descrbed n Defnton 5 stays alve f N t c B, (23) T =1 where B depends on dfferent schedulng algorthms and equals 1orN(2 1/N 1) f earlest deadlne frst (EDF) or rate monotonc (RM) algorthm s used [27], respectvely. Notce that, usng EDF, the condton of (23) s necessary and suffcent, whle, usng RM, t s only suffcent condton. For nterested readers, please refer to [26]fordetals. Equaton (23) shows that smaller rechargng tme t c helps mprove the schedulablty of the system. Wth respect to (1), ths argument strengthens (17) Selecton of e nt. From (2) and(21), we know that e nt mustbesetntherangeof[e mn +E (T ), E max E c (T )].Its worth mentonng that the schedulablty condton of (23)s not affected by the tmng when e nt s reached. Ths means whenever the energy level of node n reaches the decded value e nt, t mmedately jons the chargng schedulng and starts ts frst perod. Also notce that keepng e nt at a not-too-hgh level also helps reducng the leakage of the battery. As shown n [4], ultracapactors lose electrcal energy even when there s no operaton, and the problem of leakage s more serous when the energy level s hgh. Therefore, keepng the energy level at a moderate level benefts the reservaton of battery power Selecton of the Perod T. In ths secton, we dscuss the selecton of the perod T for a node n.weselectt accordng to the perodcty of the node s actvtes. However, a node may repeat ts actvtes wth hgh frequency, resultng n a short perod. Ths may lead to a nonneglgble context swtch cost f preemptve schedulng s appled. Therefore, n practce, T could be set multple tmes the actual perod of the repettve actvtes of node n. On the other hand, from (17), (2), and (21), we have resultng n E mn +E (T ) e nt E max E (T ) (24) E (T ) Emax E mn 2 (25) whch puts a constrant on the maxmum value of T. Meanwhle, certan energy-harvestng methods, for example, solar power, pose a lmtaton on the avalable chargng tme. Furthermore, when the characterstcs of the power source or node workload (and accordngly the power consumpton) change, the schedule must adapt to the varaton. It would then not be sutable to choose a too long perod n such crcumstance. In a word, perod T should be wsely selected, makng balance between several factors Energy Hub Coverage. As descrbed n [28], the problem of coverng all sensor nodes usng a number of charger nodes s not smlar to the wdely studed coverage problem n Wreless Sensor Networks. The latter one can be manly of two types: area coverage and target/pont coverage, whch ntend to fnd the requred number and exact locatons of sensor nodes that cover an entre pece of area or a lmted number of targets. Whle our problem, on the other hand, s to fnd a mnmum number of energy hubs and ther correspondng locatons such that all sensor nodes n the network wll receve contnuous energy replenshment. We adopt the framework proposed n [28], whch utlzed grd approxmaton, mnmum set cover problem, and greedy

10 1 Internatonal Journal of Dstrbuted Sensor Networks functon LOCATE CHARGERS(X,Y,s,R,N) N uncovered N, H whle N uncovered not equal H j, N j LOCATE NEXT(X,Y,s,R,N uncovered ) f N j equals ext Not feasble end f N uncovered N uncovered N j H H +{H j } end whle return H end functon Pseudocode 1: Pseudocode of the algorthm for fndng the locatons of energy-hubs to cover all sensor nodes. approach. The dfference les n the fact that we do not neglect the chargng tme and consder real-tme chargng scheme, whle n, [28],thechargngtmesassumedneglgbleandan energy hub s able to charge all sensor nodes wthn a certan range. Thereby, n our algorthm, sensor nodes that wll be covered by an energy hub must satsfy that (1) they are wthn the energy hub s effectve chargng range R and (2) they do not volate (23). To convert the nfnte search space to fnte one, the grd approxmaton technque s appled, where the deployment area of sensor nodes wth sze X Ys dvded nto grds wth sze s s, and all energy hubs wll only be placed on the grd ponts. At the cost of losng optmalty, ths converson sgnfcantly reduces complexty and saves computaton tme of the algorthm. Energyhubsarelocatedonebyone.Ateachstep,the best locaton of an energy hub s obtaned by testng n the fnte search space. Upon fndng the best locaton for the energy hub, ts covered sensor nodes are removed. The next energy hub s to be found n the same manner but consderng only sensor nodes that have not yet been covered by any energy hub. The procedure contnues untl all sensor nodes are covered. The pseudocode s shown n Pseudocode 1. It s shown that at each step of the whle loop, a functon LOCATE NEXT s nvoked to return the next energy hub H j and ts covered sensor nodes N j.pseudocode 1 also shows an nterestng pont that f, at any step, an energy hub s not able to cover any sensor node, t can be concluded that there s no feasble soluton to the problem. In that case, a smaller grd sze s s desrable. In fact, a smaller value of s, though makngthesearchmoretmeconsumng,usuallyleadsto a closer soluton to the optmal one, because t gves a fner approxmaton of the nfnty search space. Each step of fndng the locaton of the next energy hub H j, accomplshed by functon LOCATE NEXT, s proceeded n a greedy way. That means we search every grd pont andpcktheoneonwhch,fanenergyhubsdeployed, the maxmum number of sensor nodes wll be covered. In partcular, from each grd pont (x, y), the energy hub H j sassumedtobeplacedtocheckhowmanysensornodest s able to cover. Ths s acheved by calculatng the dstance Dst,j between each sensor node n, located at (x,y ),andths grd pont, and then comparng t wth the chargng range R. It s assumed that the locaton of each sensor node s known usng GPS or other localzaton methods [29]. To guarantee the condton gven by Lemma 7, onlyasubsetofallthe sensor nodes that are wthn the chargng range R and do not volate (23) wllbeselectedasthenodescoveredbythe energy hub. For each such placement of energy hub on a grd pont, we pck the one where the maxmum number of sensor nodes wll be covered, denotng the rato between t c and T c as U. If there are several such grd ponts gvng the same maxmum number, we select the one where a smaller sum of U s acheved. The pseudocode s shown n Pseudocode 2. Notce that, n the pseudocode shown n Pseudocode 2 at each grd pont (x, y), the set of sensor nodes that are wthn the chargng range R, denoted by N j, are ordered wth respect to U n nonncreasng order, and the frst several nodes that lead to volaton of (23) wll be removed from the set. Ths s to make sure that a larger number of nodes wll be selected to be covered. On the other hand, nodes wth smaller U arepreferableduetothefactthatsmalleru mples better chargng effcency. 6. Evaluaton and Experments We both use smulaton and testbed to evaluate our proposed methods ehub Coverage. Ths secton wll demonstrate the effectveness and performance of the proposed algorthm. In addton, the method presented n [28] wllalsobetestedand compared wth our algorthm. In partcular, we denote the two methods as follows. () Greedy-R-based refers to the method proposed n [28] where, at each step of locatng the next charger, the greedy search selects the grd pont where maxmum number of sensor nodes wll be covered based on the crteron that they are wthn the chargng range R. () Greedy-U-based sthemethodpresentednthspaper anddffersfromtheabovemethodnthecrteron of determnng whch sensor nodes wll be covered, where the condtonof (23) must be also consdered.

11 Internatonal Journal of Dstrbuted Sensor Networks 11 functon LOCATE NEXT (X, Y, s, R, N uncovered ) #nodes covered, U, H j, N j for x =toxstep s for y =toystep s H j (x,y) N j for each n N uncovered f Dst,j <= R N j N j +{n } end f end for each order N j w.r.t. U n non-ncreasng order for each n N j f nk N U k >B j N j N j {n } end f end for each U = n N U j f N j > #nodes covered or ( N j equals #nodes covered and U <U) #nodes covered N j U U H j H j N j N j end f end for end for return H j, N j end functon Pseudocode 2: Pseudocode of the functon to fnd the next energy-hub. In the frst smulaton, we randomly deploy 12 sensor nodes n an area wth dmenson of 5 m 5m. The grd approxmaton of ths area uses a step sze of s = 5m. Each node s battery has the same E max = 1JandE mn = J. (In practce, E mn s requred to be slghtly above for a sensor node to be operatonal. However, E mn = wll not jeopardze the correctness of our smulaton results, but mproves readablty.) For each node n,thechargng perod T c s gven the value randomly dstrbuted n the range [5, 1] 1 3 sec, the energy consumpton rate cr s evaluated as E max /2/T, and the ntal energy level equals E max /2, whch are set n accordance wth (6)and(14). For each energy hub to be deployed, t has chargng power E c = 1. W and chargng range R = 1.5 m. The chargng effcency η,j of each node depends on ts dstance to the energy hub H j andsobtanedusngthefollowngequaton: η,j = Dst,j (26) The functon of η,j s bult from experence and mples that η,j decreases as Dst,j ncreases and reaches 1.5%when Dst,j =R. In addton, the maxmum value of η,j s assumed to be 5%. The smulaton s performed n Matlab, and ts results of fndng deployment of energy hubs usng Greedy-R-based method and Greedy-U-based method, respectvely, are shown n Fgure 11.Inthefgures,dashedlnesmarkthegrd,small sold squares denote sensor nodes, and each crcle denotes an energy hub s chargng range, whle an energy hub s located at the center of a crcle that concdes wth a grd pont. The locaton of each energy hub s labeled as H,wherethesuffx s n accordance wth the orderng gven by the greedy search procedure. It s shown n Fgure 11(a) that, usng Greedy-R-based method, the frst energy hub H 1 s put on the grd pont of (1, 15), coverng 4 sensor nodes. The second energy hub H 2 s put at (1, 4) and also covers 4 sensor nodes. The 3rd and 4th energy hubs then cover 3 and 1 sensor nodes, respectvely. Dfferently, the Greedy-U-based method generates the deployment, shown n Fgure 11(b), where H 1 s located at (1, 1), coverng 3 nodes, H 2 s put at (5, 4), coverng 3 nodes, H 3 covers 3 nodes, and H 4 covers 2 nodes, whle an extra H 5 covers the last node. The dfference of the results produced by two methods s due to the fact that, when chargng tme s non-neglgble, an energy hub s only capable of chargng a lmted number of ts neghborng sensor nodes. To clarfy our pont, we have

12 12 Internatonal Journal of Dstrbuted Sensor Networks n 3 45 n 3 4 n 1 H 2 4 n 1 H 2 H 5 35 n 2 H 4 35 n H 4 Y (m) 25 H 3 Y (m) 25 H H H X (m) (a) X (m) (b) Fgure 11: Deployment of energy hubs usng Greedy-R-based and Greedy-U-basedmethods. Energy level (kj) Tme (hour) Energy level (kj) Tme (hour) Fgure 12: Energy level of node n 1 usng Greedy-R-based and FIFO schedulng. Fgure 13: Energy level of node n 2 usng Greedy-R-based and EDF schedulng. performed another smulaton of chargng the 4 sensor nodes covered by energy hub H 2 n Fgure 11(a). The smulaton s wrtten wth TrueTme [3], a real-tme schedulng toolbox n Matlab, and s smulated for seconds. Fgure 12 shows the energy level of node n 1, located at (3.47, 41.37), when the energy hub schedules the chargng sequence of ts 4 covered sensor nodes n FIFO order. It s shown that around 65 hours, the battery of the sensor node s depleted and ts energy level reaches. Furthermore, even when a real-tme schedulng algorthm s appled for the chargng sequence schedulng, t s stll not able to guarantee that all sensor nodes stay alve. In Fgure 13, we show the energy level of node n 2, located at (11.77, 36.65), when the energy hub schedules the chargng sequence usng EDF algorthm [27]. It clearly shows that the energy of n 2 s exhausted around 54 hours. In contrast, the deployment of energy hubs obtaned byourproposedmethod(greedy-u-based) guaranteesthe requred chargng tme of each sensor node and manages to keep the energy level of all sensor nodes above E mn.in Fgure 14, energy levels of the 3 sensor nodes that are covered by energy hub H 2 have been plotted. The fgure shows that, Energy level (kj) Tme (hour) n 1 n 2 n 3 Fgure 14: Energy levels of the 3 sensor nodes covered by energy hub H 2 usng Greedy-U-based and EDF schedulng. usng Greedy-U-based methodfor deployng theenergy hubs and EDF algorthm for schedulng the chargng sequence, no sensor node s energy level drops below zero durng the smulaton. Due to the lmted space, energy levels of the rest sensor nodes wll not be shown n ths paper.

13 Internatonal Journal of Dstrbuted Sensor Networks 13 Number of deployed energy hubs Greedy-U-based Greedy-R-based Step sze (m) Fgure 15: Comparson of the number of deployed energy hubs w.r.t. thestepszeofgrdapproxmaton. Toevaluatetheperformanceoftheproposedalgorthm, the thrd smulaton s performed. In ths smulaton, 8 sensor nodes are randomly deployed n a 1 m 1 m area. Each energy hub has chargng power E c = 2. W and chargng range R = 15m. The rest of the parameter settngs are the same as n the frst smulaton. We vary the step sze s of the grd approxmaton from 2 mto2 m. For each s, thelocatonsandchargngperodsofthe8 sensor nodes are randomly generated for 3 tmes. For each random generaton of sensor nodes, we perform both Greedy- R-based and Greedy-U-based methods and record the number of deployed energy hubs, respectvely. The result s summed up and averaged by 3 for each s, whchsthenplottedn Fgure 15. Fgure 15 showsthatasmallerstepszeofgrdapproxmatonleadstoasmallernumberofrequredenergyhubs,for both methods. On the other hand, the number of deployed energy hubs acqured by Greedy-U-based method s larger than that by Greedy-R-based method. Ths s true because Greedy-U-based method consders not only the chargng range R but also the schedulablty of an energy hub s covered sensor nodes and hence requres more energy hubs to cover all sensor nodes. Notce that the gap between two methods s not sgnfcant wth respect to the number of sensor nodes and decreases as step sze enlarges Chargng Schedulng. Two smulatons have been conducted to verfy the effectveness of the schedulng polcy descrbed n Secton5. In the frst smulaton, we consder a system wth 3 sensor nodes, each powered by a super capactor wth the maxmum = 2.5 V, whle ts mnmum voltage level s assumedtobe.everyt tme, each sensor node performs a specfc actvty (samplng, transmsson, etc.) that takes C tme and consumes energy at consumpton rate cr.takng voltage level E max Voltage (V) Voltage (V) Voltage (V) Tme (s) (a) Node Tme (s) (b) Node Tme (s) (c) Node 3 Fgure 16: Rechargng three sensor nodes usng EDF schedulng algorthm. nto account the constrant of (25), we let the chargng perod T for each node be N tmes T. Table 1 lsted the settng s of the above parameters, as well as the resultng E (T ) and the settng of the effcent chargng rates η E c and chargng tme t c.notcethatthe effcent chargng rates are set dfferent to reflect the dfferent lengths between sensor nodes and the energy hub. Usng EDF as the schedulng algorthm, we have 3 = (27) T =1 t c whch guarantees that all three sensor nodes wll be recharged n tme and kept operatonal. Let e nt = E (T ) for each node; the smulaton s run for 3 secusngmatlabandtruetme[31], and the result s shown n Fgure 16. It shows that, usng EDF real-tme schedulng algorthm, all three sensor nodes managed to keep ther energy level between E mn and E max and stay alve Evaluaton on Real Platform. We also buld a proof-ofconcept system to evaluate proposed model. The experments are establshed on real solar-powered sensor network. Our experment can be categorzed nto two parts. Frst, we conduct tests on how dfferent ranges of reflecton affect thesensornodewthsolarpowerharvestngcomponent. Ths s to evaluate the possblty of transfer energy by

14 14 Internatonal Journal of Dstrbuted Sensor Networks Table 1: Parameters of sensor nodes. Nodes T C cr N T E (T ) η E c t c T, C, T, t c are n sec; cr, η E c are n V/sec; E (T ) s n V 2 Sunlght reflected by plane mrrors 1 cm 2 cm Voltage (V) cm Tme (s) Fgure 17: Experment of three dfferent reflecton dstance. 2.5 m 1 m 2 m Sunlght reflected by concave mrrors reflectng sunlght and ts effcency. Furthermore, we obtan chargng pattern when dstance and lumnous ntensty vary n dfferent scenaro. Second, we use a flashlght as smulated sun lght and form a real network to examne node locatng algorthm and performance of schedulng. As seen n Fgure 17, wesetsupthreesetofsunlght reflecton systems whch are 5 cm, 1 cm, and 2 cm, respectvely. The experments are conducted usng plane mrrors and concave mrrors, respectvely. From Fgure 18 we can know that usng concave wll get better effcency. It wll only take 3 mnutes to charge the super capactor whle plane mrrors need about 1 mnutes to acheve the same goal. No matter what knd of methods are used, the chargng effcency s better than that of magnetc resonance. As expected, the dstance between energy hub and recever wll affect the effcency. Energy experence loss due to scatterng whle rangencreases.however,tshardtotellthedfference between three dstances when concave mrrors are employed. We also evaluated our model n an envronmental deployment as shown n Fgure 19. Node2sdrectlyfacngthe sunandothersarentheshadowoftrees.eachnodewll report ts status perodcally. If ther energy drops to a certan degree, they wll ask for energy transfer. Then, the energy hub wll turn to transfer energy smulated as n the lght of the flashlght. If more than one node s n shadow, the energy hub wll turn to the one whch has hgher prorty computed usng schedulng algorthm. If a new node s added, the energy hub wll locate t by RSSI. After that a test reflecton wll be conducted to measure the exact locaton. Then, the coordnaton of the new node wll be recorded by the energy hub. To evaluate node allocatng and functonalty of schedulng algorthm we performed ndoor experments as shown Voltage (V) m 1 m 2 m Tme (s) Fgure 18: Chargng facts wth dfferent mrror type and dstance. n Fgure 2. We set up an energy hub wth eght statonary nodes and use flashlght to smulate reflected sunlght. Each node wll report ts status perodcally. When a node s shaded t wll ask for energy transfer. We use boxes to shade one node at a tme. The energy hub wll turn to and drect the flashlght to the node to smulate the energy transferrng procedure. If multple nodes are n shadow, the energy hub wll turn to the one whch has hgher prorty computed usng schedulng algorthm. We also dd experment of addng a new node. When a new node s deployed, the energy hub wll locate t by RSSI and camera. After that, a test reflecton wll be conducted to measure the exact locaton. The chargng effcency of flashlght s shown n Fgure 21. The experments show that the proposed model s feasble. 7. Concluson Ths paper has presented an energy share system to coordnate the energy harvestng wreless sensor network to acheve

15 Internatonal Journal of Dstrbuted Sensor Networks 15 Node 3 Node 2 Node 1 Node 4 Fgure 19: Outdoor deployment of the elghthouse system. lfetme mantenance and energy balance n Wreless Sensor Networks has been ponted out wth hgh potental. elghthouse system has ts own lmtaton durng nght orunfavorableweather.inthefuture,weplantoextendthe work by consderng cooperatve charge, n the sense that a sensor node can be charged by more than one energy hub. In fact, we have assumed n ths paper that energy hubs have unlmted energy. Although ths s a reasonable assumpton when energy hubs are located n envronment where the energy for harvestng s abundant, ths may not hold n certan practcal scenaros. In that case, assgnng a set of sensor nodes to only one energy hub for chargng s not suffcent, and a multple coverage scheme s requred to handle ths ssue. Acknowledgments Ths work s supported by the Natonal Natural Scence Foundaton of Chna (Grant no , , and ),theZhejangProvnceNaturalScenceFoundaton (Grant no. Y111831), the subfoundaton of Zhejang Provncal Key Innovaton Team on Sensor Networks under Grant no. 29R546-4, and the Scentfc Research Foundaton for the Returned Overseas Chnese Scholars, State Educaton Mnstry. Voltage (V) Fgure 2: Indoor testbed usng flashlght. Charged by flashlght Tme (s) Fgure 21: Chargng facts of flashlght. energy balance and lfetme prolongng. It has clarfed that, by usng solar power reflecton system, effcent ntal state energy transfer can be acheved wth nexpensve energy hubequpment.asmplebuteffectveschemeforshort dstance sensor node accurate localzaton has been ntroduced and demonstrated as well as the sunlght retroreflector postonng method. The practcal experments prove the correctness of the desgn crtera and provde an outstandng performance. An ntroducton of the energy-aware schedulng mechansm has been carred out along wth the dervaton of the accuracy, whch enables the network to be energy balanced and sustanable. Through the proposed approach and correspondng algorthm, a novel way of achevng References [1] F.Ye,G.Zhong,J.Cheng,S.Lu,andL.Zhang, PEAS:arobust energy conservng protocol for long-lved sensor networks, n Proceedngs of the 23th IEEE Internatonal Conference on Dstrbuted Computng Systems,pp.28 37,May23. [2] T. Van Dam and K. Langendoen, An adaptve energy-effcent MAC protocol for wreless sensor networks, n Proceedngs of the 1st Internatonal Conference on Embedded Networked Sensor Systems (SenSys 3),pp ,November23. [3] M. Carde and D.-Z. Du, Improvng wreless sensor network lfetme through power aware organzaton, Wreless Networks, vol. 11, no. 3, pp , 25. [4] J. Qu, B. Ln, P. Lu, S. Zhang, and G. Da, Energy level based transmsson power control scheme for energy harvestng WSNs, n Proceedngs of the 54th Annual IEEE Global Telecommuncatons Conference: Energzng Global Communcatons (GLOBECOM 11), pp. 1 6, Houston, Tex, USA, December 211. [5] G.W.Challen,J.Waterman,andM.Welsh, IDEA:ntegrated dstrbuted energy awareness for wreless sensor networks, n Proceedngs of the 8th Annual Internatonal Conference on Moble Systems, Applcatons and Servces (MobSys 1),pp.35 48, ACM, June 21. [6] J. Yun, S. Patel, M. Reynolds, and G. Abowd, A quanttatve nvestgaton of nertal power harvestng for human-powered devces, n Proceedngs of the 1th Internatonal Conference on Ubqutous Computng (UbComp 8), pp.74 83,ACM, September 28. [7] J. Taneja, J. Jeong, and D. Culler, Desgn, modelng and capacty plannng for mcro-solar power sensor networks, n Proceedngs of the Internatonal Conference on Informaton Processng n Sensor Networks (IPSN 8), pp , St. Lous, Mo,USA,Aprl28.

16 16 Internatonal Journal of Dstrbuted Sensor Networks [8] A. Kurs, A. Karals, R. Moffatt, J. D. Joannopoulos, P. Fsher, and M. Soljačć, Wreless power transfer va strongly coupled magnetc resonances, Scence, vol. 317, no. 5834, pp , 27. [9] Y.Sh,L.Xe,Y.T.Hou,andH.D.Sheral, Onrenewablesensor networks wth wreless energy transfer, n Proceedngs of the IEEE (INFOCOM 11), pp , Shangha, Chna, Aprl 211. [1] Y. Peng, Z. L, W. Zhang, and D. Qao, Prolongng sensor network lfetme throughwreless chargng, n Proceedngs of the31stieeereal-tmesystemssymposum(rtss 1),pp , San Dego, Calf, USA, December 21. [11] T. Zhu, Y. Gu, T. He, and Z.-L. Zhang, EShare: a capactordrven energy storage and sharng network for long-term operaton, n Proceedngs of the 8th ACM Internatonal Conference on Embedded Networked Sensor Systems (SenSys 1), pp , ACM, November 21. [12] A. Savvdes, C.-C. Han, and M. B. Strvastava, Dynamc fne-graned localzaton n ad-hoc networks of sensors, n Proceedngs of the 7th Annual Internatonal Conference on Moble Computng and Networkng,pp ,July21. [13] D. Nculescu and B. Nath, Ad hoc postonng system (APS) usng AOA, n Proceedngs of the 22nd Annual Jont Conference on the IEEE Computer and Communcatons Socetes,pp , Aprl 23. [14] P. Bahl and V. N. Padmanabhan, RADAR: an n-buldng RFbased user locaton and trackng system, n Proceedngs of the 19th Annual Jont Conference of the IEEE Computer and Communcatons Socetes (INFOCOM ),pp ,March 2. [15] Z. Zhong and T. He, Achevng range-free localzaton beyond connectvty, n Proceedngs of the 7th ACM Conference on Embedded Networked Sensor Systems (SenSys 9),pp , ACM, November 29. [16]S.Rallapall,L.Qu,Y.Zhang,andY.-C.Chen, Explotng temporal stablty and low-rank structure for localzaton n moble networks, n Proceedngs of the 16th Annual Conference on Moble Computng and Networkng (MobCom 1), pp , ACM, September 21. [17] W.X,Y.He,Y.Luetal., Locatngsensorsnthewld:pursut of rangng qualty, n Proceedngs of the 8th ACM Internatonal Conference on Embedded Networked Sensor Systems (SenSys 1), pp ,November21. [18] M. I. Afzal, W. Mahmood, S. M. Sajd, and S. Seoyong, Optcal wreless communcaton and rechargng mechansm of wreless sensor network by usng ccrs, Internatonal Journal of Advanced Scence and Technology,vol.13,no.1,pp.49 69,29. [19] A. A. Syed, Y. Cho, and J. Hedemann, Demo abstract: energy transference for sensornets, n Proceedngs of the 8th ACM Internatonal Conference on Embedded Networked Sensor Systems (SenSys 1), pp , November 21. [2] R. Doost, K. R. Chowdhury, and M. D Felce, Routng and lnk layer protocol desgn for sensor networks wth wreless energy transfer, n Proceedngs of the 53rd IEEE Global Communcatons Conference (GLOBECOM 1), pp. 1 5, Mam, Fla, USA, December 21. [21] K. L, H. Luan, and C.-C. Shen, Q-ferry: energy-constraned wreless chargng n wreless sensor networks, n Proceedngs of the Wreless Communcatons and Networkng Conference (WCNC 12),pp ,212. [22] L. Xe, Y. Sh, Y. T. Hou, and H. D. Sheral, Makng sensor networks mmortal: an energy-renewal approach wth wreless power transfer, IEEE/ACM Transactons on Networkng,vol.2, no. 6, pp , 212. [23] C.M.Angelopoulos,S.Nkoletseas,T.P.Rapts,C.Raptopoulos, and F. Vaslaks, Effcent energy management n wreless rechargeable sensor networks, n Proceedngs of the 15th ACM Internatonal Conference on Modelng, Analyss and Smulaton of Wreless and Moble Systems (ACM 12), pp [24] M. K. Watfa, H. AlHassaneh, and S. Selman, Mult-hop wreless energy transfer n WSNs, IEEE Communcatons Letters, vol. 15, no. 12, pp , 211. [25] M. Erol-Kantarc and H. T. Mouftah, Msson-aware placement of rfbased power transmtters n wreless sensor networks, n Proceedngs of the IEEE Symposum on Computers and Communcatons (ISCC 12), pp , Cappadoca, Turkey, 212. [26] G. C. Buttazzo, Hard Real-Tme Computng Systems: Predctable Schedulng Algorthms and Applcatons, Sprnger, New York, NY, USA, 2nd edton, 24. [27] C. L. Lu and J. Layland, Schedulng algorthms for multprogrammng n a hard real-tme envronment, Journalofthe ACM,vol.2,no.1,pp.46 61,1973. [28] F. T. Jagrdar, M. M. Islam, and S. R. Huq, Grd approxmaton based nductve charger deployment technque n wreless sensor networks, Internatonal Journal of Advanced Computer Scence and Applcatons,vol.2,no.1,pp.3 37,211. [29] T.He,C.Huang,B.M.Blum,J.A.Stankovc,andT.F.Abdelzaher, Range-free localzaton and ts mpact on large scale sensor networks, ACM Transactons n Embedded Computng Systems, vol. 4, pp , 25. [3] D. Henrksson, A. Cervn, M. Andersson, and K. E. Årzén, TrueTme: smulaton of networked computer control systems, n Proceedngs of the 2nd IFAC Conference on Analyss and Desgn of Hybrd Systems,Alghero,Italy,June26. [31] A. Cervn, D. Henrksson, B. Lncoln, J. Eker, and K.-E. Årzén, How does control tmng affect performance? IEEE Control Systems Magazne,vol.23,no.3,pp.16 3,23.

17 Internatonal Journal of Rotatng Machnery Engneerng Journal of Hndaw Publshng Corporaton The Scentfc World Journal Hndaw Publshng Corporaton Internatonal Journal of Dstrbuted Sensor Networks Journal of Sensors Hndaw Publshng Corporaton Hndaw Publshng Corporaton Hndaw Publshng Corporaton Journal of Control Scence and Engneerng Advances n Cvl Engneerng Hndaw Publshng Corporaton Hndaw Publshng Corporaton Submt your manuscrpts at Journal of Journal of Electrcal and Computer Engneerng Robotcs Hndaw Publshng Corporaton Hndaw Publshng Corporaton VLSI Desgn Advances n OptoElectroncs Internatonal Journal of Navgaton and Observaton Hndaw Publshng Corporaton Hndaw Publshng Corporaton Hndaw Publshng Corporaton Chemcal Engneerng Hndaw Publshng Corporaton Actve and Passve Electronc Components Antennas and Propagaton Hndaw Publshng Corporaton Aerospace Engneerng Hndaw Publshng Corporaton Hndaw Publshng Corporaton Internatonal Journal of Internatonal Journal of Internatonal Journal of Modellng & Smulaton n Engneerng Hndaw Publshng Corporaton Shock and Vbraton Hndaw Publshng Corporaton Advances n Acoustcs and Vbraton Hndaw Publshng Corporaton