Design of an active radio frequency powered multihop wireless sensor network

Transcription

1 Design of an acive radio frequency powered mulihop wireless sensor nework J.P. Olds and Winson K.G. Seah School of Engineering and Compuer Science Vicoria Universiy of Wellingon Wellingon, New Zealand Absrac In his paper, we presen a low power soluion for muli-hop wireless sensor neworks using acive radio frequency (RF) harvesing. We show by careful design of boh hardware and nework proocol ha i is possible o design muli-hop neworks ha use minimal power. We idenified criical facors and incorporaed hem ino he proocol design, which we validaed by implemening a small esbed of nodes using commercially available RF energy harvesing devices. The speed a which sensor daa can be rerieved from he nework has been evaluaed and deemed o be viable for wireless sensor neworks used in low duy cycle monioring applicaions. Keywords- RF harvesing; wireless sensor nework; sink synchronized muli-hop; I. INTRODUCTION When deploying a wireless sensor nework (WSN), here are some obvious disadvanages when baeries are used o power he nodes. Firsly, besides he need o replace he baeries once hey run ou, here are also siuaions where replacing a baery is oally infeasible such as when a wireless sensor node is embedded wihin a concree building or some oher place ha humans canno reach. Secondly, baeries can leak heir conens ino he surrounding environmen, and his can lead o problems ranging from polluion of he environmen o corrosion of he surrounding maerial and compromising he srucure. This has led many o invesigae innovaive new soluions ha avoid requiring baeries. One such soluion is o use acive radio frequency (RF) energy harvesing. Acive RF energy harvesing is a mehod whereby an RF energy ransmier (ET) ransmis energy o he WSN nodes using RF waves. I is acive in he sense ha you need o urn on he ET before he nodes can harves energy. Besides no needing a baery and he advanages ha come wih his, anoher advanage of acive RF harvesing is having he abiliy o urn off he power o he nodes wihou ouching he nodes hemselves. However, using RF waves o power nodes also has some operaional challenges, of which, he mos criical one is he power ha a node ypically ges is very low if he nodes are o funcion a any reasonable disance from an ET ransmiing a realisic power levels. Consequenly, sar opology neworks have been used when using RF harvesing. I is desirable ha he nodes powered by RF harvesing are able o work a disances as far as possible from he ET. Bu, his presens problems as he amoun of available RF power decreases rapidly wih disance from an RF energy ransmier. Regardless of he amoun of RF energy ha he nodes are receiving, hey always have cerain unavoidable power overheads (quiescen power); as a resul of his, he efficiency wih which he nodes can obain energy for useful asks such as communicaing decreases wih disance. Boh he rapid reducion of available RF power due o propagaion loss and he reducion in efficiency wih disance significanly affec he maximum disance ha a node can operae a. Anoher problem when using RF o power nodes is due o he way RF waves propagae which makes he amoun of RF power a any paricular locaion in space difficul o deermine ahead of ime and can even change over ime. Boh RF absorpion from maerials, and reflecions off surfaces causing inerference can make he power eiher more or less han wha migh be prediced using he RF inverse square law. The power a node receives can even change wih respec o ime because of objecs such as people, animals and plans moving pas he nodes. Nodes also generally receive very differen levels of power compared o one anoher which can make proocol design difficul. Due o he low power ha one ges from RF harvesing, i is primarily used for applicaions where sensor readings are required infrequenly. In his projec, our arge applicaion lies in he agriculural domain where sensor readings need o be aken only a few imes daily. We designed a muli-hop nework ha uses acive RF energy harvesing and implemened a proof-of-concep prooype o validae he design. The low energy availabiliy from RF energy harvesing consrains our design of he wireless sensor node o consume less han 10µW. In he nex secion, we briefly discuss relaed work on WSNs powered by RF energy harvesing. Following ha, we presen he design of he node hardware and nework proocol, before discussing he experimenal validaion and conclusions. II. RELATED WORK To he bes of our knowledge, no one has invesigaed RF powered muli-hop WSNs nor implemened hem. Previous work on using RF harvesing o power WSNs has focused on sar opology neworks and mos, if no all sudies, relied only on simulaions. Simulaions while ineresing can be flawed due o incorrec assumpions and neglecing power overheads required in acual implemenaions. In addiion, he absence of research ino RF powered muli-hop WSNs neglecs anoher imporan ype of nework archiecure.

2 Obaining energy passively using RF harvesing from TV ransmiers has been researched by [1] and [2]. I has been repored in [1] ha a Yagi anenna was used o obain 60µW of usable power from a TV ransmier ower locaed 4km away. In [2], while i has been saed ha 20µW of usable power can be obained from a TV ransmier ower up o 6.6km away, i was only menioned ha heir prooype node has been esed a disances of 500m away from he TV ransmission ower, and ha he node was using several hundred microwas a his disance; he minimum power requiremens for he node was no menioned. The RF power densiy of GSM cellular phone owers in residenial areas in Germany have been invesigaed and he median power densiy has been found o be 200µW/m 2 [3]. Oher sudies, e.g. [4] and [5], have looked ino he RF o DC conversion circuiry and powering he conversion circuiry using nearby cellular owers, menioning only he volage obained bu no usable power [5]. Furhermore, he maximum experimenal disance is also somewha limied o 50m from a cell ower. Recenly, commercial producs for acive RF harvesing have been made available by Powercas [6]; hey produce a 3W RF ransmier and wireless sensor nodes ha operae up o a few ens of meers. Their node s communicaion radio, microprocessor and sensor circuiry is unpowered unil enough energy has been gahered, a which ime he node is powered up, he sensors ake readings, and he node ransmis a packe; hen, he microprocessor, he communicaion radio and sensor circuiry is unpowered once again. This approach is based on a sar nework where nodes closer o he RF ransmier ransmi heir daa more ofen han he ones furher away. A medium access conrol (MAC) proocol [7] has been proposed ha compensaes for he unfairness in power obained by he nodes due o differing posiions from he RF ransmier by allowing nodes ha receive lesser power o have a higher probabiliy of winning a ransmission conenion. However, heir resuls are based on solely on simulaions using he specificaions of Powercas s devices, and all heir graphs show ransmission power levels ha are more han five imes he legal limi for he 900Mhz ISM band in he USA. In [8], an acive RF powered wake-up radio sysem has been proposed for baery powered body sensor nodes and nodes ha deliver drugs on demand. This allows he nodes o respond o commands sen from a maser node exernal from he body wihin a few milliseconds. In addiion, he nodes do no use any energy from heir baeries lisening for his command because he command also conains he energy o power he wake-up radio. III. NETWORK STRUCTURE Of all he muli-hop neworks we considered, a sink synchronized muli-hop nework used he leas amoun of power [9]. Synchronized muli-hop neworks allow nodes o urn on heir radios for shorer periods of ime han nodes in asynchronous neworks and his saves energy. In a sink synchronized muli-hop proocol, each node mus be able o hear he synchronizer bu vice versa. Fig. 1 shows he nework srucure of our sink synchronized muli-hop nework using acive RF energy harvesing. Zone where ETSS can ransmi power and informaion o. ETSS Energy Transmier, Synchronizer and Sink (ETSS) ETSS Wireless Sensor Node Powered by Energy Transmier. Figure 1. Sink synchronized muli-hop nework using acive RF harvesing. The sink, which is also he synchronizer, is a specialized node ha has no power consrains and has a much higher power communicaion radio so as o be able o ransmi commands o all he nodes. The ET uses differen frequencies o ha of he communicaion radios so ha he ET s RF waves do no inerfere wih he communicaion radio; his means he ET can power he nodes a he same ime as he nodes can communicae wih one anoher. IV. HARDWARE For his projec, we used Powercas s 3W ET and heir P2110 RF energy harvesing chips as he basis of geing energy from he mains, over he air, o he nodes. The ET ransmis approximaely a 60 beam of 915MHz radio waves wih an Effecive Isoropic Radiaed Power (EIRP) of 3W. Wih an aerial and a 1mF inermediae capacior, he P2110 chips receive his energy urning i ino a pulsed 3.3V oupu. The duy cycle of his oupu varies depending on he power he chips are receiving from he ET and he power drawn from his 3.3V oupu. Our nodes connec o his pulsed 3.3V power supply oupu as shown in Fig. 2. Figure 2. Block diagram of node powered by RF harvesing.

3 The power managemen block fills he node s 4.7mF reservoir capacior when he boos converer is on and aemps o minimize he leakage from he 4.7mF back ino he boos converer when he boos converer is off. When volages are below 2.4V, he power managemen block disconnecs he microprocessor, whils volages above 3.1V would promp he power managemen block o signal he microprocessor ha he capacior is full by waking i up. The quiescen curren of he power managemen block is ypically less han 400nA. The oupu from he power managemen block is unregulaed and will flucuae from 2.4V o 3.3V depending on he charge lef on he 4.7mF capacior. This unregulaed volage allows he boos converer o operae a higher currens for shorer periods of ime hus increasing he efficiency wih which he energy in he 1mF is ransformed ino 3.3V. The node is able o use less han 2µW quiescen power when recharging he 4.7mF capacior. In his sae, he microprocessor swiches off all of is oscillaors and can only be woken up again when he power managemen block signals he microprocessor ha he capacior is full. Fig. 3. shows he node prooype. The sink/synchronizer consiss of a crysalconrolled microprocessor wih a MRF24J40MC communicaion module. This module, according o he daashee, can ransmi up o 1000m. A. Usable power a disance We measured he amoun of power a node obains afer he RF o pulsed 3.3V conversion ha is performed by he P2110 RF energy harvesing chips wih respec o disance from he 3W Powercas energy ransmier. This was performed in a room roughly 12m 2.4m 3.6m. The nodes used a high gain pach anenna wih a gain of 6.1dBi, which we posiioned o maximize he power each node was receiving for each disance. Fig. 4 below shows he poins we obained along wih a power leas main squares (LMS) fi. As shown, here is a general rend for less power when disance increases. Due o he sofware implemenaion, he nework proocol we designed requires he nodes o obain a leas 7µW o funcion properly in a muli-hop fashion. We can see ha nodes farher han 10m away from he ET can sill receive enough power o allow he nodes o funcion properly (7µW minimum). While here is no exac cuoff poin for when he nodes are unable o funcion properly, we found he maximum disance is generally is limied o around 12m. Power he node obains versus disance from energy Figure 3. Node prooype ransmier using he 6.1dBi pach anenna 1000 Power o node (uw) The microprocessor block has he abiliy o supply regulaed volage o boh he sensors, and a 2.45Ghz IEEE complian communicaion radio. The microprocessor uses an inernal high-speed resisor capacior (RC) oscillaor for code execuion, and has he abiliy o swich on an exernal crysal oscillaor when precise synchronizaion is needed beween iself and he res of he nework. The communicaion radio is an MRF24J40MA module from Microchip. The module, according o he daashee, uses 63mW when receiving, 76mW when ransmiing, and can ransmi up o 100m. The radio module, even in receiving mode, uses over 30,000 imes more power han when he node is recharging is reservoir capacior. Energy-wise, he radio module is he mos expensive componen of he node. 100 y = * x^( ) R= Disance (m) Figure 4. Power node obains versus disance. V. NETWORK PROTOCOL Taking ino consideraion he consrains posed by RF energy harvesing, we designed a sink synchronized muli-hop nework proocol. The nodes harves energy and when enough energy has been gahered, he power managemen block wakes up he microprocessor; he node hen checks for a command from he synchronizer o perform. The ime beween checking for commands from he synchronizer is non-synchronous beween nodes and dependen on he power ha he nodes are receiving. Such commands sen by he synchronizer o he nodes are o insruc he nodes o ake sensor readings, discover neighboring nodes, relay daa, ec. Also conained in hese commands is iming informaion ha allows he synchronizer o synchronizes all he nodes wih one anoher, so ha each node knows wha all he oher nodes are doing and his allows he nodes o effecively communicae wih one anoher by enabling he nodes o lisen only when oher nodes are sending, hus reducing energy consumpion. Due o he redundancy in he commands sen by he synchronizer, by sending each command muliple imes, any node need only hear a fracion of he command ransmission period o receive he command and synchronize iself wih he oher nodes; his allows he nodes o keep heir radios in an off sae more ofen, hus also saving energy. The proocol does no sipulae ha a node needs o check for commands from he synchronizer a paricular imes.

4 Insead he proocol sipulaes wha maximum ime nodes are allowed beween checking for commands. This happens due o he rae a which he synchronizer sends commands; if commands are sen more frequenly hen nodes have a shorer ime beween checking for commands. The flowchars in Fig. 5 below show he command sequences sen o he nodes from he synchronizer o iniialize he nework and o rerieve daa from he nework. Iniialize command sequence Adverise repea command TWO imes Assign neighbors band #1 repea command TWO imes Adverise ill number of bands is unchanged for 3 consecuive Adverise commands N Daa rerieval command sequence Record sensor daa Se iniial band equal o he number of bands. Hop wih ACKs se o lis of MAC addresses las received Go responses? Y 2 consecuive hops wih no daa reurned? Y N Increase iniial band by 1 Figure 5. Command sequences for iniializing and rerieving daa E = (2) η ( PTOT PB ) The synchronizer commands, along wih measured values of boh he node s capacior recharge recovery efficiencies of recovering from a command, and he energy used by he nodes due o each command are lised in Table 1. TABLE I. Nework Proocol Command Assign neighbors o band #1 Adverise Record sensor daa Hop Idle ENERGY CONSUMPTION AND CAPACITOR RECOVERY EFFICIENCY FOR VARIOUS COMMANDS Energy consumpion Capacior recovery efficiency 2.2 ± 0.7mJ 92 ± 2% 4.2 ± 1.0mJ 89 ± 2% 0.78 ± 0.25mJ 93 ± 1% 3 mj (ypical) 91 % (ypical) 0.87 ± 0.19mJ 93± 1% The so-called idle command is no echnically a command bu behaves similar o one. The idle command refers o he node periodically checking for a command from he synchronizer and failing o receive anyhing because he synchronizer has no sen any command a ha ime. We measured he ime a node akes o recover from he idle command ( I ) for differen levels of received power, and compared his wih he heoreical Eqn (2) for he idle command; Fig. 6 is a graph of boh our measured poins along wih Eqn (2) for he idle command as a comparison. VI. COMMAND TIMING 120 A. Command recovery ime We can model he soring of energy of our nodes wih a simple resisor capacior (RC) nework over he pulsed 3.3V power supply. We see ha here is an inheren associaed efficiency of soring energy in he reservoir capacior as given by Eqn (1), where V b is he volage over he power source, while V cini and V cfin are he iniial volage and he final volage across capacior respecively. V = cfin + V 2V b cini η (1) If we ake ino accoun he quiescen power ha is used by he node when i is in a recharging sae, we can derive an expression for he period of ime a node akes o recover from a command (in he sense of recharging is reservoir capacior) wih respec o he amoun of power he node receives. This is expressed in Eqn (2), where P B is he quiescen power used by he node when recharging, P TOT he power he node is receiving, E energy he node uses in response o he command, and period of ime he node akes o recover from he command. ime (s) power (uw) Figure 6. Measured node idle recovery command ime versus received power and heoreical recovery ime. As shown, Eqn (2) does indeed predic he period of ime i akes a node o recover for an idle command very well. B. Command period A command sen by he synchronizer consiss of a ransmission period ( TX ) where he synchronizer sends packes as fas as i can, followed by an acion period ( Acion ) where he nodes and he synchronizer perform he required acion due o he command, and finally followed by a res period ( Res ) where he synchronizer does no send any packes and allows he

5 nodes o res. The minimum period of ime he synchronizer ransmis for when sending a command ( TX ) mus be equal o he period of ime a node akes o recover from an idle command ( I ); his ensures he node is able o receive he command, as well as minimizing he period of ime he synchronizer needs o ransmi for. Similarly, he minimum period of ime ha he synchronizer needs o wai before sending he nex command ( Res ) mus be equal o he period of ime he node akes o recover from he command minus he period of ime i akes o recover from he idle command. The period of ime ( Acion ) required o perform he acion due o a command is dependen on proocol parameers, hardware parameers, nework opology, and he command iself; i is independen of he amoun of power a node is receiving. For our paricular implemenaion, all commands excep for he hop command, Acion is less han 50ms, while Acion for he hop command is less han 4.8s. As we are ineresed in he synchronizer being able o send commands o all nodes in he nework, i is sufficien o calculae TX and Res by considering only he node in he nework ha receives he leas amoun of power P MIN ; any oher node receives more power and hus is guaraneed o have is radio on when he synchronizer sends a command o he node. This means we can calculae TX and Res for he synchronizer by using he following wo formulae obained from Eqn (2). leas amoun of power P MIN. Any node receiving more power han his mus have filled is reservoir capacior. From Eqn (1), we see ha by iniially charging he capacior from an empy sae, he efficiency of soring energy by his process is 50%. Treaing his iniial charging like a recovery from a command ha oally deplees he capacior, and using Eqn (2), he ime he capacior akes o charge from an empy sae is given by he following: Inial 2 CVcfin = P Pʹ where Pʹ is he quiescen power used when he node is B iniially charging, C is he capacior raing and V cfin is he final volage across he capacior. This is less han he quiescen power used when he node is recovering from a command sen by he synchronizer as he power managemen block disconnecs he microprocessor, which in urn disconnecs he communicaion radio and sensors when he reservoir volage is less han 2.4V. For our nodes P B ʹ 1uW and can be assumed o be negligible. Therefore, we assume he period of ime we mus wai afer urning he energy ransmier on before sending commands can be approximaed wih he following: MIN B (6) TX I = (3) η I E ( P P ) MIN B Inial 45mJ P (7) MIN E Res = TX (4) η ( PMIN PB ) For simpliciy, raher han adjusing Res and Acion for every command sen, we use a wors-case scenario and use consan values for Res and Acion ; in addiion o simpler implemenaion his allows a simple conversion beween oal ime aken and oal number of commands performed. The command period τ is simply TX + Acion + Res. Using a wors-case scenario of E = 5mJ, η = 0.9 and Acion = 4.8s, we se he command period in he nework o be Eqn (5) by using Eqns (3) and (4): VII. EXPERIMENTAL TEST BED We consruced a es bed consising of seven nodes in a room approximaely 6m 3.4m 2.4m. We used one energy ransmier and one synchronizer as he sink conneced o a compuer for daa rerieval from he nework. Due o he physical space consrains, o allow more han one band o form, he communicaion radios of he nodes were se o he absolue minimum ransmission power ha he radios allowed. This had almos no effec on he amoun of energy he radios used as he radios are around 99% inefficien when ransmiing. This ransmission power level allowed he radios o communicae wih one anoher up o a maximum of around 2m τ = (5) P MIN C. Iniial capacior charging Before sending a command o he nodes, all nodes mus fully charge heir reservoir capaciors. Here, we calculae he period of ime we mus wai before we can send commands o he nework due o nodes having no fully recharged heir capaciors. As we are ineresed sending commands o all nodes in he nework, we jus consider he node in he nework receiving he ETSS Figure 7. Mos common nework opology 6

6 The iniializaion command sequence as shown in Fig. 5 caused he nodes o group hemselves in a non-pre-deerminisic way. However, he mos common opology for he nodes o group hemselves ino is shown in Fig. 7. Also shown in Fig. 7 are he nodes approximae relaive posiions o one anoher. A value of 10µW was used as an iniial esimae for P MIN as Fig. 4 suggess ha all nodes should be receiving more power han his because he nodes are in a room smaller han 10m. Afer he iniial capacior charging Inial as given in Eqn (6), iniializing he nework and rerieving daa using his iniial P MIN esimae resuled in he nodes reurning he approximae power hey were obaining; his allowed us o boos P MIN o 80µW increasing he speed a which daa could be gahered from he nework. Performing 700 rials of iniializing he nework and rerieving daa from he nework resuled in requiring 15.2 ± 0.1 commands on average for each rial. As all our commands had he same period τ, he oal period of ime aken on average per rial was 15.2τ ± 0.7%. Using his along wih Eqn (5) allowed us o calculae how long i would ake for his 7-node nework o be iniialized and for all he daa o be rerieved wih respec o he amoun of power being obained by he node obaining he leas amoun of power (minimum node power); Fig. 8 shows a plo of his. Time (minues) Power (uw) Figure 8. Time aken o iniialize and rerieve daa from he nework versus minimum node power VIII. CONCLUSIONS In his paper, we have shown ha indeed i is possible o design low power muli-hop neworks ha use RF harvesing as heir source of power. The use of unregulaed volage and he energy harvesing process iself waking up he microprocessor significanly conribued o he nodes abiliy o work a such low power. Even wih nodes receiving vasly differen levels of power as one anoher, he sink synchronized command based muli-hop nework proocol successfully was able o handle his large discrepancy by solely considering he node in he nework ha was obaining he leas amoun of power. Implemening a small 7-node esbed we saw ha he rae a which sensor daa can be obained from he nework was dependen on he minimum node power. Times ranged from a few minues o a few hours depending on his minimum power. This makes our curren implemenaion suiable for WSNs ha require sensor daa relaively infrequenly and where laency is no a significan issue. To he bes of our knowledge, his is he firs design and implemenaion of a muli-hop nework especially for use wih RF harvesing. While we have focused on acive RF harvesing, he nework proocol and he hardware can be adaped o work wih oher ypes of energy harvesing such as solar, vibraion, ec. For WSNs where baeries are no an opion, acive RF harvesing is a poenial avenue for furher research. Wih he curren sae of echnology, using low power ETs like he ones used in his paper, he disance a which nodes will operae from an ET is sill limied. However, as echnology advances and improves he efficiency of he conversion beween RF energy and usable elecrical energy, along wih a reducion in power needed for elecronic componens (paricularly, he communicaion radios) he disance a which nodes can operae from a given ET is expeced o increase over he years. REFERENCES [1] A.P. Sample and J.R. Smih, Experimenal Resuls wih wo Wireless Power Transfer Sysems, Proceedings of he IEEE Radio and Wireless Symposium (RWS), Jan 2009, San Diego, CA, USA. [2] H. Nishimoo, Y. Kawahara and T. Asami, Prooype implemenaion of ambien RF energy harvesing wireless sensor neworks, Proceedings of IEEE Sensors Conference, 1-4 Nov 2010, Kona, HI, USA. [3] T. Haumann, U. Münzenburg, W. Maes and P. Sierck, HF radiaion levels of GSM cellular phone owers in residenial areas, Proceedings of he 2 nd Inernaional Workshop on Biological effecs of EMFS, Rhodes, Greece, [4] A. Dolgov, R. Zane, Z. Popovic, Power managemen sysem for online low power RF energy harvesing opimizaion, IEEE Transacions on Circuis and Sysems I: Regular Papers, pp , vol 57, issue 7, Jan 2010 [5] M. Arrawaia, M.S. Baghini, G. Kumar, RF Energy Harvesing Sysem from Cell Towers in 900MHz Band, Proceedings of he Naional Conference on Communicaions (NCC), Jan 2011, Bangalore, India. [6] Powercas Corporaion, hp:// [7] J. Kim, J.W. Lee, Energy adapive MAC proocol for wireless sensor neworks wih RF energy ransfer, Proceedings of he 3 rd Inernaional Conference on Ubiquious and Fuure Neworks (ICUFN), June 2011, Dalian, Chian. [8] Z. Xiaoyu e al. An energy efficien implemenaion of on-demand Mac proocol in medical wireless body sensor neworks, Proceedings of he IEEE Inernaional Symposium on Circuis and Sysems (ISCAS), May 2009, Taipei, Taiwan. [9] J.P. Olds, and W.K.G. Seah, Power consideraions for very low duy cycle wireless sensor neworks powered by energy harvesing, Technical Repor ECSTR 11-11, School of Engineering and Compuer Science, Vicoria Universiy of Wellingon, New Zealand, 16 Oc 2011.