Collaboraive communicaion proocols for wireless sensor neworks Sefan Dulman, Lodewijk v. Hoesel, Tim Nieberg, Paul Havinga Faculy of Elecrical Engineering, Mahemaics and Compuer Science Universiy of Twene Enschede, The Neherlands {S.O.Dulman, L.F.W.VanHoesel, T.Nieberg, P.J.M.Havinga}@uwene.nl Absrac In his documen, he design of communicaion wihin a wireless sensor nework is discussed. The resource limiaions of such a nework, especially in erms of energy, require an inegraed approach for all (radiional) layers of communicaion. We presen such an inegraed, collaboraive approach which is par of curren research in he European research projec EYES on energy-efficien sensor neworks. In paricular, energy-efficien soluions for medium access conrol, cluserbased rouing and mulipah rouing are discussed. As par of he ongoing projec, hese approaches work ogeher and are designed o suppor each oher. I. INTRODUCTION Wireless sensor neworks (WSNs) are an emerging field of research which combines many challenges of modern compuer science, wireless communicaion and mobile compuing. WSNs are one of he prime examples of Ambien Inelligence, also known as ubiquious compuing. Ambien sysems are neworked embed sysems inimaely inegraed wih everyday environmen and supporing people in heir aciviies. These sysems are quie differen han hose of curren compuer sysems, and will have o be based on radically new archiecures and use novel proocols. The vision of ubiquious compuing requires he developmen of devices and echnologies, which can be pervasive wihou being inrusive. The basic componens of such a smar environmen will be small nodes wih sensing and wireless communicaions capabiliies, able o organize flexibly ino a nework for daa collecion and delivery. Realising such a nework presens very significan challenges, especially a he archiecural and proocol/sofware level. Major seps forward are required in he field of communicaions proocol, daa processing, and applicaion suppor. Recen advances in sensor echnology, low power analog and digial elecronics and low-power radio frequency design have enabled he developmen of cheap, small, low-power sensor nodes, inegraing sensing, processing and wireless communicaion capabiliies. Embedding millions of sensors ino an environmen creaes a digial skin or wireless nework of sensors. These massively disribued sensor neworks, communicae wih one anoher and summarize he immense amouns of low-level informaion o produce daa represenaive of he overall environmen. From collaboraion beween (large) groups of sensor nodes, inelligen behaviour can emerge ha surpasses he limied capabiliies of individual sensor nodes. Sensor nodes collaborae o be able o cope wih he environmen: sensor nodes operae compleely wireless, and are able o sponaneously creae an imprompu nework, assemble he nework hemselves, dynamically adap o device failure and degradaion, manage movemen of sensor nodes, and reac o changes in ask and nework requiremens. Despie hese dynamic changes in configuraion of he sensor nework, criical real-ime informaion mus sill be disseminaed dynamically from mobile sensor daa sources hrough he self-organising nework infrasrucure o he applicaions (services). This paper deals wih he neworking proocols involved in a WSN as being developed wihin he EYES projec. We address radiional layers, bu unlike hose well-known varians, we use a more inegraed view. The EYES projec (IST-2001-34734, hp://eyes.eu.org) is a hree year European research projec on self-organizing and collaboraive energy-efficien sensor neworks. I addresses he convergence of disribued informaion processing, wireless communicaions, and mobile compuing. The goal of he projec is o develop he archiecure and he echnology, which enables he creaion of a new generaion of sensors ha can effecively nework ogeher so as o provide a flexible plaform for he suppor of a large variey of mobile sensor nework applicaions. II. ENERGY EFFICIENT MAC PROTOCOL FOR SENSOR NETWORKS (EMACS) A. Goals The ype of applicaions makes ha WSNs differ grealy from convenional neworks like daa or elecommunicaion neworks. WSNs are buil of auonomous nework nodes ha each have limied processing power, memory and energy available. The neworks have o operae in a self-organizing adhoc fashion, since none of he nodes is likely o be capable of delivering he resources o ac as basesaion or cenral manager. In general wo ypes of nework nodes are recognized: nodes ha mainly ransmi heir own sensor readings (sensor nodes) and nodes ha mainly relay messages from oher nodes (relay nodes). Sensor readings will flow from source node o sink node in he nework via relay nodes. The ype of a nework node may change during he lifeime of he nework. The MAC proocol should suppor he ypical communicaion
Frames : Time slos: Secions: CR Fig. 1. n 1 2 3 TC n+1 Daa TDMA based MAC proocol beween sensor nodes and relay nodes, while i minimizes he overhead necessary o se up connecions. Moreover, i should allow for nodes o urn off heir energy consuming ransceiver as ofen as possible. Laency is in general less imporan in WSNs, alhough he suppor for mobile node communicaion requires a fas connecion seup. These characerisics make a MAC proocol for WSNs fundamenally differen from MAC proocols for mobile adhoc neworks. These limiaions and demands form a challenge for designing a power efficien communicaion proocol. B. The designed MAC proocol The main ask of he MAC proocol is o organize how he nodes in he WSN access he radio channel. The MAC proocol only deals wih communicaions beween nodes ha are in radio range of each oher. Higher layers, especially rouing, should cope wih he fac ha he sink of he daa is no always direcly reachable. This par of our combined approach will be presened in he proceeding secions. For he EYES projec, nework nodes are designed o demonsrae he funcionaliy of he designed adhoc neworking proocols. Basically he nework node design faciliaes a physical layer o he air inerface, which is a single radio channel a 868.35 MHz. The physical layer is capable of ransporing bis a a rae of 115.2 kbps. In he EYES projec we explore a TDMA based MAC scheme, since code division muliple access (CDMA) or carrier sense muliple access (CSMA) based proocols imply consan or very frequen lisening o he radio channel. This lisening o he channel consumes a large amoun of energy, which is cerainly no available in he nework nodes. The TDMA EMACS proocol also eases he (local) synchronizaion beween nodes. Time is divided ino so called frames and each frame is divided ino imeslos (see Figure 1). Each imeslo can be owned by only one nework node. This nework node decides wha communicaion should ake place in is imeslo. Oher nodes can ask for daa or noify he availabiliy of daa for he owner of he imeslo in he communicaion reques (CR) secion. The owner of he slo ransmis is schedule for is daa secion and broadcass a able in he raffic conrol (TC) secion, which ells o which oher TC secions he node is lisening. Afer he TC secion, he ransmission of he acual daa packe follows. Since ransmiing and receiving are boh very power consuming operaions, he nework nodes should urn off heir ransceivers as ofen as possible. The EMACS proocol suppors wo sleep modes of he nework nodes: 1) Sandby mode: This sleep mode is used when a a cerain ime no ransmissions are expeced. The node releases is slo and sars periodically lisening o a TC secion of a frame o keep up wih he nework. When he node has o ransmi some daa (even driven sensor node), i can jus fill up a CR secion of anoher nework node and agree on he daa ransmission, complee i and go back o sleep. 2) Dorman mode: This sleep mode is agreed on higher layers. The sensor node goes o low power mode for an agreed amoun of ime. Then i wakes, synchronizes (rediscovers he nework) and performs he communicaion. This sleep mode is especially useful o exploi he redundancy in he nework. No every node in he nework has o own a imeslo. I is clear ha a node does no own a imeslo, when i is in one of he sleep modes, since being in a sleep mode is inheren o no ransmiing a TC secion every frame. Bu even driven nodes migh also no redeem heir righ o own a imeslo. A draw back of no owning a imeslo is ha he node will only be able o receive mulicas messages and no messages direcly addressed o he node. Transmiing daa o nodes ha own a imeslo is no problem. Oher proocol layers in he nework may invoke lisening o/ransmiing in an a priory agreed (and free or no owned) daa secion. Before a node decides, ha i does no wan o own a imeslo, i should check ha sufficien TC secions are ransmied by neighbors o keep he nework conneced and o mainain synchronizaion. The fac ha nodes do no necessarily need o own a imeslo, eases he scalabiliy of he nework and reduces he power consumpion of he nodes. The proposed MAC proocol is designed o suppor he ypical communicaion paerns of a WSN. I minimizes he uilizaion of he ransceivers of he nodes o save power. Laency in he nework is reduced by allowing ransmissions in no owned or released daa secions. The raffic conrol secion can be deployed o make wake-up calls for sleeping nodes. III. CLUSTERBASED ROUTING When dealing wih large scale, ad-hoc sensor neworks, clusering offers some benefis. Grouping nodes ino clusers conrolled by a designaed node, he cluserhead, offers a good framework for he developmen of imporan feaures like rouing and coordinaed channel access for he EMACS proocol. In addiion, he hierarchical view of he nework achieved by clusering helps o decrease he complexiy of he underlying nework. A highly dynamic sensor nework appears more saic and hus some effecs of mobiliy can be miigaed. Clusering is hus one of he services useful in
a WSN and should be offered o upper layers such as he applicaions running on op of he neworking layer. A. Clusering Scheme In he clusering proocol, he se of cluserheads creaed and mainained by he proocol forms an independen se. The advanage of he independen se is ha no wo cluserheads are wihin direc ransmission range of one anoher, hus no inerfering direcly. Each node is o be wihin direc ransmission range of a leas one cluserhead, i.e. he cluserheads also form a dominaing se. The mainenance par of he proocol ensures his srucure in face of opological changes. Wih each node, here is a posiive weigh associaed. This weigh corresponds o a node s capabiliies o perform he addiional duies ha come wih being he conrolling insance of a cluser. The weigh is deermined aking ino accoun he residual energy of a node ogeher wih he conneciviy, i.e. node degree. The basic algorihm is derived from he greedy algorihm for he maximum weigh independen se [10]. This cenrally execued algorihm can also be performed equivalenly in a disribued fashion. In order for a node o decide on is saus o become cluserhead, only local informaion is needed. Basagni [1] gives a se of disribued procedures for he clusering proocol. On order o ge a conneced srucure, here are addiional gaeway nodes inroduced o enable communicaion beween he cluserheads of wo adjacen clusers. The assignmen of hese roles is impicily par of he clusering proocol, so ha each node is aware of is role in he clusering srucure, i.e. cluserhead, gaeway or regular node. Our iniial simulaion showed, however, ha he proocol has a severe drawback in he sizes of he creaed clusers. We herefore exend he general proocol by procedures o conrol he clusersizes [8]. Obviously, when he cluserheads conrol evenly sized clusers, he overall energy consumpion is disribued more evenly as well. In his se of procedures, gaeways are allowed o change heir cluser membership when he curren cluser grows oo large and an adjacen cluserhead is capable of conrolling more nodes. B. Cluserbased DSR The proocol o creae and mainain he clusered srucure of he nework comes a he cos of addiional conrol messages for he proocol, which consumes addiional energy. However, he clusered srucure of he nework can be used o limi he number of ransmissions of oher services. One of he basic services for an ad-hoc nework is muli-hop rouing. Using he clusered srucure obained from he above proocol, he roue discovery process of he dynamic source rouing (DSR, [6]) can be done more efficien han on a fla, non-clusered nework. As every node is in direc ransmission range of a leas one cluserhead, and each cluserhead has knowledge of is members, he roue discovery process ha reaches all cluserheads suffices o creae a feasible roue in he ad hoc nework. In order o only reach each cluserhead, nodes decide on no o rebroadcas a roue creaion message. These decisions can Fig. 2. Performance of Cluserbased Roue Creaion Fig. 3. Effecs of Cluser Size Bounds be done using only knowledge of he local neighborhood and he role in he clusered srucure. If a node has no deermined is role in he clusering scheme, e.g. when i has recenly been added o he nework, i will rebroadcas according o he rules given by he DSR. This abiliy o fall back o he DSR on a fla, non-clusered nework shows ha he proposed scheme does no represen backbone rouing. The algorihm does no rely on he clusered scheme. However, our simulaions show ha i is advanageous o use he scheme. C. Resuls As a resul, simulaions of he clusering proocol and he cluserbased rouing scheme on op of i showed ha in all ypes of scenarios (including varying mobiliy, node densiy and daa-raes) he amoun of raffic used for he creaion and mainenance of he proposed clusering scheme, ogeher wih he raffic needed for he roue creaion and daa-delivery, is less han using he fla rouing scheme of dynamic source rouing. In Figure 2, he number of needed conrol messages, including hose for seing up and mainaining he clusered srucure, o successfully send a daa packe from a randomly chosen source o a randomly chosen desinaion is given. The reference algorihm, DSR, on he same opology is also given for comparison. The simulaion scenario consiss of 50 nodes moving according o he random waypoin model wih fixed speed in a 800 by 500m recangle. The ransmission range of each node is 250m. In he nodes, a new messages is creaed every 500ms and simulaion is run for 200s on various differen opologies. In Figure 3, he effecs of differen bounds on he desired cluser sizes for he same seing as in Figure 2, a a fixed speed of 11m/s, is given. In almos all simulaions performed, he cluserbased rouing ouperforms he original DSR. For he procedures o conrol he cluser size, he approach aken
remains superior, and using reasonable bounds for he cluser size a srucure wih evenly spread ou cluserheads ha conrol evenly sized members can be obained. A reasonable bound for he cluser size is given by abou half he size of a node s neighborhood. The benefis of a clusered srucure can, in erms of energy consumpion, already be harvesed in he rouing process alone. However, a clusered srucure of he nework can help in oher services as well and can be assumed here neglecing he addiional coss of creaion and mainenance. Also, he amoun of raffic o be roued hrough a nework by muliple hops affecs he performance of he algorihms. More messages in he roue creaion process lead o more congesion in he nework. From his poin of view, he cluserbased approach lessens he nework congesions in wo ways: he flooding used for roue discovery is limied (i.e. absolue number of messages), and since par of he mainenance does no occur during he rouing process, he messages ha have o be broadcas are disribued more evenly in ime. The presened scheme can easily be adaped o oher algorihms ha rely on flooding. IV. MULTIPATH ROUTING The WSN opology is highly dynamic caused by frequen node mobiliy. As he nework diameer grows, daa generaed by one or more sources usually has o be roued hrough several inermediae nodes o reach he desinaion due o he limied range of each node s wireless ransmissions. Problems arise when inermediae nodes fail o forward he incoming messages. To preven his, usually, acknowledgemens and reransmissions are implemened o recover he los daa. However, his generaes large amoun of addiional raffic and delays in he nework. Wihou using hese schemes, reliabiliy of he sysem can also be increased by using mulipah rouing. Mulipah rouing allows he esablishmen of more han one pahs beween source and desinaion and provides an easy mechanism o increase he likelihood of reliable daa delivery by sending muliple copies of daa along differen pahs wihou acknowledgemen schemes. Several differen rouing algorihms for sensor neworks have been sudied unil now. The Temporally Ordered Rouing Algorihm (TORA) [9], Dynamic Source Rouing (DSR) [6] and Direced Diffusion [5] are only some of hem. All hese algorihms focus on reliable delivery of daa o desinaions. Bu hey are sensiive o a large number of communicaion failures and o high average speed of he nodes. To diminish he effecs of node failures (boh communicaion and hardware failures) mulipah rouing schemes have been developed based on hese algorihms [7][4]. They are a soluion agains failures, bu he amoun of conrol and daa raffic usually increases. A. Descripion of he Mulipah On-Demand Rouing In his secion we briefly describe a new mulipah rouing algorihm - Mulipah On-Demand Rouing (MDR). I is a Fig. 4. Daa spliing across muliple pahs viable soluion agains mobiliy and failures in wireless sensor neworks and in ad-hoc neworks in general. We have designed MDR wih he goal of providing several disjoin pahs beween daa sources and desinaions. I proved ha i is oleran o failures and more han ha, i is almos immune o opology changes due o mobiliy. High average speeds of he nodes produce negligible negaive effecs. MDR can be used in combinaion wih a daa spliing mehod. We have already shown in [3] ha daa spliing echniques are a way of reducing he high amoun of raffic associaed wih mulipah rouing. The main idea is o spli he original daa packe ha has o be roued in n subpackes (n is he number of pahs available from he source o he desinaion) in such a way ha only a k subpackes (k < n) are necessary o reconsruc he original daa packe. Even if some subpackes are los on he way, he desinaion can reconsruc he informaion (see Figure 4). MDR follows he main ideas behind he DSR algorihm. I is based on an iniial flooding of he nework wih he roue reques and hen generaes roue replies from he desinaion back o he source. There is no roue mainenance phase and he conrol messages have fixed lengh. The descripion of he wo phases is: Roue Reques Phase When he source wans o find a desinaion, i floods he nework wih a shor message announcing his. The message conains he source ID, he desinaion ID and he ID of he reques. Thus, he lengh of he message remains consan during he roue reques. Roue Reply Phase The desinaion will evenually receive one of he roue reques messages. I only knows ha here exiss a pah wihou knowing he inermediae nodes from he message. The desinaion reurns a roue reply o he neighbor from which i received he roue reques message. Each node repeas his process unil he message is received by he source. Afer receiving several differen roue replies, he source can sar sending he daa packe using hose. The main difference, when compared o DSR is moving he informaion sored inside he messages o he sensor nodes hemselves. The sensor nodes are responsible o remember where he roue reques message came from. B. Implemenaion and Resuls We have implemened he MDR algorihm in order o ge a beer undersanding of i. In Figure 5 we show a
Fig. 5. Comparison MDR - DSR comparison beween DSR and MDR. The firs observaion is ha he algorihms works for mobiliy raes a which he DSR algorihm gives an unaccepable amoun of errors. The rade-off lies in he amoun of conrol raffic. A closer look no jus a he number of messages, bu he message sizes, shows ha he oal amoun of MDR raffic compared o he amoun of DSR raffic varies beween 4.04:1 for low mobiliy raes o 1.02:1 for scenarios wih highly mobile nodes. Our sudy involved several parameers of he algorihm [2]. We have sudied he influence of mobiliy on he number of conrol messages, on he number of pahs discovered, on he daa laency, and on he number of cases he daa packe does no reach he desinaion. Oher parameers such as he period for which he source wais for roue replies and differen failure modes were also aken ino consideraion. The conclusion of our sudy is ha MDR improves he reliabiliy of daa rouing in a wireless mobile nework while mainaining he amoun of overhead raffic a a low value. Anoher imporan feaure of MDR is ha i is very robus agains he average speed of he nodes in he nework. Even for very high mobiliy raes, he algorihm succeeds in delivering he daa o he desinaion. By uning several of is inernal parameers, he algorihm can be adaped o various nework configuraions and degrees of mobiliy. The fuure work for mulipah rouing will focus mainly on improving MDR by modifying he Roue Reply phase o beer deal wih failures. The caching of roues will be aken ino consideraion as well, and he rade off beween he number of conrol messages and he reliabiliy has o be invesigaed. V. CONCLUSIONS Sensor neworks may be one of he bes examples in which he pervasiveness of energy efficien design crieria is desirable, due o he inheren resource limiaion, which makes energy he mos valuable resource. Sensor nodes should be able o esablish self-assembling neworks ha are incremenally exensible and dynamically adapable o mobiliy of sensor nodes, changes in ask and nework requiremens, device failure and degradaion of sensor nodes. Therefore, each sensor node mus be auonomous and capable of organising iself in he overall communiy of sensors o perform coordinaed aciviies wih global objecives. In order o achieve hese common goals, communicaion beween individual nodes and groups is mos imporan. We presened energy efficien soluions o an inegraed approach for wireless communicaion in a sensor nework. We have shown ha he developmen of energy efficien communicaion proocols requires a sysemwide view of he whole nework proocol sack. Fuure work will focus on he inegraion of hese proocols for medium access conrol, clusering and mulipah rouing in a nework of EYES-nodes. REFERENCES [1] S. Basagni. Finding a maximal independen se in wireless neworks. Telecommunicaion Sysems, 18(1/3):155 168, 2001. [2] S. Dulman and P. Havinga. Mulipah rouing for dynamic ad-hoc neworks. Inerne draf, submied o Personal Wireless Communicaions Conference, 2003. [3] S. Dulman, T. Nieberg, J. Wu, and P. Havinga. Trade-off beween raffic overhead and reliabiliy in mulipah rouing for wireless sensor neworks. In Proceedings of he Wireless Communicaions and Neworking Conference, 2003. [4] D. Ganesan, R. Govindan, S. Shenker, and D. Esrin. Highly-resilien, energy-efficien mulipah rouing in wireless sensor neworks. ACM SIGMOBILE Mobile Compuing and Communicaions Review, 5(4):11 25, 2001. [5] C. Inanagonwiwa, R. Govindan, and D. Esrin. Direced diffusion: A scalable and robus communicaion paradigm for sensor neworks. In Proc. Sixh Annual Inernaional Conference on Mobile Compuing and Neworks, 2000. [6] D. B. Johnson and D. A. Malz. Dynamic source rouing in ad hoc wireless neworks. In Imielinski and Korh, ediors, Mobile Compuing, volume 353. Kluwer Academic Publishers, 1996. [7] Nasipuri and S. Das. On-Demand Mulipah Rouing for Mobile Ad Hoc Neworks. In 8h Inl. Conference on Compuer Communicaions and Neworks (IC3N 99), 1999. [8] T. Nieberg, P. Havinga, and J. Hurink. On he advanages of cluserbased rouing in wireless sensor neworks. Inerne draf, submied o PWC2003, 2003. [9] V. D. Park and M. S. Corson. A highly adapive disribued rouing algorihm for mobile wireless neworks. In Proceedings of IEEE INFOCOM 97 Conf., April 1997. [10] S. Sakai, M. Togasaki, and K. Yamazaki. A noe on greedy algorihms for maximum weighed independen se problem.