5 The Intenational Aab Jounal of Infomation Technology, Vol. 6, o. 5, ovembe 9 Pefomance Optimization in Stuctued Wieless Senso etwoks Amine Moussa and Hoda Maalouf Compute Science Depatment, ote Dame Univesity, Lebanon Abstact: This pape pesents a pefomance-optimization technique in a stuctued wieless senso netwok that has a cicula fom and consists of seveal ings with many clustes. Optimization is done by vaying the numbe of ings and by applying aggegation on the data tansfeed acoss the netwok up to the destination node. The main eason fo this study is that the enegy-limited senso nodes may need to opeate in time-constained envionments, in which both enegy and time must be conseved. In this pape, we analyze the impact of ing density and data aggegation on enegy consumption and tansfe time. We also find an optimum stuctue consisting of a cetain numbe of ings, wheeby the senso netwok opeates with low consumption of enegy and little tansfe time. Keywods: Wieless senso netwok, pefomance optimization, cicula netwok model, data aggegation, enegy consumption Received Decembe 8, 8; accepted June, 9. Intoduction Wieless Senso etwoks (WSs) have emeged as a new infomation-gatheing pototype based on the collaboative effot of a lage numbe of sensos that could be used in many commecial and militay applications. Because of the equiement of unattended opeation in emote o even hostile locations, it is impeative to make good use of enegy though appopiate optimization of communication and opeation management. Hence, the main dawback in such netwoks is enegy consumption []. The sensos battey powe is limited and should be managed to keep the netwok alive thoughout the sensing pocess, which is bound by a cetain time limit. This is the time needed to pefom sensing, collect the sensed data, and fowad the data though the netwok up to the sink. Enegy consumption is affected by the tansmission time and the distance between souce nodes and destination [6]. To minimize the distance between tansceives, many nodes need to be deployed. On the othe hand minimizing the time equies less data to be deliveed, which could be achieved though aggegation; that is, educing the numbe of tansmitted packets by emoving the edundancy with in-netwok pocessing. A WS can be divided into levels, accoding to the distance sepaating each goup of sensos fom the sink. Since aggegation is pefomed on a level basis, it would be moe effective when the senso density is highe at each level. The total tansfe time theefoe deceases. Although by deceasing the tansmission time we will decease enegy consumption, having fewe levels implies longe distances between communicating sensos and highe enegy consumption. In this pape, we analyze diffeent pefomance paametes such as enegy consumption and tansfe time in a stuctued WS whee the sensos ae placed in a well defined stuctue as in industial envionments and petoleum platfoms. In paticula, this pape aims to achieve a time-effective connectivity while educing enegy consumption, by finding a elation among netwok clusteing, enegy management and time limitations. Clusteing is used to divide the netwok into goups, each of which has a single Cluste Head (CH) []. We assume that each CH is located at the cente of its cluste, etieves sensed data fom neighboing sensos (of the same cluste) and fowads it into the next CH, close to the base station. This pocess goes on until the data finally eaches its destination. To minimize the size of data tansmitted among CHs, we will use data aggegation which eliminates edundancy and deceases ovehead, eventually achieving shote delays. This pape is divided into six main sections. Section descibes the taffic model of the analyzed WS. Section intoduces the enegy model used. While section pesents some mathematical analysis of the netwok pefomance paametes, section 5 discusses the numeical esults. Finally, section 6 summaizes the pape and suggests some potential futue wok.. etwok Model The WS that we ae analyzing consists of a cicula sensing aea of adius R s. It is divided into a vaiable numbe of ings, esulting fom the fomation of concentic cicles aound the cente whee the sink lies
Pefomance Optimization in Stuctued Wieless Senso etwoks 5 []. Rings ae of adii, k whee < < < k =R s. Clustes of senso nodes ae fomed in these ings. The clustes ae located next to each othe with a size that fits exactly in the ing and that doesn t ovelap with neighboing clustes. Each cluste has one CH which communicates with the CH of the neighboing ing. The diffeence in adii between two neighboing concentic cicles, called the thickness of a ing, is constant to all ings. To espect popotionality, we set the coveage aea of the sink to be equal to twice the coveage aea of any CH. This can be justified by the fact that the sink node could be main-poweed and hence does not have any enegy consumption constaint. As the numbe of ings inceases, so does the numbe of clustes. Consideing communication issues, each CH in ing i, is assigned a coesponding neighbo, in ing i-, based on shotest distance. CHs in ing i outnumbe those in ing i-, and so some CHs can communicate with one and only one CH, while othes can communicate with seveal CHs, o even shae a CH between them. To solve this poblem we use the shotest distance elay between communicating CHs. We assign an ID numbe to each CH, stating fom ID numbe such as CH on the x axis. Once we obtain the Catesian coodinates of each CH, we save them in a matix fo the pupose of calculating the distances between CHs. We then identify CHs that ae closest to each othe (Table ) and connect them as in Figue. Table. Shotest distances sepaating the CHs. (CHID, Distance) Paent CHID (,75);(,75); ;(9,75) (Sink) (,5) (,5);(,5) the sink. We notice that the total numbe of clustes fomed in i ings is equal to c c = i (I + ) () We fist stat by analyzing the numbe of ings in ou model. When the numbe of ings is, the distance sepaating two collinea CHs is R s +, whee is the numbe of ings in a given scenaio. This value is less than the maximum sensing ange of the used nodes (75 m). Yet, the maximum ange is still less than the distance sepaating the CHs on the fist ing and the sink, i.e.,.5 R s +. Since this distance is lage than 75 m, the communication between the sensos lying on the fist ing and the sink is impossible. Futhe investigations wee done when the numbe of ings. In these cases we find that the distances between CHs, as well as the distance between CHs on the fist ing and the sink ae all less than the maximum tansmission ange. In scenaios whee collision occus, we use the following coodination scheme. Conside Figue in which we have CH A at ing n within a distance d < max fom CHs B and all thee neighboing CHs at ing n-. odes at ing n- can eceive data fom one souce at a time; hence while B tansmits its data, CHs A and C should be in sleep mode, while the othe CHs, such as D and E, can tansmit concuently with B since they do not collide. Also when a CH finishes its tansmission, it puts itself and all its cluste nodes to sleep to avoid collision and save enegy. 5 5-5 - -5 5 6 7 7 8 8 6 9 9 5 6 5 7 - - -5 - -5 5 5 8 Figue. A netwok model of thee ings with connected CHs. In this pape, we assume that the total numbe of nodes n is distibuted ove the clustes in the WS, and that the nodes ae scatteed evenly in the ings aound the sink. We assume that the maximum tansmission ange max of a given senso node is 75 m. We also assume that the aea immediately aound the sink does not contain any nodes and is coveed by 9 9 8 7 6 5 5 Figue. Coodination between neighboing sensos.. etwok Model The enegy consumed by a senso node depends on the amount of data being tansfeed and the distance that this data tavels befoe it eaches its destination. Theefoe, the amount of enegy equied to tansmit an l-bit packet ove a distance x is given by k E = l ( e+ ux ) () t whee e is the amount of enegy spent in the adio electonics cicuity and ux k is the amount of enegy spent in the adio amplifies to counte the popagation loss. We shall assume l is equal to 6 bits, e is equal to 5 nj/bit and u equal to 5 pj/bit/m. When eceiving a packet, only the eceive cicuity is invoked, and so the enegy spent on eception is given by
5 The Intenational Aab Jounal of Infomation Technology, Vol. 6, o. 5, ovembe 9 E ec = le () and finally, the amount of enegy consumed when elaying a packet ove a distance x is given by: k E = E + E = l (e+ ux ) () elay t ec Applying this enegy model to ou cicula senso netwok, we notice the following esults: The CHs in the outemost ing ae leaf nodes and hence consume enegy only when tansmitting packets to the next level. The CHs in all the inne ings consume enegy while tansmitting thei packets and while elaying those packets coming fom oute CHs. The sink consumes enegy on eceiving packets coming fom all CHs. The inta-cluste enegy consumption is the enegy consumed by the cluste nodes to tansfe thei data to the CH. Since distance is a majo issue in enegy calculation, we assume that the cluste nodes ae located at an aveage distance fom the CH that is, half the distance between the CH and the peimete edge of the cluste. We should note that the amount of enegy spent on sensing is not consideed in this pape, as its value is constant and does not affect the esults. We also assume that each senso geneates one packet and sends it to its CH.. Tansfe Time Calculation We define the tansfe time TT as the time taken to pass data fom one extemity of the netwok to the sink. We calculate the aveage tansfe time among all the fomed tees in evey simulation. Accoding to [], a packet consists of 8% ovehead and 5% payload. Aggegation is used at intemediate CHs to aggegate data geneated at a given cluste. We assume that when packets ae fowaded towads the sink, the ovehead is dopped and only the payload is added to the next packet. Also, % ( bits) of the oiginal packet size is eseved to some identification. Hence, if n packets wee fowaded up to thei paent, the esulting packet would incease by.65 n l whee l is the oiginal packet size. Each CH then sends its aggegated packets with an ovehead to the next destination on the tee. The eceiving CH pefoms the same aggegation technique by emoving any edundant oveheads. Figue. A tee fomed among fou ings. Fo the calculation of the inte-cluste TT, we conside the example given in Figue. Let nb max n be the maximum numbe of nodes pe banch at level n and let nbn be the total numbe of nodes at level n. Also let t be the tansmission time of an oiginal packet. The tansfe time between consecutive levels n+ and n is: tt n =.56t nb + nb t. (5) i= n+ i max n Fo example, the tansfe time fom level to is: tt =.65 tt+ nbmax t =.65 t + t =. 95t. Also, the inta-cluste TT is the time taken by the CH to eceive all packets fom its nodes. To avoid collision, the nodes send thei data in tun. Hence this TT becomes the sum of all tansfe times of all nodes in a cluste. 5. umeical Results In this pape, we have used MATLAB to evaluate the effect of clusteing on enegy consumption and Tansfe Time (TT) in stuctued WSs. Tees of shotest paths between nodes and the sink ae fomed. In these simulations, we have consideed that the total numbe of nodes is, the channel bit ate is Kbps [7], and the adius of the WS is m. We also assumed that CSMA/CA is used between sensos to avoid collision. Fist, we have calculated the total amount of enegy consumed as a function of the numbe of ings, whee. The simulation did not conside the enegy consumed by the sink, as it is supposed to un on electicity. We can notice fom Figue (a) that when using aggegation, the minimum amount of enegy is consumed when = 5. The total amount of enegy consumed follows a deceasing shape when is small ( to 5). Hence, clustes that ae fomed in these ings have the optimum consumption elative to thei distance fom thei child nodes. On the othe hand, as the numbe of ings inceases that is, when > 5, the cuve is found to have an inceasing shape. This is due to the pedominance of enegy consumption ove distance minimization. The amount of enegy consumed due to the fomation of numeous CHs which tansmit and elay data is much geate than the amount of enegy saved due to minimizing the tansmission distance. We have also un this simulation without applying aggegation on the tansfeed data. Since the aggegation facto is constant, the esulting paagaph (b) would be simila to paagaph (a), except fo an incease in the enegy consumption. Consideing the minimum values in both paagaphs lying at ing 5, we have found that data aggegation deceases enegy consumption by.%. Second, we have calculated the aveage time it takes the nodes on the outemost
Pefomance Optimization in Stuctued Wieless Senso etwoks 55 ings to tansmit thei data to the sink. The TT at the fist ing is always equal to the sum of all TTs of the CHs at this ing. This is because the CHs at the fist ing need to tansmit thei data consecutively, since the sink can eceive only one signal at a given time. We can notice cuve (a) in Figue 5 has an inceasing shape between ings 5 and, affected diectly by the aggegation facto. Typically, the aggegation facto deceases as the numbe of ings inceases, since clustes diminish in size, eventually pefoming fewe aggegations on the tansfeed data. Howeve, the gaph shows that when the numbe of ings is fewe than 5, the aveage tansfe time TT deceased as the numbe of ings inceased. This is because TT is dominated by the coodination behavio occuing at the fist ing. Also, in compaison with othe ings, this ing tansfes the maximum amount of data and hence its latency affects geatly the value of TT as shown in Figues 6 and 7. Figue. Impact of aggegation on enegy consumption. Ta n sfe Tim e (s) Figue 5. Impact of aggegation on tansfe time..5.5.5.5.6.7..7.9 umbe of ings a b c. Figue 6. Coodination effect when numbe of ings < 5. Tansfe Time (s).5.5.5.5....5.6.9 8 9 umbe of ings a b c Figue 7. Coodination effect when numbe of ings > 5. Geneally speaking, we have TT = c i= tt i + j= τ (6) c being the numbe of CHs at ing andτ = tt. By consideing in Figues 6 and 7, a j = i= = c tt i, b τ and j c= TT, we can notice that fo 5 the tem b is pedominant in the calculation of TT, while fo > 5 the tem a becomes the pedominant facto. To investigate the impact of aggegation on TT we conside paagaph (b) in Figue 5. This paagaph diffes fom paagaph (a) in the minimum value. In paagaph (a), the minimum value lies at ing 5, while in paagaph (b) the minimum value lies at ing. This is basically due to the aggegation facto and the loadbalancing behavio distibuted acoss the ings. By compaing both paagaph, we have found that data aggegation deceases the aveage tansfe time by.%. Thid, the wake-up time calculation is pefomed and is defined as the time one ing stays opeational until it completes the eception and tansmission pocesses. Hence, the wake-up time (T O ) of CHs at a given ing n is ( TO) = tt n n+ + ttn fo n< ( TO) = tt fo n n = j= (7) We have calculated in the following chats the wake up times fo a WS with and without aggegation. Obviously T O is lage when we have no aggegation in the WS, leading to highe enegy consumption. Wake-up Times (s) 5..6.78.8. with aggegation Ring numbe without aggegation Figue 8. Compaison of wake-up times using thee ings..8
56 The Intenational Aab Jounal of Infomation Technology, Vol. 6, o. 5, ovembe 9 Wake-up Times (s) 5..6.6.85.5.67.7... 5 with aggegation Ring numbe without aggegation Figue 9. Compaison of wake-up times using five ings. Finally, the optimum ing configuation is defined as the ing o ange of ings whee the enegy consumption and the tansfe time ae both minimal. This ensues an appopiate netwok pefomance that does not consume excessive enegy eventually dying out the nodes batteies, no does it waste time on data tansfe, leading to big netwok latency. In ode to detemine the optimum configuation we have supeimposed paagaphs of nomalized enegy consumption and tansfe time with simila aggegation popeties. Both paagaphs in Figue. Have a minimum lying on ing 5. Hence, when ou netwok is set to opeate with five ings, and with data aggegation enabled, the enegy consumed dops to a minimum, along with an aveage tansfe time that also eaches a minimum. Howeve, when data aggegation was disabled, Figue shows that the optimum aea lies between two ings namely, ings and 5. Since the enegy consumption gaph has a steepe incease in values, we chose to be equal to 5 since it gives a minimal value fo the enegy consumption and an acceptable low value fo the tansfe time. Figue. Optimum configuation with aggegation effect. 6. Conclusion In this pape, we have poposed an optimum configuation fo a cicula stuctued WS in tems of enegy consumption and tansfe time. We have suggested a clusteing configuation that impoves data collection, tansfe time and enegy consumption. We have built outing tees that use aggegation techniques to tansmit data packets. We have achieved ou goals by using an aggegation function to educe the packet size tansmitted fom one level to anothe and by vaying the distance between CHs to educe the enegy consumed fo tansmission. We have consideed diffeent scenaios based on diffeent ings and cluste sizes. We found that ou model is adjustable since by changing the numbe of ings in the WS we can change the aggegation pocesses in the netwok. This in tun will affect the total enegy consumption and the aveage tansfe time. Futue wok could involve investigating the impact of andom sensing on ou system, whee only andomly selected nodes pefom the equied sensing. Refeences [] Feentinos K., Tsiligiidis T., and Avanitis K., Enegy Optimization of Wieless Senso etwoks fo Envionmental Measuements, in Poceedings of the 5 IEEE Intenational Confeence on Computational Intelligence fo Measuement Systems and Applications, Italy, pp. 5-55, 5. [] Mhate V. and Rosenbeg C., Design Guidelines fo Wieless Senso etwoks: Communication, Clusteing and Aggegation, Compute Jounal of Ad Hoc etwoks, vol., no., pp. 5-6,. [] Saab S. and Maalouf H., Data MULE Aggegation Potocol fo Wieless Senso etwoks, in Poceedings of the Cuent Tends in the Theoy and Applications of Compute Science CTTACS7, Beiut, pp. 7-8, 7. [] Shi J., Zhong X., and Chen S., Study on Communication Mode of Wieless Senso etwoks Based on Effective Result, Jounal of Physics: Confeence Seies, vol. 8, no., pp. 7-, 6. [5] Shu T., Kunz M., and Vudhula S., Powe Balanced Coveage-Time Optimization fo Clusteed Wieless Senso etwoks, in Poceedings of the 6 th ACM intenational symposium on Mobile ad Hoc etwoking and Computing, USA, pp. -, 5. [6] Wistom., Optimization of Wieless Senso etwoks using Machine Leaning, MSc Thesis, Sweden, 6. Figue. Optimum configuation without aggegation effect.
Pefomance Optimization in Stuctued Wieless Senso etwoks 57 [7] ZigBee A., Woking with Wieless Contol, http://www.zigbee.og,5. Amine Moussa eceived his BSc in compute science, in, and the MSc in compute infomation systems in 7, both fom ote Dame Univesity, Lebanon. Hoda Maalouf eceived the diploma of engineeing in electical engineeing fom Ecole Supeieue d Ingenieus de Beyouth, Lebanon, in 987, the MSc degee in communications and signal pocessings in 99, and the PhD degee in electical engineeing in 998, fom Impeial College of Science, Technology and Medicine, London, UK.
58 The Intenational Aab Jounal of Infomation Technology, Vol. 6, o. 5, ovembe 9