A Prelmnary Study of Informaton ollecton n a Moble Sensor Network Yuemng Hu, Qng L ollege of Informaton South hna Agrcultural Unversty {ymhu@, lqng1004@stu.}scau.edu.cn Fangmng Lu, Gabrel Y. Keung, Bo L Dept omputer Scence and Engneerng Hong Kong Unversty of Scence and Technology {lfxad, gabrel, bl}@cse.ust.hk ABSTRAT Moble sensor networks are desrable n a varety of applcaton scenaros, n whch nformaton collecton s no doubt of great mportance. In ths paper, we present a moble sensor network archtecture consstng of a potentally large number of moble sensors and a sngle or multple statonary snk nodes for sensng nformaton collecton. We formulate a dstnct coverage measurement problem n term of sensng nformaton collecton; we study the relevant performance and examne the effect from a varety of relevant factors through extensve smulatons. We demonstrate that the performance s not only affected by the sensor moblty and the transmsson range between moble sensors and snk node(s), but also by the dstrbuton of moble sensors and the number and locatons of snk nodes. Based on the observaton and analyss, we also provde some prelmnary understandngs and mplcatons for mprovng the nformaton collecton performance. 1. INTRODUTION Wreless sensor networks (WSNs) have been wdely studed n recent years and are expected to be appled n a varety of applcaton scenaros such as battlefeld survellance and event detectons, hostle envronment montorng, and anmal behavor understandng. In typcal sensor networks, only statc sensors are used, n whch the performance of such systems such as feld coverage, hghly depends on the ntal deployment of sensors across a geographc area (called the regon of nterest). Gven uneven sensor dstrbutons, some regons often reman uncovered. In addton, ths can be explored by adversares once they gan knowledge about the deployment strategy and sensng characterstcs, uncovered path(s) can be found to render the statc sensor networks neffectual [1-2]. Recent advances n robotcs and low power embedded systems have made moble sensors possble [1-2], whch s beleved to be capable to construct moble sensor networks [3]. In such networks, sensors are mounted by robots, anmals or other movng obects, whch can sense and collect relevant nformaton. Moble sensors can report sensng nformaton to snk nodes wthn the coverage. The randomzed moblty s appealng for several reasons: 1) there s no pror knowledge of the regon of nterest assumed [3]; 2) t would be dffcult for an adversary or ntruder to reman undetected [2-3]; 3) perhaps more challengng n an unfrendly envronment, moble sensors may be not aware of the locatons of snk nodes beforehand. Among varous aspects of challenges posed by such moble sensor networks, the feld coverage (or called area coverage) by moble sensors has been studed [1-3]; the nformaton collecton (or called sensng data gatherng) from moble sensors, however, has not receved adequate attenton. Specfcally, how to capture the nformaton collecton of a moble sensor network? What factors can affect the nformaton collecton performance? What effect and senstvty from such factors? What s the mplcaton? We beleve t s mportant to understand the above questons n order to make a better use of moble sensor networks for dfferent applcaton scenaros. In ths paper, we study the nformaton collecton performance and examne the effect from relevant factors n a random walk moble sensor network. Our man contrbutons n ths study are: 1) We present a moble sensor network archtecture consstng of a potentally large number of moble sensors wth random walk moblty, and a sngle or multple statonary snk nodes collectng nformaton from the moble sensors; 2) we ntroduce and formulate a dstnct coverage measure (n terms of dstnct moble sensors used to be collected by snk nodes over a perod of tme) to capture the nformaton collecton performance; 3) we show through extensve smulaton that the nformaton collecton performance s not only affected by the sensor moblty and the transmsson range between the moble sensor and snk node, but also by the ntal dstrbuton of moble sensors, as well as the number and locatons of snk nodes. Further, we fnd that: 1) sensor moblty and the transmsson range between the moble sensor and snk node can be exploted to mprove nformaton collecton performance, whle they are constraned by the lmted moble speed and transmsson capablty of moble sensor; 2) n order to obtan certan level of nformaton collecton performance, more snk nodes can be deployed to compensate for the lmtaton of
sensor moblty and transmsson range; 3) for grd and random ntal dstrbutons of moble sensors, the placement of snk nodes should take nto account the area boundary. The rest of the paper s organzed as follows. Secton 2 revews the relevant work. Secton 3 presents basc system archtecture, moblty model and coverage n term of sensng nformaton collecton. Secton 4 provdes smulaton results and analyss. Secton 5 concludes the paper wth dscusson on future work. 2. RELATED WORK Recently, moble sensor networks and relevant ssues such as moblty strategy, feld coverage, and nformaton collecton, have receved ncreasng attenton. Lu et al. [3] studed the dynamc aspects of the feld coverage of a moble sensor network that depends on the contnuous movement of moble sensors. ompared to statc sensor networks, they showed that moble sensors followng a random walk can compensate for the lack of sensors and better feld coverage. A more recent study on how the qualty of feld coverage scales wth the moton velocty and strateges of moble sensors can be found n [2]. Wang et al. [1] proposed a hybrd network of statc sensors and moble sensors wth a random walk model and showed that a small set of moble sensors can effectvely address the uneven dstrbuton of the statc sensors so as to mprove the feld coverage qualty. A comprehensve dscusson on the moblty model ncludng random walk, random waypont, as well as Gauss-Markov can be found n [4]. The feld coverage descrbes how well a regon of nterest s montored by sensors; the coverage can have another nterpretaton from an nformaton collecton perspectve. Lma et al. [5] ntroduced the node coverage to descrbe the sensng data gatherng performance of a statc sensor network wth a sngle moble patrol node n terms of the expected number of sensors captured wthn a gven tme frame. Shah et al. [6] utlzed randomly movng Data Mules to help collect the sensng data. Kalpaks et al. [7] studed the problem of fndng an effcent manner to collect data from all the sensors and transmt data to the base staton, such that the system lfetme s maxmzed. There have been also studes on relable and powereffcent data transmsson and gatherng [8-9], on statc sensor network wth moble snks can be found n [10-11], n whch statc sensors send out data when the snk s movng around. The focus n ths paper s dfferent from all pror works n that we consder the nformaton collecton and relevant factors n a moble sensor network composed of potentally large number of moble sensors and statonary snk node(s). We are partcularly nterested n the key factors that affect the performance of nformaton collecton. 3. SYSTEM ARHITETURE In ths secton, we present the basc system archtecture, moblty model and coverage n term of sensng nformaton collecton. 3.1 The System Model We consder a moble sensor network consstng of a potentally large number of moble sensors (M). The moblty of sensor nodes follows a random walk model wthn a 2-D geographc area A to sense the envronment or detect events and store relevant nformaton. There exsts a sngle or multple statonary snk nodes collectng nformaton from the moble sensors. In order to capture the traectory of the moble sensor movement, the ntal locatons become relevant. Specfcally, we assume that, at tme t 0, the ntal dstrbuton D(t 0 ) of moble sensors across the area follows a certan pattern accordng to dfferent applcaton scenaros and requrements. In ths study, we consder two typcal ntal dstrbutons as follows: Grd dstrbuton: moble sensors are arranged usng a grd-based fashon [12] across the area, and the separaton between adacent sensors s A / M. The grd layout s a natural way for the cases n whch t s possble and preferable to place the sensors n partcular locatons at the begnnng. Random dstrbuton: moble sensors are randomly and ndependently dstrbuted n the area. Such an ntal deployment s sutable n scenaros where pror knowledge of the area s not avalable [3] or the area s not under control such as ardrop n an unfrendly area. Startng from one type of the ntal dstrbutons, we assume that each moble sensor performs the 2-D random walk movement, whch s one of the most common and wdely used moblty models [4]. Wth ths moblty model, each moble sensor travels from ts current locaton to a new locaton by randomly choosng a drecton θ [0, 2π) and a speed v [v mn, v max ] respectvely, n each dscrete tme nterval t. We further make one smplfcaton that all moble sensors move at a constant speed v = v max, so that the dstance traveled n each dscrete tme nterval can be denoted by r = v t. Ths s reasonable for the applcaton scenaro n that each moble sensor prefers to speed up ts msson progress (e.g., searchng a target n a vast area), and reports ts nformaton to the snk node as soon as possble; on the other hand, as ponted out n [3], more general speed dstrbutons can be approxmated usng the fxed speed scenaro. Dependng on dfferent types of mobles and applcaton context, sensors can have dfferent levels of speed represented by r. For example,
Fg. 1. A smple example for understandng the probablty of a moble sensor enterng the transmsson regon of a snk node. Table 1 summarzes the notatons used n ths paper. Symbols A M D(t 0 ) θ v t r Table 1. Notaton Defntons A vast 2-D geographc area called the regon of nterest (ROI) The number of moble sensors The ntal dstrbuton of moble sensors across the area A The movng drecton of moble sensors, whch s randomly chosen wthn [0, 2π) The speed of moble sensors, and n ths paper we assume a constant speed The dscrete tme nterval n whch each random walk movement occurs The travel dstance of moble sensors at each t, whch represents the speed. It depends on the moble platform of sensors and r << A N The number of snk nodes, and N 1 S R T The poston of statonary snk node, and S A, 1 N The transmsson range between the moble sensor and the snk node A perod of tme Fg. 2. Grd and random ntal dstrbuton. sensors can be mounted on robots or anmals. We show n secton 5 that the nformaton collecton performance s very relevant to ths factor. Besdes moble sensors, there exst a sngle or multple statonary snk nodes {S : S A, 1 N} for collectng nformaton. Let the transmsson range between a moble sensor and the snk node R,.e., each snk node s capable of communcatng wth or collectng nformaton from those moble sensors located wthn the dsk of radus R centered at the snk node. Hence, once a moble sensor enterng the transmsson regon of a snk node, we say t s covered. Although we do not explctly consder energy and storage constrans n ths paper, the lmted transmsson range R can be vewed as energy constran. For example, the commercally avalable sensors usng ZgBee standards [13-14] has the data transmsson capablty of 30m~90m n outdoor envronment. We wll show n secton 5 that dfferent values of R can lead to dfferent levels of nformaton collecton performance. Wth lmted transmsson range, the snk nodes mght not cover all the moble sensors at specfc tme nstants, but wth the random walk moblty, more dstnct moble sensors can enter the transmsson regon of snk nodes to be collected over a perod tme [t 0, t 0 +T]. 3.2 overage from an Informaton ollecton Perspectve We now defne the coverage measure from an nformaton collecton perspectve as follows. Defnton 1. overage of dstnct moble sensors over a perod of tme T (denoted by overage(t)): The total number of dstnct moble sensors that enter the transmsson regon of statonary snk nodes (.e., the nformaton of moble sensors s collected by the statonary snk nodes and the coverage count s ncreased) over a perod of tme [t 0, t 0 +T]. Ths coverage measure reflect the nformaton collecton n the sense that gven a requred tme perod, hgher coverage value mples better collecton; alternatvely, n order to collect certan amount of nformaton, the shorter tme used to meet the requrement, the hgher performance we could obtan. We next use a smple case of sngle snk node to llustrate the factors affectng the coverage performance. At the begnnng tme t 0, suppose there are M 0 ( 0) moble sensors already covered by the snk node due to the ntal dstrbuton D(t 0 ). Accordng to the Eucldean dstance from the poston of snk node, the remanng (M - M 0 ) moble sensors, whch are ntally outsdes the transmsson regon of the snk node, can be classfed to subsets of U M, where M denotes the set of moble
sensors that ntally have the same Eucldean dstance from the poston of snk node, and M M M = 0 For any moble sensor ntally belongng to M, the possble locaton of the sensor at tme t = t 0 + t can be characterzed by the followng normalzed probablty functon: k= 1 ( k 1 2 Where those crcle areas k (k 1) denote the area of possble locatons of the sensor as tme goes by, as llustrated n Fg. 1. Note that at tme t the farthest Eucldean dstance traveled by the sensor cannot exceed under the random walk moblty model. The probablty term P ( k, t ) s the probablty of the sensor lays wthn k at tme t, and the exact probablty densty functon s descrbed n [15]. Thus, the probablty of the moble sensor to be covered by the transmsson regon of the snk node wthn a perod of tme [t 0, t 0 +T] can be expressed as follows: Q( M, T ) P, t ) = P(, t ) + P(, t ) + L + P(, t ) = 1. (2) = P(, t ) L( S) + + = P(, t ) L( = + 1 P( + 1 S)., t ) L( + 1 S) + L Where the probablty terms L( k S) 0 s related to the cross secton area between the possble locatons of moble sensor and the transmsson regon of snk node, as shown n Fg 1. Due to the symmetry, all the moble sensors that ntally belongng to the same M would have equal probablty to be covered by the transmsson regon of the snk node. Therefore, the expectaton of coverage over a perod of tme [t 0, t 0 +T] can be expressed as follows: E [ overage ( T )] = M Q( M, T ). The above smple dervaton reveals that the coverage performance s related to several factors. For example, the M are relevant to the number and ntal dstrbuton status of moble sensors, as well as the poston of snk node and the transmsson range between moble sensor and snk node; the possble locatons of moble sensor denoted by the k and so as the probablty terms Q( M, T ) are relevant to the random walk movement of moble sensors and the transmsson range between moble sensor and snk node, as well as the length of the tme perod. (1) (3) (4) 4. SIMULATION AND ANALYSIS In ths secton, we frst descrbe the smulaton wth relevant settngs and then carry out a seres of experments to nvestgate the effect and senstvty from varous factors. 4.1 Smulaton Settng We develop a smulator that captures the essental aspects of the network and moblty model descrbed n secton 3. Startng from a specfc ntal dstrbuton of moble sensors, the smulator contnuously calculates the coverage measure along wth dstnct moble sensors enterng the transmsson regon of snk nodes over a perod of tme (T). Specfcally, f a moble sensor reaches the area boundary, t bounces off the area border accordng to the ncomng drecton [4]. Our smulator provdes the flexblty of selectvely controllng the confguraton of varous parameters ncludng: 1) the length and wdth of the area (l*w= A ); 2) the number of moble sensors (M); 3) dfferent types of ntal dstrbuton, e.g., grd and random dstrbutons; 4) the speed of the moble sensor (r); 5) the transmsson range between moble sensor and snk node (R); 6) the number (N=1 or N>1) and postons of snk nodes; 7) the length of the tme perod (T). Unless otherwse specfed, we use the followng default settngs: for grd dstrbuton, we defne the ntal separaton between adacent sensors to be 10 unts resultng n 20301 moble sensors evenly dstrbuted n an area of sze 2000 1000. For comparson, the same number of moble sensors s used n random dstrbuton scenaro. The results are averaged over multple runs for each correspondng set of parameter confguraton. 4.2 Sngle Snk We frst consder a sngle snk scenaro wth grd and random ntal dstrbutons of moble sensors as shown n Fg. 2. We study the nformaton collecton performance by varyng two mportant parameters: 1) the speed of moble sensors (r) and 2) the transmsson range between moble sensor and snk node (R). Fg. 3 plots the percentage of moble sensors covered aganst tme by varyng the transmsson range between moble sensor and snk node (R), begnnng from grd ntal dstrbuton of moble sensors. The fgure shows fve dstnct ncreasng curves wth dfferent growth speed separated by transmsson ranges from R = 20 to 400 respectvely 1. The result demonstrates that the ncrease of the transmsson range can suffcently mprove the nformaton collecton performance wthn a tme perod. 1 We choose the rato between r, R and A by consderng ther magntudes n realstc stuatons [13-14].
Fg. 3. Percentage of moble sensors covered aganst tme by varyng the transmsson range between moble sensor and snk (R), begnnng from grd dstrbuton. Fg. 4 plots the percentage of moble sensors covered aganst tme by varyng the speed of moble sensors (r), begnnng from grd ntal dstrbuton of moble sensors. As we ncrease the parameter from r = 2 to 20 respectvely 1, the coverage curve rses sharply wth hgh senstvty, whch means sensor moblty can sgnfcantly affect the nformaton collecton performance. Ths ndcates that we can explot the speed of moble sensors to mprove the nformaton collecton performance. ompared wth Fg. 3, the parameter r has hgher senstvty than the parameter R. Under a random ntal dstrbuton, we obtan smlar conclusons that the ncreases of the values of R and r both result n ncreasng the coverage percentage wthn a fxed amount of tme perod, and r s more senstve than R. Fg. 5 shows consstent results wth that n Fg. 3, confrmng that the ncrease of transmsson range (R) can mprove the nformaton collecton performance,.e., reduce the tme to reach certan level of coverage percentage. Specfcally, under a fxed confguraton of r and R, the tme to acheve 75% coverage percentage s twce of the tme to acheve 50%; t would take longer tme to reach hgher coverage percentage. Lkewse, Fg. 6 shows consstent results wth that n Fg. 4, confrmng that ncreasng the speed of moble sensors (r) can mprove the nformaton collecton performance. In summary, for both of the grd and random dstrbutons, ncreasng the speed of moble sensors (r) and the transmsson range between moble sensor and snk node (R) can help to mprove the nformaton collecton performance. However, snce r and R represent the moble platform speed and transmsson capablty of moble sensor respectvely, they are both constraned n realstc applcatons. For example, the commercally avalable sensors usng ZgBee standards [13-14] has a lmted data transmsson capablty of 30m~90m n outdoor envronment, and the Bluetooth standards [14] has even Fg. 4. Percentage of moble sensors covered aganst tme by varyng the speed of moble sensors (r), begnnng from grd dstrbuton. shorter transmsson range; on the other hand, the speed of moble sensor depends on the speed lmtaton of dfferent types of moble platform, such as robots or dfferent knds of anmals. 4.3 Multple Snks In ths subsecton, we consder the multple snk nodes scenaro wth grd and random ntal dstrbutons of moble sensors. Specfcally, we use the followng snk nodes placement: (a) two snk nodes wth lne placement,.e., at coordnates (500, 500) and (1500, 500); (b) two snk nodes wth dagonal placement,.e., at (500, 250), (1500, 750); (c) four snk nodes wth square placement at (500, 250), (500, 750), (1500, 250), and (1500, 750), respectvely. For grd ntal dstrbuton, Fgs. 7 and 8 plot the percentage of moble sensors covered aganst transmsson range (R) and the speed of moble sensor (r) respectvely, at a partcular tme (t=20000) when the percentage gap between sngle and multple snk nodes clearly shown. Frst, smlar to sngle snk node scenaro, the nformaton collecton performance wth multple snk nodes can be mproved wth the ncrease of R and r. Second, ncreasng the number of snk nodes (N) can mprove the nformaton collecton performance, whch ndcates that f r or R s constraned by a specfc value due to physcal lmtaton or applcaton requrement. In order to acheve certan level of nformaton collecton performance, alternatvely we can utlze more snk nodes to compensate for the lmtaton of r or R. Thrd, the placement of snk nodes can greatly affect the nformaton collecton performance. For example, the coverage curve of two snk nodes wth lne placement outperforms the two snk nodes wth dagonal placement. Ths s because the latter s relatvely closer to the area boundary; ths leads to unbalanced dstrbuton of moble
Fg. 5. Tme to obtan 50% and 75% coverage aganst transmsson range between moble sensor and snk (R), begnnng from grd and random dstrbuton. Fg. 6. Tme to obtan 50% and 75% coverage aganst the speed of moble sensors (r), begnnng from grd and random dstrbuton. Fg. 7. Percentage of moble sensors covered aganst transmsson range (R), at a partcular tme (t=20000). sensors around the snk and cause more sensors to travel longer dstance to be covered by the snk (.e., less probablty of beng covered wthn a gven perod of tme under random walk moblty model). Smlar results are observed under random dstrbuton. 5. ONLUSION AND FUTURE WORK In ths paper, we present a moble sensor network archtecture composed of a potentally large number of moble sensors wth random walk moblty, and a sngle or multple statonary snk nodes collectng nformaton from the moble sensors. To descrbe the nformaton collecton performance of such a moble sensor network, we ntroduced and formulated a dstnct coverage measure n term of sensng nformaton collecton. We demonstrated through extensve smulaton that the nformaton collecton performance s not only affected by the sensor moblty and the transmsson range between the moble Fg. 8. Percentage of moble sensors covered aganst the speed of moble sensors (r), at a partcular tme (t=20000). sensor and snk node, but also by the ntal dstrbuton of moble sensors, as well as the number and locatons of snk nodes. There are several avenues for further studes: 1) to consder dfferent moblty models such as the one [4]; 2) to study the relatonshp between the feld coverage and the coverage n term of sensng nformaton collecton; 3) snk node placement strategy for certan sensor dstrbuton and moblty; 4) consder other ntal sensor dstrbutons, more realstc area shapes and stuatons (e.g., wth obstructons). 6. REFERENES [1] D. Wang, J. Lu, and Q. Zhang, Probablstc feld coverage usng a hybrd network of statc and moble sensors, n Proc. of the IWQoS'07, hcago, IL, USA, 20-22 June, 2007. [2] N. Bsnk, A. Abouzed, and B. Isler, Stochastc event capture usng moble sensors subect to a qualty metrc, n
Proc. of the AM MOBIOM 06, Los Angeles, September, 2006. [3] B. Lu, P. Brass, O. Dousse, P. Nan, and D. Towsley, Moblty mproves coverage of sensor networks, n Proc. of the AM MOBIHO 05, Urbana hampagn, IL, May, 2005. [4] T. amp, J. Boleng, and V. Daves, "A Survey of Moblty Models for Ad Hoc Network Research," n Wreless ommuncaton & Moble omputng (WM): Specal Issue on Moble Ad Hoc Networkng Research, Trends and Applcatons, vol. 2, no. 5, pp. 483-502, September, 2002. [5] L. Lma and J. Barros, Random walks on sensor networks, n Proc. of the 5th Internatonal Syposum on Modelng and Optmzaton n Moble, Ad hoc, and Wreless Networks (WOpt 2007), Lmassol, yprus, Aprl, 2007. [6] R. Shah, S. Roy, S. Jan, and W. Brunette, Data mules: Modelng a three-ter archtecture for sparse sensor networks, n Proc. of the IEEE Workshop on Sensor Network Protocols and Applcatons (SNPA), May, 2003. [7] K. Kalpaks, K. Dasgupta, and P. Namosh, Effcent algorthms for maxmum lfetme data gatherng and aggregaton n wreless sensor networks, n omputer. Networks, vol. 42, no. 6, pp. 697 716, August, 2003. [8] S. Lndsey and. S. Raghavendra, Pegass: Power-effcent gatherng n sensor nformaton system, n Proc. of the IEEE Aerospace onference, March, 2002. [9] H. O. Tan and I. Korpeoglu, Power effcent data gatherng and aggregaton n wreless sensor networks, n SIGMOD Rec., vol. 32, no. 4, pp. 66 71, December, 2003. [10] A. Kansal, A. Somasundara, D. Jea, M. Srvastava, and D. Estrn, Intellgent flud nfrastructure for embedded networks, n Proc. of the 2nd nternatonal conference on Moble systems, applcatons, and servces (MobSYS), June, 2004. [11] W. Zhao, M. Ammar, and E. Zegura, A message ferryng approach for data delvery n sparse moble ad hoc networks, n Proc. of the 5th AM nternatonal symposum on Moble ad hoc networkng and computng (MobHoc), AM Press, pp. 187 198, May, 2004. [12] S. Shakkotta, R. Srkant, and N. Shroff, Unrelable sensor grds: overage, connectvty and dameter, n Proc. of the 22nd Annual Jont onference of the IEEE omputer and ommuncatons Socetes (INFOOM 03), Aprl, 2003. [13] http://www.rabbt.com/products/zgbee_app_kt/zgbee.pdf [14] http://en.wkpeda.org/wk/sensor_node [15] W. Stade, The exact probablty dstrbuton of a twodmensonal random walk, n Journal of Statstcal Physcs, vol. 46, no. 1-2, pp. 207-216, January, 1987.