Fast and Efficient Data Forwarding Scheme for Tracking Mobile Targets in Sensor Networks

Transcription

1 Artcle Fast and Effcent Data Forwardng Scheme for Trackng Moble Targets n Sensor etworks M Zhou 1, Mng Zhao, Anfeng Lu 1, *, Mng Ma 3, Tang Wang 4 and Changqn Huang 5 1 School of Informaton Scence and Engneerng, Central South Unversty, Changsha , Chna; mzhou@csu.edu.cn School of Software, Central South Unversty, Changsha , Chna; meanzhao@csu.edu.cn 3 Department of Computer Scence, Stony Brook Unversty, Stony Brook, Y 11794, USA; mnma@cs.stonybrook.edu 4 Department of Computer Scence and Technology, Huaqao Unversty, Quanzhou Chna; wangtan@hqu.edu.cn 5 School of Informaton Technology n Educaton, South Chna ormal Unversty, Guangzhou , Chna; cqhuang@zju.edu.cn * Correspondence: afenglu@mal.csu.edu.cn; Tel.: Receved: 3 October 017; Accepted: 6 ovember 017; Publshed: 9 ovember 017 Abstract: Transferrng emergent target trackng data to snks s a major challenge n the Industral Internet of Thngs (IIoT), because neffcent data transmsson can cause sgnfcant personnel and property loss. For trackng a constantly movng moble target, sensng data should be delvered to snks contnuously and quckly. Although there s some related research, the end to end trackng delay s stll unsatsfyng. In ths paper, we propose a Fast and Effcent Data Forwardng (FEDF) scheme for trackng moble targets n sensor networks to reduce trackng delay and mantan a long lfetme. Innovatons of the FEDF scheme that dffer from tradtonal scheme are as follows: frstly, we propose a scheme to transmt sensng data through a Quckly Reacted Routng (QRR) path whch can reduce delay effcently. Duty cycles of most nodes on a QRR path are set to 1, so that sleep delay of most nodes turn 0. In ths way, end to end delay can be reduced sgnfcantly. Secondly, we propose a perfect method to buld QRR path and optmze t, whch can make QRR path work more effcently. Target sensng data routng scheme n ths paper belongs to a knd of tral-based routng scheme, so as the target moves, the routng path becomes ncreasngly long, reducng the workng effcency. We propose a QRR path optmzaton algorthm, n whch the rato of the routng path length to the optmal path s mantaned at a smaller constant n the worst case. Thrdly, t has a long lfetme. In FEDF scheme duty cycles of nodes near snk n a QRR path are the same as that n tradtonal scheme, but duty cycles of nodes n an energy-rch area are 1. Therefore, not only s the rest energy of network fully made use of, but also the network lfetme stays relatvely long. Fnally, comprehensve performance analyss shows that the FEDF scheme can realze an optmal end to end delay and energy utlzaton at the same tme, reduce end to end delay by 87.4%, mprove network energy utlzaton by.65%, and ensure that network lfetme s not less than prevous research. Keywords: Industral Internet of Thngs; trackng moble target; quckly reacted routng; delay; lfetme 1. Introducton Industral Internet of Thngs (IIoT) [1 5] as well as cloud computng [6 10] leverage the ubquty of sensor-equpped devces such as smart portable devces, and smart sensor nodes to collect nformaton at a low cost, provdng a new paradgm for solvng the complex sensng applcatons from the sgnfcant demands of crtcal nfrastructure such as survellance systems [11 15], remote patent care systems n healthcare [4,16,17], ntellgent traffc management [18 0], automated Symmetry 017, 9, 69; do: /sym

2 Symmetry 017, 9, 69 of 31 vehcles n transportaton envronmental [19,0] and weather montorng systems [1,]. One of the mportant applcatons n ths area s target trackng, whch has a varety of applcatons such as trackng enemes, humans, anmals and cars on hghways, and many other cases [3,4]. In trackng applcatons, when nodes sense the target, they report the state nformaton to a snk through multhop routng [3,4]. When the target moves to the next locaton, the closest node to the target contnues to montor the target, and the target contnues to be montored through the collaboraton of such multple nodes. In such moble target montorng applcatons, there are several key ssues that are worth studyng. (1) Delay. The delay can be dvded nto two categores: one s called sensng delay, referrng to the tme dfference between nodes sensng the target or event and the target appears or the event occurs [5]. The other category s called end to end delay (or communcaton delay) [5]. It refers to the tme that sensor nodes wll be routed to snk after recevng montorng data. In applcatons such as emergency and dangerous moble target montorng, t s mportant to send the status nformaton of a moble target to the snk quckly and contnuously [6 8]. Therefore, excessve delay n moble target montorng applcaton can affect the tmelness and accuracy of the decson, whch can cause very serous losses [6 8]. For example, n the survellance of an enemy, t s necessary to contnuously send the status nformaton of the enemy to the snk, so the alles can make approprate responses to the nvason of the enemy. A longer delay wll cause great danger to one s own sde. () Another key ssue s energy consumpton and network lfetme. Sensor nodes are usually powered by batteres, so ther energy s extremely lmted. Replacement and rechargng of batteres s costly and sometmes mpossble [9 31]. Therefore, n wreless networks, one of the man challenges n target trackng s reducng the energy consumpton of each node and balancng the energy consumpton n all nodes at the same tme to optmze the energy consumpton and maxmze the network lfetme [3 34]. Generally speakng, sensor nodes are restrcted by low capacty processng capabltes, battery operated devces, lmted transmsson ranges as well as lmted data transmsson capacty and other attrbutes [35 38], among whch lmted energy s the most nfluental constrant. In order to reduce energy consumpton, sensor nodes often take a perodc sleep/awake rotaton workng model to save energy [5,7]. The rato of the length of tme the node s n the workng state to the length of the cycle tme s called the duty cycle. Snce the energy consumpton of the node durng work s 100 to 1000 tmes that of the energy consumpton n the state of sleep, nodes should be n the state of sleep as possble to save energy whch means duty cycles of nodes should be small as much as possble to prolong lfetme. However, when the node s n the state of sleep, t cannot montor the target anymore [5,7]. Therefore, a small duty cycle can have a negatve effect on network montorng and ncrease sensng delay and communcaton delay. Sensng delay and communcaton delay also ncrease wth longer sleep tme. Because sensng data s transferred to snk through a mult-hop routng path, each hop on routng path needs to wat for ts forwardng nodes to be awakened from sleep to transmt data when a sender has data to be transferred. The tme spent watng for forwardng nodes to awaken from sleep s called sleep delay. Sleep delay s much longer than the actual amount of tme t takes to send the data, so sleep delay of mult-hop relays becomes the man component of communcaton delay. Although there s much research about reducng delay, most of them concentrate on reducng sensng delay. For nstance, n target montorng, the method whch s proposed n Ref. [3] s more about how to proactvate nodes n whch the target s ready to move, so that when the target moves to ths regon, t can greatly reduce ts sensng delay. However, t can be seen from the prevous demonstraton that there s sleep delay n every hop of communcaton delay, and each sleep delay s comparable to the montorng delay [5,5]. Therefore, communcaton delay s much more than sensng delay. In addton, the system cannot make decsons untl snk receves the sensng data even f the sensng delay s very small. Thus, t s mportant to reduce the sensng delay, but reducng communcaton s even more mportant. However, there s lttle research on reducng communcaton, whch can affect the performance of the entre system. Reducng communcaton delay between target and snk s a challengng ssue. Prevous research concentrates on buldng an effectve path from target to snk. And the same routng method n the network s adopted after establshng the path to make the communcaton delay larger [4], so that the

3 Symmetry 017, 9, 69 3 of 31 performance of the whole target montorng s greatly affected. Therefore, n ths paper, a Fast and Effcent Data Forwardng (FEDF) scheme for trackng moble targets n sensor networks s proposed to reduce trackng delay and mantan long lfetme. Innovatons n ths paper are as follows: (1) We propose a method to transmt target sensng data through a QRR path whch can reduce delay effcently. In wreless sensor networks, energy consumpton s not balanced, because n the area near the snk ahead of all sensng nodes, ts energy consumpton s far greater than that of the far snk area. And network lfetme depends on lfetme of the frst node n the network. Accordng to related research, the network stll has as much as 90% of the energy left when t des. Therefore, n FEDF scheme, a QRR path s created from target to snk. On QRR path, the duty cycle of nodes near the snk s the same as that n tradtonal scheme whle the duty cycle of nodes far from snk s set to 1. Snce most of the area of the network has surplus energy, on a QRR path, normally only one node's duty cycle s the same as the tradtonal one, whle the duty cycles of other nodes are 1. In ths way, when data s forwardng n ths QRR path, ts sleep delay wll be reduced to 0, whch can reduce the communcaton delay n target montorng greatly. () Ths paper presents a comprehensve approach to fast routng establshment and routng optmzaton n order to mprove the effcency of fast routng. The target data routng scheme n the paper belongs to tral-based routng, so the routng path gets longer and longer wth the movement of target, leadng to lower effcency. We proposed QRR path optmzaton algorthm n ths paper, n whch the rato of the routng path length to the optmal path s mantaned at a smaller constant n the worst case. (3) Fnally, comprehensve performance analyss shows that FEDF scheme can realze the optmzaton of end to end delay and energy utlzaton at the same tme, reduce end to end delay by 87.4%, mprove network energy utlzaton by.65%, and ensure that network lfetme s not less than prevous research. The rest of ths paper s organzed as follows: n Secton, a lterature revew related to ths work s ntroduced. Then the system model and problem statement are descrbed n Secton 3. In Secton 4, we propose an effcent FEDF scheme. The performance analyss of the FEDF scheme s provded n Secton 5. Fnally, Secton 6 presents the concluson and future perspectves of our work.. Related Work There are already qute a few studes on trackng targets. These studes manly focus on the followng aspects..1. Target Detecton n Statonary Snk etwork In terms of statonary target or event montorng [5,5], as shown n the network model shown n Fgure 1, n ths knd of research, sensor nodes are deployed n advance n the network, the event or target can be sensed by nodes nearby when a pre-defned event or target appears n the network. The sensng data s then sent to the snk through the shortest routng path. Apparently, n such statonary sensor and snk networks, communcaton delay depends manly on factors such as the dstance between the locaton of event or targtet and snk and the method used n data transmsson. In a gven network, f common data transmsson methods are adopted, the communcaton delay s usually determned, so the man concern s to reduce the sensng delay n such networks.

4 Symmetry 017, 9, 69 4 of 31 b snk B Internet a event A Interestng target User Fgure 1. The network model. Factors that affect delay nvolve several layers, manly on the MAC layer, network layer, and applcaton layer. Effectve MAC protocol s a way to reduce energy consumpton and delay. MAC protocol adopted by sensor network can be dvded nto synchronous MAC protocol [39,40] and asynchronous MAC protocol [41] accordng to applcaton network t s targetng. In a synchronous wreless sensor network, nodes have the same clock frequency. The nodes wake up only when they need to work, whle at other tmes they sleep. Such protocols are manly TDMA protocol. The TDMA protocol [4] usually mnmzes power consumpton whle ensurng bounded delay and farness. However, these protocols requre precse synchronzaton, whch lmts the scalablty of the system. Synchronzaton n a large-scale network s a very dffcult thng. Therefore, n wreless sensor networks, most applcatons adopt asynchronous mode. In ths way, each node only needs to select ts own work slots ndependently wthout synchronzaton, thus ncreasng ts applcablty. In general, however, the performance of delay n a wreless sensor network s not as good as that n a synchronous network when workng asynchronously. It s because, n the asynchronous mode, the wake/sleep cycle rotaton of nodes s ndependently determned by each node. Perodc cycle s the most mportant factor n determnng delay, so there s more research on duty cycle than delay [43,44]. Generally speakng, ncreasng duty cycle of the node can reduce sleep delay so as to reduce communcaton delay sgnfcantly. However, ncreasng the duty cycle of nodes ncreases the energy consumpton and reduces the network lfetme. Therefore, exstng researches manly focuses on how to meet the requrement of applcatons delay n the case of mnmzng the duty cycle. A dynamc duty cycle method s commonly used. In the method, we often set small duty cycle to nodes to extend network lfetme. In addton, when the amount of data ncreases, we ncrease the duty cycle of nodes, whch can meet the applcaton requrements, reduce the delay and mantan a relatvely hgh network lfe. The man types that belong to the research are Demand Wakeup MAC (DW-MAC) [45] and Adaptve Schedulng MAC (AS-MAC) [46]. In the network where nodes have the same duty cycle, the larger the node densty s, the larger the amount of nodes perceved by target. In addton, the target cannot be detected when sensor nodes are n the state of sleep. Therefore, the larger the node densty, the smaller the sensng delay s. However, on the one hand, ncreasng node s duty cycle ncreases the energy consumpton. On the other hand, ncreasng the densty of nodes ncreases the deployment cost... Trackng Moble Target In moble target networks, nodes n the network are statonary after deployment, whle the target s moble. As shown n Fgure 1, the elephant s a moble target. When the target moves nto the

5 Symmetry 017, 9, 69 5 of 31 network, the sensor node s requred to perceve the target n the shortest possble tme. The tme dfference between target appearng n the network and target beng perceved s represented as sensng delay. In some research, sensng delay s also represented as the dstance the target moves from ts access to the network to when t s perceved by the sensor node. Meanwhle, t s necessary to send the target s sensng nformaton to snk contnually as the target keeps movng. As s shown n Fgure 1, when the target s n place A, the sensor node A perceves ts nformaton to be sent to the snk, whle target moves to B, and the sensor node B perceves ts nformaton and sends t to the snk. In fact, the percepton of moble target s far more smple than as descrbed above, whch can be dvded nto two methods: () non-collaboratve target montorng; and () collaboratve target montorng. In a non-cooperatve target montorng method, nodes are randomly deployed n areas that need to be montored. Therefore, one of the man research contents of non-cooperatve target montorng s target coverage. The purpose s to acheve thata stuaton n whch, the target enters the montorng area, at least one node can montor the target. Huang et al. [17] studed the optmal placement of sensors wth the goal of mnmzng the number of nstalled devces, whle ensurng coverage of target ponts; and wreless connectvty among sensors. Collaboratve target montorng s acheved to montor the target through collaboraton between nodes. Such montorng s manly appled to montorng moble targets [5,6,3 5]. In such studes, when the moble target s montored by a node, the node wll notfy nodes of the next regon that the target may move to. So contnuous montorng of the target s realzed through collaboraton between nodes. Obvously, ths knd of montorng method only sets the duty cycle of nodes to 1 at the possble locaton of the target, and the duty cycle of node n other areas s very small. Consequently, t can save energy and mantan hgh montorng qualty..3. Target Detecton n Moble Snk etwork In such a network, sensor nodes are statonary after deployment, whle the snk s moble. The sensng delay n ths knd of network s the same as the second network mentoned earler. However, the communcaton delay s qute dfferent from the prevous network. In the prevous network, sensor nodes and snks are statonary, so we can know the path data routng to snk after sensor nodes sense data. However, n the moble snk network, even f a routng path to the snk s establshed n the prevous perod, a snk may move to another locaton wthn the next perod. Therefore, how the data montored by a sensor node can be effectvely routed to a moble snk s more complcated n ths knd of network, and the communcaton delay s larger. In such networks, because communcaton delay s the largest component of delay, such networks manly focus on how to reduce communcaton delay. Ths type of network s usually appled to a moble user, whch requre attractve events or targets to be sent to these moble users (snks)..4. Track Moble Target wth Moble Snk etwork Multple moble targets and multple snks. There are also varous applcatons of ths type of network n practce [39]. As Fgure shows, n safar parks, vstors are equvalent to a moble snk (user n Fgure ), and the object that attract toursts lke the elephant s moble target. Toursts s walkng all the tme. When sensor perceves the elephant, t nforms toursts (namely moble target) and tells them the locaton of elephant. Such studes generally adopt tral-based routng to mantan the route between target and snk so that the snk and target can keep up correspondence wth each other. In target trackng lke ths, the routng tral between target and moble snk and snk s tral are stored n the routng path. As a result, nodes can always route to snk through these trals after sensng the target. And regardless of how target and snk moves, through tral routng, a routng path from source node to snk can always be establshed successfully. As s shown n Fgure, when the target moves from A to B, user (namely moble snk) moves from C to D. The method of establshng routng s, when moble target (elephant) s movng, t keeps track of how t gets to A. In ths way, when moble target moves to B, the sensor node wll stll be able to send ts sensng nformaton to A through tral. Because routng between A and B has been establshed already, and when the user moves from C to D, user also retans tral between C to D. Thus, sensng data that reaches C can

6 Symmetry 017, 9, 69 6 of 31 contnue to route to user n poston D. In ths method, the routng from target to snk s routed through the orgnal locaton, so ts routng path s not optmzed. As Fgure shows, the best routng s a straght lne for routng from B to D. However, due to the moblty of target and snk, the actual routng path establshed s B A C D whose length s twce as long as the straght lne from B to D. Obvously, due to the erratc moton of target and snk, the routng path from source to snk wll be longer and longer after a longer perod of tme, resultng n poor routng effcency. The path to snk may actually be k hops, but the current path of source to snk could be n tmes as much as k. The commonly used method s, after every perod of tme, when the routng effcency becomes very poor, we reestablsh the straght lne route from target to snk, reduce nvald path, and make ts path length close to k. Ths method of path optmazaton s also dscussed n lterature [4], In ther method, when a detour appear on the routng path from target to snk, there wll be a certan way to fnd the path shortcut, reduce the path length, and mprove the routng effcency. C b c user B d Internet a D event A Interestng target User Fgure. The moble target and snk network. Accordng to the target s attrbutes, the network can be dvded nto dscrete target or event percepton network and contnuous target montorng network. In dscrete target or event percepton network, event occurs randomly n the network. In ths type of network, sensor nodes typcally transmt the data perceved to snk whle n contnuous target montorng network, moble target keeps movng n the network, so sensor nodes need to transmt data to snk contnuously after sensng the target. Although there s plenty of research on trackng moble targets, these studes manly concentrate on how to establsh a routng path from moble target to snk. And n terms of reducng delay, they specalze n reducng sensng delay and neglect communcaton delay and the proposed soluton s very few [47]. Furthermore, the professonal methods about reducng the communcaton delay do not take nto account the actual applcaton of moble target. For nstance, aveen and Kumar proposed Tunable Locally-Optmal Geographcal Forwardng (T-LOGF) polcy, [48] n order to reduce the communcaton delay. Ther deas are based on the followng analyss: n the network of duty cycle workng mode, each node may have multple forwardng node sets when forwardng data, these forwardng nodes use the perodc awake/sleep mode ndependently. So, when the sender has data to be transferred, the frst node that wakes up s not necessarly close to snk. At that tme, the sender can contnue to wat for a node closer to snk to wake up, or send t mmedately. The dsadvantages of mmedate dlvery are: every tme the routng dstance to snk s short, more hops are needed to be

7 Symmetry 017, 9, 69 7 of 31 routed to snk, whch can cause large delay. And watng for nodes closer to snk to wake up can make the number of hops from snk smaller. It s possble to reduce the total delay but ncrease the watng delay. T-LOGF proposed an algorthm of optmzng forwardng nodes choce to reduce delay [48]. Ths method that only consder about routng can adopted by target trackng method. However, these routes stll don t take nto account factors n trackng target, such as how to mantan the unqueness of routng tral, and also lack of ways to optmze the routng on the bass of mantanng the characterstcs of the tral. Although some studes have suggested that the adopton of mproved duty cycle methods can effectvely reduce delay, but mprovng the duty cycles of the entre network wll have a sgnfcant mpact on network lfetme. And as a matter of fact, target occurs locally and sporadcally. If we just ncrease the duty cycle of nodes n these areas and on routng path, t can sgnfcantly reduce the delay and has lttle mpact on the network lfetme. The scheme demonstrated n ths paper exactly are based on ths dea. The FEDF scheme proposed n ths paper mantans hgh network lfetme on the bass of reducng delay. 3. System Model and Problem Statement 3.1. System Model The network model n ths paper belongs to a typcal planar perodc data gatherng wreless sensor network, whch s smlar to [13,3,4], and ts model structure s as follows: (1) homogeneous sensor nodes are randomly deployed n a two-dmensonal planar network whose center s snk. The network radus s R, and the node densty s ρ. Each node n the network montors the surroundng envronment contnuously, and once the event or target s detected, the next hop s searched n the range of the communcaton radus r, and the sensng data should be sent to snk through mult-hop relays. () Sensor nodes adopts asynchronous sleep/wake workng mode n ths paper, and nodes montor the target and transmt the data only when they are n the wakng state. (3) All the montorng targets are randomly dstrbuted n the network, so the probablty that each node montors target, whch leads to the probablty that each node generates data s equal. 3.. etwork Parameters The component of a sensor node ncludes a sensng unt and a communcaton unt [16,19,41], n whch the sensng unt s n charge of sensng the montorng target or event and moblzng the communcaton unt after sensng the target or event, then the communcaton unt ntates ts nternal communcaton mechansm to transmt data to snk. The sensor nodes adopt sleep/wake perodc mode to save energy for lackng of energy [17]. In a unt cycle, the nodes sleep/wake work perodcally. The node only transmts data and senses targets or events when t s n the wakng state. In a unt cycle, the rato of the node n the wakng state to the whole cycle called duty cycle, assume that Q Sen s the sensng duty cycle and Q Com s the communcaton duty cycle, then: Q Sen = Q Com = s T SE w T SE w + T SE = T w SE (1) T S T C w T s w C + T = T w C () C T Com where T SE w s the tme that the node s n the wakng state durng the sensng cycle and T SE s s the tme that the node s n the sleepng state durng the sensng cycle; T S s the sensng duraton of the node and T Com s the communcaton duraton of the node; T C w s the tme that the node s n the wakng state durng the communcaton cycle, and T C s s the tme that the node s n the sleepng state durng the communcaton cycle. The energy consumpton model of ths paper s smlar to [14,15], and the energy consumpton of nodes s manly composed of event sensng, data transmsson, data recevng and low power lstenng. Therefore, the energy consumpton model can be expressed as:

8 Symmetry 017, 9, 69 8 of 31 E = ε Sen + ε Tra + ε Re + ε Low (3) The man parameters of the system model used n ths paper are smlar to [7], and the parameters values are derved from the nternal data tables of the prototype sensor nodes. Tables 1 and lst the relevant parameters used n ths paper, the remanng parameters not n tables wll be descrbed n the specfc calculaton. Table 1. etwork Parameters. Parameter Value Descrpton E n 0.5 Intal energy (J) T S 15 Sensng duraton (s) T Com 100 Communcaton duraton (ms) T REC 0.6 Preamble duraton (ms) T AFF 0.6 Acknowledge wndow duraton (ms) T D 0.93 Data packet duraton (ms) Ρ Tra Power consumpton n transmsson (w) Ρ Rec Power consumpton n recevng (w) Ρ SE Power consumpton n sensng (w) Ρ Sleep Power consumpton n sleepng (w) Table. Parameters Related to Calculaton. Symbol Q Sen Q Com ε S ε TRA ε SE ε REC Descrpton Sensng duty cycle Communcaton duty cycle Energy consumpton n low power lstenng Energy consumpton n data transmsson Energy consumpton n event sensng Energy consumpton n data recevng 3.3. Problem Statement Desgnng an effcent communcaton scheme that s sutable for wreless sensor networks s a major goal of ths paper. Wth regard to network performance, the scheme should be able to optmze the overall performance of the network, reduce the communcaton delay, mprove the energy utlzaton and mantan the network lfetme. It can be expressed as follows: (1) Mnmze communcaton delay In ths paper, the communcaton delay refers to the tme t takes for data to be transferred from the sendng node to the snk va mult-hop relays [19]. Mn(D com ) = Mn ( d h 1 ) (4) where d h stands for the transmsson delay of -th hop, and the number of relay hop s, then the mnmzed communcaton delay can be expressed as Formula (4). () Maxmze energy utlzaton Energy utlzaton refers to the rato of the energy consumed by the network to the ntal energy of the network. Max(R E_U ) = Max [( E ) ( E n )] (5) 1 1

9 Symmetry 017, 9, 69 9 of 31 where E n stands for the ntal energy of the node and the energy consumpton of node s E so the maxmzed energy utlzaton can be expressed as Formula (6). (3) Maxmze network lfetme etwork lfetme s defned as the death tme of the frst node n the network n most studes [1,0]. In wreless sensor networks, f the energy of the node s exhausted, the node wll de. As a result, the network lfetme s closely related to the maxmum energy consumpton of the network. Assume that there are nodes n the network, the energy consumpton of node φ s ε φ, ts ntal energy s E n. Therefore, maxmzng the network lfetme s equvalent to maxmzng the lfetme of the node wth the largest energy consumpton, that s: Max(LF) = Max [E n Max (ε φ 1 )] (6) In a nutshell, the research objectves of ths paper are as follows: { Mn(D com ) = Mn ( d h ) 1 Max(R E_U ) = Max [( E ) ( E n )] 1 1 Max(LF) = Max [E n Max (ε φ 1 )] (7) 4. The Desgn of FEDF Scheme 4.1. Introducton of QRR Path In a wreless sensor network, the sensor node has the functon of sendng and recevng data. In addton, n the process, there s a lot of energy consumpton. Therefore, sensor nodes often take the perodc sleep/awake rotaton workng model n order to reduce energy consumpton. Data can be sent and receved only when the node s n the wakng state. Therefore, when a node needs to send data to the snk, t has relatvely large delay by usng the tradtonal method, because t has to wat for the next node to wake up. In ths case, the delay wll be ncreasngly large as the routng path gets longer. We propose a method n ths paper that s to create a Quckly Reacted Routng (QRR) path. On the path, the duty cycles of nodes far away from snk are set to 1, whch means they are n the work state all the tme. In ths way, the effcency of data transmsson has been greatly mproved. Consderng that most nodes on a QRR path are workng all the tme, the energy consumpton s pretty huge. In addton, there s a phenomenon n the feld of wreless sensor network called an energy hole. The energy consumpton of nodes close to the snk s greater; nodes near the snk are dead n the end so as to form energy hole. The duty cycle of the nodes near the snk s set as normal. As a result, the network mantans relatvely hgh network lfetme. Fgure 3 shows the communcaton delay of the network by usng normal path and a QRR path. The duty cycle s 0.. The delay of the node 1 hop away from the snk s equal. As the dstance from the snk s farther and farther, the communcaton delay s ncreasng whether on the normal path or the QRR path. However, t s obvous that the communcaton delay on normal path s far greater than on the QRR path. Fgure 4 shows the communcaton delay on the normal path and QRR path under dfferent duty cycles. R n Fgure 4 s short for normal path. Path under the same duty cycle have the same communcaton delay n the range of one hop from the snk. In addton, as the duty cycle s larger, the communcaton delay s smaller. Furthermore, the communcaton delay of QRR path s sgnfcantly smaller than that of normal path. It can be seen that usng a QRR path to transmt data can greatly reduce the delay and mprove the workng effcency of the network.

10 Symmetry 017, 9, of P, DR=0. QRRP,DR=0. communcaton delay Dstance from snk(hop) Fgure 3. Communcaton delay on normal path and QRR path. communcaton delay P, DR=0.4 P, DR=0.4 QRRP, DR=0. QRRP, DR= Dstance from snk(hop) Fgure 4. Communcaton delay on P and QRRP under dfferent duty cycles. 4.. General Desgn of FEDF Scheme In a wreless sensor network, the target keeps movng, and when the node senses the target, t wll send the data to snk. As s shown n Fgure 5, each tme the node perceves the target, t goes straght through the shortest path and sends the data to the snk. However, the FEDF scheme s proposed n ths paper, whch can transmt data effcently and save resources. Target Snk Target Fgure 5. General method for data transmsson.

11 Symmetry 017, 9, of 31 Before ntroducng the scheme, we suppose that: Frstly, accordng to Global Postonng System (GPS) or poston assessment devces, each node knows ts own coordnates. Secondly, the target moves randomly. Thrdly, each tme a node s passed, the node records two messages: number of hops from snk (hopcount) and a set of nodes that have been vsted (vsted_lst). The scheme conssts of three steps. The frst step s ntalzng the sensor network: In a crcular area of radus R, set and locate the coordnates of each node and the dstance between nodes and ntalze the shortest dstance from them to the snk. The second step s creatng an ntal QRR path when the sensor node senses the target n the begnnng. The thrd step s creatng a QRR path accordng to the random movement of moble target. The last step s optmzaton of the path. (1) Intalze the network: determne the mnmum hopcount between node and snk In the ntal stuaton, there are many sensor nodes n the provded montorng area. In ths paper, we wll assume that nodes n the network are unformly dstrbuted, whch means the dstances between nodes are equal. Each node knows ts poston n the coordnate system. In the process of ntalzaton, we set the number of hops from snk to snk tself (hopcount snk snk ) to 0, whle set other nodes to the snk to. Then the snk broadcast hopcount snk snk s 0. When other adjacent nodes receved the value, they add one to ths value and contnue to broadcast t. When the value of hopcount s smaller than that of tself, update the value to the smaller one and add one. Update the value of each node s hopcount accordng to ths method untl hopcount of every node n the network s no longer changed. The network ntalzaton s completed. Fgure 6 shows the network after ntalzng. How to ntalze the network s shown n Algorthm snk Fgure 6. Intalzed network. Algorthm 1 Intalze the network 1: Intalze hopcount snk snk = 0 : Intalze every node n the network: hopcount snk = 3: snk broadcast ts hopcount, and assume node s hopcount receved from others s H bro 4: For every node n the network Do 5: If hopcount snk > H bro + 1 Then 6: hopcount snk = H bro + 1 7: End f 8: Else 9: keep ts orgnal value 10: End else 11: End for

12 Symmetry 017, 9, 69 1 of 31 () Create an ntal QRR path When the sensor node senses the target, t must check ts recordng table (ncludng the last node and the next node) and ensure whether there has been a QRR path. If any, the node just transmts data through t. If no, we create a QRR path accordng to Algorthm, and set duty cycles of nodes far from the snk to 1. In ths way, an ntal QRR path s created, as s shown n Fgure 7. event snk Fgure 7. Intal Quckly Reacted Routng Path. The next node of s represented as next, the current node s c, and hopcount snk the number of hops of the node from snk. Algorthm create a quckly reacted routng path 1: Whle hopcount snk > 0 Do : next = false 3: For each node n the network Do 4: If (hopcount snk < c hopcountsnk ) Then 5: next = true 6: c = 7: hopcount c snk = hopcountsnk 8: End f 9: End For 10: End whle (3) Trackng target represents In the sensor network, a moble target wll move randomly. If we do not record ts moton tral, the data wll be transmtted a totally new way every tme the target moves, whch cannot take advantage of the QRR path that was already establshed. Therefore, we appled the recordng table. The recordng table wll be updated wth the movement of the target to record ts tral. In addton, as every movement of the target s the hopcount from the snk plus 1, add the node vsted to vsted_lst, and set the duty cycle of the node to 1. Algorthm 3 updates the recordng table, the Current ode s represented as C, and the Last ode s L.

13 Symmetry 017, 9, of 31 Algorthm 3 updatng path recordng table(rrt) 1:begn : For(L.next) L.next = C L = C C 5: hopcount snk ++ 6: C.DR = 1 7: add C to vsted_lst 8: End for 9:End (4) Takng Shortcuts Because the locaton of the target s constantly changng, the transmsson path wll be complcated and tortuous when the target moves fast. In ths case, both the delay and the energy consumpton are very costly. So we need to smplfy the path (shortcut) when necessary, whch can reduce the delay and energy consumpton effectvely. Wth regard to the nodes n the wreless sensor network, the energy consumpton s manly composed of event sensng, data transmsson, data recevng and low-power lstenng. So the total energy consumpton of a node s as follows: ε sum = ε SE T S + ε TRA T + ε REC R +ε S T C (8) In Formula (8), ε sum stands for total energy consumpton, ε SE stands for the energy consumpton when the node n sensng, ε TRA stand for that n sendng data, ε REC stands for that n recevng data, and ε S stand for that n sleepng. And T, R stands for the node s data amount for sendng and recevng data. T S and T C are sendng duraton and communcaton duraton respectvely. Energy consumpton n sensng s as follows: ε SE = P Sen Q Sen + P Sleep (1 Q Sen ) (9) where P Sen s power consumpton n sensng, P Sleep s power consumpton n sleepng and Q Sen s sensng duty cycle. Energy consumpton n sendng data s as follows: Q Com T C ε TRA = P Tra T D + [ 4(T REC + T AFF ) + 1 ] (P TraT REC + P Rec T AFF ) (10) where T D stands for the data packet duraton, T REC s the preamble duraton and T AFF s acknowledge wndow duraton. Energy consumpton n recevng data s as follows: ε REC =P Rec T REC + P Rec T D + P Tra T AFF (11) where T D stands for the data packet duraton, T REC s the preamble duraton and T AFF s acknowledge wndow duraton. Energy consumpton n recevng data s as follows: ε S =P Rec Q Com + P Sleep (1 Q Com ) φ T φ R (1) where the frst tem n the formula represents energy consumpton n recevng data, and the second represents that n sleepng. φ T, φ R n the formula can be expressed as follows: φ T = {P Sleep [ (1 Q Com)T C + T REC + T AFF ] + P Rec T REC } T T COM (13)

14 Symmetry 017, 9, of 31 φ R = [P Sleep (T D + T AFF ) + P Rec T REC ] R T C (14) In sensor network, the closer the dstance to snk s, the greater the energy consumpton s. Supposng that the radus of the network s R, the communcaton radus of the node s r, and the probablty of generatng data s beta β, so the data amount of the node that s meters away from snk can be represented as follows: R =[(Μ + 1) + Μ(Μ+1)r ] β (15) where + Μr < R Data amount of the node when sendng data s equal to data amount when recevng data plus data amount produced by the node tself: T = R + β (16) Supposng that communcaton duty cycle n the network s Q Com, communcaton duraton s T Com, so one hop transmsson delay of a node s as follows: D nqr = (1 Q Com) T Com + T REC + T AFF + T D (17) Theorem 1. In the network, supposng that number of hops from node to snk s hopcount snk, the communcaton delay of the node s as follows: D ETE D nqr =hopcount snk =hopcount snk ( (1 Q Com) T Com + T REC + T AFF + T D ) Proof. Accordng to Formula (17), we have already known one hop transmsson delay s D nqr, that s (1 Q Com ) T Com + T REC + T AFF + T D, and number of hops from snk s hopcount snk, s communcaton delay. The product of the two s communcaton delay from node to snk. In the plane, dstance between a (x a, x b ) and b (y a, y b ) D a b can be expressed as (x a x b ) + (y a y b ). So the dstance between two ponts can be expressed as follows: (18) D b a = (x a x b ) + (y a y b ) (19) Theorem. In a wreless network wth unformly dstrbuted nodes, dstance between two adjacent nodes s d, so the communcaton delay from node ε 1 to node ε s as follows: D ε ε ε1 = ε (Dε1 /d) D nqr = (D ε1 /d) ( (1 Q Com) T Com + T REC + T AFF + T D ) (0) ε Proof. Accordng to Formula (19), the dstance between ε 1 and ε s D ε1, and dstance between two ε adjacent nodes s d, so D ε1 /d represents number of hops from node ε 1 to ε, the communcaton delay between two nodes s the product of the hopcount from snk and one hop delay. Theorem 3. In FEDF scheme, one hop delay of a hotpot s expressed as follows: qr D 1_hop = T REC + T AFF + T D (1) Proof. Accordng to Formula (17), n FEDF scheme, communcaton duty cycle of the node s set to 1, so the value of (1 Q Com) T Com s 0. One hop delay of the hotspot s only related to T REC, T AFF, T D.

15 Symmetry 017, 9, of 31 ε Theorem 4. In FEDF scheme, number of hops from node ε 1 to ε s hopcount ε1, the end to end delay between two nodes wth hgh communcaton duty cycle s expressed as follows: ε Dl ε ε1 = qr ε hopcountε1 D 1_hop = hopcount ε1 (T REC + T AFF + T D ) () ε Proof. As for nodes separated by hopcount ε1 hops, the communcaton delay s the product of one hop delay and hopcount. And one hop delay of a hotpot s T REC + T AFF + T D accordng to Formula ε (1), so the result s hopcount ε1 (T REC + T AFF + T D ). Theorem 5. In FEDF scheme, total cost n the process of node 1 sendng data to can be expressed as follows: C qr =Dl ε sum =hopcount 1 (T REC + T AFF + T D )+ ε sum 1 (3) ε Proof. Communcaton delay between two nodes wth hgh duty cycle s Dl ε1 accordng to Formula (), and energy consumpton of node s ε sum accordng to Formula (8), so the total cost of a node wth hgh communcaton duty cycle s the sum of cost on delay and on energy consumpton, that s + εsum Dl 1 1. Theorem 6. Smlarly, the transmsson cost from node 1 to node wth nomal communcaton duty cycle s as follows: C nqr =D ε ε1 + ε 1 εsum =(D ε1 /d)( (1 Q Com) T Com + T REC + T AFF + T D )+ ε sum 1 (4) Proof. Accordng to Formula (0), the communcaton delay between nodes wth normal ε communcaton duty cycle s (D ε1 /d)( (1 Q Com) T Com + T REC + T AFF + T D ), and energy consumpton of the node s ε sum accordng to Formula (8). So when node 1 send data through nodes to node, the total energy consumpton s 1 ε sum. Compared to QRR path, t has less energy consumpton. Theorem 7. In FEDF Scheme, n order to measure the cost of sendng data, we set nfluence factor δ 1, δ ndcates the nfluence level of takng the orgnal path and creatng new path n the process of data transmsson. Total cost of transmsson of a node s expressed as follows: C tot =C qr δ 1 +C nqr δ =[hopcount 1 ( TREC + T AFF + T D )+ 1 ε sum ] δ 1 +[(D (1 Q 1 /d)( Com ) T Com + T REC + T AFF + T D )+ 1 ε sum ] δ (5) Proof. For a node that needs to send data, ts cost n the entre process of transferrng data s the sum of the cost through exstng path and create a new path. Accordng to Formula (3), the cost for the node transferrng data through the exstng path s hopcount 1 (T REC + T AFF + T D )+ 1 ε sum, and the cost for the node creatng a new path s (D (1 Q 1 /d)( Com ) T Com + T REC + T AFF + T D )+ 1 ε sum. In addton, δ 1 and δ stands for the degree to whch these two tems matter. Therefore, the total cost s expressed as Formula (5).

16 Symmetry 017, 9, of 31 (1) Pre-Shortcuts We do not know whch stuaton s the best before actually take shortcuts. Therefore, explorng a relatvely approprate path s necessary. We synthesze a varety of stuatons and fnally choose the best as the fnal transmsson path. The process s called pre-shortcuts n ths paper. The endpont of a shortcut s called EDP. Every tme the sensor node perceves the target, the cost of every node n ts vsted_lst wll be calculated and analyzed through Algorthm 4, ncludng the delay and energy consumpton. Thus, the node that has the mnmum cost s the EDP. The mplementaton of the procedure wll be analyzed n detal. When the target moves to the poston of node source, dstance between every node p (ncludng snk) n vsted_lst and source s calculated and the result s D p source. And t s known to all that dstance between two adjacent nodes s d, so D p source /d stands for the mnmum hopcount from node p to source (hopcount p source ). In the condton that hopcount p source reaches a threshold η, the cost of creatng a path from source to p can be calculated accordng to Formula (4), and the cost of transmttng data from source to snk can also be caculated accordng to Formula (3). Therefore, the total cost s C tot = C qr δ 1 +C nqr δ accordng to Formula (5). However, t s unnecessary to create a new path f transmttng data through the orgnal path, and ts cost s C old = C qr δ. In Algorthm 4, t shows the method about how to fnd possble EDP (pedp). Algorthm 4 Identfyng the possble EDPs 1: Fnd_pEDP (ode source, Threshold η) : Begn 3: //η: threshold for shortcuts 4: //pre_lst: the set of pedp 5: For each node p n source.vsted _lst Do 6: If (hopcount p source η) Then 7: Calculate C tot of p 8: add p to pre_lst 9: End f 10: End for 11: End In Algorthm 5, for every node n pre_lst, the node that has the mnmum C tot wll be found and t s exactly EDP. Algorthm 5 Determnng the EDP 1: begn : For each node n pre_lst 3: Fnd mn{ C tot = C nqr δ 1 + C qr δ } 4: return p 5: End for 6: End In order to reduce unnecessary delay and energy consumpton, we take the shortest path when creatng path, as s shown n Algorthm 6. Supposng the locaton of node 0 s (x 0, y 0 ), the locaton of source s (x, y). Accordng to Formula (6), the offset of the abscssa and the ordnate of A (the next hop of source) can be calculated as (x.offset, y.offset) = (a, b). So nexthop(x, y) = (x-x.offset, y-y.offset). { b a = y y 0 x x 0 = k a + b = d (6)

17 Symmetry 017, 9, of 31 In Formula (6), k s the slope of the straght lne formed by snk and source, the locaton of snk s (x 0, y 0 ), the locaton of source s (x, y), a, b stands for the offset of the abscssa and the ordnate of source respectvely, and d stands for the shortest dstance from source to 0. Algorthm 6 Creatng a straght lne path 1: Fnd_exthop (H) : begn 3: //H: next hop 4: If (H s not snk) Then 5: nexthop(x, y) = (x-x.offset, y-y.offset) 6: If (search_neghbor (nexthop)!= ULL) Then 7: return nexthop 8: End f 9: End f 10: End () Takng Shortcuts After the fnal EDP s determned, the target sends a data packets from the source to the EDP to nform t of the two ends of the shortcut. Accordng to Algorthm 6, the next hop s found contnuously and every tme a node s vsted, ts recordng table s updated, and ts communcaton duty cycle s set to 1. Untl the next hop s the EDP, the orgnal path can be cancelled, whch means restorng the duty cycle of nodes on the path and deletng relevant tems of the recordng table. Before that, data s transmtted through the orgnal path. As s shown n Fgure 8, the target sends data from source to snk through A-B-C-D-E-F-G. Assumng that the value of η s, the cost of nodes (except for node A an B) n vsted_lst can be calculated accordng to Algorthm 6, and add pedps to pre_lst. E D C B F I H A G Source snk L K J Fgure 8. Possble shortcuts. We compare the cost of all nodes n pre_lst accordng to Algorthm 5, assumng that the node that has the mnmum cost s F, F s EDP n the example. Therefore, the target send a nformaton packet from source to F to nform F, accordng to Algorthm 6, the sendng path s H I F. Before the arrval of data packet, target transmt data through the orgnal path. When the packet arrves, cancel path A B C D E F, restore ther duty cycle, remove them from vsted_lst and delete the relevant nformaton of recordng table. Snce then, the target wll send data data through path H I F G.

18 Symmetry 017, 9, of Performance Analyss of FEDF Scheme In ths paper, we measure the performance of a FEDF scheme from three aspects: delay, energy utlzaton and network lfe. By analyzng and comparng the performance of the FEDF scheme and the tradtonal routng scheme (every tme take the shortest way), compared to tradtonal scheme, the FEDF scheme reduces communcaton delay by 87.4%, and mproves energy utlzaton by.65%. It s obvous that the FEDF scheme demonstrated n ths paper performs extremely effcently. In the sensor network, the dstance between two adjacent nodes s d. A moble target keeps movng and moves randomly n all drectons n the network, whch produces a varety of paths. So n the mathematcal statstcs pont of vew, the probablty that each node moves n every drecton s equal. That s, the actual path (D) of the moble target and ts dstance (d) to the snk are proportonal. Therefore, we can set a path smplfcaton coeffcent λ (λ 1) and D = d λ. For dfferent scenaros, the value of λ s dfferent. For nstance, when the moble target moves fast and the path bendng degree s large, the value of λ s larger. However, when the velocty s slow, the path s smlar to a straght lne, and the value of λ s small and close to 1. We wll analyze the delay and energy consumpton n the case of dfferent values of λ below Transmsson Delay Theorem 8. In FEDF scheme, assume that the network radus s R, the communcaton radus s r, the set of nodes wth communcaton duty cycle of 1 s Q, the set of nodes wth common duty cycle s F, and one hop delay of node n can be expressed as follows: (1 Q Com ) T Com + T D 1_hop = { REC + T AFF + T D, n F T REC + T AFF + T D, n Q (7) Proof. accordng to Formula (17), one hop delay of node wth normal communcaton duty cycle s (1 Q Com ) T Com + T REC + T AFF + T D, and accordng to Formula (1), one hop delay of node wth duty cycle of 1 s T REC + T AFF + T D. In summary, one hop delay of a node can be expressed as Formula (7). Theorem 9. In FEDF scheme, on QRR path, the duty cycle of node 0 near snk s normal, others s 1. n HopCount from node n to 0 s hopcount 0, and these nodes has hgh duty cycle. Therefore, the communcaton delay from node n to snk can be expressed as follows: D ete =D 1_hop +hopcount n 0 (T REC + T AFF + T D ) = (1 Q Com) T Com +(hopcount n 0 +1) (T REC + T AFF + T D ) Proof. One hop delay of nodes wth normal duty cycle s (1 Q Com) T Com + T REC + T AFF + T D from Theorem 8, so one hop delay of 0 s (1 Q Com) T Com + T REC + T AFF + T D. And the duty cycle of other n hopcount 0 nodes s 1, the sum of the delay of these nodes s hopcount n 0 (T REC + T AFF + T D ). Therefore, the communcaton delay from node n to snk s as Formula (8). In the FEDF scheme, on a QRR path, the frst hop from the snk has a normal duty cycle, whle other nodes have a duty cycle of 1. However, n the tradtonal routng path (TRP) scheme, the communcaton delay s drectly proportonal to dstance from the snk. Fgure 9 shows the communcaton delay comparson of the FEDF scheme and the TRP scheme, n whch the routng path s straght; the value of λ s 1. (8)

19 end-to-end delay (ms) communcaton delay Symmetry 017, 9, of FEDF TRP Dstance from snk (m) Fgure 9. End to end delay n FEDF and TRP. When the degree of bendng of the path s dfferent, the communcaton delay s dfferent. In general, the longer the path, the larger the delay. Fgure 10 shows the performance of the communcaton delay n the FEDF scheme from dfferent λ as the dstance from the snk becomes larger. It s obvous that the communcaton delay s large f the value of λ s large =1 =1. =1.4 = Dstance from snk (m) Fgure 10. Communcaton delay n FEDF from dfferent λ. Fgure 11 shows the communcaton delay n FEDF from dfferent λ and n TRP. When λ s, the communcaton delay n the FEDF scheme s greatly less than that of the TRP scheme. In general, the value of λ wll be mantaned at a relatvely small number, because every tme the value of λ becomes relatvely large, that s, when the degree of the path bendng gets large, the path wll be updated accordng the algorthm. In a word, wth regard to delay, the FEDF scheme performs very well.