Architecture Design of Mobile Access Coordinated Wireless Sensor Networks

Architecture Design of Mobile Access Coorinate Wireless Sensor Networks Mai Abelhakim 1 Leonar E. Lightfoot Jian Ren 1 Tongtong Li 1 1 Department of Electrical & Computer Engineering, Michigan State University, East Lansing, MI 4884, USA. Air Force Research Laboratory, Wright-Patterson Air Force Base, Dayton, OH, 45433 USA. Abstract This paper consiers architecture esign of mobile access coorinate wireless sensor networks (MC-WSN) for reliable an efficient information exchange. In sensor networks with mobile access points (SENMA), the mobile access points collect information irectly from iniviual sensors as they traverse the network, such that no routing is neee in ata transmission. While being energy efficient, a major limitation with SENMA is the large elay in ata collection, making it unesirable for timesensitive applications. In the propose MC-WSN architecture, the sensor network is coorinate by powerful mobile access points (MA), such that the number of hops from each sensor to the MA is minimize an limite to a pre-specifie number through active network eployment an network topology esign. Unlike in SENMA, where the ata collection elay epens on the physical spee of the MA, in MC-WSN, the elay epens on the number of hops an the electromagnetic wave spee, an is inepenent of the physical spee of the MA. This innovative architecture is energy efficient, resilient, fast reacting an can actively prolong the lifetime of sensor networks. Our simulations show that the propose MC-WSN can achieve higher energyefficiency an orers of magnitue lower elay over SENMA, especially for large-scale networks. Inex Terms Mobile access coorinator, wireless sensor networks, sensor network security an reliability. I. INTRODUCTION Wireless sensor networks have receive significant attention from the research community ue to their potential impact on various military an civilian applications. For efficient an reliable communication over large-scale networks, sensor networks with mobile access points (SENMA) was propose in [1]. In SENMA, the mobile access points (MAs) traverse the network to collect the sensing information irectly from the sensor noes; when the energy consumption at the MAs is not of a concern, SENMA improves the energy-efficiency of the iniviual sensor noes over a-hoc networks by relieving sensors from the energy-consuming routing functions [1], []. While being energy efficient, a major limitation with SENMA is the large elay in the ata collection process. This elay epens on the physical spee of the MA an the length of the MA trajectory, which woul increase ramatically as the network size increases. Large elay makes SENMA unesirable for time-sensitive applications. Along with the recent avances in the remote control technologies, UAVs have been use for management an coorination functions in wireless networks. For example, network eployment through UAV has been recently explore in the literature [3], [4]. For sensor eployment, the UAV basically carries one or more sensor noes, then flies to the require location an gets own to a specific altitue where it is safe to rop the sensor for eployment. A possible network eployment metho using UAV was experimente in [4]. In this paper, mobile access coorinate wireless sensor networks (MC-WSN) architecture is consiere for energyefficient, reliable, an time-sensitive information exchange. In MC-WSN, the whole network is ivie into cells, each is covere by one MA. The MAs coorinate the network through eploying, replacing an recharging noes. They are also responsible for enhancing the network security, by etecting compromise noes then replacing them [5]. Data transmission from sensor noes to the MA goes through simple routing with the cluster heas along a ring or a powerful center cluster hea locate at the mile of each cell. Base on active network eployment an topology esign, the number of hops from any sensor to the MA is minimize an limite to a pre-specifie number. Unlike in SENMA, the elay in MC-WSN epens on the number of hops an the electromagnetic wave spee, an is inepenent of the physical spee of the MA. For MC-WSN, the energy consumption at the iniviual sensors is mainly etermine by the istance from the nearest cluster hea, which is one hop away, an is inepenent of the coverage area of the MA an the noe ensity. We emonstrate the effectiveness of the propose architecture through simulation examples, which show that the MC-WSN architecture achieves higher energy-efficiency an orers of magnitue lower elay over SENMA, especially for large-scale networks. II. REVISIT THE SENMA SYSTEM To reuce the routing buren on iniviual sensors, SENMA networks utilize mobile access points to collect the sensing reports from all sensors noes [1]. The access point traverses the network at a height H S broacasting beacon signals at ranom locations. The coverage area of the access point is moele as a circle of raius r. The access point activates sensors within its coverage area, an each time only a single sensor respons to the beacon message by reporting its sensing information. In SENMA, since sensors communicate irectly to the mobile access point without any routing, the energy consumption at the iniviual sensors is significantly reuce over a-hoc networks [1]. For this to happen, the mobile access point nees to traverse its footprint exhaustively to cover all sensors, resulting in a very long mobile access trajectory an consequently

huge elay. The elay epens on the physical spee of the MA an the length of the MA trajectory, which woul increase significantly as the network size increases. Thus, SENMA coul be unesirable for time-sensitive applications. Motivate by this observation, we propose a mobile access coorinate wireless sensor networks architecture. III. THE PROPOSEOBILE ACCESS COORDINATED WIRELESS SENSOR NETWORK (MC-WSN) In this section, we escribe the propose MC-WSN architecture that aims at proviing reliable, energy-efficient an scalable network structure for prolonge-network lifetime an time-sensitive ata exchange. We assume the network is ivie into hexagonally shape cells, with sies of length. Each cell contains a single powerful mobile access point (MA) an K SN uniformly eploye sensor noes (SNs) that are arrange into K CH clusters, each is of raius R CH. Each cluster is manage by a cluster hea (CH), to which all the cluster members report their ata. CHs then route the ata to the MA. A powerful center cluster hea (CCH) is employe in the mile of each cell. The CCH can establish irect communication with the MA as long as it is insie the cell. After receiving the ata of the sensors, the MA elivers it to the Base Station (BS), which in turn makes the final ecisions. To improve the network reliability, efficiency an scalability, multiple BSs can be employe. The overall network architecture is illustrate in Figure 1. sensor noes an cluster heas. It moves physically for ata collection only in the case when the routing paths o not work. Otherwise, for security reasons, it stays at a ranom location on a circular path of a raius an at an angle φ from the CCH as shown in Figure. φ is uniformly istribute with PDF f φ (φ) = 1 π. The CHs along the circular path forms a ring, through which the ata can be elivere to the MA. The ring is esigne such that there is at least one CH on it can communicate irectly with the MA, which we refer to as the RCH. This shoul be the one nearest to the MA location. Note that, since the MA moves ranomly over the ring, the RCH is not fixe. It is also note that the ring may not be a strict circle, an it epens on the coverage area of the MA. Data transmission from any SN to the MA goes through simple routing, either with the CCH or the RCH. Let the communication range of each sensor noe be r c. SNs only communicate with their corresponing CH, which then route the ata to the MA. CHs have larger storage capacity an longer communication range than SNs. To minimize the elay in ata transmission from the sensors to the MA, the number of hops neee in routing shoul be minimize. Therefore, we consier iviing the CHs in the cell into two groups base on their location. The first group contains CHs within the area of raius R o from the CCH. While the secon group contains all CHs locate outsie the raius R o, where R o <. CHs in the first group will mainly route their information to the MA through the CCH, which can eliver the ata irectly to the MA. While CHs in the secon group route their information to the MA either irectly or through the RCH. The latter case happens by first routing the ata to the nearest CH on the ring, which then broacasts the ata in both irections along the ring until it reaches the RCH as shown in Figure. The RCH then eliver the ata to the MA. To minimize the number of hops, if a CHs in the secon group is very far from RCH, it can irectly forwars its ata to the CCH, as will be illustrate in Section V. Cluster hea Sensor noe Center cluster hea Mobile access Point Base Station Fig. 1. Propose MC-WSN architecture. c In the propose MC-WSN architecture, the MA coorinates the sensor network an resolves the noe eployment issue as well as the energy consumption problem of wireless sensor networks. More specifically, the MAs are responsible for (i) eploying noes, (ii) replacing an recharging noes, (iii) etecting malicious sensors, then removing an replacing them, (iv) collecting the information from sensors an elivering it to the BS. When an MA nees to be recharge or reloae, it sens a request to the MA base. The base will sen a new MA to the cell, an the ol MA will be taken back to the base for maintenance services. The MAs can move on the groun, an can also fly. Each MA traverses its cell mainly for removing the malicious noes an replacing or recharging low-energy R o RCH RCH Cluster hea Mobile access point Sensor noe Center cluster hea Fig.. The topology use in the MC-WSN architecture. Here CH c has a packet to eliver to the MA. 171

In the architecture esign, we limit the average number of hops from any SN to its corresponing CH to N 1, an limit the average number of hops from any CH to the MA to N, where N 1 an N are pre-specifie numbers. N 1 is controlle through active network eployment, while N is controlle through the network topology esign (i.e. select the ring raius an the raius R o ). The main features of the MC-WSN architecture are: Resolve the network eployment problem an actively prolong the network lifetime The propose MC-WSN allows the MAs to manage the eployment of SNs an CHs. That is, the MA can a more noes, relocate or replace exiting noes. In aition, it can recharge or replace low energy noes. When a noe has low remaining energy, it sens a control message to the MA notifying it with its energy level. The MA can then make the ecision to replace the noe or recharge it. Being coorinate by the MA, the MC-WSN architecture resolves the network eployment issue an can actively prolong the network lifetime. Minimize the elay Unlike in SENMA, where the ata collection elay epens on the physical velocity of the MA, in MC-WSN the elay epens on the number of hops an the electromagnetic wave spee, an is inepenent of the physical spee of the MA. Therefore, the elay is significantly lower than that in SENMA. The elay is further reuce by minimizing the number of hops require to reach the MA; this is achieve through network topology esign an active network eployment. Provie high energy efficiency The SNs have the most limite resources in wireless sensor networks. In the propose MC-WSN, SNs only communicate with their nearest CHs, an are not involve in any inter-cluster routing. Also, in contrast to SENMA, SNs in MC-WSN o not nee to receive the beacon signal from the MA. Enhance network security The MAs can etect malicious SNs an CHs an replace them [5]. It is ifficult to get the MA itself compromise or estroye, since it is much more powerful than other network noes, an it moves ranomly in the network where its location can be kept private [6]. Enhance network resilience, reliability an scalability: The MC-WSN is a self-healing architecture, where the CCH an RCH represent two options for relaying the ata to the MA. Each option can act as an alternative for the other. In the case when the routing paths o not work, the MA can traverse its cell for ata collection. Overall, MC-WSN is a resilient, reliable an scalable architecture. Discussions on Feasibility: Sensor noes an MAs eployment in ifferent cells is possible through the UAV technology. A helicopter, or a larger unmanne aerial vehicle, can eploy the MAs in the network, an can replace existing MAs when they are out of energy. The sensor eployment is initiate by the UAV an then tune by the MA. IV. NETWORK OPERATIONS In this section, we illustrate the main network proceures use in MC-WSN. A. Network Set-Up We assume that the CHs an the MAs are equippe with Global Positioning System (GPS) to obtain their location information. The network set-up is establishe through the following steps: 1) Cluster formation: All CHs broacast Hello messages containing their IDs an locations. Each SN etects it nearest CH, to which it sens a Request to Join (RT J). Upon reception of RT J, CH replies with Confirm to Join (CTJ). ) Ring set-up: The MA traverses the cell along a circular ring of raius broacasting Start messages to the CHs in its coverage area. Denote the set of CHs that are along the ring an within the MA coverage area as χ. All the CHs in χ receive the Start message, an reply to the MA with an ACK. 3) Discover links to the ring: broacast InitCH message to all their neighboring CHs. The InitCH message inclues the ID an location of the sener CH. When a CH receives the InitCH, it will in turn broacast InitCH to its neighbors. This happens until all CHs receive at least a single InitCH message. Confirmation is mae backwar through the same links. 4) Discover links to the CCH: CCH also broacasts a Reference signal to all its neighboring CHs. The Reference signal is forware by CHs, using the same manner iscusse above, until it reaches the CHs on the ring. 5) Establish the links to the ring or the CCH: CHs then establish connections with the CCH an/or the closest CH on the ring by sening Request to Connect (RT C) message. The process is complete when the intene receiver replies with Confirm to Connect (CTC). B. Sensing an Collecting The sensing an collecting stage is performe perioically, where the iniviual sensors monitor the environment an report their information to the CHs. When TDMA is use within clusters, each SN reports to its corresponing CH a ata message in its allotte time slot. In orer to minimize the interference between clusters (inter-cluster interference), Direct Sequence Sprea Spectrum (DSSS) can be use, where the noes of the ifferent clusters use ifferent spreaing coes [7]. Also, ata transmissions from SNs to CHs, between CHs, from CHs to the MA, an from the CCH to the MA are mae over ifferent frequencies to avoi interference between the ifferent communication links. CHs route the sensing information to the MA through their establishe links with the CCH or any noe in χ. When the MA visits a region, it sens a beacon signal to the CHs within its coverage area. If a CH receives the beacon signal, then it can respon irectly by sening its ata to the MA. If the ata is receive correctly, the MA respons with an ACK. It

is note that the ata a CH sens to the MA can be information from its own cluster members, or from other clusters that relay their ata through it. C. Malicious Noe Detection When the MA receives ata from a noe, it first authenticates the source an checks its ientity. If the noe passes the authentication proceure, its ata woul be use in the final ecision making process. Some authenticate sensors can be compromise an may report fictitious ata. This is known as Byzantine attacks [5]. The MA shoul be able to etect these malicious noes an avoi their harmful effect. One way to etect compromise noes is to use a reliable ata fusion scheme [5], on the information collecte from many sensors, an obtain the final ecision. The MA monitors the reports of each iniviual noe an compares it with the final ecision obtaine by the ata fusion. Base on the observations over several sensing perios, the malicious noes can be etecte an remove. V. THE NETWORK TOPOLOGY DESIGN In this section, we obtain the optimal raius R o an the ring raius that minimize the require number of hops from any CH to the MA. Define θ as the smaller angle between the MA position an the CH on the ring that first receive the ata; the PDF of θ is f θ (θ) = 1 π. Assume that two cluster heas separate by a istance can communicate irectly. Then, for a given θ, routing along the ring requires θr approximately t hops to reach the RCH. While, the R routing from the ring to the CCH requires t hops. Therefore, to minimize the number of hops for noes outsie the ring, the routing over the ring is mae uner the conition that θ < 1, i.e., θ < 57.3. If this conition is not satisfie, noes reach the MA mainly through the CCH. Noes in the region between R o an route their ata through the ring if θ Rt x < x. That is, if θ < x 1, the noes at istance R o x< route their ata to the RCH. Otherwise, they route the ata to the CCH. We assume that CHs along the ring can estimate θ from the beacon signal they receive. The RCH notifies the noes in the region between R o an that are connecte to it. If the noes in this region o not receive a notification from the RCH, then they forwar their ata to the CCH. Let N hops be the average number of hops require to route the ata from any CH in the network to the CH that can establish irect communication with the MA, i.e., the CCH or the RCH. Note that the number of hops to reach the MA is N = N hops 1. N hops is given by: N hops = 1 Rt [ Ro 0 π x=r o θ= R x 1 t x 1 Rt x=r o θ=0 xf X(x)x xf X(x)f θ (θ)θx ( x θ) f X(x)f θ (θ)θx π x= θ=1 1 x= θ=0 xf X(x)f θ (θ)θx ] (x θ) f X(x)f θ (θ)θx,(1) where x is the istance from any CH to the center of the cell, f X (x) is the PDF of x an can be approximate by f X (x) = x assuming that the CHs are uniformly istribute in a circle of raius. By setting N hops R o =0, we obtain the optimal R o =0.5. Then, we substitute in (1) with the optimal R o, an obtain the optimal by setting N hops =0. We get =0.686 ; it then follows that R o =0.343. Proposition: To minimize the ata collection elay in MC- WSN, the following conitions shoul be met. (1) The CHs within a istance R o =0.343 from the center of the cell eliver their ata to the MA through the CCH. () Noes at a istance x from CCH, where R o x<, eliver their ata to the MA through the ring if θ < x 1, or through the CCH if θ > x 1. (3) Other noes at a istance x from the CCH, eliver their ata to the MA through the ring if θ < 1, or through the CCH if θ > 1. The ring raius is =0.686. The elay is proportional to the number of hops require for routing the ata to the MA. The average istance travele corresponing to N hops is N hops ; therefore, the elay in packet elivery is N hops V EM, where V EM =3 10 8 m/s is the electromagnetic wave (EM) propagation spee. VI. NUMERICAL RESULTS Example 1: Delay comparisons For elay calculations, we assume that there are no collisions or retransmissions. We also assume that each SN reaches its nearest CH in one hop, which can be ensure with the active network eployment, an the number of hops from the RCH or CCH to the MA is one. Therefore, we use the number of hops to reach the RCH or the CCH as an inication of the elay. To show the effectiveness of our propose architecture, we compare two topologies: (a) Topology 1: all noes in the cell use the CCH to eliver their ata to the MA. (b) Topology : both the CCH an the RCH are employe to eliver the ata to the MA. For topology 1, let N hops,1 be the average number of hops for a CH to reach the CCH. Then, N hops,1 = x 0 f X (x)x = 3R CH. For topology, the number of hops is obtaine using (1). By substituting with =0.686, we get N hops, = N hops =0.97 R CH. Comparing N hops, with N hops,1, it can be seen that lower elay can be achieve with the propose topology (topology ). This is further illustrate in Figure 3, where the number of hops is plotte versus for both topologies. The elay in ata elivery ( ) can be expresse as: = C 1 V EM, where C 1 is a constant. In SENMA, the SNs wait for the MA visit to report their ata; hence, the elay epens on the velocity of the MA, as well as the cell size; that is, the average elay for a noe to report its ata to the MA in SENMA is D S = C V MA [1], where V MA is the MA spee

an C is a constant. Therefore, we have D S = C V EM V MA, where C is a constant. This implies that the propose MC- WSN architecture coul result in several orers of magnitue lower elay over SENMA. The elay reuction is proportional to the ratio between the spee of light (EM spee) an the physical spee of the MA. Table I shows the elay ratio D S for ifferent cell sizes. TABLE I DELAY COMPARISON WITH V MA =30m/s. Energy (J/bit) 6 x 10 6 5 4 3 1 MC WSN SENMA Cell ege length (m): 100 1000 D Delay ratio: S 10 9 10 10 D S : Average elay in SENMA, : Average elay in MC-WSN. Average number of hops 10 9 8 7 6 5 4 3 CCH only Ring an CCH 100 10 140 160 180 00 0 40 60 80 300 (m) Fig. 3. The average number of hops in the cases when the communication with the MA is establishe through the CCH only (N hops,1 ), an both the CCH an the ring (N hops, ). Here we use R CH =10m. Example : Energy efficiency We focus on the energy issipate in the iniviual SNs, since they have the most limite resources. We use the circuitry raio energy issipation moel to evaluate the energy efficiency [7]. We assume that the raius of the cluster is R CH = r c. Let E tx an E rx be the energy issipate in the transmitter an receiver electronics of the SNs, respectively. Then, in MC-WSN, the maximum energy issipate in SN to transmit a bit to the CH is E SN,M = E tx ϵ pa rc γ (J/bit), where ϵ pa is the energy consume by the power amplifier, γ is the path loss exponent. In SENMA, each SN must first receive a beacon signal from the MA in orer to report its ata. Therefore, the energy issipate by a noe to report a single bit to the MA is E SN,S = E tx ϵ pa H γ S E rxπr K SN A c [1], where A c is the area of the cell. Figure 4 shows E SN,M an E SN,S as the number of sensor noes increases. It can be seen from the figure that the MC-WSN is significantly more energy-efficient than SENMA, especially when the ensity of the sensors increases. It shoul be note that energy issipation in the CHs an MAs are ignore here. However, if their energy issipation is taken into account, the MC-WSN woul still be more efficient than SENMA architecture; since in SENMA, the MA is assume to traverse the network continuously leaing to a very high energy consumption. 0 0 1 3 4 5 6 7 8 9 Number of sensors K SN x 10 4 Fig. 4. The energy issipation (J/bit) vs. the number of SNs in the MC-WSN an SENMA networks, when =100m, r c = r =10m, H S =10m, γ =, E tx = E rx =50pJ/bit an ϵ =10pJ/bit/m VII. CONCLUSIONS In this paper, we propose a reliable an energy-efficient architecture esign for mobile access coorinate wireless sensor networks (MC-WSN). In the propose architecture, the network exploits the mobile access points to actively eploy noes, perform ata collection, etect malicious sensors an enhance the network security. Not only oes the MC-WSN resolve the network eployment problem, but it also prolongs the network lifetime actively an provies an efficient framework for time-sensitive information exchange. Simulation results showe that the propose MC-WSN architecture can achieve higher energy efficiency an several orers of magnitue lower elay over SENMA. The gains achieve by the propose architecture increase as the network size increases. REFERENCES [1] G. Mergen, Z. Qing, an L. Tong, Sensor networks with mobile access: Energy an capacity consierations, IEEE Transactions on Communications,, vol. 54, no. 11, pp. 033 044, Nov. 006. [] A. Poornima an B. Amberker, Agent base secure ata collection in heterogeneous sensor networks, 010 Secon International Conference on Machine Learning an Computing (ICMLC), pp. 116 10, Feb. 010. [3] I. Maza, F. Caballero, J. Capitan, J. Martinez-e Dios, an A. Ollero, A istribute architecture for a robotic platform with aerial sensor transportation an self-eployment capabilities, Journal of Fiel Robotics, vol. 8, no. 3, pp. 303 38, 011. [Online]. Available: http://x.oi.org/10.100/rob.0383 [4] P. Corke, S. Hrabar, R. Peterson, D. Rus, S. Saripalli, an G. Sukhatme, Autonomous eployment an repair of a sensor network using an unmanne aerial vehicle, IEEE International Conference on Robotics an Automation, ICRA 04, vol. 4, pp. 360 3608 Vol.4, 6-May 1, 004. [5] M. Abelhakim, L. Lightfoot, an T. Li, Reliable ata fusion in wireless sensor networks uner byzantine attacks, IEEE Military Communications Conference, MILCOM, Nov. 011. [6] J. Ren, Y. Li, an T. Li, Spm: Source privacy for mobile a hoc networks, EURASIP J. Wireless Comm. an Networking, 010. [7] W. Heinzelman, A. Chanrakasan, an H. Balakrishnan, An applicationspecific protocol architecture for wireless microsensor networks, IEEE Transactions on Wireless Communications, vol. 1, no. 4, pp. 660 670, Oct. 00.