aggregation in single hop cell phone based sensor network incorporating mobility of cell phone users. We present a human mobility aware weighted clustering ...
2011 International Conference on Advanced Information Networking and Applications
Human Mobility Based Stable Clustering for Data Aggregation in Singlehop Cell Phone Based Wireless Sensor Network M. B. Shah, P. P. Verma, S. N. Merchant, U. B. Desai Electrical Engineering Department Indian Institute of Technology Bombay, India Email: (mehulshah,verma,merchant,ubdesai)@ee.iitb.ac.in
Abstract—Advances in 3G and 4G technology have offered many possibilities for developing novel applications using sensors embedded in hand held devices like cell phones. Mobility of cell phone based wireless sensor network has a critical issue of gathering sensed information in an energy efficient and delay sensitive manner. In this paper we provide a human mobility based stable clustering algorithm for data aggregation in single hop cell phone based sensor network incorporating mobility of cell phone users. We present a human mobility aware weighted clustering algorithm for data aggregation under Truncated Levy Walk (TLW) mobility model. Our approach is to select stable Cluster Heads (CH) to save the energy expenditure of network back bone formation. We have compared our algorithm with WCA [9] of mobile adhoc network and with MRECA [6] algorithm of mobile adhoc sensor network which we consider to be closely related with our work. WCA algorithm’s mobility parameter is not effectively capturing mobility of human walk. Our Human mobility aware Weighted Cluster based Data Aggregation algorithm (Hm-WCDA) effectively captures human walk characteristics and thereby stabilizes the back bone network. We have evaluated performance of our algorithm primarily with stability related parameters such as number of dominant set (DS) updates, number of reaffiliations and number of cluster heads; which directly effects the energy consumption of the algorithm. The simulation results show that our algorithm is more energy-efficient and reduces the energy consumption by 18 percent as compared to MRECA and by 9 percent as compared to WCA for cluster radius of 400m.
Figure 1.
Networks provides a superior performance over conventional cellular networks in terms of network capacity and coverage. This is also termed as Multihop Cellular Networks [3][4][5]. Embedded sensors within cell phones create a Multihop Cellular Sensor Network (MCSN) scenario [7][8]. In our work we assume single hop communication between mobile nodes and Direct Sink Access (DSA)[20] communication between mobile nodes and Base Station (BS). We call this architecture as Singlehop Cellular Sensor Network (SCSN). This allows us to avoid much of the overhead associated with medium access control and routing. This paper focuses on data gathering for SCSN. Traditional wireless sensor network (WSN) has a centralized sink(BS) where the data has to be forwarded, thus forming the many-to-one communication paradigm which results in unequal energy consumption by sensor nodes. The main challenge in WSN is to design the data gathering protocol in such a way that all the nodes consume almost equal energy. A lot of research has been done to prolong the network’s lifetime by periodically selecting CH from a group of nodes [15][16][17]. These algorithms normally work in rounds so as to dissipate equal amounts of energy from all the nodes under the assumption that the CH consumes more energy. The SCSN scenario closely relates with MANET and WSN although it has some major differences. Table-1 shows the major differences between WSN, MANET and SCSN. The node mobility creates difficulties in SCSN for designing the data gathering protocol in an effective manner as the
Keywords: Weighted Clustering, Data Aggregation, Human Mobility, Truncated Levy Walk
I. I NTRODUCTION Ubiquitous use of cell phones has inspired the idea of participatory sensing with sensors embedded in cell phones. Sensed data can be temperature, pressure, pollutant or toxicity levels. Efficient means are required to gather the sensed data in the network prior to storage at the base station. This is often known as data aggregation. Data aggregation helps to avoid congestion in the network by allowing innetwork processing at intermediate nodes and reducing buffer requirements for sensed data at the mobile nodes. Cell phones are already operating in a cellular infrastructure so the issue of deployment does not arise. Further the mobility of cell phone users provides improved coverage and energy efficiency [1][2]. Multihoping capability in Cellular 1550-445X/11 $26.00 © 2011 IEEE DOI 10.1109/AINA.2011.60
Singlehop cell-phone based sensor network
427
Table I C OMPARISON OF DIFFERENT NETWORKS WSN Nodes are stationary or quasi stationary. Many to one communication scenario. Energy resource is limited and recharging is not possible. Critical node (Nodes near to sink) issue. Data aggregation protocol operates in rounds to extend the network life time. Periodic rotation of CH role.
Mobile Adhoc Network Nodes are moving with moderate to high velocity One to one communication scenario. Energy resource is limited, but replacing of resource is possible. Node consume almost uniform energy. Data routing protocols are proactive,reactive or hybrid type. Node’s mobility influences selection of CH for cluster based routing.
as Weighted Cluster based Data Aggregation (WCDA) algorithm. Out of four parameters of WCDA, we have prioritized the mobility parameter for selecting stable CH. The non CH nodes directly communicate with CH and thereby form onehop clusters. In each BS-initiated round of data collection, the selected CHs receive data samples from their cluster members by 802.11 based RTS-CTS mechanism; and after data aggregation CHs send the data back to the BS on their dedicated data channel. We have given the related work in section II. In section III system model is described. In section IV we have discussed the WCDA algorithm. In the same section we have proposed the Hm-WCDA algorithm. The performance metrics and simulation set up are described in section V and VI followed by the results and conclusion in section VII and VIII respectively.
Singlehop Cellular Sensor Network Nodes mobility patterns correspond to human walk characteristics. Both types of communication scenario. Energy resource is limited and recharging is possible. Sensing operation should not overload primary communication. Challenge in gathering data with mobility of nodes in an energy efficient manner. CH changes due to node mobility. Stable CH should be selected.
II. R ELATED W ORK Weighted clustering algorithms have been proposed in adhoc and wireless sensor networks due to their advantages such as hierarchical organization, energy efficiency, bandwidth efficiency etc[9][10][11][12]. Many cluster based protocols had been proposed for data aggregation task in static wireless sensor network[15][16]. In [17][18] the effect of mobility on clustering has been investigated using a simple mobility (SM) model for mobile sensor networks (MSN). The aggregation techniques of MSN are closely related to that of SCSN. The sensor node’s mobility is preprogrammed in MSN while it is unpredictable in SCSN. The use of mobile phone for sensing was advocated in [19]. Authors of [7] have used layered approach for query based data aggregation for monitoring applications in MCSN under Random Way Point (RWP) mobility model. However the nodes were assumed stationary for the duration of data aggregation. We present method for data aggregation for queries spanning sufficiently long periods of time and under human mobility. In [20] the aggregation technique for WSN having mobile sink has been investigated. Our scenario is complementary to that of [20]. Our work is closely related with algorithms proposed in [6][9]. For counteracting node mobility, the authors of [6] have proposed a local maintenance mechanism, authors of [9] have proposed mobility parameter, while we propose human walk based mobility parameter to select the next set of CHs of the network.
topology of the network changes frequently. Our approach is to partition the network into clusters and stabilize the clusters by selecting Cluster Heads (CH) in such a way that these selected nodes suppose to remain more stable compare to other nodes of the network. Cluster’s member node’s movement will be handled by changing their affiliation to different CH. Due to CH’s movement two clusters may merge which accounts for reclustering. Reclustering or Dominant Set (DS) update will incur message exchanges through out the whole network and thereby is an energy consuming process. However each change of affiliation or reaffiliation process will incur only a single message overhead per node. DS update or reclustering process in mobile scenario will reduce collision of message and data packets. This entails an aggregation algorithm which maintains maximum stability under node movements. Our approach to clustering is to select a CH in a manner similar to [9]. our proposed algorithm selects more stable CHs compare to [9]. In [9] weights to mobile nodes have been assigned based on parameters like node degree, transmission power, mobility and CH count value. We have introduced one more parameter based on human walk characteristics and then we select the CH in a manner similar to that in [9]. As we are introducing a new system parameter based on human mobility, we call it Human mobility aware Weighted Cluster based Data Aggregation (Hm-WCDA) algorithm. We compare the performance of Hm-WCDA with WCA of mobile adhoc network and MRECA[6] of mobile adhoc sensor network. WCA[9] for SCSN scenario is termed
III. S YSTEM M ODEL Mobility models should emulate the behavior of targeted real life mobile user. Traditionally the RWP mobility model [14] is used for analysis of protocols in mobile ad hoc networks. For SCSN scenario the consideration of Truncated Levy Walk (TLW)[13] mobility model is more realistic. This model is based on about one thousand hours of GPS traces involving 44 volunteers in various outdoor settings and captures movement of the people in various outdoor scenarios
428
1) All N nodes are uniformly distributed and have variable transmission range. The Base Station (BS) is located at the center of the region. 2) All the nodes can operate in dual power mode. 3) The monitoring application requests for samples at the BS specifying the area of interest, the frequency of data collection and the maximum tolerable delay constraint on the reception of data at the BS. 4) The BS sends the query simultaneously to all nodes for data aggregation. 5) The BS also gives location information to all nodes through a separate control channel. The BS has a dedicated control channel for each node (akin to the existing networks). 6) The intra cluster communication is single hop and inter cluster communication is multihop. 7) Each step of the proposed algorithm takes a finite amount of time. Therefore, the algorithm terminates in finite amount of time. 8) The algorithm proposed here, take into consideration the fact that future technology (3G, 4G) promises further empowerment of hand-held devices, considering their ubiquitousness and potential use for various urban applications
such as campus area, theme park, fair, metropolitan area etc[13]. We have therefor considered the Truncated Levy Walk mobility model [13] for comparing our algorithms. In TLW mobility model a mobile node moves from its current location to a new location by randomly choosing a direction. The straight line distance between two consecutive visit is called a flight and this flight length follows a truncated power law distribution. The pause time between two consecutive visit also follows truncated power law or truncated pareto distribution. • Flight lengths follow a truncated power law P (l) ∝ |l|−(1+α) ; 0 < l < lmax •
(1)
where lmax is the maximum flight length. Pause times follow a truncated power law φ(t) ∝ t−(1+β) ; 0 < t < tmax
(2)
where tmax is the maximum pause time. Turning angles follow a uniform distribution. • Velocity increases as flight lengths increase. One of the properties of a power law distribution is that long flights are expected to be more frequent as compared to Gaussian or an Exponential distribution used in simple random walks. Evidently for humans the duration of moves or the distance moved in one ‘step’ (i.e flight length) is limited owing to physical constraints. This suggests that the power law characteristics hold true only in a limited range and that a truncated power law is more suitable for human walk[13]. The tail of the distribution may be long •
IV. S TABILITY AWARE DATA AGGREGATION A LGORITHMS This section describes the WCDA and Hm-WCDA algorithm for SCSN. A. Weighted Cluster based Data Aggregation Algorithm for SCSN
Figure 2.
The WCA algorithm is popular for Mobile Adhoc Network, but for SCSN environment we are modifying it as below. The important steps of the algorithm are as follows. Cluster Formation: 1) The Base Station broadcast (id,Rtx ,angle) packet to each node through a dedicated control channel. We assume that the BS has a dedicated control channel with each user, as in existing networks. 2) After receiving broadcast message each node finds all its neighbors (with transmission range Rtx ). Rtx is the expected transmission range for each node which depends upon the required resolution for the query. Each node periodically broadcasts N BRdis (Neighbor Discovery) message containing its own ID in order to handle mobile scenario. 3) Upon receiving N BRdis message from all its neighbors, each node can also estimate the sum of their mutual distances Su by received signal strength; where X Su = dist(v, v 0 ) (3)
Truncated Levy Walk mobility model’s waypoints
or short depending upon context of the location. The TLW distribution has a heavy tail up to a few kilometers[13]. This is a natural consequence of truncations due to geographical constraints. The main assumptions made for the system are as under:
v 0 ∈N (V )
429
4) After a certain time Tboostrap , each node receives all N BRdis message from its neighbors. Each node calculates the degree difference Du by the equation: Du = |nu − δ|
Cluster Maintenance: 12) Due to node mobility some of the nodes currently attached with a particular CH may join the cluster of some other CH. This is called reaffiliation. Reaffiliation is detected both at cluster head node and at ordinary node. If CH does not get ’Hello’ message periodically from its member node then it strikes that node off from its neighbor list. Similarly, if a node doesn’t hear a periodical ’Hello’ message from the CH, then it assumes that it has become a free node. If a free ordinary node hears a ’Hello’ message from some other CH then it will join it. Cluster Termination(DS Update): 13) Due to node mobility two CH may come within transmission range of each other. This situation is detected in a similar fashion as in step 12. Now these CHs no longer remain the dominant nodes. This situation should be avoided in order to reduce collisions at MAC layer. Hence in our algorithm this will end the current round and again invoke cluster formation as in step 1.
(4)
where nu is the number of neighbors of mobile node u and δ is the resolution parameter derived from Rtx . 5) Each node calculates the velocity parameter (Mu ) from consecutive location estimates obtained from the BS. 1p (Xt − Xt−1 )2 + (Yt − Yt−1 )2 (5) Mu (t) = Ts where Ts is the sampling interval. 6) The number of times the node has served as a CH is stored by the node as a parameter Pu ; which is indirectly related to the available battery power of the node. CH node consumes more energy as compared to non cluster head node as per equation 13 and 14. 7) All the nodes initialize their counter based on the count value Cu given by: Cu = w1 Du + w2 Su + w3 Mu + w4 Pu
8)
9)
10)
11)
(6)
where w1 , w2 , w3 , and w4 are arbitrarily chosen weighing factors for the clustering process satisfying P4 w i=1 i = 1. The lesser is the count value (Cu ), sooner will the node announce itself as a CH. The BS transmits information about the maximum value of the above parameters attained in the network to all the nodes. All the nodes normalize their parameters with respect to this value. The counter with the lowest count value expires first. Then it declares itself as a Cluster Head by transmitting ‘CHdeclare ’ message to its neighbors. This message consists of node id field, cluster id field and score field. In the case of CH, the first two fields are the same. Upon receiving ’CHdeclare ’ packet the nodes in the neighborhood of CH reset their counters and join the cluster formed by the CH by transmitting back a ‘CHaccept ’ message. This also contains the node id and the cluster id field. If the neighboring node is already a member of some other cluster then it will ignore the CHdeclare message. After the clustering starts, the procedure will terminate after Tstop (which is finite). The maximum value of Tstop is proportional to the maximum value a counter can take (denoted by Cmax ). After finite time Tstop all the cluster member nodes transmit their sensed data to the CH as per 802.11 based RTS-CTS messaging scheme. Then the CH aggregates the data packet and sends it to the BS as shown in figure 1. This communication is done with high transmitting power.
Figure 3.
Flow-chart of proposed clustering algorithm
B. Human mobility aware Weighted Cluster based Data Aggregation (Hm-WCDA) algorithm Clustering technique of network organization is a hierarchical technique in which cluster heads form the Dominant Set (DS). We have modified the WCA protocol by adding one human walk context mobility parameter called Hu . 1 H(u) = (Slen(u) −4) 1+e
430
Slen(u) is the super length counter value. The super length counter value is incremented every time a node traverses a super length. Super length: The TLW distribution has a heavy tail up to a few kilometers[13]. This is a natural consequence of truncations due to geographical constraints. The truncations are excessive if the area is small and/or highly crowded, which encourages short flights and discourages long flights. Thus a small crowded area will not have a heavy tail and the flight length distribution can fit well even to a short-tailed distribution. However with a really large area, the truncations will not be excessive. When truncations have less impact on flight lengths, the mobility of the nodes has a stronger powerlaw tendency[13]. This behavior is incorporated as a “Super Length”. The super length parameter won’t hold for small simulation areas (approx. 200 meters)[13]. The threshold for determining whether a given flight is a super length or not is a subjective one. It is very much influenced by geographical conditions and mobility environment. However it is always a large percent of simulation area. For our simulation, we have taken 85 percent as an empirical super length threshold, which has been observed through experimentation and observation on TLW mobility model. Thus Super length is that flight (continuous walk without any pause) of the node whose length is more than 85 percent of the width of the area. The parameter Hu is inversely proportional to super length count value. The weighing parameter of Mu selects the node with lowest velocity as a CH. In addition to current node mobility, the history of mobile node’s flight length should be taken into account for selecting a stable CH. That is done by parameter Hu . This newly introduced parameter by us captures the power-law tendency of human walk. To do so it remembers the history of mobile node’s flight lengths. Incorporation of Hu helps in selecting more stable CHs as evident from experimental results. Due to the truncated power law property, the node which has traversed super lengths few times should have lesser priority for being selected as a CH. The Hm-WCDA algorithm is as follows: Steps 1-7 are similar to the steps described in part-A. 1) Each node finds Hu where 1 1 + e(Slen(u) −4) = 0
H(u) =
; if Slen(u) > 0
H(u)
; if Slen(u) = 0
C.
Start of Algorithm 1. 2. 3. 4. 5.
D.
Reception of CHdeclare message 1. 2. 3. 4.
E.
if (Cid == ownid) then if (own Cid == N U LL) CHaccept in RT X [ownid, Cid , weight] end
Up Reaffiliation 1. 2. 3. 4.
F.
count = w1 D + +w2 Su + w3 M + w3 P + w30 H if (counter < 0) then CHdeclare with Rtx [ownid, own Cid = ownid, weight] else delay end
if (D is CH) then invite to join cluster (send cluster info) else do nothing end
Down Reaffiliation 1. 2. 3. 4. 5.
if (D is CH) then if node in Neighbour level remove from Neighbour list else do nothing end
Figure 4.
Pseudo-code of the algorithm
Remaining steps are the same as steps 9-14 described in part-A. V. P ERFORMANCE METRICS The performance of data aggregation schemes for monitoring applications can be evaluated based upon: 1) Delay Analysis: The delay is time difference between the time interval of query generation event and aggregated data reception event at the BS. Delay =
Cluster Formation Time +Aggregation Time + Propagation Time
Within a cluster, the maximum delay is given by: max(disti→CHi ) (9) c The second term is the time taken for the last data packet within the cluster k to reach CHk . c is the velocity of radiowave propagation and max(disti→CHi ) is the distance of the farthest node amongst all i ∈ Kk from CHk . Thus the maximum delay (ignoring processing delays and control packet exchanges) in propagating the L data packets to BS is: Tcluster (k) = Tstop +
distCHk →BS (10) c As the other terms are significantly small, the maximum delay is primarily dependant upon the value of Tstop (which is a constant for the network). Hence equation 10 clearly describes the delay in the network. 2) Energy Consumption: The algorithm must be energy efficient so that it does not drain resources away from the primary intended use of cell phones for voice/data communication. Unlike WSN, there is no first node Tmax = max Tcluster (k) +
(7)
Each node finds its super length by estimating the distance from received signal strength in some time interval. 2) Initialize counters of all the nodes with the following equation. Cu = w1 Du + w2 Su + w3 Mu + w4 Pu + w30 Hu (8)
431
DS update will consume Ed units of energy, where
death phenomenon in SCSN due to recharging capability of cell phones. We are calculating the total energy consumption of all nodes for various cluster radii. Our energy model for the sensors is based on the first order radio model described in [15]. A sensor consumes Eelec = 50nJ/bit to run the transmitter or receiver circuitry and Eamp = 100pJ/bit/m2 for the transmitter amplifier. Thus, the energy consumed by a sensor i in receiving a k-bit data packet is given by ERx = (Eelec · k)
Ed =
i=1
Etot
=
(11)
(12)
(Eelec · k · CHdegree + EDA · k) +(Eelec · k + Eamp · d4toBS · k) (13)
where k is the number of bits in each data message, CHdegree is the number of ordinary nodes in the cluster, dtoBS is the average distance from a CH to the base station, and we assume lossy data aggregation with the energy for aggregation to be EDA . As for each non-CH node, it only needs to transmit its data to the CH once during a round. Since the distance to the CH is small, the energy dissipation follows the Friss free space model. Thus, the energy used in each non-CH node is EnonCH = (Eelec · k) + Eamp · Rtx · k
VI. S IMULATION S CENARIO The simulation is implemented in MATLAB. We have applied MRECA, WCDA and Hm-WCDA on the TLW model. Monte Carlo simulations have been carried out for 50 random seeds and for varying cluster radii (50m to 500m). Due to the importance of showing robustness against node mobility, the weight w2 associated with Mu was chosen high. The values used for MRECA algorithm are w1 = 0.5, w2 = 0.5 and w3 = 0.0. The values of weights for WCDA are w1 = 0.2, w2 = 0.0, w3 = 0.7 and w4 = 0.1. The values used for Hm-WCDA algorithm are w1 = 0.0, w2 = 0.0, w3 = 0.4, w4 = 0.1 and w3 0 = 0.5. To show the relative effect newly introduced parameter, we have kept weight values of all parameter except w3 and w5 same and had analyzed the effect. This weight combination has been termed as Hm-R WCDA. The values used for Hm-R WCDA algorithm are w1 = 0.2, w2 = 0.0, w3 = 0.5, w4 = 0.1 and w3 0 = 0.2. Next, we apply Hm-WCDA algorithm on TLW for 200 nodes for various weight combinations in such a way that weights w1 = 0.1, w2 = 0.1 and w4 = 0.1 are constant amongst various weight vectors and weight w3 and w3 0 are relatively changed so that overall sum remains 1. This is for observation of relative importance of weight vectors w3 and w3 0 .
(14)
Table II S YSTEM PARAMETERS Parameters Sensor Field Number of Nodes (N) BS location Packet header size Initial Energy Data packet size Eelec Ef s Emp Ef usion Sampling Time Total Simulation Time
Values 1000 × 1000 rectangle area 400 (500, 500) 25bytes 2J/battery 500bytes 50nJ/bit 10pJ/bit/m2 0.0013pJ/bit/m4 5nJ/bit/signal 1 minute 120 minutes
Reaffiliation will consume Er units of energy, where Er Er
= =
(18)
where K is the total number of CH, R is the total number of reaffiliations, D is the total number of DS updates, k’ is size of ’Hello’ packet. 3) Network Lifetime (Energy Efficiency): It is a primary metric for evaluating the performance of a sensor network. The common definitions include the time until the first or the last node in the network depletes its energy. Periodic reclustering improves network lifetime and thereby energy efficiency. 4) Number of reaffiliations: The reaffiliation count is incremented when a node gets dissociated from its cluster head and becomes a member of another cluster within the current dominant set. 5) Number of DS updates: The dominant set update takes place when a node can no longer be a neighbour of any of the existing cluster heads. These parameters are studied for varying cluster radii for TLW mobility model.
’Rtx ’ is the transmission range. Each CH uses energy in receiving data signals from its members, aggregating the data and transmitting it to the BS represented using the multi path model. Thus, the energy spent by a CH node during a single round is ECH
K X = E(k) + (Er · R) + (Ed · D) l=1
While the energy consumed in transmitting a data packet to sensor j is given by 2 ET x = Eelec · k + Eamp · Rtx ·k
N X (Eelec · k 0 ) + (Eamp · Rtx )2 · k 0(17)
VII. S IMULATION R ESULTS
(Eelec · k) + (Eamp · d2 ) at CH (15) (Eelec · k) at ordinary node. (16)
Figure 5 shows reaffiliation of mobile nodes due to mobility for MRECA, WCDA and Hm-WCDA algorithms.
432
Figure 5.
Average no. of reaffiliations vs cluster radius
Number of reaffiliations are less for smaller cluster radius due to frequent dominating set updates at these radii.
Figure 7.
Figure 8.
Figure 6.
Dominant set updates vs cluster radius
Total energy consumption of nodes
Number of nodes alive for 2 hour simulation time
node death is considered as definition for network lifetime. If 50% node death is considered as network lifetime, then Hm-WCDA has 22% improvement over MRECA and 12% improvement over WCDA.
Figure 6 shows DS updates for MRECA, WCDA and Hm-WCDA algorithms. When the cluster radius is less, the clusters are of smaller size and nodes have a higher probability of moving out of the cluster. As the radius increases, the number of set updates required decreases, as the nodes tend to stay within the same cluster in spite of their movements. The number of DS updates for Hm-WCDA reduces by 31% as compared to MRECA and by 15% as compared to WCDA for cluster radius of 300m. Figure 7 shows energy consumption for MRECA, WCDA and Hm-WCDA algorithms. Energy consumption increases with increasing cluster radius as nodes have to send data to the CH over a longer distance. The energy consumption for Hm-WCDA reduces by 18% as compared to MRECA and by 9% as compared to WCDA for cluster radius of 400m. Figure 8 shows the No. of nodes alive vs simulation time. Hm-WCDA algorithm consumes more uniform energy but at the expense of a few nodes having high energy consumption. These high energy consumption nodes correspond to stable CH nodes and this has resulted in early first node death. Hm-WCDA algorithm has higher network life time if 20%
Figure 9.
Average no. of reaffiliations vs cluster radius
Figure 9 shows reaffiliations for Hm-WCDA for different weight combinations of w3 & w5 . Figure 10 shows DS updates for Hm-WCDA algorithm for various combinations of weights w3 and w3 0 . It is observed
433
[6] J.H. Li, M. Yu, R. Levy and A. Teittinen, A mobility resistant efficient clustering approach for ad hoc and sensor networks, Mobile Computing and Communications Review, 10 (2), pp. 1-12, 2006. [7] D.Chander,B.Jagyasi,U.B.Desai,and S. Merchant, Layered data aggregation in cell-phone based wireless sensor networks, International Conference on Telecommunications, 2008. [8] D.Chander,B.Jagyasi,U.B.Desai,and S. Merchant, DVD Based Moving Event Localization in Multihop Cellular Sensor Networks, IEEE International Conference on Communications, 2009. Figure 10.
[9] M.Chatterjee, S. Das, and D.Turgut, WCA: A Weighted Clustering Algorithm for Mobile Ad Hoc Networks, Cluster Computing, vol. 5, pp. 193-204, 2002.
Dominant set updates vs cluster radius
that DS update is minimum for w3 =0.5 and w3 0 =0.2. Hence this weight combination should be preferred.
[10] M. Gerla and J. T. Tsai, Multiuser, Mobile, Multimedia Radio Network, Wireless Networks, vol. 1, pp. 25565, Oct. 1995
VIII. C ONCLUSION
[11] S.K. Dhurandher and G.V. Singh, Weight-based adaptive clustering in wireless ad hoc networks, IEEE International Conference on Personal Wireless Communications, New Delhi, pp. 95-100, 2005.
This paper presents a human mobility based stable clustering algorithm for data aggregation in SCSN. Timer based cluster formation has resulted in global optimum selection of CHs without significant communication overhead. Incorporation of human mobility based parameter has improved the performance of existing WCDA & MRECA algorithms. The present framework targets stability of cluster structure for data aggregation against node mobility. It would be interesting to modify the algorithm further in order to incorporate more characteristics of human mobility such as power law intercontact time and pause time distributions.
[12] Brust M.R.; Andronache A.; Rothkugel S., WACA: A Hierarchical Weighted Clustering Algorithm Optimized for Mobile Hybrid Networks, Third International Conference on Wireless and Mobile Communications, 2007. [13] I. Rhee, M. Shin, S. Hong, K. Lee, and S. Chong, On the levy walk nature of human mobility, IEEE Conference on Computer Communications, Phoenix, AZ, April 2008. [14] J.-Y. L. Boudec and M. Vojnovic, Perfect simulation and stationarity of a class of mobility models, IEEE Conference on Computer Communications, Miami, FL, 2005.
IX. ACKNOWLEDGMENT This work has been supported by the Department of Science and Technology (DST)-Government of India, and the India-UK Advanced Technology Center of Excellence in Next Generation Networks (IUATC) Project. R EFERENCES
[15] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, An applicationspecific protocol architecture for wireless microsensor networks, IEEE Transactions on Wireless Communications vol. 1, no. 4, pp. 660670, 2002. [16] O. Younis, S. Fahmy, HEED: A Hybrid, Energy-Efficient, Distributed clustering approach for Ad Hoc sensor networks, IEEE Transactions on Mobile Computing vol. 3, no. 4, pp. 366379, 2004.
[1] X. Du and F. Lin, Improving sensor network performance by deploying mobile sensors, IEEE International Performance, Computing, and Communications Conference, pp. 67-71, 2005. [2] M. Zhang, X. Du, and K. Nygard, Improving coverage performance in sensor networks by using mobile sensors, Military Communications Conference, 2005.
[17] C.-M. Liu, C.-H. Lee, Power efficient communication protocols for data gathering on mobile sensor networks, Proceedings of the 60th IEEE Vehicular Technology Conference, 2004.
[3] Y.-D. Lin and Y.-C. Hsu, Multihop cellular: A new architecture for wireless communications, IEEE Conference on Computer Communications, pp. 1273-1282, 2000.
[18] Liu, C.M., Lee, C.H., Wang, L.C.,Distributed clustering algorithms for data-gathering in wireless mobile sensor networks, Journal of parallel and distributed computing, 2007. [19] A. Kansal, M. Goraczko, and F. Zhao, Building a sensor network of mobile phones, Information Processing in Sensor Networks, April 2007.
[4] L. Le and E. Hossain, Multihop cellular networks: Potential gains,research challenges, and a resource allocation framework, IEEE Communications Magazine, vol. 45, no. 9, pp. 66-73, September 2007.
[20] M. Lotfinezhad, B. Liang, E. Sousa, Adaptive cluster-based data collection in sensor networks with direct sink access, IEEE Transactions on Mobile Computing 7 (7) 884-897, 2008.
[5] G.Kannan, S. Merchant, and U.B.Desai, Cross-layer routing for multihop cellular networks, Advanced Information Networking and Applications Workshops, pp. 165-170, 2007.
434
