An Optimized Lifetime Enhancement Scheme for Data Gathering in Wireless Sensor Networks

Transcription

1 An Optimized Lifetime Enhancement Scheme for Data Gathering in Wireless Sensor Networks Ayon Chakraborty, Kaushik Chakraborty, Swarup Kumar Mitra 2, M.K. Naskar 3 Department of Computer Science and Engineering, Jadavpur University, Kolkata 32, India. 2 Department of Electronics &Communication Engineering, MCKV Institute of Engineering, Howrah, India. 3 Advanced Digital and Embedded Systems Lab, Department of ETCE, Jadavpur University, Kolkata 32,India. Abstract Design of energy efficient schemes for data gathering is an important concern for lifetime enhancement of wireless sensor networks. Variation in the distances of nodes from the Base Station and differences in inter-nodal distances are primary factors causing unequal energy dissipation among the nodes. Thus energy difference between the various nodes increases with time resulting in degraded network performance. The LEACH and PEGASIS schemes which provided elegant solutions to the problem suffer basic drawbacks due to randomization of cluster heads and greedy chain formation respectively. In this paper, we propose an Optimized Lifetime Enhancement (OLE) Scheme which shows enhanced performance over these schemes. OLE increases the network performance by ensuring a sub-optimal energy dissipation of the individual nodes despite their random deployment. It employs modern heuristics like particle swarm optimization instead of the greedy algorithm as in PEGASIS to construct energy efficient routing paths. Extensive simulations validate the improved performance of OLE. Keywords: Network Lifetime, Wireless sensor networks, Data Gathering, Particle Swarm Optimization, Simulated Annealing I. INTRODUCTION Wireless sensor networks (WSNs) consist of sensor nodes that are randomly deployed in a large area, collecting important information from the sensor field. These sensor nodes have very limited energy resources and hence, the energy consuming operations such as data collection, transmission and reception, must be kept at a minimum []. Also, it is often infeasible to replace or recharge the sensors nodes deployed in inaccessible terrains. The sensor networks are required to transmit gathered data to a base station (BS) or sink, often distantly located from the sensor field. Network lifetime thus becomes an important metric for efficiency of a sensor network. In case of WSNs, the definition of network lifetime is application specific [2]. It may be taken as the time from inception to the time when the network becomes nonfunctional. A network may become non-functional when a single node dies or when a particular percentage of nodes perishes depending on requirement. However, it is universally acknowledged that equal energy dissipation for equalizing the residual energy of the nodes is one of the keys for prolonging the lifetime of the network [2]. Each sensor node is provided with transmit power control and omni-directional antenna and therefore can vary the area of its coverage [3]. It has been established in [4] that communication requires significant amount of energy as compared to computations. In this paper, we consider a wireless sensor network where the base station is fixed and located far off from the sensed area. Furthermore all the nodes are static, homogeneous, energy constrained and capable of communicating to the BS. The LEACH protocol [4] presents an elegant solution to this energy utilization problem where nodes are randomly selected to collaborate to form small number of clusters and the cluster heads take turn in transmitting to the base station during a data gathering cycle. The PEGASIS protocol [5] is a further improvement upon the LEACH protocol where a greedy chain of nodes is formed which take rounds in transmitting data to the base station. This problem was even approached by modern heuristic techniques like Ant colony Optimization [6], trying to optimize energy dissipation. [6] tried to form an optimized chain for data gathering. In this paper, we approach the problem from a new viewpoint. In our Optimized Lifetime Enhancement (OLE) scheme a chain is formed, but instead of allowing all nodes to become the leader, to communicate with the base station the same number of times, the network lifetime is increased by allowing the individual nodes to transmit unequal number of times to the base station depending on their residual energy and location. Furthermore, instead of forming a greedy chain, which may not always ensure minimum energy dissipation, we make use of modern heuristic optimization techniques like Particle Swarm

2 Optimization (PSO) [7] and Simulated Annealing (SA) [8]. This results in an enhanced network performance as balanced energy dissipation by the individual nodes is achieved in the network. The results obtained in the OLE scheme shows encouraging improvements over PEGASIS, LEACH and ACO schemes. The algorithm for the OLE scheme was implemented in nesc [9] for the TinyOS [0] software platform. This not only signifies the coding feasibility of our scheme, but also verifies it for running on real hardware platforms (embedded systems like MicaZ or Mica2 sensor motes). The packet reception ratios in these schemes has also been studied using the interference model offered by TOSSIM [] environment. The rest of the paper is organized as follows: Section II describes the energy dissipation model and Section III judges the emergence of an energy-efficient data gathering protocol followed by the gradual development of our scheme. Our scheme is evaluated by results obtained from extensive simulation in Section IV. Finally, we conclude in Section V. II. THE ENERGY DISSIPATION MODEL Our aim in this paper is to minimize the energy usage in the sensor nodes by formation of an optimal chain through which the data gathering will take place. For this purpose, we assume the radio model as discussed in [4] for the radio hardware dissipation. This is one of the most widely used models in sensor network simulation analysis. The energy spent in transmitting a k- bit packet over a distance of d meters, is given by: E tx(k,d)=(ξ elec + ξ amp * d n ) *k () and that for receiving the packet is, E rx(k)= ξ elec* k (2) Here ξ elec (45nJ/bit) is the energy dissipated per bit to run the radio electronics and ξ amp is the energy expended to run the power amplifier for transmitting a bit over unit distance. n is the path loss exponent, whose value enhances with increasing channel non-linearities (usually, 2.0 n 4.0). In our approach, we have used both the free space (distance 2 power loss) and the multipath fading (distance 4 power loss) channel modes. In our model, we assume, that inter-nodal distances are small compared to distance between the nodes and the Base Station (BS). Thus for communication among sensors we take n = 2, and that between the leader and BS, we take n = 4, in equation (). Value of ξ amp = 0pJ/bit/m 2 for n = 2 and 0.00pJ/bit/ m 4 for n = 4. Now for all practical purposes, we can assume that the computational energy is much less than the communicational energy and thus can be neglected. Thus for the chain of length n, the total energy expended in data gathering is the summation of the energy used by the individual sensor nodes and the leader. Assuming a constant packet size of k = 2000 bits, n E total= [ (ξelec+ξamp*di2 ) +(ξ elec+ξ amp* D 4 ) ]* k (3) i = In equation (3) d i denotes the distance between the (i+) th node and the i th node in the data gathering chain. D is the distance between the leader and the base station. The values of ξ elec and ξ amp are stated earlier. Here we impose a threshold value on d i as d TH. This ensures reliable communication in between the nodes reducing unwanted noise and packet loss probability. It is also assumed that the channel is symmetric so that the energy spent in transmitting from node i to j is the same as that of transmitting from node j to i for any given value of SNR. III. PROPOSED OPTIMIZED LIFETIME ENHANCEMENT SCHEME The PEGASIS scheme [5] depends upon a greedy chain formation whereas the LEACH scheme [4] randomizes the leader selection in the network. While the greedy chain can not always guarantee minimal energy consumption, the randomized leader selection does not take into account the node's capability in being the leader, in terms of its energy content and transmit distance. Keeping the above drawbacks in mind, we proceed to form a suboptimal chain for data gathering and device a scheme to choose an efficient leader for communicating to the base station. A. Chain formation based on Particle Swarm Optimization In order to avoid trapping at the local minima and increase the diversity of the swarm, the simulated annealing algorithm [8] is applied to PSO [7] to solve the problem of chain formation. The total number of nodes being n, the solution space U can be said to be a collection of arrangements of {,2,3,,n} where two consecutive numbers denote a direct link between those nodes. Thus every arrangement C j represents a chain, where U = {C j C j is a permutation of (,2,.. n)}. So C j denotes the j th particle in our n-dimensional system. Energy Function: The energy function for the SA algorithm is designed as, f(c j) = n di2 (4) i= The above equation is derived from equation (3), we have considered the terms (in the chain) related to distance only. When a particle C old updates its position to C new, f = f(c new) f(c old) is used as E representing the energy difference the two energy states. We start by guessing an initial solution, and proceed, resulting in a solution with smaller energy value, than the last solution. In case of a larger energy value, the decision to accept

3 or reject the solution, is determined by the probability function below, in equation (5), P = if E 0 = exp (- f/ Ө) if E > 0 (5) If E is not positive, implying lower energy value of the new solution, it is accepted; else the acceptable probability P is calculated as in equation (5). If P > rand (0, ), a random number between 0 and, the new solution is also accepted else it is rejected. Cooling Schedule: One of the most important control parameter in equation (5) is Ө, called the annealing temperature; a parameter which is decremented, every time the system of particles approaches a better solution (or a low energy state). If Ө i be the initial temperature and Ө f be the final temperature, and t be the cooling time, Ө(t) = Ө f + (Ө i - Ө f )*α t (6),is the designed cooling schedule. Here α is the rate of cooling, (usually, 0.7 α <.0) and t is the cooling time. For our purpose we considered t as the number of iterations. The SA algorithm incorporates the concept of probability through the Metropolis acceptance rule [2] into the fast optimal search ability of PSO, a new algorithm is proposed for the optimal energy chain formation. B. Proposed algorithm for chain formation Input: A set of N sensor nodes, randomly deployed in the sensor field, and a base station. Step : Initialization :- At first, a swarm of m particles selected at random is initialized which are expressed as: C, C 2, C 3. C m. Ci={node[], node[2],...,node[n]}, where node[i] = j means that the i th member (node) of the chain has id j. The parameters Ө i (initial temperature), Ө f (final temperature), α (cooling rate) are initialized. The higher the initial temperature, the better the result is. At low temperature, every particle finds its local best chain C ilbest in its local area. L is maximum number of iterations at a certain temperature and t is the maximum number of iterations for the total process (t is analogous to the maximum cooling time). Step 2 : Finding a local best chain :- For all the m particles, each one searches for its local best chain, at a particular temperature Ө, for L iterations. This searching is done by a random binary swapping, where two positions in a chain (C old) are randomly selected and exchanged, resulting in a new chain (C new). The new chain is checked such that distance between the individual nodes do not exceed d TH or else another swapping is done. The old chain is updated by the newly formed chain according to the acceptance rule as stated in equation (5). Step 3 : Updating the pbest and gbest values :- For each particle the C ilbest chain obtained in Step 2 is compared to the historically obtained best chain C ipbest for that particle. Again, C ipbest is updated by C ilbest according to the following rule : C ipbest = C ilbest if { f( Cilbest) f(c ipbest) } < 0 = C ipbest if { f( Cilbest) f(c ipbest) } 0 (7) Now, comparing all the C ipbest values, C gbest is updated by that C ipbest which has minimum energy state i.e. f(c ipbest). Step 4 : Formation of a new chain :- Based on the global knowledge of the swarm each particle forms a new chain from its original best chain (C ipbest) and the globally obtained best chain (C gbest) by the crossing method as discussed in [3]. For eg. say, C ipbest = {4,5,2,3,6,} and C gbest = {5,2,,4,3,6}. The slot {2,,4} is randomly chosen from C gbest and inserted in the same position in C ipbest and the node ids that are repeated are deleted. Thus C inew becomes {5,2,,4,3,6}. After the crossover, the energy state of C inew is compared with that of C ipbest and the best one (i.e. with lower energy state) is taken as the new individual best position. The crossover can help the particles jump out of the local optimization by sharing the global information about the swarm. Step 5: Loop :- The temperature Ө(t) is calculated. If its value is less than or equal to Ө f or the total number of iterations up to now exceeds the value of t, the algorithm comes to a halt. The best chain formed is C gbest. Else go to Step 2. C. Leader selection phase Once the sub-optimal chain is formed we look for the node which has the maximum value of E resi /D 4. Here E resi denotes the residual energy of an individual node before starting a data gathering round and D is the distance of the base station from that node. The node with the maximum value of E resi /D 4 becomes the leader. Here we consider the multipath fading (distance 4 power loss) channel mode, as the leader is concerned with communicating to the distant base station. IV. SIMULATION A.. Hardware Implementation The OLE scheme was implemented in nesc [9], a component based dialect of the C programming language, meant to be hosted on the TinyOS [0] software platform, an operating system which runs in the Berkeley Mica Motes like Mica2 or MicaZ. But, in absence of real motes, we simulated OLE in the TOSSIM [] environment, which acts as a simulator for TinyOS platform. Thus the implementation and coding of the algorithm in

4 nesc justifies the feasibility of OLE for real hardware platforms, with limited memory and resources. Since, this algorithm requires centralized knowledge about the sensor network, it would be best to carry out the algorithm in the Base Station (BS) and disseminate the result in the network before initiating data gathering tasks. However, this could well be dependent upon the application itself. In case, frequent communication with the BS is not feasible for all the nodes, this chain formation algorithm can also be applied in individual clusters in the sensor field, where these computations can be done by a local leader in each cluster. This will not only use up less resources as the number of nodes in a cluster is limited but also result in equalized energy dissipation among the local leaders. Secondly, this sort of distributed computation will speed up the process of self-organization of the network. Finally the BS could connect these local leaders to form the final optimized chain. B. Simulation Overview To evaluate the performance of the OLE scheme extensive simulations were performed on several random 00 node networks in a 50m*50m field as in [5]. Simulations performed in MATLAB show that OLE scheme outperforms the data gathering schemes like PEGASIS [5] and ACO [6]. This readily implies the efficiency of our method over LEACH [4]. As mentioned in Section III, for implementing our energy efficient data gathering protocol the chain formation was done by Particle Swarm Optimization with Simulated Annealing. Simulation results are shown in Table I. The base station was located at (25m, 50m) and energy per node was varied. As mentioned earlier, while comparing PEGASIS, ACO and OLE schemes, a common threshold was introduced as the inter-nodal distance. A second simulation was conducted in TOSSIM to study the packet reception ratios in the three schemes. The Simulation process in TOSSIM[] considers the TOSSIM radio loss model, shown in figure, which is based on the empirical data. The loss probability captures transmitter interference using original trace that yielded the model. More detailed measurements would be required to simulate the exact transmitter characteristics; however experiment have shown the model to be very accurate. Here a fixed number of 20 packets were considered for transmission from a network of 20 nodes, where each node had a packet to send, to the BS. But due to the factor of packet loss and unreliability of the links, all the packets could not ultimately reach the BS successfully, if no retransmission attempts are made. Retransmission attempt simply means that when node A sends a packet to node B, and the process fails, node A tries to resend the packet to B. The maximum number of times this process can happen is called the Maximum Retransmission Attempts (MRA). We intend to show here, how the number of successful packet transmission increases with increasing number of MRAs for the different schemes. Packet Reception Ratio (PRR) is defined as the ratio of the number of packets delivered successfully to the total number of packets sent. Hence, as the successful number of packet transmission increases, an increase in PRR also takes place. Thus this simulation also helped us to take into account any interference caused by noise, as if in a real life environment. Also the radio interference model used for simulation purposes helped us to study the problem from the perspective of a more realistic physical layer. Detailed results for the outcome are shown in Table II. Fig.. The mean packet loss rate versus distance is shown, with error bars indicating one standard deviation from the mean. The model is highly variable at intermediate distances.tossim radio loss model based on empirical data C. Simulation Results In this section we show the results obtained in simulating our algorithm. Table I demonstrates the enhancement of network lifetime compared to the other schemes. From figure 2, we find that OLE largely outperforms PEGASIS, and also the chain obtained by ACO as in [6]. It also reveals that OLE performs better than both ACO and PEGASIS till about 70% of nodes in the network are dead. Networks with over 70% of nodes dead are very inefficient and therefore the degradation of performance of our schemes under these conditions can easily be ignored keeping in mind the superior performance with lesser percentage of dead nodes.

5 TABLE I NUMBER OF DATA GATHERING ROUNDS FOR VARIOUS SCHEMES WITH PERCENTAGE Intial Energy (J/node) N U M B E R O F R O U N D S Protocol PEGASIS ACO OLE PEGASIS ACO OLE PEGASIS ACO OLE OF DEAD NODES PERCENTAGE OF DEAD NODES TABLE II MAXIMUM AND MINIMUM NUMBER OF SUCCESSFUL TRANSMISSIONS WITH INCREASING MRAs OLE ACO PEGASIS MRA Min Max Min Max Min Max Figures 3. and 3.2 portray the chains formed by greedy and OLE algorithm respectively, for the same node distribution in a 50m x 50m field. It depicts clearly how the inter-nodal distances are bound to increase when the greedy algorithm is used. The bold lines in red indicate the inter-nodal distances which are larger than the threshold. The leader node is also shown as a red circle. Fig. 3. : Greedy Chain Fig. 2. Performance analysis of different protocols with Energy/node J and base station at (25,50). The packet loss rates has been compared in Table II. It shows how the number of successful packet transmissions (out of the total of 20 packets) increase with increasing the retransmission attempts. For a particular value of MRA the simulation has been conducted for 0 times. The figures show the maximum and minimum number of packets, delivered successfully to the BS for different values of MRAs. Here also, we see, that packet loss in OLE scheme is less compared to PEGASIS or ACO schemes. Thus less number of retransmissions are necessary on the average. This is also an important aspect of OLE in terms of energy efficiency. Fig. 3.2 : Chain by OLE Scheme V.CONCLUSION AND FUTURE WORK The protocols considered in this paper ensures that a near optimal energy utilization occurs thereby increasing network lifetime as is validated by simulation results. The Particle Swarm Optimization

6 along with Simulated Annealing helps to enhance the performance of our scheme. Reports of applications of using these meta-heuristic tools have been widely published, thus forming a solid background. Developing solutions with these tools offers two major advantages: (i) Development time is much shorter rather than using more traditional approaches. (ii) The systems are very robust, being relatively insensitive to noisy and/or missing data. Moreover, the OLE scheme has been coded in nesc, which justifies it to be feasible on real motes. Also, we have considered the TOSSIM interference model, while simulating packet loss rates for the various schemes. This simulation helped us to compare the reliability of the schemes for successful packet delivery, as if in a real life environment. All the results we obtained are in total compliance with our objective. Thus our OLE scheme in true sense plays a good role in enhancing the lifetime of a sensor network by optimizing the routing paths. We have already developed the chain using SA-PSO, and also have compared it to the ACO technique. Our future goal is to study the problem using Genetic Algorithms and compare it to the OLE scheme. [9] David Gay, Philip Levis, David Culler, Eric Brewer, nesc. Language Reference Manual,May [0] Philip Levis,TinyOS Programming, June 28, [] P. Levis, N. Lee, M. Welsh, and D. Culler. TOSSIM: Accurate and Scalable Simulation of Entire TinyOS, [2] N. Metropolis et. al. J. Chem. Phys (953). [3] Zhi-Feng Hao, Zhi-Gang Wang; Han Huang, A Particle Swarm Optimization Algorithm with Crossover Operator, International Conference on Machine Learning and Cybernetics 2007, pp -9-22, Aug REFERENCES [] Clare, Pottie, and Agre, Self-Organizing Distributed Sensor Networks,In SPIE Conference on Unattended Ground Sensor Technologies and Applications, pages , Apr [2] Yunxia Chen and Qing Zhao, On the Lifetime of Wireless Sensor Networks, Communications Letters, IEEE, Volume 9, Issue, pp: , DigitalObjectIdentifier0.09/ LCOMM , Nov [3] S. Lindsey, C. S. Raghavendra and K. Sivalingam, Data Gathering in Sensor Networks using energy*delay metric, In Proceedings of the 5 th International Parallel and Distributed Processing Symposium, pp , 200. [4] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, Energy- Efficient Communication Protocol for Wireless Microsensor Networks, IEEE Proc. Of the Hawaii International Conf. on System Sciences, pp. -0,January [5] S. Lindsey, C.S. Raghavendra, PEGASIS: Power Efficient Gathering in Sensor Information Systems, In Proceedings of IEEE ICC 200, pp , June 200. [6] Ayan Acharya, Anand Seetharam, Abhishek Bhattacharyya, Mrinal Kanti Naskar, Balancing Energy Dissipation in Data Gathering Wireless Sensor Networks Using Ant Colony optimization,0th International Conference on Distributed Computing and Networking-ICDCN 2009, pp , January 3-6, [7] Eberhart, R. C, Kennedy, J. A new optimizer using particle swarm theory, 995. [8] Kirkpatrick S, Simulated Annealing, Sci, Vol 220, 983.