AISTC: A new Artificial Immune System-based Topology Control Protocol for Wireless Sensor Networks

Transcription

1 AISTC: A new Artificial Immune System-based Topology Control Protocol for Wireless Sensor Networks Amir Massoud Bidgoli 1, Arash Nikdel 2 1 Department of computer engineering, Islamic Azad University, Tehran North Branch, MIEEE, Ph. d. Manchester University, Tehran, Iran 2 Department of computer, Science and Research Branch, Islamic Azad University, Khouzestan, Iran Abstract Topology control protocols try to decrease average of node s transition radius without decreasing network connectivity. In this paper, we propose a new Artificial Immune System-based topology control protocol for wireless sensor networks named AISTC. In this protocol, proper transition radius can be determined using Artificial Immune System algorithm. This protocol is simulated and compared its functionality to some other protocols. Simulation results show high efficiency of the proposed protocol. Keywords: wireless sensor network, topology control, artificial immune system, network lifetime, energy consumption. 1. Introduction The wireless sensor network (WSN) has emerged as a promising tool for monitoring the physical world. This kind of networks consists of sensors that can sense, process and communicate. Sensors can be deployed rapidly and cheaply, thereby enabling large-scale, on-demands monitoring and tracking over a wide range of applications such as danger alarm, vehicle tracking, battle field surveillance, habitat monitoring, etc [1]. Due to their portability and deployment, nodes are usually powered by batteries with finite capacity. Although the energy of sensor networks is scarce, it is always inconvenient or even impossible to replenish the power. Thus, one design challenge in sensor networks is to save limited energy resources to prolong the lifetime of the WSN [2]. A number of studies for reducing the power consumption of sensor networks have been performed in recent years. These studies mainly focused on energy efficient MAC protocols, data aggregated routing algorithms, and the applications of level transmission control. Power saving techniques can generally be classified in two categories: scheduling the sensor nodes to alternate between active and sleep mode, and adjusting the transmission or sensing range of the wireless nodes [2]. Topology control in sensor networks is coordination art of nodes by decision making about transition radius [3].Choosing appropriate topology for a sensor network has much effect on networks performance, especially considering power consumption and lifetime network. In this paper, we propose a new topology control protocol based on Artificial Immune System, named AISTC. In this protocol, each node adjusts its transition radius using Artificial Immune System algorithm and considers the transition radius of its neighbors and the network status. The transition radius will be between minimum transition radius and maximum transition radius. The remaining of this paper is organized as follow: Related works are explained in section 2. In section 3 problem definition is introduced. Artificial Immune System will be discussed in Section 4. Proposed protocol is explained in section 5. Simulation results are shown in section 6 and a final conclusion is discussed in Section Related Works So far, many protocols have been introduced for topology control in sensor networks. Topology control protocols are divided into homogeneous and heterogeneous control topologies. In homogeneous control topology, all network nodes use the same transition radius and topology control problem is to find a minimum value for transition radius considering the network characteristic such as in heterogeneous control topology in which network nodes can have non uniform transition radius. In this group, protocols with information used for making topology are divided into three groups. First group are methods based on location. In this group, nodes are informed of their location and by using this information, a proper topology for network is made [4], [5]. Second group are methods based on orientation. In these methods, nodes don t have exact information of their location, but they can identify direction of their neighbors. Protocol CBTC 1 [6] is an example of these methods. Third group are methods based on neighbors. In these methods, nodes have limited information about their neighbors. This 1 Cone Based Topology Control

2 information consists of ID number, and distance or quality of node s neighbors. Kneigh 2 [7] and XTC 3 [8] are examples in this group. RAA-2L 4 is another topology control protocol. In this protocol, each node chooses one of two transition areas R S or R W (R W <R S ) [9]. If a node with transition area of R W could communicate with a neighbor with transition area R S, it chooses transition area node R W, else it chooses transition area node R S. In RAA-3L 5, each node chooses one of three transition areas: R t, R S or R W (R W < R t <R S ). In [10], Cellular Automata is used for topology control, but it didn t consider the nodes relations and their residual energies. In our proposed topology control protocol, each node will consider the transition radius of its neighbors, its residual energy and network connectivity. 3. The Model And The Assumptions In this section, we present the model and the assumptions used in this paper Adjustable transition radius We assume that each node has adjustable transition radius that can be between a minimum and a maximum area. R min is transition area with minimum power, R max is transition area with maximum power and R S is selective transition area of node. The value of R T should be between the R min and R max (R min R T R max ). Value of transition area R min and R max will be calculated based on R t. Value of transition area R t is determined proportional to the network density [11]. When distance of both nodes is less than R max, we assume they are neighbors. Both of neighbor nodes are in four different groups. Sets of A min, A S and R max are obtained by (1). In (1), n i is neighbor node number, and D ni is the distance between current node and n i. n i A min if D ni R min n i A S if R min D ni R S (1) n i A max if R S D ni R max Therefore: A max A S A min = All neighbor (2) A max A S A min = {} (3) A C consists of neighbor nodes that are in selective transition area or are accessible through nodes that are in selective node area. A min is proper subset of A C, because each node has minimum transition area A min. Transition area and set of nodes are depicted for node n in Fig. 1. The main problem in this study is choosing minimum transition area R T between R min and R max for each node without decreasing the network connectivity. 2 k-neighbors 3 Extreme Topology Control 4 Radius Adaptation Algorithm_2 Level 5 Radius Adaptation Algorithm_3 Level Figure1. Ttransition area and set of nodes The transition radius of each node is coded in binary format and the required B bit is calculated using (4): B = log 2 (R max - R min +1) (4) The transition radius of each node is calculated using (5): R T =R min +(R max - R min +1) b=1 to B 2 b-1 a b-1 / b=1 to B 2 b-1 (5) 3.2. The cluster-based architecture We introduce a cluster-based coverage control scheme in this paper, which is scheduled into rounds. In each round, firstly, the target area is divided into several equal squares. Then the node in each square having the largest energy will be chosen as the cluster-head, and the procedure of selecting the cluster-head is the same work in [15]. This cluster-based architecture is shown in Fig. 2. The nodes of cluster-heads are those asterisked ones. The black nodes represent the active ones which are working in the target area. And the red sensor nodes are these inactive ones in sleeping mode. The cluster-head has full control of the square and it will choose transition radius of nodes. In the next round, another sensor set will be turned on. It is done in a random way, so the energy consumption among all the sensors can be balanced well. Figure2. The cluster-based architecture of AISTC protocol 3.3. Energy consumption analysis For the brief of the energy consumption analysis, here we only consider the energy consumed by the transmission function, and do not include the power consumption of sensing and calculation.

3 Define that the size of the monitoring area is A area, the working sensor set is n={n^1, n^2,, n^n}and the sensing radius set is R={R T^1, R T^2,, R T^n}, where R T^i is the transition radius of node n^i, and R T^i [R min, R max ]. According to different energy consumption models, the energy consumed by a node to deal with a transmission task is proportional to R T 2 or R T 4, where R T is the transition radius of node [16]. In this paper, we take the transmission energy consumption as u.r T 2, where u is the factor. Thus, the coverage energy consumption of the sensor set, which is related to the sum of the sensor's transition radius squared, is defined as: E total = u. i=1 to n R T^i 2 So, the energy consumption per area is shown as the following: (6) E total /A area = u. i=1 to n R T^i 2 / A area (7) 4. Artifical Immune Systems Models AIS are distributed adaptive systems for problem solving using models and principles derived from the Human Immune System [13]. The capabilities of the AIS is mainly the inner working and cooperation between the mature T-Cells and B-Cells that are responsible for the secretion of antibodies as an immune response to antigens [14]. The different theories regarding the functioning and organizational behavior of the natural immune system (NIS) are discussed in literature. These theories inspired the modeling of the NIS into an artificial immune system (AIS) for application in non-biological environments [14].Many different AIS algorithm models have been built, including Classical View Models, Clonal Selection Theory Models, Network Theory Models, Danger Theory Models [14]. Artificial immune systems have been successfully applied to many problem domains. Some of these domains range from network intrusion and anomaly detection, to data classification models, virus detection, concept learning, data clustering, robotics, pattern recognition and data mining [14]. 5. Proposed Protocol In this section, we try to decrease average of node s transition radius without decreasing network connectivity. In proposed algorithm, at first the primary population of nodes transition radius are selected randomly. Then, the affinity rate of nodes is evaluated and based on this evaluation some of nodes are selected as memory cells and the transition radius of other nodes is mutated. The main loop of algorithm continues until the number of its repeats exceeds from threshold rate or the affinity rate of all nodes become better than threshold rate. Therefore, the proposed algorithm includes six steps as follows: Pahse1. Problem and algorithm parameter initialization: Step1: Initializing A min, A S and A max sets for each node. Step2: producing a transition radius mask and a transition radius mask operation for each node. Step3: Initializing transition radius node, RT, for each node randomly. Pahse2. Repeating main loop of algorithm until meeting termination criteria: Step4: Calculating the affinity rate of nodes. Step5: Selecting the nodes with more affinity rate as memory cells and mutating the transition radius of other nodes. Step6: Checking the loop termination criteria and jumping to step Algorithm Details Description In this section, we describe the proposed algorithm in detail: Step1: Initializing A min, A S and A max sets for each node. At first, according to (8), the transition radius of each node is set between R min and R max. R T = (R T^1, R T^2, R T^3,, R T^n ) (8) R T^i R T : R T ^i = R min + (R min - R max ) / Such a way that is a constant factor (e.g. =2) which its rate can be determined regarding nodes density. According to (9), if is considered much more than R max, R T^i will almost equal R min. If = R T^i R min (9) Then according to (1), (2) and (3) as mentioned before, A min, A S and A max sets for each node is created. Step2: producing a transition radius mask and a transition radius mask operation for each node. Then A min, A S and A max sets are updated for each node. Whenever one node of A S and A max sets becomes a member of the set of another node, that will be removed from these sets. For this purpose, at first, we initialize A C set by A min set content and then the sets are updated according to (10): A C (N) A min (N) (10) n i A min (N) n j A min (n i ) AND ( n j A S (N) OR n j A max (N)) A min (N) = A min (N) + n j A S (N)=A S (N) n j OR A max (N)=A max (N) n j Then: n i A S (N) n j A min (n i ) AND (n j A max (N)) A S (N) = A S (N) + n j A max (N)=A max (N) n j

4 Regarding to A S and A max condition, the node performs a transition radius mask (Mask transition ) and determines a transition radius mask operation (Operation mask_transition ) with OR/AND. The method of determining transition radius mask and transition radius mask operator is calculated according to the four conditions: Both A S and A max sets are empty The transition radius of node is equal to A min. So, radius mask is as below: (the mask operator is AND) If A S = and A max = Masktransition = 0 (11) Operation mask_transition = AND A S set is not empty and A max set is empty The node can select its transition radius between both R min and R T. The transition radius mask is as below: (the mask operator is AND) If A S and A max = X = log 2 R T (12) Mask transition =2 X -1 Operation mask_transition = AND A S set is empty and A max set is not empty The node can select its transition radius between R T and R max. Mask of this transition radius is as below: (the mask operator is OR) If A S = and A max X=[log 2 R T ] (13) Mask transition =2 X Operation mask_transition = OR Both A S and A max sets are not empty The node can select its transition radius between A max and A min. This transition radius mask is as below: (the mask operator is OR) If A S And A max Mask transition =0 (14) Operation mask_transition = OR Step3: Initializing transition radius node, A T, for each node randomly. In this step, according to (15), the transition radius node for each node is initialized randomly. R T = (R T^1, R T^2, R T^3,, R T^n ) (15) R T^i R T : R T ^i = random number between R min to R max Then, according to (16), for each node, the transition radius mask is applied to transition radius of node by transition radius mask operator (AND/OR). R T = (R T^1, R T^2, R T^3,, R T^n ) (16) R T^i R T : R T ^i R T ^i Operation mask_transition (AND/OR) Mask transition Step4: Calculating the affinity rate of nodes. The process of calculating the affinity rate for each node N is as follows: Whenever one node of A C becomes a member of A S or A max, that node is removed from A S or A max and adds to the A C set. See more details in (17): n i A C (N) n j A C (n i ) And ( n j A S (N) Or n j A max (N)) (17) A C (N) =A C (N)+n j A S (N)=A S (N) n j Or A max (N)=A max (N) n j In (17), A x (y) shows A x set of node y. After updating the sets, the node determines the affinity of its selected transition radius regarding to neighbor s selected transition radius. For this purpose, the node considers a temporary TA C set. As can be seen in (18), this set, at first, is equal to A C. TA C (N) = A C (18) Then, according to (19), the node adds the A S set of its neighbors to the same neighbor A C set: n i : A C (n i ) = A C (n i ) +A S (n i ) (19) After that, regarding the (20), the node updates TA C set: n i TA C )N( n j TA C ) n i ( and ( n j A S (N) or n j A max (N) ) (20) TA C (N) = TA C (N) + n j After updating TA C set, the process of determining transition radius affinity of node is as below: If A S TA C and A max TA C At this situation, more closely the transition radius rate to R min, more fit the transition radius. So: affinity = *(my-node-a/max-node-a)(1/(r T -R min + )) (21) Where shows very small positive number, λ 1 shows the minimum acceptable rate for affinity of node and ψ 1 is selected as the affinity rate doesn t exceed a given limit.

5 If A S TA C or A max TA C The node adds A S set to TA C set and updates TA C set again by (20). Then, If A max TA C (evidently A S TA C ), so the transition radius will be fit and can be smaller. The details can be seen in (22): affinity = *(my-node-a/max-node-a)(1/(r T -R min + )) (22) Fig. 4 shows the process of nodes mutation. In Fig. 4, the nodes with yellow transition radius are those that are selected as memory cells. As mentioned before, the nodes which are selected as memory cells don t mutate and their transition radius doesn t change. The nodes mutation rate is different as shown in Fig. 4. If A max TA C, more closer the node transition radius to R max, more affinity of it. So, affinity can be defined as (23): affinity = *(my-node-a/max-node-a)(1/(r max -R T + )) (23) Where shows very small positive number, λ 2 shows the minimum acceptable rate for affinity node, and ψ 2 is selected as the affinity rate doesn t exceed a given limit. In (21), (22) and (23), my-nodes-a shows all member of two A S, A max sets of node. The rate of max-nodes-a is calculated by (24). In this relation A x (y) shows the A x set of node y. For each node N : max-node-a = max (a,b,c) a = max ( A S (n j ) + A max (n j ) n j nebr max ) b = max ( A S (n j ) n j nebr S ) (24) c = my-node-a = A s (N) + A max (N) Figure4. Process of mutating the transition radiuses The Fig. 5 shows a big binary number with four different mutation ranges. Regarding the binary rate come before and after mutation, we observe that the performance of mutation operator makes the small number bigger and big ones smaller very likely. Regarding the reverse ratio of mutation rate to affinity rate, the less node s affinity have the more node s mutation and if a number is big, It will become smaller and rice versa. The nodes with more affinity have less mutation and also less change. Step5: Selecting the nodes with more affinity rate as memory cells and mutating the transition radius of other nodes. After calculating the affinity rate of the nodes, percent of them (e.g. 50%) are selected as memory cells. It means that their transition radius doesn t mutate until next cycles (in the easiest mode, 1 cycle). For this purpose, the nodes are arranged in ascending order. Then percent of nodes with more affinity rate are selected as memory cells. After determining the memory cells, transition radius of other nodes is mutated. The ratio of mutation rate to affinity rate is inverses, as a result the node with more affinity rate will have less mutation and with less affinity rate, they have more mutation. The bits mutate and are selected randomly the selected bit will be inverted (zero change to 1 and vice versa). The only operator in Artificial Immune System algorithm is mutation operator. The mutation rate is in reverse ratio to affinity rate. As a result the mutation rate for each node ( node ) is in reverse ratio to that node affinity rate (affinity node ). node is calculated for node according to (25). node = / ( affinity node + ) (25) Where is a constant number that is calculated in the way that mutation rate doesn t become less than the determined level. Also is a constant number that should be selected properly in order that mutation rate doesn t exceed the determined level. Figure5. A transition radius with four different mutation rates Step6: Checking the loop termination criteria and jumping to step 4. The main loop of algorithm (steps 4 and 5) continues until meeting one of the conditions stated bellow: The affinity rate of all nodes, become better than TA 6 threshold rate. And also this transition radiuses can provides the full connectivity of network. The number of performing the main loop of the algorithm exceed TC 7 threshold rate. Then, the final transition radius of nodes regarding the condition of their sets is determined. In the way that if the A max set is not empty, R max transition radius is selected. If A S set is not empty, R T transition radius is selected, otherwise R min transition radius will be selected. 6. Simulation Results In this section, our proposed protocol, AISTC, is simulated and compared to RAA-2L, RAA-3L [9] and homogeneous mode (HOM) [11] using NS2 simulator. We considered 6 Threshold Affinity 7 Threshold Cycles

6 m 2 area for these simulations. The number of nodes considered equal to 200, 300, 400, 500, and 600. Each node has a transition range between R min and R max. Transition ranges R min and R max are proportional to network density. Transition range R min and R max considered equal 87 and 136 (for 200 nodes), 69 and 108 (for 300 nodes), 59 and 93 (for 400 nodes), 54 and 84 (for 500 nodes), 48 and 75 (for 600 nodes) respectively. The parameter is equal 1, λ 1 and λ 2 parameters are equal 0.2, ψ 1 and ψ 2 are equal 0.3 and threshold ranges values, TP and TS considered 0.98 and 300 respectively. The energy model for these simulations is similar to energy model used in [12]. Three metrics are used for evaluations. These metrics are: average of transition area, average number of neighbors for each node, and Probability of complete connection of network nodes. In the first experiment, we measured the average of transition area of network with 100 nodes for AISTC, RAA-2L, RAA-3L and HOM protocols. The result of this simulation is depicted in Fig. 6. As can be seen, AISTC has minimum average of transition area and HOM has maximum average of transition area. In the last experiment, network connectivity in AISTC is measured and compared to RAA-2L, RAA-3L, HOM and MAX-RANGE protocols. Note that, in MAX-RANGE, all of nodes have maximum transition radius. Network connectivity means ability of communicating with all of network nodes. For this purpose, we will define the concept of complete connectivity of network probability, P C as (26): P C = i=1 to Nd C i / N d C i = 1 if mcp=n n 0 other (26) In (26), mcp is the biggest component connected to network and N n is the number of network nodes and N d is the number of different configuration of network nodes. In this experiment, we suppose N d is equal to 100. Probability of complete connecting of network nodes is depicted in Fig. 8 for AISTC, RAA-2L, RAA-3L, HOM and MAX-RANGE protocols. Figure6. Average of transition area Vs number of nodes In Second experiment, average number of neighbor nodes for AISTC, RAA-2L, RAA-3L and HOM protocols is measured. The result of this experiment is depicted in Fig. 7. Figure7. Average number of neighbors Protocol AISTC has minimum average number of neighbors compared to other protocols. Note that number of neighbors has a direct effect on interference between nodes and so, lower number of neighbors is better. Figure8. Probability of complete connecting of network nodes As can be seen in Fig 8, probabilities of complete connecting of network nodes for AISTC, RAA-2L, RAA-3L and MAX-RANGE are almost equal. So, network connectivity in our protocol is acceptable. This result can show prominence of our considered mechanism. While maintaining network connectivity, they could decrease the average transition radius and the average number of network neighboring nodes and consequently it decreases energy consumption and interference between network nodes. 7. Conclusion In this paper, we proposed a topology control protocol based on Artificial Immune System. In this protocol, nodes can select proper transition radius. Simulation results showed that the proposed protocol has some advantages compared to previous protocols. First advantage is minimum average of transition area and dynamic adjustment of the radius ratios, unlike previous protocols that should select radius ratios among predefined values. Second advantage is that our protocol has minimum average number of neighbors

7 compared to existing protocols. So, the energy consumption in our protocol is less than others and the network lifetime will be prolonged. In addition, we showed that the network connectivity in our protocol is in the acceptable level. 8. References [1] A. Jie Jia, C. Jian, C. Guiran, C. Zhenhua, Energy efficient coverage control in wireless sensor networks based on multi-objective genetic algorithm,elsevier, Computers and Mathematics with Applications 57 (2009) 1756_1766 [2] A. Jie Jia, C. Jian, B. Guiran, W. Yingyou, S. Jingping, Multi-objective optimization for coverage control in wireless sensor network with adjustable sensing radius, Elsevier, Computers and Mathematics with Applications 57 (2009) 1767_1775 [3] P. Santi, Topology Control in Wireless Ad Hoc and Sensor Networks, Wiley, [4] V. Rodoplu and T. H. Meng, "Minimum energy mobile wireless networks", in: Proceedings of the IEEE Journal on Selected Areas in Communications, Vol. 17, pp , [5] N. Li, J. Hou and L. Sha., Design and analysis of an mst-based topology control algorithm, in: Proceedings of the IEEE Infocom, Vol. 4, pp , May [11] D. Stauffer and A. Aharony, Introduction to Percolation Theory, London: Taylor & Francis, [12] W. Heinzelman, A. Chandrakasan and H. Balakrishnan, Energy Efficient Communication Protocol for Wireless Microsensor Networks, Intl. Conf. on System Sciences, Hawaii, vol. 2, pp January [13] Amir Massoud Bidgoli, Abdol Karim Javanmardi, Amir Masoud Rahmani, Application of AIS algorithm for optimization of TORA protocol in ad hoc network, in: IEEE 2010 [14] Andries P. Engelbrecht, Computational Intelligence An Introduction,second edition, wiley 2007 [15] J. Jia, J. Chen, Y. Wen, G. Chang, An extensible corecontrol routing protocol in large scale ad-hoc networks, in: Proc. of the 6th International Conference on ITS Telecommunications, Chengdu, China, 2006, pp. 955_958. [16] M. Lu, J. Wu, M. Cardei, M. Li, Energy-efficient connected coverage of discrete targets in wireless sensor networks, in: Proc. of the International Conference on Computer Networks and Mobile Computing, ICCNMC, Zhangjiajie, China, 2005, pp. 43_52. [6] R. Wattenhofer, L. Li, P. Bahl and Y. Wang, Distributed topology control for power efficient operation in multihop wireless ad hoc networks, in: Proceedings of the IEEE Infocom, Vol. 3, pp , [7] D. Blough, M. Leoncini, G. Resta. and P. Santi, The k-neighbors protocol for symmetric topology control in ad hoc networks, in: Proceedings of the ACM MobiHoc 03, pp , [8] R. Wattenhofer and A. Zollinger, XTC: a practical topology control algorithm for ad-hoc networks. in: Proceedings of the 18th International Parallel and Distributed Processing Symposium, pp. 2-16, April [9] A.Venuturumilli and A. Minai., Obtaining Robust Wireless Sensor Networks Throuh Self-Organization of Heterogeneous Connectivity, Proceedings of the 2006 International Conference on Complex Systems (ICCS'06), Boston, MA, June [10] M.R. Meybodi and S.Abolhasani, Usage of Learning Automata for Topology Control in Wireless Sensor Network,2008.