A Node Deployment Strategy for Blindness Avoiding in ... - IEEE Xplore

IEEE COMMUNICATIONS LETTERS, VOL. 19, NO. 6, JUNE 2015

1005

A Node Deployment Strategy for Blindness Avoiding in Wireless Sensor Networks Guohang Huang, Dongming Chen, and Xuxun Liu, Member, IEEE

Abstract—In this letter, a node deployment strategy is proposed for blindness avoiding in wireless sensor networks (WSNs) on the basis of ant colony optimization (ACO). Firstly, by group division of nodes, a group-based connection mechanism is designed to avoid blindness of connection and reduce deployment cost. Secondly, by actual load evaluation for a critical region, a load-balancing deployment mechanism is designed to prevent blindness of load-balancing deployment and realize on-demand load balancing in the true sense. Finally, simulations show the effectiveness and superiority of our findings. Index Terms—Wireless sensor networks, node deployment, blindness avoiding, ant colony optimization. Fig. 1. Connection methods: (a) blind connection and (b) non-blind connection.

I. I NTRODUCTION

N

ODE deployment is one of the most critical issues in wireless sensor networks (WSNs). Node deployment can be divided into continued-point based deployment and gridpoint based one. Due to specific advantages, the latter becomes necessary in a broad range of applications [1], [2]. The basic requirement of node placement is to achieve desired coverage and connectivity, where coverage is to guarantee that every point of interest (PoI) is monitored by at least one sensor, and connectivity is to ensure sufficient routing paths [3]. To do that, the problem of grid-based coverage with lowcost and connectivity-guarantee (GCLC) is newly introduced, and some research based on ant colony optimization (ACO) [4], which is one of the most useful swarm intelligence [5], has been done. Li et al.[6] proposed the EasiDesign, which formulates coverage as a minimum-cost connectivity problem and optimizes the routing hops. Liu [7] put forward the ACOTCAT, in which three classes of transition modes are used to decrease coverage cost. Liu and He [8] also presented the ACO-greedy, where the mechanism of greedy migration and sensing/communication radius adjustment was used to reduce deployment cost and prolong network lifetime. However, there exist two open issues: blindness of connection and blindness of load-balancing deployment. Firstly, such as [6]–[8], when the ant can’t cover new PoIs on the local area, the ant randomly moves step by step until finding one or more new PoIs to guarantee connectivity. Thus, such a random search manner is of blindness of connection, which leads to winding

Manuscript received July 8, 2014; revised October 19, 2014; accepted December 1, 2014. Date of publication December 12, 2014; date of current version June 5, 2015. This work was supported in part by the National Natural Science Foundation of China (Grant No. 61001112, 61372082), the Guangdong Natural Science Foundation, China (Grant No. S2013010012141), the Cultivation Program for Major Projects and Important Achievements of Guangdong Province, China (Grant No. 2014KTSCX012), and the Fundamental Research Funds for the Central Universities, China (Grant No. 2011ZM0030, 2013ZZ0042). G. Huang and D. Chen contributed to this work equally. The associate editor coordinating the review of this paper and approving it for publication was E. Liu. (Corresponding author: Xuxun Liu.) The authors are with the School of Electronic and Information Engineering, South China University of Technology, Guangzhou 510640, China (e-mail: [email protected]; [email protected]). Digital Object Identifier 10.1109/LCOMM.2014.2379713

Fig. 2. Load-balancing deployment methods: (a) blind node density control and (b) non-blind node density control.

path and large deployment cost, as shown in Fig. 1(a). Secondly, such as [8], despite uneven distribution of nodes, this leads to new load imbalance, because the node density is determined only by node location rather that real load distribution. For example, in Fig. 2(a), there are many more PoIs and traffic loads on the left-half plane than that on the right-half one. So, the left one needs many more nodes to share workload. Nevertheless, there is an identical node density distribution and same deployment cost on the two parts. Motivated by the above, we propose a node deployment strategy with blindness avoiding (NDSBA) on the basis of ACO, which possesses the following distinctive characteristics: 1) For avoiding blindness of connection and reducing deployment cost, a non-blind group-based connection mechanism is designed, in which nodes are divided into different groups. By deterministic search, these groups are connected only between their most close regions [see Fig. 1(b)]. 2) For avoiding blindness of load-balancing deployment and extending network lifetime, a non-blind load-balancing deployment mechanism is designed, in which the node distribution is determined by actual load level and extra nodes are deployed only on where really necessary [see Fig. 2(b)].

1558-2558 © 2014 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

1006


Fig. 3. Network model.

II. S YSTEM M ODEL

Fig. 4.

In the binary rectangle network, there is a stationary sink and multiple stationary PoIs randomly located on the grid points, as shown in Fig. 3. The network is divided into multiple same regular-shaped areas, denoted as A1, A2, A3, etc., and any such point belongs to a specific area. The area size is determined by the accuracy requirements of users. Any node with initial energy ε0 and communication radius Rc can detect the PoIs within the sensing radius Rs . Similar to that in [8], the detection rate of a target with distance d is 1, if d ≤ Rs P(d) = (1) 0, otherwise. The same radio model and related parameters in [8] is adopted. To transmit an l-bit message over a distance d, the transmission radio and reception radio are respectively leelec + lε f s d 2 , d < d0 ET x (l, d) = (2) leelec + lεamp d 4 , d ≥ d0 and ERx (l) = leelec .

(3)

III. P ROPOSED D EPLOYMENT S TRATEGY This deployment strategy is performed by ACO, in which an ant is initially located on any point on which a sensor node can cover at least one PoI. The ant moves from a point to another step by step, and places one node on each point visited by it. The set of all visited points is a solution of the GCLC problem. So, each visited point is a member of solution (MoS), as shown in Fig. 3. The details are described as follows. A. Basic Transition Methods of the Ant There are two types of basic transition ways for the ant, i.e. ordinary transition and greedy migration. 1) Ordinary Transition: Similar to that in [6]–[8], the ant chooses the next point according to the probability pi j (t) =

[τi j (t)]α [ηi j (t)]β α β ∑ [τir (t)] [ηir (t)]

(4)

r∈Si

where τi j (t) and ηi j (t) are respectively the pheromone intensity and heuristic factor on path (i, j). The parameters α and β respectively determine the relative influence of them. Si is the set of candidate points of point i, and consists of all points within the communication radius of this point.

Greedy migration: (a) greed migration and (b) local deployment.

The ant chooses the point that can cover more uncovered PoIs, since the heuristic factor ηi j (t) is defined as ηi j (t) = sum( j) + 1

(5)

where the function sum( j) is the summation of uncovered PoIs within the sensing radius of point j. If, point j is called an effective candidate point (ECP). After a complete tour, the pheromone intensity on every visited path (i, j) is updated according to τi j (t + 1) = (1 − ρ)τi j (t) + Δτi j (t)

(6)

where ρ ∈ (0, 1) is the pheromone evaporation parameter, and the added amount of pheromone Δτi j (t) is given by Δτi j (t) =

Q total(t)

(7)

where the function total(t) is the total number of MoSs in the solution, Q is a constant and Q > 0. It is revealed from the heuristic factor definition and the pheromone updating rule that the both have the potential capability of making deployed nodes as less as possible. 2) Greedy Migration: Similar to that in [8], if a candidate point can’t cover any uncovered PoI, the ant migrates to one visited place with most ECPs within the communication radius of this point. As shown in Fig. 4(a), the ant moves from point F to point A around which there are a few ECPs. This transition way is called greedy migration. After that, the ant continues to cover other PoIs, such as moving to point G and H in Fig. 4(b). Ordinary transition and greedy migration are generally used in alternation. To be specific, if any candidate point can cover uncovered PoIs, ordinary transition is performed. Otherwise, greedy migration is selected. When there is no ECP within the communication radius of any visited point, the two transition methods are no longer performed, such as when the ant is on point H in Fig. 4(b). Clearly, all the visited points are connected together, such as point A, B,C, D, E, F, G, and H in Fig. 4(b). These grid points can be formed as a local connected group (LCG). B. Non-Blind Group-Based Connection Mechanism This connection mechanism consists of the following steps: 1) Long-Distance Jumping: If a LCG is just formed and there still exist uncovered PoIs, the ant randomly jumps to an unvisited place with ECP. For example, as shown in Fig. 5(a), when the ant is on point D, it selects a random ECP of point H and jumps directly to it. Generally the ant needs jumps quite far, so this jumping way is called long-distance jumping.

HUANG et al.: A NODE DEPLOYMENT STRATEGY FOR BLINDNESS AVOIDING IN WIRELESS SENSOR NETWORKS

1007

Fig. 6. Non-blind load-balancing deployment mechanism: (a) before the deployment and (b) after the deployment.

C. Non-Blind Load-Balancing Deployment Mechanism

Fig. 5. Non-blind group-based connection mechanism: (a) long-distance jumping, (b) creation of new LCGs, and (c) short-distance connection.

2) Creation of New LCGs: After jumping from a LCG to an ECP, the ant begins to perform coverage task according to the two the above transition methods. Thus, another LCG is formed after several steps of transitions, such as point H, G, E, and F in Fig. 5(b). Similarly, multiple separated LCGs can be formed when all PoIs have been covered. 3) Short-Distance Connection: When all LCGs are formed, they need to be connected. As mentioned above, any point belongs to a defined area, so any LCG is located on specific areas. By comparison of distances among different areas, it is easy to know which two LCGs should be directly connected and which areas of these LCGs should be connected directly. For instance, in Fig. 5(b), area A7 and A9 are selected to achieve connection, as shown in Fig. 5(c). Thus, the two LCGs are merged into a new larger one. The sink is a special LCG, and finally any LCG must be connected to the sink. When all LCGs are connected to the sink, the task of connection has been finished. At this point, the total deployment cost is achieved and the pheromone is updated by the ant. Such a connection way is generally performed between close areas, so it is called short-distance connection. During the course of long-distance jumping and creation of new LCGs, the ant must measure the number of uncovered PoIs for each next-hop candidate point, update the total deployment cost, and update the residual uncovered PoIs. However, it only needs to update the total deployment cost during the course of short-distance connection, because the coverage task is finished and there is no any uncovered PoI at this time. Apparently, on account of close range of connection, the deployment cost can be reduced compared with that of other methods. Moreover, area division is similar to clustering [9], so this connection mechanism can enhance scalability.

Herein, we design a non-blind load-balancing deployment mechanism to realize on-demand load balancing for scenario with random distribution of PoIs. This deployment mechanism includes the following two steps. 1) Load Level Evaluation: Generally energy hole exists near the sink, so it is necessary to evaluate the load level of such places. A circle centered at the sink is defined as a fragile region (FR), which is used to assess the load level within it. As demonstrated in Fig. 6, the large circle with dotted line is the FR, whose radius R0 is determined by the requirements of users. For a LCG Ωi , its load level is simply defined by ψi =

Mi NiFR

(8)

where Mi is the number of PoIs covered by LCG Ωi , NiFR is the number of MoSs of this LCG within the FR. Take Fig. 6(a) for example, for the upper LCG, ψi = 7/2 = 3.5, while for the lower one, ψi = 19/2 = 9.5. For any LCG Ωi , the value of Mi and NiFR are initialized to zero in the beginning. During the course of moving of the ant, these values are updated. When two LCGs Ωi and Ω j are connected to a new one Ωk , the ant performs the corresponding computing tasks: Mk = Mi + M j NkFR

= NiFR + N FR j .

(9) (10)

As all tasks of connection are completed, the load level of each LCG is computed by the ant according to formula (8). 2) Extra Node Deployment: If the load level exceeds the predefined threshold δ0 , extra nodes will be deployed to create extra paths and share workload. The number of extra paths is determined by the load level of the LCG. The origin of the extra path is selected by the distances between the sink and different areas of the LCG. Outside the FR, the area with minimal distance to the sink is selected the origin of the extra

1008


Fig. 7. Deployment cost with different ratios of PoIs: (a) 15 m × 15 m network and (b) 25 m × 25 m network.

Fig. 8. Network lifetime with different ratios of PoIs: (a) 15 m × 15 m network and (b) 25 m × 25 m network.

path. The ant moves from the origin to the destination, i.e. the sink, according to the shortest distance between the two areas. For instance, in Fig. 6(b), only the lower LCG needs extra node deployment. Point E and F are selected as the origins of the extra paths and extra nodes are placed on point G and H to create two extra paths. The radius R0 of FR is determined by different requirements of users. If the user requires lower cost, R0 is set to a smaller value and fewer extra nodes are placed to share workload. Otherwise, R0 is lager and more extra nodes are placed. It is revealed that this deployment mechanism can achieve load balance on demand in the true sense, especially for irregular distribution of PoIs. Moreover, the origins of the extra paths are not far from the sink, so the extra deployment cost is small. Additionally, by adjustment of radius R0 of FR, this deployment method can satisfy different requirements of users.

Fig. 8 compares network lifetime achievable in different deployment strategies. Results show that, the lifetime, represented by the maximum transmission round, in EasiDesign and ACO-TCAT is much lower that that in ACO-Greedy and NDSBA, since the former two have no any load balancing measure. Although extra nodes are deployed for load balancing in ACO-Greedy and NDSBA, NDSBA performs better. This is because the load balance mechanism of NDSBA is based on real demand and extra nodes are placed only on where really necessary.

IV. P ERFORMANCE E VALUATION ACO-TCAT shares the same computational complexity O(n2 ) with EasiDesign [7]. The differences between NDSBA and ACO-TCAT mainly exist in the two extra tasks, shortdistance connection and extra node deployment. However, NDSBA has the same computational complexity O(n2 ) as ACO-TCAT, because such tasks have the largest computational complexity, which is not more than O(n2 ). The performance of NDSBA is further assessed through extensive simulations. Nodes are distributed on 15 m × 15 m and 25 m × 25 m square regions respectively. For the 15 m × 15 m network, ε0 = 1.4 × 105 nJ and R0 = 5 m, while ε0 = 3.6 × 105 nJ and R0 = 8 m for the 25 m × 25 m network. Other parameters are set as follows: Rs = Rc = 2 m, ρ = 0.05, Q = 20, α = 2, β = 3. For performance comparison in terms of deployment cost and network lifetime, we consider other deployment algorithms: EasiDesign [6], ACO-TCAT [7], and ACO-Greedy [8]. Fig. 7 plots the deployment cost of the four deployment algorithms with different ratio of PoIs. It can be seen that, the total deployment cost of all deployment methods is increasing with the rising of PoI ratio. This is because that more PoIs need more nodes to cover them. ACO-TCAT and ACO-Greedy generate less deployment cost than that of EasiDesign, and this is due to more transition types. NDSBA needs the least deployment cost due to its specific connection mechanism with random moving avoiding and trip length reduction in ACO.

V. C ONCLUSION This letter explores the problem of GCLC for WSNs, and put forward a deployment strategy with blindness avoiding on the basis of ACO, which consists of two salient features. First, a non-blind group-based connection mechanism is designed to reduce deployment cost. Second, a non-blind load-balancing deployment mechanism is designed to achieve load balancing on demand in the true sense. Finally, simulations prove that our findings significantly outperform the state-of-the-art solutions in terms of deployment cost and network lifespan. R EFERENCES [1] G. Moro and G. Monti, “W-grid: A scalable and efficient self organizing infrastructure for multi-dimensional data management, querying and routing in wireless data-centric sensor networks,” J. Netw. Comput. Appl., vol. 35, no. 4, pp. 1218–1234, Jul. 2012. [2] F. Al-Turjman, H. Hassanein, and M. Ibnkahla, “Quantifying connectivity in wireless sensor networks with grid-based deployments,” J. Netw. Comput. Appl., vol. 36, no. 1, pp. 368–377, Jan. 2013. [3] Y. Wang, C. Hu, and Y. Tseng, “Efficient placement and dispatch of sensors in a wireless sensor network,” IEEE Trans. Mobile Comput., vol. 7, no. 2, pp. 262–274, Feb. 2008. [4] M. Dorigo and T. Stutzle, Ant Colony Optimization. Cambridge, MA, USA: MIT Press, 2004. [5] J. Kennedy, “Swarm intelligence,” in Handbook of Nature-Inspired and Innovative Computing. New York, NY, USA: Springer-Verlag, Jan. 2006. [6] D. Li, W. Liu, and L. Cui, “EasiDesign: An improved ant colony algorithm for sensor deployment in real sensor network system,” in Proc. IEEE Globecom, Dec. 2010, pp. 1–5. [7] X. Liu, “Sensor deployment of wireless sensor networks based on ant colony optimization with three classes of ant transitions,” IEEE Commun. Lett., vol. 16, no. 10, pp. 1604–1607, Oct. 2012. [8] X. Liu and D. He, “Ant colony optimization with greedy migration mechanism for node deployment in wireless sensor networks,” J. Netw. Comput. Appl., vol. 39, pp. 310–318, Mar. 2014. [9] X. Liu, “A survey on clustering routing protocols in wireless sensor networks,” Sensors, vol. 12, no. 8, pp. 11113–11153, Aug. 2012.