Hybrid movement strategy in self-orienting directional sensor networks

Ad Hoc Networks 11 (2013) 1075–1090

Contents lists available at SciVerse ScienceDirect

Ad Hoc Networks journal homepage: www.elsevier.com/locate/adhoc

Hybrid movement strategy in self-orienting directional sensor networks q M. Amac Guvensan ⇑, A. Gokhan Yavuz Department of Computer Engineering, Yildiz Technical University, Istanbul, Turkey

a r t i c l e

i n f o

Article history: Received 5 May 2012 Received in revised form 14 September 2012 Accepted 22 November 2012 Available online 4 December 2012 Keywords: Sensing coverage Directional sensor networks Motility Mobility Hybrid movement strategy Motility assisted mobility Energy consumption

a b s t r a c t The coverage problem in directional sensor networks (DSNs) introduces new challenges especially for randomly deployed networks. As many overlapped regions and coverage holes might occur after the initial deployment, self-orientation of the nodes is a necessity for randomly deployed DSNs. There exist two main approaches for the self-orientation of directional sensor nodes in DSNs [1], motility and mobility. Motility refers to the adjustment of the working direction of the nodes, whereas mobility describes the physical movement of the nodes. Most existing studies propose solutions based on the motility capability of the directional sensor nodes. On the other hand, mobility is a powerful feature offering great flexibility. Nevertheless, the high energy consumption of mobility discourages researchers to utilize this approach in their solutions. In this study, we propose a novel approach, a hybrid movement strategy (HMS), where we exploit motility/mobility in a cascaded manner for the coverage improvement in DSNs. The HMS improves the initial coverage up to 47% and achieves up to 7% more coverage than the motility only solution. Besides, it has provided at least 40% energy-saving compared to the mobility only solution in our scenarios. Ó 2012 Elsevier B.V. All rights reserved.

1. Introduction The coverage problem is a fundamental problem in sensor networks [3,4], especially after random deployment. The research community has thoroughly explored this problem for omni-directional sensor networks [5,6]. Proposed solutions basically depend on two principles: redeployment of additional nodes and use of mobile sensors. On the other hand, a new network type, the so called ‘‘Directional Sensor Network’’ became popular within the last five years and brought new challenges to the traditional coverage problem [1,7,8]. Directional sensor networks typically consist of nodes equipped with direction-sensitive video,

q A preliminary version of this paper was accepted for publication in Proc. of 7th Intl. Conf. on Wireless and Mobile Communications (ICWMC), Luxembourg, June 2011 [2]. ⇑ Corresponding author. Tel.: +90 212 373 57 65. E-mail addresses: [email protected] (M.A. Guvensan), [email protected] (A.G. Yavuz).

1570-8705/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.adhoc.2012.11.011

infrared and/or ultrasound sensors. A directional sensor node has a sectoral perception within its sensing radius. Thus, a node could work in several directions around itself. The adjustment capability of the working direction of a directional sensor is called as motility. Random deployment in DSNs mostly results in several overlapped regions, therefore, directional sensor nodes need to be repositioned after their initial deployment in order to increase the total area coverage. In DSNs, the utilization of motility is the most energy-efficient way to minimize the overlapped regions and to cover some coverage holes, since motility does not cause the node to change its physical location. Thus, a great number of existing solutions exploit this motility capability for the coverage problem [9–12]. Although, the efficiency of a DSN heavily depends on the correct working direction of its individual sensory units in the field, motility could not be sufficient enough in some scenarios. At this point, mobility capability of a sensor node offers a great advantage to decrease existing overlapped areas and to prevent coverage holes [2].

1076

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

However, exploiting mobility alone is a highly expensive solution compared to motility. Thus, we propose a hybrid movement strategy (HMS) for the coverage problem in DSNs, where both motility and mobility are utilized in a cascaded manner. In the HMS, we first exploit the motility capability of the nodes to minimize the overlapping regions without moving the nodes to a different location. With appropriate working directions of the nodes determined, if there exist overlapped regions and coverage holes, we engage the mobility capability in our solution. Thus, in this study, a new coverage enhancement model, hybrid movement strategy, has been introduced in order to maximize the coverage after the initial deployment in an energy-efficient way. The key contributions of this paper are given as follows,  To demonstrate the efficiency of the HMS, we exploit three novel algorithms for both motility and mobility parts of the proposed solution. Two novel algorithms, the Attractive Forces of Uncovered Points (AFUP) [13] and the weighted AFUP (W-AFUP) [14], are implemented for utilizing the motility capability. Besides, one unique algorithm, Window-based Neighborhood Exploring (WNE), has been introduced in order to exploit the mobility capability. The proposed algorithms run fully distributed using local information only, thus, communication overhead is incurred only between neighboring nodes. We also introduce Motility Assisted Mobility (MAM) algorithm, a particular implementation of the HMS, where motility and mobility parts of the algorithm consist of the W-AFUP and the WNE algorithms respectively.  Though AFUP and W-AFUP outperforms several existing solutions [15–17] based on the motility capability, we show that motility could improve the total coverage up to a limit. Especially for dense networks, we demonstrate that the total area coverage could be improved further by exploiting mobility.  We also show that exploiting the mobility alone is highly expensive than the HMS, in terms of the number of moved nodes, total travel distance and energy consumption. The HMS achieves in maximizing the area coverage with possible minimum energy by combining the advantages of motility and mobility. The remainder of the paper is organized as follows. In Section 2, we discuss the available studies about directional coverage problem in DSNs. Section 3 gives the notations and definitions to the area coverage problem. In Section 4, we present our new coverage enhancement model, hybrid movement strategy and explain the details of the proposed AFUP, W-AFUP and WNE algorithms. AFUP and W-AFUP utilize the motility capability, whereas WNE exploits the mobility capability. In Section 5, experimental results are given in detail. The results demonstrate that the HMS could improve the total coverage much more than a motility only solution. Moreover, it achieves this improvement in an energy-efficient way compared to the mobility only solution. Section 6 concludes the paper.

2. Related work Maintaining and maximizing sensing coverage of scattered sensor nodes have been studied in great depth in traditional sensor networks for the past decade [5]. Available studies aim at covering a plane by arranging circles on the plane. However, the proposed solutions for omni-directional coverage cannot be used for the coverage of direction-sensitive nodes, since these nodes capture only a predefined sector within their sensing radius according to their angle of view and working direction. This predefined sector is described as the Field-of-View (FoV) of a node. Especially after random deployment, some FoVs might overlap in the observed area. Besides, some targets might be left uncovered. To overcome these difficulties, researchers opt for the motility capability of directional sensors. On the other hand, to the best of our knowledge, only one study [18] proposes to utilize the mobility capability of directional nodes. However, in this study, the authors do not take the motility capability into account and they do not change the working directions of the nodes. In [1], we have described the directionality problem in DSNs and surveyed existing solutions to the coverage problem. Available studies could be categorized into four groups.    

Target-based coverage enhancement solutions. Area-based coverage enhancement solutions. Coverage enhancement with guaranteed connectivity. Network lifetime prolonging solutions.

Some sensor applications are only interested in stationary target points, such as buildings, doors, flags, and boxes, whereas other applications aim at tracking mobile targets like intruders. Stationary targets can be located anywhere in the observed area. To cover only the interested targets instead of the whole area, researchers have defined target-based coverage problems. Unlike the area coverage, this issue puts emphasis on how to cover the maximum number of targets. Several solutions have been presented for target-based coverage enhancement [19–21]. On the other hand, enhancing area coverage is an very important issue for DSNs to fulfill the specified sensing tasks. The objective is to achieve maximal sensing region with a finite number of sensors. As we aim at maximizing the area coverage, we will discuss only the area-based coverage enhancement solutions in details. Since a small unmonitored sub-area defeats the whole purpose of the network, sensor nodes need to be spread as uniformly as possible over the entire sensing region with minimum gaps. However, random deployment may cause several problems, such as overlapped and occluded regions, uncovered areas, and broken sensor nodes. Therefore, three basic (well-known) principles have been utilized by the research community to overcome these difficulties. First approach is based on the redeployment of new sensors after the initial deployment. This approach has taken too little interest due to its several drawbacks. For instance, deploying additional nodes to the estimated positions is extremely difficult. Thus, in the monitored

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

area, there will be many redundant nodes. Moreover, each attempt for redeployment will cost more due to the nature of available redeployment methods. Second approach for coverage enhancement in DSNs is to adjust the working directions of the directional sensor nodes to improve the field coverage. Several studies [15,17,22–25] exploit this approach, i.e. the motility capability, to maximize the total area coverage. The last approach offers the utilization of the mobility capability. To the best of our knowledge, in [18], Liang et al. have primarily proposed to relocate the directional nodes for the coverage improvement. The study [15] is one of the pioneering works on coverage enhancement. The authors present a new method based on a rotatable sensing model. To achieve less overlapping area, a directional node repositions itself on the reverse direction of the interior angle-bisector occurring between two neighboring directional nodes. On the other hand, Cheng et al. describe the area-coverage enhancement problem as the Maximum Directional Area Coverage (MDAC) problem and prove the MDAC to be NP-complete [24]. In their study, the authors define two new concepts, virtual sensor and virtual field. A virtual sensor represents one working direction of a directional sensor, whereas a virtual field is a minimal region that is formed by the intersection of the sensing regions of a number of virtual sensors. The distributed solution for MDAC problem, DGreedy algorithm, chooses the least overlapped direction as the new working direction. The authors observe that scarce sensors are highly critical to achieve maximal coverage, thus they utilize the number of sensing neighbors to differentiate the priority which represents the decision slot of the sensor nodes. On the other hand, Liang et al. have recently investigated the same MDAC problem and proposed two new algorithms [9], both of them outperforming the DGreedy algorithm. They basically introduce a new scheduling algorithm for the decision order of the nodes. They utilize the overlapped ratios of the nodes for the prioritization. Nodes with non-overlapped FoVs decide with top priority. Hence, their solutions achieve more coverage gain than DGreedy. Zhao and Zeng [17] have adapted the theory of the virtual potential field to wireless multimedia sensor networks for coverage improvement. They propose an electrostatic field-based coverage-enhancing algorithm (EFCEA) to enhance the area coverage of WMSNs by turning sensors to the correct orientation and decreasing the coverage overlap of active sensors. They also aim at maximizing the network lifetime by shutting off as much redundant sensors as possible using the grid-based approach, and waking them up according to a correlation degree. In [23], the authors name the above mentioned coverage problem as the optimal coverage problem in directional sensor networks (OCDSNs). They propose a greedy approximation algorithm to the solution of the OCDSNs problem, based on the boundary Voronoi diagram. By constructing the Voronoi diagram of a directional sensor network, one could find the maximal breach path of this network. The authors introduce an assistant sensor that can obtain the global information by traveling the edges of the Voronoi diagram. While moving, the assistant sensor

1077

determines which sensor to wake up in order to ensure the uncovered boundaries to be covered. In [11,16], Tezcan and Wang have studied the problem of self-orientation in WMSNs, that is finding the most beneficial orientation for all multimedia sensors to maximize the multimedia coverage. Unlike previous works, they define obstacles in the observed area and propose a solution for occluded regions. They aim at both minimizing the overlapping areas and enabling occlusion-free viewpoints. Their simulation results show that the occlusion-free viewpoint approach increases the multimedia coverage significantly. Apart from the existing approaches, in this study, we present a novel approach, named hybrid movement strategy, which utilizes motility and mobility in a cascaded manner. Our hybrid solution first enables the directional nodes to adjust their poses into appropriate directions consuming a reasonable energy. Afterwards, regarding the size of the redundant FoVs and the coverage holes, our proposed solution determines some nodes to change their physical location to enhance the area coverage further. Centralized solutions of DSN systems may not clearly scale since DSNs are usually composed of a large number directional nodes. On the other hand, nodes may fail due to battery outage or external effects, which should be handled by a dynamic update of the working directions. This ! update of the working direction ðW d Þ can be performed via local information exchange among neighboring sensors. Thus, the proposed parts of the hybrid solution is designed as distributed algorithms, making it a suitable candidate for deployment in practical systems. 3. Directional coverage A directional sensor network with N sensors represented by S = {s1, s2, . . . , sN} can be deployed in a polygonal sensing field, denoted by A. We also assume that each node is equipped with some hardware to learn its location information via any lightweight localization technique for wireless sensor networks [26]. In this context, some definitions that will be used in the rest of the paper are as follows. Definition 1. Sensing radius (Rs). The sensing radius of a directional node gives the depth of its FoV. The greater Rs, the greater the FoV.

Definition 2. Communication radius (Rc). The communication radius indicates how far a node could transmit/ receive a signal to/from another node without using the help of any intermediate nodes. d(i, j), in Eq. (1), refers to the distance between si and sj. Assuming Rc P kRs (k P 2) in a network, sensor nodes si and sj are counted as neighboring nodes, if d(i,j) 6 kRs. These sensors use neighbor information to compute non-overlapping viewpoints. A communication link exists between sensor si and sensor sj if a single-hop transmission from si to sj and sj to si can be performed successfully. Note that, in several studies, k has been chosen as 2. However, there exist some special nodes whose communication radius is at least three times greater than its sensing radius. In

1078

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

this study, we assume that Rc = 2Rs and nodes communicate omni-directional.

dði; jÞ ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðxi  xj Þ2 þ ðyi  yj Þ2

ð1Þ

Definition 3. Angle of View (AoV). AoV denotes the largeness of the view of a directional sensor node. Directional sensors have a limited angle of sensing coverage due to technical constraints and/or cost considerations. The size of the AoV may theoretically range from 1° to 360°. If the angle equals to 360°, the sensing model of the node can be described as omni-directional. DSNs consisting of sensor nodes with smaller AoVs require excessive number of nodes to achieve a given coverage ratio. ! Definition 4. Working direction ðW d Þ. The direction to which a directional sensor faces is the working direction of this sensor. In DSNs, sensors may have different working directions after a random deployment. In this case, re-orientation of sensors is required to maximize the coverage. Moreover, due to external effects or application-specific requirements, sensor nodes may need to change/re-orient their working direction over time. Also, nodes may fail due to battery outage or hostile effects, which should be handled by a dynamic update of their working directions. Adjusting working directions can be performed via local information exchange among sensors. Definition 5. Field of View (FoV). The term field of view refers to the directional view of a sensor and is assumed to be an isosceles sectoral unit (two-dimensional approximation) as shown in Fig. 1. The FoV of a directional sensor node ! s is denoted by a 4-tuple ðP; Rs ; W d ; aÞ, where P is the loca! tion, Rs is the sensing radius, W d is the working direction, and a is the AoV of the sensor node s. The common directional sensing capability for 2D spaces is illustrated in Fig. 1. 3.1. Target in sector test for grid-based area coverage In this study, we exploit the grid-based approach [27] and the Target In Sector (TIS) test to model and solve the

Fig. 1. A directional sensor node senses a unit of sector described with the position (P), the working direction (Wd), the sensing radius (Rs), and the AoV (a). A target (T) may be covered if it is located within the FoV of the node. It is found by the TIS test.

area coverage problem. In grid-based approach, the area is divided into several cells with the help of grids. Each cell represents a point in the observed area. According to the binary detection model, a sensor node theoretically covers each point in its FoV. Fig. 2 models the initial FoV and the possible FoV of a node using grid-based approach. Target In Sector (TIS) test helps to find out whether a given target is in the FoV of a sensor node or not. The two conditions, given in Eqs. (2) and (3), are tested in order to determine whether a target is covered by a directional sensor s.

dðPs ; Pt Þ 6 Rs a ! ! Ps Pt :W d P dðPs ; Pt Þ cos 2

ð2Þ ð3Þ

d(Ps, Pt) denotes the distance between the target t and the sensor node s. Eq. (2) ensures that the target is within the sensing range of s, whereas Eq. (3) performs the actual FoV test. According to the binary model, the target is sensed if both conditions are satisfied. This approach is commonly used in target-based coverage problems [27,19,20,28]. For area coverage problems, researchers opt for gridbased approach to adapt this test model for indicating the (un)covered points in the observed area. Each point around the sensor node s is tested with the TIS test. The coverage map of the sensor node s is then created according to the test results, as shown in Fig. 2. 4. Hybrid movement strategy for coverage enhancement in DSNs HMS is a novel approach for the coverage problem in DSNs. It encourages the use of motility and mobility in a consecutive order, i.e. this approach consists of two parts: motility and mobility. Thus, any solution based on motility/ mobility can be easily adapted to both parts. The main reason why motility precedes mobility in the HMS is the high energy consumption values of the mobility [29,30]. In motility, compared to mobility, energy requirements are low, since only the sensor transducer has to be moved while the bulkier parts such as the motors, the battery, and the processor board, remains stationary. Therefore,

Fig. 2. A directional sensor node covers a unit sector within its sensing radius. In this figure, a coverage map of a sensor node is illustrated using the grid-based approach where zeros and ones represent uncovered and covered points, respectively. According to the binary detection model, we assume that each point within the FoV of the node is covered.

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

the key idea of the HMS is to use first motility due to its low energy requirements, to arrange the directions of the nodes to provide as much coverage as possible and only then to move the nodes, if there can be some further coverage gain. On the other hand, applying first mobility would yield eventually the same coverage improvement as in HMS but at the expense of the movement of unnecessary number of nodes and the depletion of too much energy. Thus, after the initial deployment, we first exploit motility to minimize the overlapped areas and occluded regions. Then, we check for possible coverage holes. If there are redundant sensor nodes and coverage holes, we redirect these nodes to these holes. A block diagram of this hybrid movement strategy is given in Fig. 3. In this study, we have designed our own motility-based and mobilitybased algorithms (AFUP,W-AFUP,WNE) and hybrid movement strategy algorithm (MAM), to show the powerful aspects of the hybrid movement strategy. In the following subsections, we discuss the algorithms, AFUP, W-AFUP, WNE, and MAM (W-AFUP + WNE) in details. 4.1. Motility This capability aims at rotating the sensor nodes with higher overlap ratio towards uncovered regions to minimize the overlapped regions in the sensing area. We propose two new distributed algorithms with details given in the following subsections. Both algorithms utilize local information to determine the final working directions of the sensor nodes. 4.1.1. Attraction Force of Uncovered Points (AFUP) algorithm After the initial deployment, directional sensor nodes need to be positioned towards uncovered areas both to minimize possible overlapping/occlusion and to cover ! those uncovered points. To adjust the W d of a directional sensor node towards such an uncovered area, we may assume that each point in this area exerts a positive attractive force on the given node. AFUP exploits these ! attractive forces to determine a more appropriate W d for a given node. It uses a grid-based approach, where each cell in the grid represents the points of the observed area. AFUP basically utilizes two different maps: coverage map and attractive force map. In its coverage map, each node saves its FoV and the points within its Rs belonging to the FoVs of its neighbors. On the other hand, the attractive force map is generated using the coverage map. The uncovered points are considered as attraction forces and marked with zeros, whereas the covered points by neighboring nodes are indicated by ones. Besides, three message formats are defined for the AFUP algorithm:

1079

DISCOVER_MSG, NEIGHBORHOOD_MSG and WD_MSG. DISCOVER_MSGs and NEIGHBORHOOD_MSGs are utilized in the AFUP_INIT phase, whereas WD_MSGs are exchanged in each iteration in the AFUP_NODE phase among the nodes. The algorithm consists of two main phases, AFUP_INIT and AFUP_NODE. PHASE 1. AFUP_INIT phase aims at both forming subgroups in the network and prioritizing the nodes with lesser number of neighboring nodes. Given that N directional sensor nodes are deployed, each sensor node discovers its neighborhood within its communication radius (Rc P 2Rs) via exchanging DISCOVER_MSGs. DISCOVER_MSG carries off the id, the location (P), and ! the current working direction ðW d Þ information of the related node. Although Rs and a are necessary for the calculation of the FoVs of the nodes and should be included in DISCOVER_MSG, they can be safely left out in homogenous sensor networks in order to reduce network communication, since all the deployed nodes have identical Rs and a values. However, in networks consisting of heterogeneous nodes, DISCOVER_MSG should definitely include both Rs and a values. After the discovery phase, N subgroups are formed within the network, where each sensor node belongs to at least one subgroup. Then, nodes start exchanging the number of their neighbors via NEIGHBORHOOD_MSGs. According to the number of neighbors, each node is assigned to a priority. A sensor node with lesser number of neighboring nodes has higher priority and it is the first node to decide its new working ! direction. Since the nodes only change their W d , and not their physical location, the number of their neighboring nodes remains the same. Thus, the prioritization is needed only once at the beginning of the AFUP_INIT phase. The prioritization is necessary for planning the running order of the nodes during the iteration process. Therefore, each node maintains a priority table (PTABLE). PTABLE holds the ids and the number of neighbors of the neighboring nodes. Each node updates its PTABLE according to the received NEIGHBORHOOD_MSGs. After the last NEIGHBORHOOD_MSG, which is detected by a time-out mechanism in the underlying network layer, each node sorts its PTABLE by the number of neighbors and ids. The sorting process always yields a unique running order for each participating node even with the same number of neighboring nodes, as each node is assigned to a unique id. Once the first node finishes its AFUP_NODE phase, each correctly received WD_MSG causes the next node in the PTABLE to start its AFUP_NODE phase. Thus, contention between the nodes could be easily prevented. PHASE 2. AFUP_NODE phase is the core of the proposed algorithm and takes place iteratively. At the beginning of

Fig. 3. A block diagram of the hybrid movement strategy to the coverage problem in DSNs.

1080

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

this phase, each node is in the so called ‘‘unbalanced’’ state. In each iteration, sensor nodes determine their new candidate working directions. However, this new direction is not considered as final until the node reaches the ‘‘balanced’’ state. Some sensor nodes may not find an appropriate working direction after several iterations, especially when the node density is too high. In such scenarios, to prevent infinite oscillations, the corresponding sensor nodes update their status as ‘‘balanced’’ after a predefined number of iterations. Once the status of a node is changed to ‘‘balanced’’, the node finalizes the AFUP_NODE phase. AFUP_NODE phase runs iteratively and consists of three main steps. Each node starts the iteration with ‘‘Detection of Overlapped Regions’’ and finalizes it if the condition in ‘‘Threshold-Value Test’’ is satisfied. Each node, which fails the test, runs the third step to find a new appropriate working direction. The elaboration of the AFUP_NODE phase is given below in the following items. 1. Detection of overlapped regions. First, the node marks the points covered by itself in its coverage map (2Rsx2Rs) ! using its Rs, a and W d (Fig. 4a). Then, it obtains an up! to-date W d for each of its neighbors using WD_MSGs. Afterwards, the node calculates the FoVs of its neigh! boring nodes using their respective P; a; Rs ; W d values and updates the coverage map using the points located within its sensing radius. Finally, the node counts the number of overlapped points (NOP) to determine the state of the node, shown as in Fig. 4b. 2. Threshold-Value Test. Threshold-Value Test is performed to determine the state of the node. The values in the coverage map represent the number of sensors, which cover a given grid cell. Grid cells with a value greater than 1 indicate overlapped cells (points). The node sums up the number of overlapped points in its coverage map, shown as in Fig. 5. Next, the sensor node compares the total number of overlapped points to a predefined threshold value (Th). If the value is less than this threshold value, the sensor node finishes the AFUP_NODE phase and it physically turns to the last determined new working direction. If a node could

Fig. 4. Zeros represent uncovered points. In the initial coverage map of sensor A, covered points are denoted by ones, which get updated after node A receives coverage information from nodes B and C. As a result in (b), the number in each cell shows how many nodes cover the corresponding cell. The default width and height of the maps of a sensor node are denoted by 2Rsx2Rs.

not find an appropriate working direction after several iterations (fixedIterationNumber: its value is determined according to the density of the network. However, the algorithm mostly converges after 5 or 6 iterations), this node is regarded to be oscillating and its state is forcibly changed to ‘‘balanced’’. The threshold value has been considered as the k percentage of the FoV. After several tests, k has been chosen as 10%. Eq. (4) gives the formulation of the Threshold-Value Test.

( Sstate ¼

balanced

NOP < k  a2 R2s

unbalanced NOP P k  a2 R2s

ð4Þ

3. Determination of the new working direction. A sensor node, which cannot pass the Threshold-Value Test, marks the points covered by its neighbors in its attractive force map. Then, it calculates the center of the uncovered points. The uncovered points are represented with a value of ‘‘0’’. Each point is considered to perform an equal positive attractive force on the working direction of the node. The center (xm,ym) of the attractive forces is calculated using Eq. (5), where Fij represents the attractive force at the point represented by x and y. Given that the sensor node is located at the   m point ðx0 ; y0 Þ; arctan yx00 y gives the new working xm direction of the sensor node.

PRs

PRs

i¼Rs F ij  x PRs j¼Rs i¼Rs F ij PRs PRs j¼R i¼Rs F ij  y ¼ PRs s PR s j¼Rs i¼Rs F ij

xm ¼

j¼Rs

PR s

ym ð5Þ

4. Informing the neighborhood. After calculating the new working direction, the sensor node informs the neighboring nodes via WD_MSG about its candidate new working direction. The node waits then for the next round. WD_MSG involves only the exchange of the working directions of the neighboring nodes. Thus, no exchange of coverage maps is required.

Fig. 5. In the attractive force map of sensor A, a value of ‘‘0’’ in a cell indicates that the given cell is covered by any of the neighboring nodes. Thus, this cell exerts an attractive force to the sensor A. All the remaining cells are denoted by a ‘‘x’’ in the map and they do not contribute to the attractive force map calculation. Zeros represent the attractive forces on the node A, whereas x0 s apply no force at all to A.

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

Algorithm 1. The pseudo-code of the AFUP Algorithm ! Each sensor node knows its location (P(x0,y0)), W d ; Rs and a. /* Parallel computation */ /* AFUP_INIT */ set the parameter Sstatus = unbalanced; set the parameter forceThreshold (Th); ! exchange DISCOVER_MSG ðP; W d ; Rs ; aÞ create neighboring sensor list; count the number of neighbor nodes exchange NEIGHBORHOOD_MSG (Number of Neighbors) /* Prioritization */ while (NEIGHBORHOOD_MSG!=numberOfNeighbors) do update the priority table; end while sort the priority table by numberOfNeighbors and ids if (node_id == PTABLE(1)) then start AFUP_NODE phase; else wait for the WD_MSG from the previous node in the PTABLE and then start the AFUP_NODE phase end if /* AFUP_NODE */ oscillation = 0 while (Sstatus = unbalanced) AND (oscillation < fixedIterationNumber) do sum covered points; ! collect W d from neighbor sensors; calculate NOP; if NOP < Th then Sstatus = balanced; else find UNCOVERED points within the FoV; calculate the CENTER (xm,ym) of uncovered points; ! set the W d towards the CENTER point; inform neighboring sensors; end if oscillation ++; end while

1081

The main difference of W-AFUP from AFUP lies in two additional sub-steps included in the determination of the new working direction. Therefore, we will explain this step in detail to make the difference more precise. It should be emphasized that this difference is the key factor of the superior performance of W-AFUP. Determination of the new working direction. The key idea of W-AFUP is to amplify the attractive effect of the points which are far away from the points covered by other sensor nodes. To this effect, W-AFUP assigns a weight to each attractive force exerted by the aforementioned points (Fig. 7). In the AFUP algorithm, this weight was chosen to be ‘‘1’’s for all points under consideration. Once a weight is given to each point, W-AFUP calculates the center of these uncovered points, which in turn is considered to be the center of the positive attractive force to set the new ! W d of the node. To determine the weight of attractive force of any given uncovered point, we search for the closest covered point. Then, the Euclidean distance to the closest covered point becomes the weight of the attractive force (Eq. (6)). Searching for the closest point is computationally intensive. However, instead of scanning the whole map, we designed an efficient Spiral Search (SS) algorithm. The SS algorithm runs circularly starting from the uncovered point until either it reaches the borders of the map or a matching point is found. The search radius r is set to 1 initially, and incremented by one at each iteration. All candidate points on the circle, defined by r, are tested in order to find a covered point around the uncovered point. At the first match, the SS terminates and the distance is calculated and assigned as the weight of the attractive force of the related point. Thus, several comparison operation could be saved. The principle of SS algorithm is shown in Fig. 6.

wUP ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðxup  xcp Þ2 þ ðyup  ycp Þ2

ð6Þ

The center (xm, ym) of the attractive forces is calculated using Eq. (7), where Fij is equal to one and wij refers to the weight of the force at that location (i, j). Given that a   m sensor node is located at the point ðx0 ; y0 Þ; arctan yx00 y xm gives the new working direction of the sensor node.

4.1.2. W-AFUP W-AFUP is the enhanced version of the AFUP algorithm where W refers to ‘‘weighted’’. W-AFUP consists of the same steps as the AFUP algorithm.  W-AFUP_INIT Neighborhood discovery Exchange of number of neighbors Priority determination  W-AFUP_NODE Detection of overlapped regions Threshold-Value Test Determination of the new working direction Informing the neighborhood

Fig. 6. In the AFUP algorithm, each ‘‘0’’ cell has equal weight and thus it is assumed that it exerts an attractive force of weight ‘‘1 ’’ to the sensor S. On the other hand, SS algorithm is run to calculate the weight of attractive force of each ‘‘0’’ cell exerted to the sensor S. In this example, the weight of the zero with red color is equal to 2, since the closest covered point denoted with a red circle is located 2 units away.

1082

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

PRs xm ¼

PRs ¼

PRs

i¼Rs wij F ij  x Rs X Rs X wij F ij

j¼Rs

ym

j¼Rs i¼Rs

PRs

i¼Rs wij F ij  y Rs X Rs X wij F ij

j¼Rs

ð7Þ

proposed to apply mobility after motility and named this approach as hybrid movement strategy. To demonstrate the efficiency of the HMS, we designed our own mobility algorithm and called it Window-based Neighborhood Exploring (WNE). It runs in a distributed manner and utilizes local information. The details of WNE algorithm are discussed in the following subsection.

j¼Rs i¼Rs

4.2. Mobility Mobility is very important in sensor networks, since it may heal several network problems, including coverage and connectivity [31]. For example, random deployment is expected to cause several coverage holes within the observed area especially in harsh/dangerous deployment regions (scenarios). Moreover, a certain number of nodes may loose their functionality due to node-specific reasons, such as running out of battery or damages originating from the environment. Thus, there may occur uncovered regions both during the initial deployment and the network lifetime. One of the preferred solutions to cover these regions is to relocate the nearest mobile nodes. There are several such solutions using the mobility for the coverage problem in omni-directional sensor networks [32–36]. Although mobility is the main solution for coverage enhancement in omni-directional sensor networks, in DSNs it is not as energy-efficient solution as motility. To the best of our knowledge, only one study [18] has explored mobility in DSNs. Nevertheless, they did not exploit motility. On the other hand, researchers propose several solutions exploiting only motility due to its advantages. However, our simulation results show that motility could improve the coverage only up to a limit especially above a certain threshold of node density. After implementing several scenarios, we observed that in uniformly deployed DSNs, where nodes could cover up to 55% of the observed area when they placed manually, motility could satisfactorily minimize the overlapped areas. However, DSNs above this threshold and non-uniformly distributed DSNs need to exploit mobility capability of the nodes. Thus, we

4.2.1. Window-based Neighborhood Exploring (WNE) WNE algorithm consists of two main phases. WNE_INIT and WNE_NODE. In the first phase, each nodes discovers its neighborhood via using DISCOVER_MSG. After obtaining ! the P(x,y), W d , a and Rs information of neighboring nodes, each node calculates the number of neighboring nodes and propagates this information via NEIGHBORHOOD_MSG. Thus, each node determines its running order based on the number of neighboring nodes. This sub-step is called prioritization and it occurs at each round of the algorithm, if the physical position of a node changes within the network. The priority table (PTABLE) of each node is maintained by the node via three messages: DISCOVER_MSG, NEIGHBORHOOD_MSG, and REMOVE_MSG. When a node decides to move to a new location, it announces this condition with a REMOVE_MSG. As a result, the PTABLEs of all the remaining nodes get updated and new running order is calculated. Whenever this every node reaches its target location, it again announces its presence via a DISCOVER_MSG. This discovery message acts like the REMOVE_MSG and causes a respective update of PTABLEs at the target location. Thus, PTABLEs of all the nodes are synchronized to the physical movements of the nodes and they always reflect the current prioritized node information at both the source and destination locations. The node with lesser number of neighboring nodes starts the second part of the algorithm, called WNE_NODE. This phase includes five main steps and will be explained in the following paragraphs in detail. The psuedo-code of the WNE algorithm is given in Algorithm 2. WNE algorithm basically aims at finding uncovered regions around the neighborhood of a given node. If a node has more overlapped points than a predefined value (Th), after several test scenarios Th is chosen as 1% of the FoV

Fig. 7. At initial step, the attractive force map consists of zeros and ‘‘x’’s, where zeros represent attractive forces. In the AFUP algorithm, each uncovered point performs an a positive attractive force and the forces are equal to one, as shown in (b). On the other hand, in W-AFUP, each uncovered point has a weight, as shown in (c). SS algorithm calculates the weight of each forces according to the Eq. (6) and stores the new values to attractive force map. Note that, each uncovered point has a weight which is calculated according to its distance to the closest covered point (Fig. 6).

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

of a sensor node, a horizontal and a vertical window are created. In each iteration, these windows are utilized to scan the neighborhood map horizontally and vertically in order to determine the largest uncovered area on the map. The width and height of the windows are set to Rs and 2Rs cos a2, respectively. Within the windows only cells (points) corresponding to the AoV of the deployed nodes are set to ‘‘1’’. All the remaining points are set to zero. Thus, during the scanning process, only the cells with a value of ‘‘1’’ contribute to the calculation of an uncovered area. The direction of the AoV is set as follows. The horizontal window is assumed to have an imaginary node located at the middle of its left edge ! with a W d of 0°. Similarly, the vertical window has an ! imaginary node at the middle of its top edge with a W d ! of 270°. The W d s of these imaginary nodes are used to represent the direction of the windows. The aforementioned horizontal and vertical windows are illustrated in Fig. 8. WNE could benefit from other types of windows, such as diagonal window, to increase the performance. However, with the increased number of windows, the run-time of the WNE algorithm gets longer. Algorithm 2. The pseudo-code of the WNE algorithm Each sensor node knows its location (P(x0, y0)), ! W d ; Rs and a. /* parallel computation */ /* WNE_INIT */ set the parameter Sstatus = unbalanced; set the parameter forceThreshold (Th); ! exchange DISCOVER_MSG (P; W d ; Rs ; a); create neighboring sensor list; count the number of neighboring nodes; exchange NEIGHBORHOOD_MSG (Number of Neighbors); set the priority; /* WNE_NODE */ iterationNumber = 0 while (Sstatus = unbalanced) AND ( iterationNumber < fixedIterationNumber) do sum covered points; ! collect W d and P (x,y) of one-hop and two-hop neighbors, from one-hop neighboring sensors; calculate NOP; if (NOP < Th) then Sstatus = balanced; else create the windows for the scanning process; /* scanning the Neighborhood Map */ while (NOP < NFoV) do max = NEP; maxLocation = (x, y); end while if (NEP > PEP) then inform the current NEIGHBORS; move to the NEW LOCATION (maxLocation); ! set the W d as the DIRECTION of the

1083

SCANNED WINDOW; after moving the new location inform the new neighbors and collect their information; Prioritization end if end if iterationNumber++; end while

WNE uses a grid-based approach like in (W-) AFUP algorithms, where each cell in the grid represents the points of the observed area. The WNE_NODE phase takes place iteratively like AFUP_NODE and W-AFUP_NODE, and consists of five main steps. Detection of overlapped regions. A sensor node first ! builds its coverage map using its W d ; Rs and a and then it marks the overlapped points covered by its neighboring nodes. The necessary information including the position and the current working direction of the neighboring nodes are exchanged via DISCOVER_MSGs. Afterwards, each node calculates the number of overlapped points in its map. Threshold-Value Test. According to the result of the Threshold-Value Test, the node decides whether to search a new location within its neighborhood area or not. This test is identical to the tests utilized both in AFUP and WAFUP. The node compares the total number of overlapped points to a predefined threshold value (Th). If the number of overlapped points is above the threshold (Th) value, the node fails the Threshold-Value Test and proceeds to the next step. Otherwise, it finalizes the WNE algorithm and maintains its physical position. Scanning the neighborhood. After a failed Threshold-Value Test, the node scans its neighborhood map using the windows created. The basic idea behind this window scanning is to find the greatest uncovered region within the neighborhood map. The scanning process, shown in Fig. 9, involves the bitwise-AND operation. During the scanning process, the respective window is used as a sliding window starting from the left top edge of the neighborhood map. In each iteration, the window is shifted one cell

Fig. 8. Two windows (horizontal and vertical) are utilized during the scanning process for finding possible empty regions within the window. As we only seek for a sectoral empty region, we mark inside points of the sector with ‘‘1’’s and outside points ‘‘0’’s. Afterwards, we apply the bitwise-AND operation to each point in the neighborhood map.

1084

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

5. Performance evaluation We implemented a simulation environment using MATLAB 7.8 in order to model the self-orientation capabilities of directional sensor nodes in two dimensional terrains. In the following subsections, we first give our simulation parameters and their values. Then, we present the performances of our motility-based (AFUP, W-AFUP), mobilitybased (WNE) solutions, and the HMS (MAM) comparatively. MAM is the particular implementation of the HMS and consists of W-AFUP and WNE algorithms. In MAM, first, W-AFUP is run on the sensor nodes. Afterwards, once all nodes are ‘‘balanced’’, W-AFUP finalizes itself and WNE algorithm starts to run on the nodes. In WNE, some nodes might change their physical locations and move to a more appropriate (empty) region. After a predefined number of iterations, the nodes finalize the WNE algorithm, thus concluding the MAM algorithm too. 5.1. Simulation settings

Fig. 9. Scanning process is done on the neighborhood map of the node S to find the most empty region. Horizontal and vertical windows are used to scan the neighborhood map both horizontally and vertically.

to the right until the right edge is reached. Afterwards, the window is reset to the left edge with one cell shifted down. This procedure is repeated for the entire neighborhood map. At each iteration, each point of the window is bitwise AND’ed with the complement of the underlying point of the neighborhood map and a running sum is calculated. If the sum equals to the number of ‘‘1’’ within the window, this indicates a perfectly uncovered region, and the scanning process terminates. Then, the node moves to this new location. As it is difficult to find a perfectly uncovered region especially in dense networks, the greatest uncovered region among all the uncovered regions is recorded during the scanning process. If this region offers an empty area greater than the current location of the node, the node determines to move to this new location. Otherwise, it keeps its current location and working direction. Setting up the new Working Direction. The node assumes ! the direction of the window as its new W d . Informing the Neighborhood. There are two sub-steps. First, the node informs its neighbors before leaving its location, thus neighboring nodes remove this node from their list. Second, at its final location, it sends DISCOVER_MSGs in order to discover its neighborhood and to announce its presence in the environment.

In our simulations, sensor nodes are deployed randomly in a rectangular two-dimensional area. We use the uniform distribution function of MATLAB. The basic parameters of the nodes have been chosen similar to the configuration setups of existing solutions [15–17]. Therefore, all the nodes in the field are identical and cover with an Rs = 30 m and AoV = 60°. Their positions and working directions are determined randomly during the initial deployment. Communication between two sensors are assumed to be possible, if the distance between the transmitter and receiver is no more than 60 m. A sensing field, spanning an area of 250  250 m2 has been used, in which the number of sensors were varied to study the system performance from sparse to dense deployments. All scenarios have been run with 50 different initial deployments and the average values of them are given in the following charts and tables. In the deployments, each cell represents an area of 1  1 m2 in the field. The resolution of the sensing field could be increased. However, there is a trade-off between the computational complexity and the accuracy of the modeling of the sensing field. A higher resolution would increase not only the model accuracy of the sensing field but also increase the computational complexity. However, it is not the focus of this study. 5.2. Simulation results For the performance evaluation, we basically consider two key metrics. total coverage and energy consumption. Besides, we indicate the number of overlapped points and the number of moved sensors as two sub-keys for some scenarios. Our performance evaluation tests consists of two parts. In the first part, we demonstrate the coverage gain of the proposed motility-based solutions, AFUP and W-AFUP, over the coverage performance of random and optimum deployment. We also show that in the motility only solution, coverage improvement is still possible above a certain node density. In the second part, we evaluate the

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

Fig. 10. Coverage ratios of random deployment, AFUP, W-AFUP, and optimum deployment.

performance of the HMS over motility and mobility only solutions. In this study, optimum deployment (OD) refers to the deployment, where sensor nodes are placed manually with appropriate working directions causing zero overlapping. Given that a sensor node covers a2 R2s units, N sensor nodes would cover N a2 R2s units. Following this formula, the coverage ratio of the optimum deployment is calculated as given in Eq. (8), where A, N, a, and Rs refer to the size of total area, the number of sensor nodes, the AoV and the sensing radius of the nodes, respectively. In Fig. 10, coverage ratios of random deployment, AFUP and W-AFUP are compared against the coverage ratios of the optimum deployment. It should be noted that the coverage ratio of OD exceeds 100%. This is because according to Eq. (8) in an area A of 250  250 m2 and with AoV = 60° and Rs = 30 m the CROD becomes 113. Thus, this value shows that we have deployed an excessive number of nodes in the field.

CROD ¼

Fig. 11. Overlap minimization ratios of AFUP and W-AFUP. With 150 nodes, the AFUP algorithm fails to minimize the overlapped regions further, whereas W-AFUP could still minimize overlapped regions.

1085

N a2 R2s  100 A

ð8Þ

In Fig. 10, the coverage performances of both AFUP and W-AFUP are given. Both algorithms improve the coverage substantially after the initial deployment. However, WAFUP achieves up to 10% more coverage improvement than the AFUP algorithm. Thus, the network coverage after WAFUP gets slightly closer to the results of the optimum deployment. Moreover, applying W-AFUP after random deployment reduces the overlapped regions up to 90%

Fig. 12. Both AFUP and W-AFUP converges as quickly as in 5–6 iterations. However, denser or non-uniformly distributed networks may require higher number of iterations for the nodes to reach the balanced state.

1086

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

Fig. 13. Three figures illustrate how the coverage is improved with the HMS consecutively.

where AFUP achieves only 77% overlap minimization. Especially for dense networks, W-AFUP is still able to minimize the overlapped points (Fig. 11), where AFUP could not achieve more reduction of the overlapped areas. This fact manifests itself in Fig. 11 by a negative overlap minimization value for 150 nodes. Thus, the network became denser than AFUP can handle and it was not able to find empty areas anymore.

Analyzing both Figs. 10 and 11 shows that there is significant performance difference between AFUP and WAFUP. Thus, we have considered W-AFUP for the rest of our experiments. Another point which needs to be clarified is the iteration number chosen for preventing the oscillation. The convergence time depends on the number of nodes within subgroups. With the increasing number of nodes, it would be better to set the iteration number to

1087

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090 Table 1 Hybrid movement strategy (MAM) vs. random algorithm. Rs = 30 m a = 60°

Random deployment (%)

Hybrid movement strategy (MAM) (%)

Increment of total coverage

Coverage gain (%)

N = 25 N = 50 N = 75 N = 100 N = 125 N = 150

15.90 26.90 39.70 48.71 56.49 63.34

18.74 37.36 55.41 70.31 81.36 88.38

2.84 10.46 15.71 22.60 24.87 25.04

17.86 38.88 39.57 46.40 43.97 39.53

Fig. 14. Comparison of random deployment, the HMS, motility only solution, mobility only solution, and optimum deployment in terms of coverage ratios.

higher values. However, simulation results show that AFUP and W-AFUP algorithm mostly converges in 5–6 iterations, shown as in Fig. 12. Since the main goal of this study is to show the advantages of the hybrid movement strategy, the next experimental results include the comparison of motility, mobility and HMS. Fig. 13a–c illustrate the significant coverage improvement obtained with the HMS. In Table 1, we give the numerical results of the HMS. Then, in Fig. 14, we demonstrate the coverage ratios of random deployment alone, with motility (W-AFUP), with mobility (WNE), and with HMS (MAM) compared to the optimum deployment for both sparse and dense sensor networks. The HMS

improves the coverage up to 7% more than the motility only solution due to the advantages of the mobility. Moreover, the HMS could slightly improve the total coverage up to 2% more than the mobility only solution. From sparse to moderate densities, the HMS almost achieves the same coverage gain as the mobility only solution, and it outperforms the mobility only solution in dense networks. Overlap minimization ratio is another important metric for the evaluation of the hybrid solution. Analyzing Fig. 15 shows that the hybrid solution could reduce the overlapped regions up to 98%. Especially in dense networks, the HMS performs significantly better than the motility and mobility only solutions at minimizing overlapped regions (Fig. 15). We have also analyzed the effect of changing the sensing radius and the angle of view of the nodes. Figs. 16 and 17 demonstrate the coverage ratios of the random deployment, the HMS, motility only, and mobility only solutions. With sensors having small FoVs the total coverage does not change too much, whereas nodes with large FoVs cause more contribution to the total coverage. The main advantage of the hybrid movement strategy is its reasonable energy consumption in the course of the self-orientation of the network. Physical movement is the most energy consuming activity of a sensor node. Besides, under limited mobility constraints [37,38], it is more challenging to move the nodes over long distances. As WNE maintains a list of candidate locations, with greater coverage than the current location, according to their distance, the nodes have the opportunity to move over shorter distances under limited mobility scenarios. Thus, HMS could

Fig. 15. Overlap minimization performance of the HMS, motility only and mobility only solutions.

1088

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

Fig. 16. Coverage gain performance of the HMS, motility only and mobility only solutions with different values of angle.

Fig. 17. Coverage gain performance of the HMS, motility only and mobility only solutions with different values of radius.

Fig. 18. Total travel distance of traveled sensors for the HMS, motility only and mobility only solutions.

adapt itself more easily to those constraints, since it causes in the movement of lesser nodes. Comparing motility against mobility, the consumed energy during motility is negligible [29,30]. Since in the motility only solution and in the HMS, the nodes change their working direction only once for at most 180°, we considered the total travel distance of the nodes to be used as the main criteria in the calculation of the energy consumption of the nodes. It is obvious that no physical movement of a node occurs during motility only solution. It was our

motivation to propose the HMS to find an optimum balance between coverage gain and total travel distance/energy consumption. Fig. 18 demonstrates that nodes during the HMS travel significantly less than nodes during mobility only solution especially in dense networks. We also give the total energy consumption values of the HMS, motility only and mobility only solutions in Table 2. These values have been calculated based on the energy consumption values of the sensor nodes in [29,30], where a directional node consumes only 1.5 J during 180° rota-

1089

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

Table 2 Total energy consumption of directional nodes exploiting the HMS, motility only and mobility only solutions. A directional node consumes only 1.5 J during 180° rotation, whereas 1 m physical movement of a node causes 3.6 J to deplete [29,30]. Number of nodes

25

50

75

100

125

150

Mobility (WNE) (J) Hybrid movement strategy (MAM) (J) Motility (W-AFUP) (J)

2322 147 13

7905 2311 33

14,313 8176 57

17,492 10,260 72

22,230 11,297 95

25,992 13,842 118

Table 3 Number of traveled directional sensor nodes. Mobile vs. motile/mobile sensor network. Total number of nodes

25

50

75

100

125

150

Mobility (WNE) Hybrid movement strategy (MAM)

9 1

29 8

49 28

65 35

80 39

89 43

tion, whereas 1 m physical movement of a node causes 3.6 J to deplete. Besides, Table 3 shows that the number of traveled nodes are at most 35% of the total number of nodes in the HMS. Thus, this fact causes most of the nodes to restrain their energy during the coverage improvement phase. 6. Conclusion Coverage enhancement in directional sensor networks has so far been achieved by exploiting motility or mobility only solutions. In this study, we have introduced a novel approach, the hybrid movement strategy, to this problem. In order to demonstrate the performance of the HMS in terms of the coverage improvement and energy efficiency, we utilized two new algorithms, AFUP [13] and W-AFUP, for exploiting motility, and WNE algorithm for utilizing mobility capability of the nodes. We also called the combination of W-AFUP and WNE as MAM, which is a particular version of the HMS. With MAM, we showed that up to 47% coverage gain is possible by exploiting motility and mobility consecutively. Moreover, compared to the mobility only solution, DSNs exploiting our hybrid movement strategy could save at least 40% energy, though obtaining a similar or even more coverage improvement depending on the density of the given network. Acknowledgement This research has been supported by Yildiz Technical University Scientific Research Projects Coordination Department under the grant number 2011-04-01-DOP03. References [1] M.A. Guvensan, A.G. Yavuz, On coverage issues in directional sensor networks: a survey, Ad Hoc Networks 9 (7) (2011) 1238–1255. [2] M.A. Guvensan, A.G. Yavuz, A hybrid solution for coverage enhancement in directional sensor networks, in: Proc. of the 7th Intl. Conf. on Wireless and Mobile Communications (ICWMC 2011), Luxembourg, 2011b. [3] I.F. Akyildiz, T. Melodia, K.R. Chowdhury, A survey on wireless multimedia sensor networks, Computer Networks 51 (4) (2007) 921–960.

[4] I. Akyildiz, T. Melodia, K. Chowdhury, Wireless multimedia sensor networks: applications and testbeds, Proceedings of the IEEE 96 (10) (2008) 1588–1605. [5] G. Fan, S. Jin, Coverage problem in wireless sensor network: a survey, Journal of Networks 5 (9) (2010). [6] C.-F. Huang, Y.-C. Tseng, The coverage problem in a wireless sensor network, Mobile Networks and Applications 10 (2005) 519–528. [7] D.G. Costa, L.A. Guedes, The coverage problem in video-based wireless sensor networks: a survey, Sensors 10 (9) (2010) 8215– 8247. [8] S. Soro, W. Heinzelman, A Survey of Visual Sensor Networks, 2009. [9] C.-K. Liang, C.-H. Tsai, M.-C. He, On area coverage problems in directional sensor networks, in: Proc. of the Intl. Conf. on Information Networking (ICOIN’11), 2011, pp. 182–187. [10] G. Fusco, H. Gupta, Selection and orientation of directional sensors for coverage maximization, in: Proc. of IEEE Intl. Conf. on Sensor, Mesh and Ad Hoc Communications and Networks (SECON’09), Rome, Italy, 2009, 1–9. [11] N. Tezcan, W. Wang, Self-orienting wireless multimedia sensor networks for occlusion-free viewpoints, Computer Networks 52 (13) (2008) 2558–2567. [12] X. Han, X. Cao, E. Lloyd, C.-C. Shen, Deploying directional sensor networks with guaranteed connectivity and coverage, in: Proc. of. IEEE Intl. Conf. on Sensor, Mesh and Ad Hoc Communications and Networks (SECON’08), San Francisco, USA, 2008, pp. 153–160. [13] M.A. Guvensan, A.G. Yavuz, A new coverage improvement algorithm based on motility capability of directional sensor nodes, in: Lecture Notes in Computer Science: Proc. of the 10th Intl. Conf. on Ad-hoc, Mobile, and Wireless Networks, ADHOC-NOW’11, Springer-Verlag, Berlin, Heidelberg, 2011, pp. 206–219. [14] M. Guvensan, A. Yavuz, Motility-based coverage enhancement in directional sensor networks, in: Proc. of the 20th Signal Processing and Communications Applications Conference (SIU’12), 2012, pp. 1– 4. [15] D. Tao, H. Ma, L. Liu, Coverage-enhancing algorithm for directional sensor networks, in: Lecture Notes in Computer Science: Mobile Adhoc and Sensor Networks, vol. 4325, 2006, pp. 256–267. [16] N. Tezcan, W. Wang, Self-orienting wireless multimedia sensor networks for maximizing multimedia coverage, in: Proc. of the IEEE Intl. Conf. on Communications (ICC’08), 2008b, pp. 2206–2210. [17] J. Zhao, J.-C. Zeng, An electrostatic field-based coverage-enhancing algorithm for wireless multimedia sensor networks, in: Proc. of IEEE Intl. Conf. on Wireless Communications, Networking and Mobile Computing (WiCom’09), Beijing, China, 2009, pp. 1–5. [18] C.-K. Liang, M.-C. He, C.-H. Tsai, Movement assisted sensor deployment in directional sensor networks, in: Proc. of Intl. Conf. on Mobile Ad-Hoc and Sensor Networks (MSN’10), 2010. [19] J. Ai, A.A. Abouzeid, Coverage by directional sensors in randomly deployed wireless sensor networks, Journal of Combinatorial Optimization 11 (1) (2006) 21–41. [20] Y. Cai, W. Lou, M. Li, Cover set problem in directional sensor networks, in: Proc. of IEEE Intl. Conf. on Future Generation Communication and Networking (FGCN’07), 2007. [21] Y. Cai, W. Lou, M. Li, X.-Y. Li, Energy efficient target-oriented scheduling in directional sensor networks, IEEE Transactions on Computers 58 (9) (2009) 1259–1274. [22] H. Ma, X. Zhang, A. Ming, A coverage-enhancing method for 3D directional sensor networks, in: Proc. of IEEE Intl. Conf. on Computer Communications (INFOCOM’09), 2009, pp. 2791–2795. [23] J. Li, R.-C. Wang, H.-P. Huang, L.-J. Sun, Voronoi based area coverage optimization for directional sensor networks, in: Proc. of Intl. Symposium on Electronic Commerce and Security (ISECS’09), vol. 1, 2009, pp. 488–493. [24] W. Cheng, S. Li, X. Liao, S. Changxiang, H. Chen, Maximal coverage scheduling in randomly deployed directional sensor networks, in: Proc. of Intl. Conf. on Parallel Processing Workshops (ICPPW’07), 2007.

1090

M.A. Guvensan, A.G. Yavuz / Ad Hoc Networks 11 (2013) 1075–1090

[25] C. Kandoth, S. Chellappan, Angular mobility assisted coverage in directional sensor networks, in: Proc. of Intl. Conf. on Network-Based Information Systems (NBIS’09), 2009, pp. 376–379. [26] T. He, C. Huang, B.M. Blum, J.A. Stankovic, T. Abdelzaher, Range-free localization schemes for large scale sensor networks, in: Proceedings of the 9th Annual Intl. Conf. on Mobile Computing and Networking, MobiCom’03, 2003, pp. 81–95. [27] H. Chen, H. Wu, N.-F. Tzeng, Grid-based approach for working node selection in wireless sensor networks, in: Proc. of the Intl. Conf. on Communications (ICC’04), 2004. [28] J. Wen, L. Fang, J. Jiang, W. Dou, Coverage optimizing and node scheduling in directional wireless sensor networks, in: Proc. of the IEEE Intl. Conf. on Wireless Communications, Networking and Mobile Computing (WiCom’08), 2008, pp. 1–4. [29] D. O’Rourke, R. Jurdak, J. Liu, D. Moore, T. Wark, On the feasibility of using servo-mechanisms in wireless multimedia sensor network deployments, in: Proc. of the IEEE 34th Conf. on Local Computer Networks (LCN’09), 2009, pp. 826–833. [30] K. Dantu, M. Rahimi, H. Shah, E. Babel, A. Dhariwal, G.S. Sukhatme, Robomote: enabling mobility in sensor networks, in: Proc. of IEEE/ ACM Intl. Conf. on Information Processing in Sensor Networks (IPSNSPOTS’05), 2005, pp. 404–409. [31] A. Ghosh, S.K. Das, Coverage and connectivity issues in wireless sensor networks: a survey, Pervasive and Mobile Computing 4 (3) (2008) 303–334. [32] N. Heo, P. Varshney, Energy-efficient deployment of intelligent mobile sensor networks, IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans 35 (1) (2005) 78–92. [33] G. Wang, G. Cao, T.L. Porta, Movement-assisted sensor deployment, IEEE Transactions on Mobile Computing 5 (6) (2006) 640–652. [34] G.G. Wang, G. Cao, P. Berman, T.F. La Porta, Bidding protocols for deploying mobile sensors, IEEE Transactions on Mobile Computing 6 (5) (2007) 563–576. [35] Y.-C. Wang, Y.-C. Tseng, Distributed deployment schemes for mobile wireless sensor networks to ensure multilevel coverage, IEEE Transactions on Parallel and Distributed Systems 19 (9) (2008) 1280–1294. [36] B. Wang, H. Lim, D. Ma, A survey of movement strategies for improving network coverage in wireless sensor networks, Computer Communications 32 (13–14) (2009) 1427–1436. [37] S. Chellappan, W. Gu, X. Bai, D. Xuan, B. Ma, K. Zhang, Deploying wireless sensor networks under limited mobility constraints, IEEE Transactions on Mobile Computing 6 (2007) 1142–1157.

[38] S. Chellappan, X. Bai, B. Ma, D. Xuan, C. Xu, Mobility limited flipbased sensor networks deployment, IEEE Transactions on Parallel and Distributed Systems 18 (2) (2007) 199–211.

M. Amac Guvensan received his B.Sc., M.Sc., and Ph.D. degrees in computer engineering from the Yildiz Technical University, Turkey, in 2002, 2006, and 2011 respectively. He is a research assistant at the Department of Computer Engineering, YTU. He is also a member of the Intelligent Systems Laboratory (ISL) at Yildiz Technical University. His current research interests include coverage optimization in directional sensor networks, deployment strategies and signal processing algorithms in wireless multimedia sensor networks. He visited the Wireless Networks and Embedded Systems Laboratory, University at Buffalo, The State University of New York, for 6 months within 2009–2010.

A. Gokhan Yavuz received his B.Sc., M.Sc., and Ph.D. degrees in computer engineering from the Yildiz Technical University, Istanbul, Turkey, in 1990, 1994, and 1999, respectively. He is currently an assistant professor at the Computer Engineering Department of Yildiz Technical University. He is also affiliated with the Scientific and Technological Research Council of Turkey as a senior researcher. He has received the IBM Faculty Award in 2008. His research interests include wireless sensor networks, operating systems, and real-time embedded systems.