An optimal coverage-preserving scheme for ... - Semantic Scholar
Computer Communications 30 (2007) 2708–2720 www.elsevier.com/locate/comcom
An optimal coverage-preserving scheme for wireless sensor networks based on local information exchange Azzedine Boukerche a
a,*
, Xin Fei a, Regina B. Araujo
a,b
PARADISE Research Laboratory, University of Ottawa, Ottawa, Ont., Canada K1N 6N5 b VRLnet Lab, Federal University of Sao Carlos, SP, Brazil Available online 15 June 2007
Abstract Coverage-preserving and energy-saving solutions have been reported in the literature and are generally based upon both quality coverage and off-duty scheme. Off-duty scheme based solutions present at least three challenging problems: (1) keeping coverage and connectivity of the network while optimizing the number of active sensor nodes; (2) resolving conflicts when determining which nodes should be turned off in order to save energy; and (3) finding optimal wake-up strategies that avoid waking up more nodes than necessary. This paper presents a novel distributed solution, the optimal coverage-preserving scheme (OCoPS), that extends the well-known Central Angle Method in order to identify fully sponsored nodes. OCoPS comprises an extended Central Angle Method, new decision algorithms devised to resolve the off-duty conflict problem under different network densities, and an energy-aware wake-up scheme that solves coverage hole problems in off-duty schemes. Compared to the widely used node scheduling scheme, our solution is based on local information exchange without the uncertainty of self-schedule algorithms. OCoPS is implemented as an extension of LEACH. A set of simulation experiments is carried out to evaluate its performance compared to other well-known schemes which are based on the Central Angle Method and self-scheduling. Our results indicate that on network lifetime OCoPS outperforms other schemes by over 20% and by over 25% when the coverage rate is higher than 80%. The experimental results also show that our coverage scheme effectively limits the number of on-duty node compared to the other schemes. 2007 Published by Elsevier B.V. Keywords: Coverage; Wireless sensor network; Energy
1. Introduction Emergency Preparedness is a stringent wireless sensor network application class in which context aware physical environments subjected to critical conditions, such as fire or leaking of toxic liquids, are required to be reliably monitored. Wireless sensor networks (WSN) to support such a class of applications must provide a fast, reliable, fault-tolerant, and energy-aware channel for event diffusion, which meets the requirements of different sensor network application scenarios such as query-based (when a sink queries the WSN for specific information), event-driven (sensor nodes *
Corresponding author. E-mail addresses: [email protected] (A. Boukerche), xfei@ site.uottawa.ca (X. Fei), [email protected] (R.B. Araujo). 0140-3664/$ - see front matter 2007 Published by Elsevier B.V. doi:10.1016/j.comcom.2007.05.018
send information to the sink upon detection of an event), and periodic (sensor nodes periodically send information to the sink) [4,6,7]. These requirements must be met even in the presence of emergency conditions that can lead to node failures and path disruptions to the sink. Partial solutions have been introduced from the physical to the application layers [1–3]. These requirements must be met even under emergency situations that can lead to node failures and path disruptions to the sink. Partial solutions have been reported from the physical to the application layer. However, an important issue that begins to be fully addressed is how well a given area can be monitored by the WSN – an issue often related to QoS and known as coverage. Other questions related to QoS also arise under the coverage perspective: How to keep good quality monitoring in face of high rates
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
of failure? What is the optimal number of nodes that meets coverage quality and scarce energy? What are the breach points in a sensor field? When is tracking people or objects required when in a search for survivors during an emergency situation? How accurate is the location? Moreover, for supervision and control applications, different types of sensors with different processing power and communication properties (radio range) might be integrated in the same WSN, as happens with actor nodes (more resourceful nodes with the capability to respond to events in real-time) in Wireless sensor and actor networks – WSAN [9]. How can coverage be met with heterogeneous sensors? These are challenging questions for a WSN to cope with, and good solutions have been introduced in literature. These solutions tackle one or more of the problems discussed above by making coverage-preserving methods either aware of connectivity or energy, among others. However, a solution that addresses all of the issues raised above simultaneously poses a great challenge. Basically, the coverage problems of WSN are defined either as deterministic or nondeterministic deployment [23]. Because most wireless sensors are typically randomly deployed in difficult-to-access environments, nondeterministic deployment attracts more attention than deterministic deployment. At the same time, distributed or decentralized solutions, which keep coverage while maintaining network longevity, become interesting research topics for nondeterministic deployment. The main approaches to such solutions are set partition and node scheduling. The representative algorithm of set partition is the disjoint set of Cardei and Du [20]. Important algorithms of node scheduling include the work of Tian and Georganas (named here as C-PNSS) described in [13]; Ye et al. [16]; and the optimal geographical density control (OGDC) proposed by Zhang and Hou [17]. The main motivation for our work is the lack of novel solutions that meet coverage quality while maintaining a long network lifetime. In the disjoint set scheme proposed by Cardei and Du [20], global location information is required and is achieved through an energy-costing message flood across the entire wireless sensor network. In the node schedule solution proposed by Tian and Georganas [13], a global clock synchronization (for time delays) is required in order to resolve conflicts when determining what nodes should be turned off in order to save energy. Even though global clock synchronization assumption is held, the random time delays of the time back-off scheme in C-PNSS are still difficult to calculate accurately. Thus, blind areas may emerge, thereby jeopardizing the quality of network coverage since the time back-off scheme of C-PNSS does not guarantee that there is no coverage breach after turning off certain sensors. This paper presents the optimal energy-aware coveragepreserving scheme (OCoPS), a distributed scheme that maintains the network coverage while saving energy. This scheme is based on the extension of the Center Angle Method described by Cardei and Du [13]. It is also part
2709
of our scheme the proposal of decision algorithms that help nodes decide their status and resolve eventual status conflicts by exchanging local information instead of node scheduling. Based on our decision algorithms, OCoPS can guarantee initial coverage after turning off redundant sensors without the potential breach problem of time scheduling. In order to avoid excessive energy consumption in actual wake-up/turn-off schemes, a new rationale wakeup scheme is devised. Our scheme has been implemented as an expansion of the LEACH protocol [26] and compared with the C-PNSS scheme. The rest of this paper is organized as follows: Section 2 gives a brief summary of related work. Section 3 reviews the basic concepts of the coverage scheme and presents our optimal coverage-preserving, energy-saving protocol with the extended Central Angle Calculation Method, decision algorithms, and wake-up strategy. Section 4 compares OCoPS with C-PNSS. Experimental results are reported and analyzed in Section 5, which is followed by the Conclusion. 2. Related work Coverage is an important issue in terms of wireless-sensor-enabled emergency applications and is narrowly related to energy saving, connectivity, fault-tolerant, network reconfiguration, etc. Current solutions are based for the most part either on node scheduling (off-duty mechanisms) or coverage quality [1,5,8,18,21,23]. For node-schedulingbased solutions, the main idea is to find the optimal number of inactive nodes that exist while maintaining connectivity and coverage (with no breach areas of monitoring). In the coverage quality approach, the goal is to find areas that are highly observable and identify the best support and guidance regions. In the worst case, the goal is to find areas that are poorly observable and detect blind regions. In [22], an analytical solution that incorporates sensor and target characteristics (sensor radius, figure noise, moving speed, and traveling distance) is derived so as to determine the number of sensors required to cover a given region. In [14], the weakest breach path is investigated through defining the breach probability as the miss probability of the weakest breach path. Based on Dijkstra’s shortest-path algorithm and the Neyman–Pearson detection model, the authors point out that the most significant parameter is the false alarm rate, which is inversely proportional to breach probability. Although a novel solution for the target tracking application is introduced, it does not address other important issues such as network breach points. In [23], a solution is presented that addresses these problems. The authors propose an optimal polynomial time algorithm which combines graph theory and computational geometry (Voronoi constructs) in order to solve best-case and worse-case coverage. In [12], a more general solution for coverage is presented whose goal is to determine whether every point in a given area is sufficiently covered by at least k sensors (k-covered, where k is a
2710
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
predefined constant). In addition to coverage, connectivity is another important factor of the sensor-enabled emergency system. Several solutions have been proposed in order to guarantee connectivity (ASCENT [10]; Span [11]; GAF [28]). However, connectivity alone does not guarantee coverage. In [16], a connectivity-aware coverage solution is presented in which coverage is achieved through a probing mechanism that controls network density. In this algorithm, a node can be in one of three states: sleeping, awake and working. When a sleeping node wakes up (after an exponentially distributed period of time), it broadcasts a probing message within a certain range and awaits a reply. If no reply is received within a timeout, it will take over the surveillance task continuously until it runs out of battery power. In this solution, the probing range and wake-up rate can be adjusted to indirectly affect the degree of coverage. However, this solution does not guard against blind points since there is no guarantee of coverage [25]. Other solutions that provide connectivity-aware coverage include PEAS [15] (which does not provide an analytical guarantee for degree coverage and connectivity) and [27], which provides a geometric analysis of the relationship between connectivity and coverage. In [20], Cardei and Du propose a protocol that partitions the set of available sensors into disjoint sets such that each set covers all targets in different rounds. The experimental evaluation reveals that such a protocol discovers about 10% more disjoint covers than the algorithm of Slijepcevic and Potkonjak [24]. However, the Cardei and Du algorithm must fetch global location information by flooding the entire network. In order to guarantee the monitoring quality of emergency systems, network density is increased. This leads to higher energy consumption. Thus, power saving coverage solutions must be devised. The random deployment of sensors may lead to redundant nodes that share the same area. In order to extend the network lifetime by saving energy while meeting the coverage requirement, a common solution is to have an optimal number of nodes active and to have redundant neighbor nodes turned off. Tian and Georganas [13] present C-PNSS, a novel solution for preserving coverage, in which a node-scheduling scheme is based on off-duty eligibility rules, which allow nodes to turn themselves off so long as neighbor nodes can cover the area in their stead. C-PNSS attracts our attention not only because of its novel approach, but also because it spurs further research. C-PNSS is based on finding the optimal number of nodes to put off-duty while maintaining coverage. This is achieved by having every node in the network determine its own status (on-duty or off-duty) by gathering information from local neighbors and using both this information and its own to check whether the neighbors can help it monitor its coverage area. If eligible neighbors are found, the node turns itself off. One potential problem with this scenario occurs when two neighbor nodes decide to turn themselves off at the same time.
One thinks that the other can cover for it and both go off-duty. This situation leads to blind areas, which jeopardize coverage. In the C-PNSS, a random delay is introduced in the decision process for each node so that no two nodes decide to go off-duty at the same time (this is known as the back-off scheme). In this scheme, each node begins determining its candidate status after a random back-off time period Tw. In order to further decrease the probability of conflicts that lead to blind areas, just before turning itself off, a candidate postpones for yet another Td time interval before going off-duty. If times Tw and Td are of sufficient length, the probability of conflict occurrence is very low. However, the problem lies in the fact that Tw and Td cannot be calculated – they are expected values and are related to the density of the network. It is also assumed that a global clock is used. Thus, an important question to be asked is: Can two candidates go off-duty at the same time without jeopardizing sensing coverage? This paper introduces a new coverage-preserving scheme that guarantees initial coverage even after eligible sensors have been turned off. The solution extends the Center Angles Calculation Method and proposes a new Decision Algorithm that is used to determine status. A smart wake-up strategy that optimizes the wake-up phase in the given solution is also described. 3. OCoPS: an optimal coverage-preserving scheme In order to maintain coverage and the connectivity of the network while optimizing the number of active nodes and avoiding waking up more nodes than necessary, an optimal coverage-preserving scheme (OCoPS) has been devised. This scheme extends the Center Angle Calculation Method used in C-PNSS and uses a devised decision algorithm that dispenses with the use of global clock synchronization by exchanging local information with neighbors. Our scheme is designed under the following assumptions: (1) Network density is high enough that only parts of the deployed sensors are necessary to monitor a finite region R. (2) Region R is large enough, compared to the sensing range, for the boundary effects to be ignored. (3) All sensors have the same disk-sensing and communication ranges. The communication range is at least twice that of the sensing range so that connectivity is guaranteed [17]. (4) All of the events that occur under the sensing range of a sensor can be detected by that sensor. (5) Every sensor has a Unique ID and is aware of its own position. No two sensors are placed in the same location. (6) Our algorithms work under an ideal network in which there is no message loss during transmission. (7) The routing protocol knows the locations of all neighbors.
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
3.1. Basic definitions for coverage The basic definitions necessary to understand the coverage scheme are described below and serve as a basis for our proposed solution. Definition 1 (Neighboring set). Consider a set of sensors {p1, . . . , pn} in a finite area d. If we denote the radius of the sensor as r, then the neighboring sensor set NEpi of sensor pi is defined as:NEpi ¼ fn 2 @jdis tan ceðpi ; pj Þ < r; pi 6¼ pj g Definition 2 (Candidate/fully sponsored sensor). We refer to a node A as a Candidate or as fully sponsored by its neighbors if the sensing area S(A) is fully covered by S(NEA), where NEA represents the neighboring set of sensor A. It sp is denoted by NEA ! A. Definition 3 (Edge sensor). We refer to node A as an Edge sensor if A is not a fully sponsored sensor. Definition 4 (Edge area). We refer to a monitor area as an Edge area if and only if such an area is covered only by Edge sensors. Definition 5 (Candidate area). We refer to a monitor area as a Candidate area if and only if such an area is covered only by Candidates.
2711
Even though we are able to identify candidates, not every candidate can be turned off. For instance, consider the two dash circles in Fig. 2. According to the definition of candidate, the two dash circles A and B are fully sponsored by their neighbors. However, A and B cannot turn themselves off simultaneously; otherwise area H would become a blind area. We call this phenomenon the Off-duty Conflict problem. As mentioned in Section 2, C-PNSS solves such a problem by means of a time back-off scheme; however, it cannot guarantee that no holes were created and that no eligible off-duty nodes were set to on-duty.
3.2. The extended Central Angle Method The method for identifying a candidate (a fully sponsored sensor) is a key issue in the coverage problem. Combined with the angle antenna, C-PNSS [13] proposed a novel method in which the candidate was identified by calculating the central angles of nodes instead of the sensing area. However, the limitation of this method is that C-PNSS only considers situation, in which the sensing range r is equal to communication range R. As shown in Fig. 3, node A, which is in the center, is fully covered by nodes B, C, D, and E. Unfortunately, such a case is ignored by the Central Angle Method mentioned in the C-PNSS
In order to identify the candidate status of a sensor, we must determine whether the sensor is fully sponsored by its neighbors (in this paper we consider only the disk sensing range). The works described in [12,13] present a simple method for determining a candidate by calculating the center angles of its neighbors. If these angles can cover the entire 360 like the gray cycles in Fig. 1, node A is considered fully sponsored. Details on how to calculate the coverage using angles can be found in [12,13].
A
H B
Fig. 2. Off-duty conflict.
A D
C A E B
Fig. 1. Angle calculation.
Fig. 3. Full coverage, communication range = 2 · sensing range.
2712
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
because nodes C, D, and E will not be considered to be neighbors. In other words, because node A will not be offduty but will instead be considered to be an edge node, holes can emerge in the intersections among A, C, D, and E when C, D, and E are out of energy. Therefore, the coverage of the entire network will drop earlier and faster. Besides [17] points out that in order to keep the network connected while preserving coverage, communication range R should be at least doubled that of sensing range r. In order to maintain the connectivity of the network and improve the accuracy of the Central Angle Method, we extend the communication range assumption of C-PNSS to twice that of the sensing range. The extended Central Angle Method is explained below after a required definition. Definition 6 (Associated sponsor). As shown in Fig. 4, sensors I and K fully cover J with the help of sensors H and G. In this case, we say that sensors I and K fully sponsor J with H and G. Sensors H and G are I’s associated Sponsor sp sp for J; these are denoted as I ! HJ and I ! GJ . We use sensor H as an example in order to find and calculate the associated sponsored angle between H and I. As can be seen in Fig. 4, when 2r > dHfiJ > r where r is the sensing range, the sensing areas of H, I, and J intersect in four points A, B, C, D and create a sponsor area RA fi B fi J fi C fi D. If the line segments J fi A and J fi C are extended, they will intersect with the sensing edge of J in points A 0 and C 0 , while we obtain another area RJ!B0 !C0 where RJ!B0 !C0 RA!B!J!C!D and can be presented by an associated sponsored angle hJ!B0 !C0 . The hJ!B0 !C0 can be calculated as follows. For example, in intersection point B of sensors J and H, the coordinates of sensors J and H are known. Therefore, we can obtain J aJ!H ¼ arctanðXY HH Y Þ. As described in C-PNSS, the central X J angle hJfiH can be calculated using formula hJ!H ¼ ;H Þ 2 arccosðdðJ2r Þ. Thus, we can compute the coordinates of point B as follows: XB = XJ + r*cos(aJfiH hJfiH) and YB = YJ + r·sin(aJfiH hJfiH), while A, C, and D coordinates can be calculated in the same way. Based on the computations of the coordinates of A, B, C, and D, we can J calculate fully sponsored angle hJ!A0 !C0 ¼ arctanðXY CC Y Þ X J Y A Y J arctanðX A X J Þ.
H
A A’
B
C’ C
D
I J
Associated Sponsored Angle
K
G
Fig. 4. Extended Central Angles Calculation.
3.3. The decision algorithms In order to resolve the off-duty conflict problem without using a back-off time schedule, a decision process has been devised in which a sensor A can decide its status by exchanging its sponsored and contribution angle information with its neighbors. During this decision process, three major factors are considered: the number of off-duty nodes; the control message cost; and the running time. The former two factors directly affect the energy cost of the entire network, while a short running time can make the algorithm easier to insert into other protocols. Three algorithms, Off-Duty Election (Off-E), On-Duty Election (On-E), and Alternate Election (AltE), are proposed and implemented using the Associated Sponsor Method. Algorithm III-C.1 – The Decision Algorithm FOR (each candidate node q) { Compare q with all neighbors; IF (q is the largest candidate) set q off-duty; IF (q is not candidate) set q on-duty; }//end of loop 3.3.1. The off-duty election algorithm The off-duty election algorithm was the first algorithm we developed to solve the off-duty conflict problem without using the node schedule. The algorithm performs a local election based on the election rules, which are defined below, in order to select the eligible off-duty nodes among the neighborhoods in each election round. Definition 7 (The fixed sponsored central angle). Consider any candidate A and its neighbor B. We say that the overlap angle FAAfiB is a Fixed Sponsored Central Angle if and only if B is an on-duty sensor; otherwise, we call such angles Pending Central Angles, which are denoted PAAfiB. Definition 8 (The off-duty decision rule). Consider any two candidates A and B. We say that A is larger than B, or A > B, if and only if FAA!NEA > FAB!NEB or PAA!NEA > PAB!NEB while FAA!NEA ¼ FAB!NEB , or IDA > IDB while PAA!NEA ¼ PAB!NEB and FAA!NEA ¼ FAB!NEB . According to assumption in Section 3, the routing protocol already has information concerning all neighbor locations. Thus, by counting the number of messages from neighbors instead of using the time schedule, we can trade off the chance of having breach points in the network by using more control messages. When a candidate receives all status messages from its neighbors, it will begin deciding its own status. In order to do this, it will compare its own status with the status of all of the neighbors’ candidates according to the election rules defined above. A node will be off-duty if and only if it is the largest node in its neighborhood. Even off-duty election algorithms can solve the
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
off-duty conflict problem well; however, their high number of control messages nevertheless cost a certain amount of energy. 3.3.2. On-duty election algorithm Reducing the cost of control messages is one of the main questions in the decision algorithm. Because the off-duty election algorithm uses too many messages and has the potential problem of having long initial time, we are looking for another election algorithm so as to avoid such problems. The number of on-duty nodes compared to the number of deployed nodes attracts our attention. Compared with the off-duty nodes, the on-duty nodes occupy no more than 20% of total deployed sensors. If we use on-duty election instead of off-duty election, the number of control messages will be remarkably reduced. As we know, our extended Central Angle Method is able to identify the edge nodes and set all of them to on-duty. Thus, the duty of the decision algorithm is to choose the fewest on-duty nodes to cover the Candidate area. Definition 9 (The Contribution Central Angle). Consider any candidate A and its neighbor set NEA. We say that the Contribution Central Angle CAA!NEA is the summation of PAA!NEA (Pending Central Angles of A as defined in Definition 7). Definition 10 (The on-duty decision rule). Consider any two candidates A and B. We say that A is lager than B, or A > B, if and only if CAA!NEA > CAB!NEB or PAA!NEA > PAB!NEB while CAA!NEA ¼ CAB!NEB or IDA(ID of node A) > IDB(ID of node B) while PAA!NEA ¼ PAB!NEB and CAA!NEA ¼ CAB!NEB . Algorithm III-C.2 – The On-duty Election Algorithm FOR (each candidate node q) { Compare q with all of its neighbors using On-duty Decision Rule; IF (q is the decision result) set q on-duty; IF (q is fully sponsored) set q off-duty; }//end of loop In the on-duty election algorithm, the sensor with the largest Contribution Angle is elected in its neighborhood. As we know, the number of such on-duty nodes is much lower than the number of off-duty nodes; thus, the on-duty election algorithm runs faster than the off-duty election algorithm even in the worst case. Moreover, the number of control messages is remarkably reduced. The on-duty election algorithm reduces the number of control messages by reducing the rounds of algorithms, but it causes another problem in that the on-duty election algorithm turns off fewer sensors than the off-duty election algorithm because on-duty election algorithm begins at an unstable status that may use more on-duty nodes to resolve the off-duty conflict problem. Thus, another alternate election algorithm has
2713
been developed in order to keep both initial time t and the number of on-duty nodes at a low level. 3.3.3. Alternate election algorithm Even though the on-duty election algorithm shows better performance on the average than off-duty algorithm, too many control messages are wasted in the worst case. In each round of this scenario, only one candidate can be turned off. Another drawback of on-duty is the Crack Effect, where the central nodes will have a high probability of being elected as the on-duty nodes and the Candidate area will generate a number of pieces of hole between onduty and off-duty nodes, which will lead to more on-duty nodes than the off-duty algorithm offers. In order to resolve such problems, an Alternate Election algorithm in which only candidates who receive stable status announcement messages will be involved in the election, has been proposed. The algorithm is separated into odd and even rounds and begins with the SAM (Status Acknowledge Message) from the on-duty nodes (edge nodes) after identifying all of the candidates and on-duty nodes in the initial phase. Two new status, on-duty candidate and off-duty candidate, are introduced as temporary status. An Odd round is triggered when a P-Off node A receives either a SAM message from all P-On neighbors, or SAM messages from all neighbors at the zero level. If A is the node selected by the election algorithm, it will be set to Off-duty; otherwise, it will be set as P-On. In the odd round, only the off-duty election will be performed. Any candidate A that receives a SAM will check the distance between itself and the sender. If this distance is smaller than sensor range r, the status of the sensor is set to off-duty candidate; otherwise, it is set to on-duty candidate. Such a candidate A will classify its neighbors into four groups: • On-duty neighbors. Neighbors that sent an On-duty SAM message; • Off-duty neighbors. Neighbors that sent an Off-duty SAM message; • Pending off-duty (P-Off) neighbors. Neighbors whose distance from any on-duty neighbor is lower than r; • Pending on-duty (P-On) neighbors. Neighbors whose distance from any on-duty neighbors is larger than r; An even round will be triggered when a P-On candidate A receives a SAM message from all of its Off-duty candidate neighbors. The center angles of A are recalculated using the new status information gathered from its P-Off neighbors. If A cannot keep its candidate status, it will be set as On-duty; otherwise, it will be set as P-Off. In the Even round, any candidate that receives a SAM will update its status to P-On and execute the Extended Coverage Calculation Method to find out whether the candidate is a qualified On-duty node. The status update rule is that the status only updates by means of the high-level SAM message.
2714
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
Algorithm III-C.3 – Alternate Election Algorithm FOR (P-On candidate q in even round) { Extended Coverage Calculation; IF (q is not candidate) set q on-duty; }//end of loop FOR (P-Off candidate q in odd round) { Extended Coverage Calculation; Off-duty Election; }//end of loop
3.4. The wake-up strategy Several node-scheduling solutions presented in literature introduce methods for turning off the largest possible number of sensor nodes. However, no solution addresses the wake-up strategy. A wake-up strategy can influence not only the energy consumption in a sensor network, but also the quality of the monitoring task. Three issues must be considered in a wake-up strategy: 3.4.1. The off-duty sensors should take over the sensing area of the dead sensor (a sensor that has run out of energy) as soon as possible in order to avoid coverage holes In the periodical wake-up strategy used in C-PNSS, if a sensor A dies between two rounds, events occurring within the sensing range of A may not be detected by other onduty sensors, and the off-duty sensors will only wake up to monitor the area watched by A at the beginning of the next round. In summary, a hole may appear between two rounds and cannot be fixed until the off-duty sensors wake up at the beginning of the next round. 3.4.2. The wake-up strategy should wake up as few necessary off-duty sensors as possible Consider the wake-up process of the two off-duty sensors A and C in Fig. 5. The following strategy would typically be used: If an on-duty sensor p is running at low battery, it will broadcast a wake-up message so as to wake up turned-off nodes in its neighbor list. Assuming that onduty sensor B (the dark circle) is running out of energy and broadcasts a wake-up message, both off-duty sensors A and C will be waken up. However, in this case, node A is fully sponsored by its neighbors; thus, in this case we say that off-duty sensor A is Over-Waken-up.
A
B
C
A
Fig. 5. Over wake-up.
B
C
3.4.3. The wake-up strategy should not be overly costly in terms of energy Because the period wake-up strategy uses too many control messages, which is very energy-consuming, an alternative solution is to have an additional communication channel (wake-up channel) that receives wake-up messages. The problem with this solution is that the channel must be listened to even when the nodes are off-duty. However, it is well known that sensing wastes much less energy than transmitting a message. If a node can be monitored in terms of its energy level, and its state of ‘‘running out of energy’’ is detected as an event (signal) that can be captured by the sensing subsystem of the off-duty sensor, a wake-up solution that avoids the problems with the current wake-up strategies and saves energy can be devised. To a typical sensor, the sensing energy cost is close to the message receiving energy cost [29]. In order to detect such events the sensing module has to be work on standby or rotation modes. Gu and Stankovic [19] have proposed a radio wake-up circuit that uses the energy in the radio communication signal to generate a wake-up signal. Such circuit does not have any regular energy supply. When the radio signal is present with a special frequency, the antenna receives energy input and drives the circuit. According to the experiments of Lin on Berkeley Mica2 motes with 5/ 8-wavelength monopole antenna, when transmission power is 10 dBm (10 mW) the output voltage increases to about 0.6 V. Even when comparing to the 5 V of the PC and 1 V of some other low-power embedded systems, 0.6 V still can be accepted as a high signal and distinguish from noise. A radio-triggered ID circuit was also mentioned in the same paper. Through assigning each sensor a special frequency, the receiver can even identify the sender of such incoming signals. By applying such wake-up circuit our wake-up scheme (the Algorithm 3.4.3) is described as follows. Algorithm III-D – Wake-up Strategy For (each node p that is out of energy) p generates an out-of-energy signal; For (each off-duty node q that captures an out-of-energy signal) { identify ID by frequency; recalculate the off-duty status; IF (q cannot keep the off-duty status){ wakeup(); q.status ‹ On-duty; }}//end of loop 4. Energy cost comparison of OCoPS and C-PNSS Because our optimal coverage calculation scheme was derived from C-PNSS, we will compare our OCoPS with C-PNSS firstly from the principles point of view, then quantitative viewpoint. C-PNSS is a novel energy-saving sensing coverage protocol that uses fewer control messages during its initial phase
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
than other protocols. However, the use of these control messages is repeated round after round. Considering the lifetime of the network and not merely its initial phase, such control messages may cost more energy than is desired. However, because of the limitations of the Central Angle Method, C-PNSS will treat certain fully sponsored nodes as edge nodes and ignore the association sponsors (mentioned in our scheme) during the wake-up phase in a high-density network. Thus, C-PNSS uses more on-duty nodes in order to maintain full coverage. Our OCoPS guarantees that our algorithm does not create a hole and is more efficient than C-PNSS in terms of the coverage issue. The price for achieving this goal is using more control messages than C-PNSS during the initial phase. Moreover, in the OCoPS scheme, off-duty nodes are aware of the wake-up signal by applying the wake-up radio circuit we mentioned in the wake-up scheme section. An off-duty node turns off all its equipment until its receive wake-up signal generated by wake-up radio circuit. Considering that energy consumption is affected by many factors, such as sensing range, the length of the message, the number of rounds for data gathering, and the power consumption model itself, OCoPS is mostly concerned with message transmission energy cost, that include both control messages and data message (the energy cost of sleeping node was ignored). Because we only consider the sensors that have overlapped area, when transmitting the control messages, the transmission range is adjust to double of the sensing range. After the initial phase, the transmission range is adjusted to reach the furthest on-duty neighbor. the energy consumption for the control messages that our scheme uses does not significantly affect network lifetime, as can be seen by the results obtained from the simulations. In order to compare C-PNSS and OCoPS on the same grounds, we use the same energy parameters and radio model as C-PNSS. Those parameters are shown in Table 2 and discussed in [26]. After the initial phase, both C-PNSS and OCoPS work in rounds on a TDMA network. Each round lasts 10 s. Each sensor sends a 2000-bit report message to the base station with a 0.5-s time interval.
2715
Consider a wireless sensor network that includes N sensors, M candidates, O off-duty sensors, and runs T rounds. In order to simplify the calculation, we assume that the average number of neighbors is XC in C-PNSS and XE in OCoPS; the energy cost of sending is S/bit; the energy cost of receiving message is R/bit; We assume the size of control message (Bc) are the same in both OCoPS and C-PNSS. the decision algorithm finishes after t rounds and the number of remaining candidates in the i rounds is Mi. During the wakeup phase, each off-duty node has XO on-duty neighbors. In Table 1, the lifetime of the network is separated into two phases: the initial phase and the wake-up phase. Both schemes will have to exchange a PAM (Position Acknowledgement Message) in order to gather the location information and broadcast a CAM (Candidate Acknowledgement Message) to publish their candidate status among neighbors. Because the information contained in these messages is almost exactly the same, we can assume that both types of messages cost the same amount of energy for both C-PNSS and OCoPS. The difference is that, in order to resolve the Off-duty Conflict, our Pt scheme requires extra messages whose energy cost is i¼1 M i ðS þ X Ei RÞ Bc to publish candidate status. In the initial phase, OCoPS consumes more energy than C-PNSS, but not in the long run. Since C-PNSS works in rounds, let us consider the period from the end of the initial phase to the time at which the network dies after T rounds, which is referred to as wake-up phase in Table 1. In OCoPS, because we assume that the energy cost of generating an out-of-energy event (signal) is same as the cost of sending a message, and because each sensor generates only one out-of-energy event, the total energy cost of generating out-of-energy events in our scheme is NS*Bc. The off-duty sensors, on the other hand, did not cost energy to catch such signal. However, because the communication range in OCoPS is doubled that of C-PNSS, in order to simplify the calculation we can assume that XE 6 2XC. Knowing that in a fixed area M < N, when N is large sufficiently we can ignore the edge nodes or say that M 6 N. By increasing the number of rounds, Mi approaches (M O) which itself approaches zero when M and O are large sufficiently. Therefore, we
Table 1 Comparison of the energy costs of control messages Our scheme Initial phase
C-PNSS PAM CAM
NS+NXER NS+NXER t P M i ðS þ X Ei RÞ
SAM Run T rounds Wake-up phase Total cost
Schedule phase
PAM SAM
NS+NXCR NS+NXCR
Wake-up phase
T*SAM
T*(NS+NXCR)
Total cost
Bc*(2 + T)(NS + NXCR
M i ð1 þ X Ei ÞÞÞ
Total cost
An optimal coverage-preserving scheme for wireless sensor networks based on local information exchange Azzedine Boukerche a
a,*
, Xin Fei a, Regina B. Araujo
a,b
PARADISE Research Laboratory, University of Ottawa, Ottawa, Ont., Canada K1N 6N5 b VRLnet Lab, Federal University of Sao Carlos, SP, Brazil Available online 15 June 2007
Abstract Coverage-preserving and energy-saving solutions have been reported in the literature and are generally based upon both quality coverage and off-duty scheme. Off-duty scheme based solutions present at least three challenging problems: (1) keeping coverage and connectivity of the network while optimizing the number of active sensor nodes; (2) resolving conflicts when determining which nodes should be turned off in order to save energy; and (3) finding optimal wake-up strategies that avoid waking up more nodes than necessary. This paper presents a novel distributed solution, the optimal coverage-preserving scheme (OCoPS), that extends the well-known Central Angle Method in order to identify fully sponsored nodes. OCoPS comprises an extended Central Angle Method, new decision algorithms devised to resolve the off-duty conflict problem under different network densities, and an energy-aware wake-up scheme that solves coverage hole problems in off-duty schemes. Compared to the widely used node scheduling scheme, our solution is based on local information exchange without the uncertainty of self-schedule algorithms. OCoPS is implemented as an extension of LEACH. A set of simulation experiments is carried out to evaluate its performance compared to other well-known schemes which are based on the Central Angle Method and self-scheduling. Our results indicate that on network lifetime OCoPS outperforms other schemes by over 20% and by over 25% when the coverage rate is higher than 80%. The experimental results also show that our coverage scheme effectively limits the number of on-duty node compared to the other schemes. 2007 Published by Elsevier B.V. Keywords: Coverage; Wireless sensor network; Energy
1. Introduction Emergency Preparedness is a stringent wireless sensor network application class in which context aware physical environments subjected to critical conditions, such as fire or leaking of toxic liquids, are required to be reliably monitored. Wireless sensor networks (WSN) to support such a class of applications must provide a fast, reliable, fault-tolerant, and energy-aware channel for event diffusion, which meets the requirements of different sensor network application scenarios such as query-based (when a sink queries the WSN for specific information), event-driven (sensor nodes *
Corresponding author. E-mail addresses: [email protected] (A. Boukerche), xfei@ site.uottawa.ca (X. Fei), [email protected] (R.B. Araujo). 0140-3664/$ - see front matter 2007 Published by Elsevier B.V. doi:10.1016/j.comcom.2007.05.018
send information to the sink upon detection of an event), and periodic (sensor nodes periodically send information to the sink) [4,6,7]. These requirements must be met even in the presence of emergency conditions that can lead to node failures and path disruptions to the sink. Partial solutions have been introduced from the physical to the application layers [1–3]. These requirements must be met even under emergency situations that can lead to node failures and path disruptions to the sink. Partial solutions have been reported from the physical to the application layer. However, an important issue that begins to be fully addressed is how well a given area can be monitored by the WSN – an issue often related to QoS and known as coverage. Other questions related to QoS also arise under the coverage perspective: How to keep good quality monitoring in face of high rates
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
of failure? What is the optimal number of nodes that meets coverage quality and scarce energy? What are the breach points in a sensor field? When is tracking people or objects required when in a search for survivors during an emergency situation? How accurate is the location? Moreover, for supervision and control applications, different types of sensors with different processing power and communication properties (radio range) might be integrated in the same WSN, as happens with actor nodes (more resourceful nodes with the capability to respond to events in real-time) in Wireless sensor and actor networks – WSAN [9]. How can coverage be met with heterogeneous sensors? These are challenging questions for a WSN to cope with, and good solutions have been introduced in literature. These solutions tackle one or more of the problems discussed above by making coverage-preserving methods either aware of connectivity or energy, among others. However, a solution that addresses all of the issues raised above simultaneously poses a great challenge. Basically, the coverage problems of WSN are defined either as deterministic or nondeterministic deployment [23]. Because most wireless sensors are typically randomly deployed in difficult-to-access environments, nondeterministic deployment attracts more attention than deterministic deployment. At the same time, distributed or decentralized solutions, which keep coverage while maintaining network longevity, become interesting research topics for nondeterministic deployment. The main approaches to such solutions are set partition and node scheduling. The representative algorithm of set partition is the disjoint set of Cardei and Du [20]. Important algorithms of node scheduling include the work of Tian and Georganas (named here as C-PNSS) described in [13]; Ye et al. [16]; and the optimal geographical density control (OGDC) proposed by Zhang and Hou [17]. The main motivation for our work is the lack of novel solutions that meet coverage quality while maintaining a long network lifetime. In the disjoint set scheme proposed by Cardei and Du [20], global location information is required and is achieved through an energy-costing message flood across the entire wireless sensor network. In the node schedule solution proposed by Tian and Georganas [13], a global clock synchronization (for time delays) is required in order to resolve conflicts when determining what nodes should be turned off in order to save energy. Even though global clock synchronization assumption is held, the random time delays of the time back-off scheme in C-PNSS are still difficult to calculate accurately. Thus, blind areas may emerge, thereby jeopardizing the quality of network coverage since the time back-off scheme of C-PNSS does not guarantee that there is no coverage breach after turning off certain sensors. This paper presents the optimal energy-aware coveragepreserving scheme (OCoPS), a distributed scheme that maintains the network coverage while saving energy. This scheme is based on the extension of the Center Angle Method described by Cardei and Du [13]. It is also part
2709
of our scheme the proposal of decision algorithms that help nodes decide their status and resolve eventual status conflicts by exchanging local information instead of node scheduling. Based on our decision algorithms, OCoPS can guarantee initial coverage after turning off redundant sensors without the potential breach problem of time scheduling. In order to avoid excessive energy consumption in actual wake-up/turn-off schemes, a new rationale wakeup scheme is devised. Our scheme has been implemented as an expansion of the LEACH protocol [26] and compared with the C-PNSS scheme. The rest of this paper is organized as follows: Section 2 gives a brief summary of related work. Section 3 reviews the basic concepts of the coverage scheme and presents our optimal coverage-preserving, energy-saving protocol with the extended Central Angle Calculation Method, decision algorithms, and wake-up strategy. Section 4 compares OCoPS with C-PNSS. Experimental results are reported and analyzed in Section 5, which is followed by the Conclusion. 2. Related work Coverage is an important issue in terms of wireless-sensor-enabled emergency applications and is narrowly related to energy saving, connectivity, fault-tolerant, network reconfiguration, etc. Current solutions are based for the most part either on node scheduling (off-duty mechanisms) or coverage quality [1,5,8,18,21,23]. For node-schedulingbased solutions, the main idea is to find the optimal number of inactive nodes that exist while maintaining connectivity and coverage (with no breach areas of monitoring). In the coverage quality approach, the goal is to find areas that are highly observable and identify the best support and guidance regions. In the worst case, the goal is to find areas that are poorly observable and detect blind regions. In [22], an analytical solution that incorporates sensor and target characteristics (sensor radius, figure noise, moving speed, and traveling distance) is derived so as to determine the number of sensors required to cover a given region. In [14], the weakest breach path is investigated through defining the breach probability as the miss probability of the weakest breach path. Based on Dijkstra’s shortest-path algorithm and the Neyman–Pearson detection model, the authors point out that the most significant parameter is the false alarm rate, which is inversely proportional to breach probability. Although a novel solution for the target tracking application is introduced, it does not address other important issues such as network breach points. In [23], a solution is presented that addresses these problems. The authors propose an optimal polynomial time algorithm which combines graph theory and computational geometry (Voronoi constructs) in order to solve best-case and worse-case coverage. In [12], a more general solution for coverage is presented whose goal is to determine whether every point in a given area is sufficiently covered by at least k sensors (k-covered, where k is a
2710
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
predefined constant). In addition to coverage, connectivity is another important factor of the sensor-enabled emergency system. Several solutions have been proposed in order to guarantee connectivity (ASCENT [10]; Span [11]; GAF [28]). However, connectivity alone does not guarantee coverage. In [16], a connectivity-aware coverage solution is presented in which coverage is achieved through a probing mechanism that controls network density. In this algorithm, a node can be in one of three states: sleeping, awake and working. When a sleeping node wakes up (after an exponentially distributed period of time), it broadcasts a probing message within a certain range and awaits a reply. If no reply is received within a timeout, it will take over the surveillance task continuously until it runs out of battery power. In this solution, the probing range and wake-up rate can be adjusted to indirectly affect the degree of coverage. However, this solution does not guard against blind points since there is no guarantee of coverage [25]. Other solutions that provide connectivity-aware coverage include PEAS [15] (which does not provide an analytical guarantee for degree coverage and connectivity) and [27], which provides a geometric analysis of the relationship between connectivity and coverage. In [20], Cardei and Du propose a protocol that partitions the set of available sensors into disjoint sets such that each set covers all targets in different rounds. The experimental evaluation reveals that such a protocol discovers about 10% more disjoint covers than the algorithm of Slijepcevic and Potkonjak [24]. However, the Cardei and Du algorithm must fetch global location information by flooding the entire network. In order to guarantee the monitoring quality of emergency systems, network density is increased. This leads to higher energy consumption. Thus, power saving coverage solutions must be devised. The random deployment of sensors may lead to redundant nodes that share the same area. In order to extend the network lifetime by saving energy while meeting the coverage requirement, a common solution is to have an optimal number of nodes active and to have redundant neighbor nodes turned off. Tian and Georganas [13] present C-PNSS, a novel solution for preserving coverage, in which a node-scheduling scheme is based on off-duty eligibility rules, which allow nodes to turn themselves off so long as neighbor nodes can cover the area in their stead. C-PNSS attracts our attention not only because of its novel approach, but also because it spurs further research. C-PNSS is based on finding the optimal number of nodes to put off-duty while maintaining coverage. This is achieved by having every node in the network determine its own status (on-duty or off-duty) by gathering information from local neighbors and using both this information and its own to check whether the neighbors can help it monitor its coverage area. If eligible neighbors are found, the node turns itself off. One potential problem with this scenario occurs when two neighbor nodes decide to turn themselves off at the same time.
One thinks that the other can cover for it and both go off-duty. This situation leads to blind areas, which jeopardize coverage. In the C-PNSS, a random delay is introduced in the decision process for each node so that no two nodes decide to go off-duty at the same time (this is known as the back-off scheme). In this scheme, each node begins determining its candidate status after a random back-off time period Tw. In order to further decrease the probability of conflicts that lead to blind areas, just before turning itself off, a candidate postpones for yet another Td time interval before going off-duty. If times Tw and Td are of sufficient length, the probability of conflict occurrence is very low. However, the problem lies in the fact that Tw and Td cannot be calculated – they are expected values and are related to the density of the network. It is also assumed that a global clock is used. Thus, an important question to be asked is: Can two candidates go off-duty at the same time without jeopardizing sensing coverage? This paper introduces a new coverage-preserving scheme that guarantees initial coverage even after eligible sensors have been turned off. The solution extends the Center Angles Calculation Method and proposes a new Decision Algorithm that is used to determine status. A smart wake-up strategy that optimizes the wake-up phase in the given solution is also described. 3. OCoPS: an optimal coverage-preserving scheme In order to maintain coverage and the connectivity of the network while optimizing the number of active nodes and avoiding waking up more nodes than necessary, an optimal coverage-preserving scheme (OCoPS) has been devised. This scheme extends the Center Angle Calculation Method used in C-PNSS and uses a devised decision algorithm that dispenses with the use of global clock synchronization by exchanging local information with neighbors. Our scheme is designed under the following assumptions: (1) Network density is high enough that only parts of the deployed sensors are necessary to monitor a finite region R. (2) Region R is large enough, compared to the sensing range, for the boundary effects to be ignored. (3) All sensors have the same disk-sensing and communication ranges. The communication range is at least twice that of the sensing range so that connectivity is guaranteed [17]. (4) All of the events that occur under the sensing range of a sensor can be detected by that sensor. (5) Every sensor has a Unique ID and is aware of its own position. No two sensors are placed in the same location. (6) Our algorithms work under an ideal network in which there is no message loss during transmission. (7) The routing protocol knows the locations of all neighbors.
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
3.1. Basic definitions for coverage The basic definitions necessary to understand the coverage scheme are described below and serve as a basis for our proposed solution. Definition 1 (Neighboring set). Consider a set of sensors {p1, . . . , pn} in a finite area d. If we denote the radius of the sensor as r, then the neighboring sensor set NEpi of sensor pi is defined as:NEpi ¼ fn 2 @jdis tan ceðpi ; pj Þ < r; pi 6¼ pj g Definition 2 (Candidate/fully sponsored sensor). We refer to a node A as a Candidate or as fully sponsored by its neighbors if the sensing area S(A) is fully covered by S(NEA), where NEA represents the neighboring set of sensor A. It sp is denoted by NEA ! A. Definition 3 (Edge sensor). We refer to node A as an Edge sensor if A is not a fully sponsored sensor. Definition 4 (Edge area). We refer to a monitor area as an Edge area if and only if such an area is covered only by Edge sensors. Definition 5 (Candidate area). We refer to a monitor area as a Candidate area if and only if such an area is covered only by Candidates.
2711
Even though we are able to identify candidates, not every candidate can be turned off. For instance, consider the two dash circles in Fig. 2. According to the definition of candidate, the two dash circles A and B are fully sponsored by their neighbors. However, A and B cannot turn themselves off simultaneously; otherwise area H would become a blind area. We call this phenomenon the Off-duty Conflict problem. As mentioned in Section 2, C-PNSS solves such a problem by means of a time back-off scheme; however, it cannot guarantee that no holes were created and that no eligible off-duty nodes were set to on-duty.
3.2. The extended Central Angle Method The method for identifying a candidate (a fully sponsored sensor) is a key issue in the coverage problem. Combined with the angle antenna, C-PNSS [13] proposed a novel method in which the candidate was identified by calculating the central angles of nodes instead of the sensing area. However, the limitation of this method is that C-PNSS only considers situation, in which the sensing range r is equal to communication range R. As shown in Fig. 3, node A, which is in the center, is fully covered by nodes B, C, D, and E. Unfortunately, such a case is ignored by the Central Angle Method mentioned in the C-PNSS
In order to identify the candidate status of a sensor, we must determine whether the sensor is fully sponsored by its neighbors (in this paper we consider only the disk sensing range). The works described in [12,13] present a simple method for determining a candidate by calculating the center angles of its neighbors. If these angles can cover the entire 360 like the gray cycles in Fig. 1, node A is considered fully sponsored. Details on how to calculate the coverage using angles can be found in [12,13].
A
H B
Fig. 2. Off-duty conflict.
A D
C A E B
Fig. 1. Angle calculation.
Fig. 3. Full coverage, communication range = 2 · sensing range.
2712
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
because nodes C, D, and E will not be considered to be neighbors. In other words, because node A will not be offduty but will instead be considered to be an edge node, holes can emerge in the intersections among A, C, D, and E when C, D, and E are out of energy. Therefore, the coverage of the entire network will drop earlier and faster. Besides [17] points out that in order to keep the network connected while preserving coverage, communication range R should be at least doubled that of sensing range r. In order to maintain the connectivity of the network and improve the accuracy of the Central Angle Method, we extend the communication range assumption of C-PNSS to twice that of the sensing range. The extended Central Angle Method is explained below after a required definition. Definition 6 (Associated sponsor). As shown in Fig. 4, sensors I and K fully cover J with the help of sensors H and G. In this case, we say that sensors I and K fully sponsor J with H and G. Sensors H and G are I’s associated Sponsor sp sp for J; these are denoted as I ! HJ and I ! GJ . We use sensor H as an example in order to find and calculate the associated sponsored angle between H and I. As can be seen in Fig. 4, when 2r > dHfiJ > r where r is the sensing range, the sensing areas of H, I, and J intersect in four points A, B, C, D and create a sponsor area RA fi B fi J fi C fi D. If the line segments J fi A and J fi C are extended, they will intersect with the sensing edge of J in points A 0 and C 0 , while we obtain another area RJ!B0 !C0 where RJ!B0 !C0 RA!B!J!C!D and can be presented by an associated sponsored angle hJ!B0 !C0 . The hJ!B0 !C0 can be calculated as follows. For example, in intersection point B of sensors J and H, the coordinates of sensors J and H are known. Therefore, we can obtain J aJ!H ¼ arctanðXY HH Y Þ. As described in C-PNSS, the central X J angle hJfiH can be calculated using formula hJ!H ¼ ;H Þ 2 arccosðdðJ2r Þ. Thus, we can compute the coordinates of point B as follows: XB = XJ + r*cos(aJfiH hJfiH) and YB = YJ + r·sin(aJfiH hJfiH), while A, C, and D coordinates can be calculated in the same way. Based on the computations of the coordinates of A, B, C, and D, we can J calculate fully sponsored angle hJ!A0 !C0 ¼ arctanðXY CC Y Þ X J Y A Y J arctanðX A X J Þ.
H
A A’
B
C’ C
D
I J
Associated Sponsored Angle
K
G
Fig. 4. Extended Central Angles Calculation.
3.3. The decision algorithms In order to resolve the off-duty conflict problem without using a back-off time schedule, a decision process has been devised in which a sensor A can decide its status by exchanging its sponsored and contribution angle information with its neighbors. During this decision process, three major factors are considered: the number of off-duty nodes; the control message cost; and the running time. The former two factors directly affect the energy cost of the entire network, while a short running time can make the algorithm easier to insert into other protocols. Three algorithms, Off-Duty Election (Off-E), On-Duty Election (On-E), and Alternate Election (AltE), are proposed and implemented using the Associated Sponsor Method. Algorithm III-C.1 – The Decision Algorithm FOR (each candidate node q) { Compare q with all neighbors; IF (q is the largest candidate) set q off-duty; IF (q is not candidate) set q on-duty; }//end of loop 3.3.1. The off-duty election algorithm The off-duty election algorithm was the first algorithm we developed to solve the off-duty conflict problem without using the node schedule. The algorithm performs a local election based on the election rules, which are defined below, in order to select the eligible off-duty nodes among the neighborhoods in each election round. Definition 7 (The fixed sponsored central angle). Consider any candidate A and its neighbor B. We say that the overlap angle FAAfiB is a Fixed Sponsored Central Angle if and only if B is an on-duty sensor; otherwise, we call such angles Pending Central Angles, which are denoted PAAfiB. Definition 8 (The off-duty decision rule). Consider any two candidates A and B. We say that A is larger than B, or A > B, if and only if FAA!NEA > FAB!NEB or PAA!NEA > PAB!NEB while FAA!NEA ¼ FAB!NEB , or IDA > IDB while PAA!NEA ¼ PAB!NEB and FAA!NEA ¼ FAB!NEB . According to assumption in Section 3, the routing protocol already has information concerning all neighbor locations. Thus, by counting the number of messages from neighbors instead of using the time schedule, we can trade off the chance of having breach points in the network by using more control messages. When a candidate receives all status messages from its neighbors, it will begin deciding its own status. In order to do this, it will compare its own status with the status of all of the neighbors’ candidates according to the election rules defined above. A node will be off-duty if and only if it is the largest node in its neighborhood. Even off-duty election algorithms can solve the
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
off-duty conflict problem well; however, their high number of control messages nevertheless cost a certain amount of energy. 3.3.2. On-duty election algorithm Reducing the cost of control messages is one of the main questions in the decision algorithm. Because the off-duty election algorithm uses too many messages and has the potential problem of having long initial time, we are looking for another election algorithm so as to avoid such problems. The number of on-duty nodes compared to the number of deployed nodes attracts our attention. Compared with the off-duty nodes, the on-duty nodes occupy no more than 20% of total deployed sensors. If we use on-duty election instead of off-duty election, the number of control messages will be remarkably reduced. As we know, our extended Central Angle Method is able to identify the edge nodes and set all of them to on-duty. Thus, the duty of the decision algorithm is to choose the fewest on-duty nodes to cover the Candidate area. Definition 9 (The Contribution Central Angle). Consider any candidate A and its neighbor set NEA. We say that the Contribution Central Angle CAA!NEA is the summation of PAA!NEA (Pending Central Angles of A as defined in Definition 7). Definition 10 (The on-duty decision rule). Consider any two candidates A and B. We say that A is lager than B, or A > B, if and only if CAA!NEA > CAB!NEB or PAA!NEA > PAB!NEB while CAA!NEA ¼ CAB!NEB or IDA(ID of node A) > IDB(ID of node B) while PAA!NEA ¼ PAB!NEB and CAA!NEA ¼ CAB!NEB . Algorithm III-C.2 – The On-duty Election Algorithm FOR (each candidate node q) { Compare q with all of its neighbors using On-duty Decision Rule; IF (q is the decision result) set q on-duty; IF (q is fully sponsored) set q off-duty; }//end of loop In the on-duty election algorithm, the sensor with the largest Contribution Angle is elected in its neighborhood. As we know, the number of such on-duty nodes is much lower than the number of off-duty nodes; thus, the on-duty election algorithm runs faster than the off-duty election algorithm even in the worst case. Moreover, the number of control messages is remarkably reduced. The on-duty election algorithm reduces the number of control messages by reducing the rounds of algorithms, but it causes another problem in that the on-duty election algorithm turns off fewer sensors than the off-duty election algorithm because on-duty election algorithm begins at an unstable status that may use more on-duty nodes to resolve the off-duty conflict problem. Thus, another alternate election algorithm has
2713
been developed in order to keep both initial time t and the number of on-duty nodes at a low level. 3.3.3. Alternate election algorithm Even though the on-duty election algorithm shows better performance on the average than off-duty algorithm, too many control messages are wasted in the worst case. In each round of this scenario, only one candidate can be turned off. Another drawback of on-duty is the Crack Effect, where the central nodes will have a high probability of being elected as the on-duty nodes and the Candidate area will generate a number of pieces of hole between onduty and off-duty nodes, which will lead to more on-duty nodes than the off-duty algorithm offers. In order to resolve such problems, an Alternate Election algorithm in which only candidates who receive stable status announcement messages will be involved in the election, has been proposed. The algorithm is separated into odd and even rounds and begins with the SAM (Status Acknowledge Message) from the on-duty nodes (edge nodes) after identifying all of the candidates and on-duty nodes in the initial phase. Two new status, on-duty candidate and off-duty candidate, are introduced as temporary status. An Odd round is triggered when a P-Off node A receives either a SAM message from all P-On neighbors, or SAM messages from all neighbors at the zero level. If A is the node selected by the election algorithm, it will be set to Off-duty; otherwise, it will be set as P-On. In the odd round, only the off-duty election will be performed. Any candidate A that receives a SAM will check the distance between itself and the sender. If this distance is smaller than sensor range r, the status of the sensor is set to off-duty candidate; otherwise, it is set to on-duty candidate. Such a candidate A will classify its neighbors into four groups: • On-duty neighbors. Neighbors that sent an On-duty SAM message; • Off-duty neighbors. Neighbors that sent an Off-duty SAM message; • Pending off-duty (P-Off) neighbors. Neighbors whose distance from any on-duty neighbor is lower than r; • Pending on-duty (P-On) neighbors. Neighbors whose distance from any on-duty neighbors is larger than r; An even round will be triggered when a P-On candidate A receives a SAM message from all of its Off-duty candidate neighbors. The center angles of A are recalculated using the new status information gathered from its P-Off neighbors. If A cannot keep its candidate status, it will be set as On-duty; otherwise, it will be set as P-Off. In the Even round, any candidate that receives a SAM will update its status to P-On and execute the Extended Coverage Calculation Method to find out whether the candidate is a qualified On-duty node. The status update rule is that the status only updates by means of the high-level SAM message.
2714
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
Algorithm III-C.3 – Alternate Election Algorithm FOR (P-On candidate q in even round) { Extended Coverage Calculation; IF (q is not candidate) set q on-duty; }//end of loop FOR (P-Off candidate q in odd round) { Extended Coverage Calculation; Off-duty Election; }//end of loop
3.4. The wake-up strategy Several node-scheduling solutions presented in literature introduce methods for turning off the largest possible number of sensor nodes. However, no solution addresses the wake-up strategy. A wake-up strategy can influence not only the energy consumption in a sensor network, but also the quality of the monitoring task. Three issues must be considered in a wake-up strategy: 3.4.1. The off-duty sensors should take over the sensing area of the dead sensor (a sensor that has run out of energy) as soon as possible in order to avoid coverage holes In the periodical wake-up strategy used in C-PNSS, if a sensor A dies between two rounds, events occurring within the sensing range of A may not be detected by other onduty sensors, and the off-duty sensors will only wake up to monitor the area watched by A at the beginning of the next round. In summary, a hole may appear between two rounds and cannot be fixed until the off-duty sensors wake up at the beginning of the next round. 3.4.2. The wake-up strategy should wake up as few necessary off-duty sensors as possible Consider the wake-up process of the two off-duty sensors A and C in Fig. 5. The following strategy would typically be used: If an on-duty sensor p is running at low battery, it will broadcast a wake-up message so as to wake up turned-off nodes in its neighbor list. Assuming that onduty sensor B (the dark circle) is running out of energy and broadcasts a wake-up message, both off-duty sensors A and C will be waken up. However, in this case, node A is fully sponsored by its neighbors; thus, in this case we say that off-duty sensor A is Over-Waken-up.
A
B
C
A
Fig. 5. Over wake-up.
B
C
3.4.3. The wake-up strategy should not be overly costly in terms of energy Because the period wake-up strategy uses too many control messages, which is very energy-consuming, an alternative solution is to have an additional communication channel (wake-up channel) that receives wake-up messages. The problem with this solution is that the channel must be listened to even when the nodes are off-duty. However, it is well known that sensing wastes much less energy than transmitting a message. If a node can be monitored in terms of its energy level, and its state of ‘‘running out of energy’’ is detected as an event (signal) that can be captured by the sensing subsystem of the off-duty sensor, a wake-up solution that avoids the problems with the current wake-up strategies and saves energy can be devised. To a typical sensor, the sensing energy cost is close to the message receiving energy cost [29]. In order to detect such events the sensing module has to be work on standby or rotation modes. Gu and Stankovic [19] have proposed a radio wake-up circuit that uses the energy in the radio communication signal to generate a wake-up signal. Such circuit does not have any regular energy supply. When the radio signal is present with a special frequency, the antenna receives energy input and drives the circuit. According to the experiments of Lin on Berkeley Mica2 motes with 5/ 8-wavelength monopole antenna, when transmission power is 10 dBm (10 mW) the output voltage increases to about 0.6 V. Even when comparing to the 5 V of the PC and 1 V of some other low-power embedded systems, 0.6 V still can be accepted as a high signal and distinguish from noise. A radio-triggered ID circuit was also mentioned in the same paper. Through assigning each sensor a special frequency, the receiver can even identify the sender of such incoming signals. By applying such wake-up circuit our wake-up scheme (the Algorithm 3.4.3) is described as follows. Algorithm III-D – Wake-up Strategy For (each node p that is out of energy) p generates an out-of-energy signal; For (each off-duty node q that captures an out-of-energy signal) { identify ID by frequency; recalculate the off-duty status; IF (q cannot keep the off-duty status){ wakeup(); q.status ‹ On-duty; }}//end of loop 4. Energy cost comparison of OCoPS and C-PNSS Because our optimal coverage calculation scheme was derived from C-PNSS, we will compare our OCoPS with C-PNSS firstly from the principles point of view, then quantitative viewpoint. C-PNSS is a novel energy-saving sensing coverage protocol that uses fewer control messages during its initial phase
A. Boukerche et al. / Computer Communications 30 (2007) 2708–2720
than other protocols. However, the use of these control messages is repeated round after round. Considering the lifetime of the network and not merely its initial phase, such control messages may cost more energy than is desired. However, because of the limitations of the Central Angle Method, C-PNSS will treat certain fully sponsored nodes as edge nodes and ignore the association sponsors (mentioned in our scheme) during the wake-up phase in a high-density network. Thus, C-PNSS uses more on-duty nodes in order to maintain full coverage. Our OCoPS guarantees that our algorithm does not create a hole and is more efficient than C-PNSS in terms of the coverage issue. The price for achieving this goal is using more control messages than C-PNSS during the initial phase. Moreover, in the OCoPS scheme, off-duty nodes are aware of the wake-up signal by applying the wake-up radio circuit we mentioned in the wake-up scheme section. An off-duty node turns off all its equipment until its receive wake-up signal generated by wake-up radio circuit. Considering that energy consumption is affected by many factors, such as sensing range, the length of the message, the number of rounds for data gathering, and the power consumption model itself, OCoPS is mostly concerned with message transmission energy cost, that include both control messages and data message (the energy cost of sleeping node was ignored). Because we only consider the sensors that have overlapped area, when transmitting the control messages, the transmission range is adjust to double of the sensing range. After the initial phase, the transmission range is adjusted to reach the furthest on-duty neighbor. the energy consumption for the control messages that our scheme uses does not significantly affect network lifetime, as can be seen by the results obtained from the simulations. In order to compare C-PNSS and OCoPS on the same grounds, we use the same energy parameters and radio model as C-PNSS. Those parameters are shown in Table 2 and discussed in [26]. After the initial phase, both C-PNSS and OCoPS work in rounds on a TDMA network. Each round lasts 10 s. Each sensor sends a 2000-bit report message to the base station with a 0.5-s time interval.
2715
Consider a wireless sensor network that includes N sensors, M candidates, O off-duty sensors, and runs T rounds. In order to simplify the calculation, we assume that the average number of neighbors is XC in C-PNSS and XE in OCoPS; the energy cost of sending is S/bit; the energy cost of receiving message is R/bit; We assume the size of control message (Bc) are the same in both OCoPS and C-PNSS. the decision algorithm finishes after t rounds and the number of remaining candidates in the i rounds is Mi. During the wakeup phase, each off-duty node has XO on-duty neighbors. In Table 1, the lifetime of the network is separated into two phases: the initial phase and the wake-up phase. Both schemes will have to exchange a PAM (Position Acknowledgement Message) in order to gather the location information and broadcast a CAM (Candidate Acknowledgement Message) to publish their candidate status among neighbors. Because the information contained in these messages is almost exactly the same, we can assume that both types of messages cost the same amount of energy for both C-PNSS and OCoPS. The difference is that, in order to resolve the Off-duty Conflict, our Pt scheme requires extra messages whose energy cost is i¼1 M i ðS þ X Ei RÞ Bc to publish candidate status. In the initial phase, OCoPS consumes more energy than C-PNSS, but not in the long run. Since C-PNSS works in rounds, let us consider the period from the end of the initial phase to the time at which the network dies after T rounds, which is referred to as wake-up phase in Table 1. In OCoPS, because we assume that the energy cost of generating an out-of-energy event (signal) is same as the cost of sending a message, and because each sensor generates only one out-of-energy event, the total energy cost of generating out-of-energy events in our scheme is NS*Bc. The off-duty sensors, on the other hand, did not cost energy to catch such signal. However, because the communication range in OCoPS is doubled that of C-PNSS, in order to simplify the calculation we can assume that XE 6 2XC. Knowing that in a fixed area M < N, when N is large sufficiently we can ignore the edge nodes or say that M 6 N. By increasing the number of rounds, Mi approaches (M O) which itself approaches zero when M and O are large sufficiently. Therefore, we
Table 1 Comparison of the energy costs of control messages Our scheme Initial phase
C-PNSS PAM CAM
NS+NXER NS+NXER t P M i ðS þ X Ei RÞ
SAM Run T rounds Wake-up phase Total cost
Schedule phase
PAM SAM
NS+NXCR NS+NXCR
Wake-up phase
T*SAM
T*(NS+NXCR)
Total cost
Bc*(2 + T)(NS + NXCR
M i ð1 þ X Ei ÞÞÞ
Total cost
