Towards prolonged lifetime for deployed WSNs in outdoor environment monitoring

Abstract

Recently, Wireless Sensor Networks (WSNs) emerged as a powerful and cost-efficient solution for unattended Outdoor Environment Monitoring (OEM) applications. These applications impose certain challenges on WSN deployment, including 3-Dimensional (3-D) settings, harsh operational conditions, and limited energy resources. To prolong lifetime of the deployed WSN, while mitigating the effects of these challenges, we propose the use of Relay Nodes (RNs) in addition to Sensor Nodes (SNs) in a distributed manner. While RNs facilitate reaching distant destinations, SNs can reserve their limited energy resources for sensing and data gathering. In addition, Mobile RNs (MRNs), which is a set of RNs capable of being reallocated (i.e. mobilized) at any point within the network lifetime, can be used to overcome possible link/node failure caused by the harsh conditions. It can also guarantee minimal energy consumption through imposing a balanced traffic distribution. This article proposes a 3-D grid-based deployment for heterogeneous WSNs (consisting of SNs, RNs, and MRNs). The problem is cast as a Mixed Integer Linear Program (MILP) optimization problem with the objective of maximizing the network lifetime while maintaining certain levels of fault-tolerance and cost-efficiency. Moreover, an Upper Bound (UB) on the deployed WSN lifetime, given that there are no unexpected node/link failures, has been driven. Based on practical/harsh experimental settings in OEM, intensive simulations show that the proposed grid-based deployment scheme can achieve an average of the expected UB.

Introduction

Wireless networking and advanced sensing technology have enabled the development of low-cost, power-efficient WSNs that can be used in various application domains such as healthcare, military, and OEM [2]. The main building block of a WSN is SNs. These nodes collect information, i.e. sense, some physical and/or chemical properties of a monitored environment and transmit their measurements to a central node known as the Base Station (BS). This transmission can either be periodic or on-demand [1]. Amongst the various application domains of WSNs, OEM attracted considerable attention due to its unique characteristics [3], [4]. These include large monitored areas, isolated and distant territories, harsh operational conditions, and high probabilities of node and link failures [5], [6]. Fortunately, a well-planned WSN deployment in such environments offers a reliable, yet cheap, means of decentralized data collection with minimal human intervention. However, in order to maintain a prolonged and reliable monitoring, the WSN needs to withstand the harsh operational conditions of outdoor environments, like heavy rains, snowfalls, sandstorms, extreme temperature variations, etc. These conditions may cause a significant percentage of node and link failure [7], [8], which can be captured and statistically quantified using the Probability of Node Failure (PNF) and Probability of Disconnected Nodes (PDN). Hence, a well-planned WSN deployment should reduce these two probabilities [9]. While PNF can partially be reduced through proper placement and packaging of the nodes; PDN can significantly be reduced by a fault-tolerant deployment [10]. Fault tolerance is a pivotal step for a sustained and reliable monitoring. It is achieved through injecting a certain level of redundancy in the network such that it can withstand a given percentage of failure while maintaining the desired monitoring level.

Due to the scarcity of energy resources in outdoor environments, network nodes are almost always battery-powered. However, since a WSN in an OEM application is envisioned to work unattended for long periods of time, a stringent constraint is imposed on the energy consumption per node. This becomes a serious challenge when the network monitoring area is huge. In this case, energy can be drastically consumed if a distant BS is to be reached directly by all the SNs with a blind knowledge of the remaining energy budget per node in the network. To overcome this, a distributed system of RNs/SNs can be used [15]. A RN is a dedicated communication node with larger energy storage, i.e. larger battery, than regular SNs, capable of collecting data from a cluster of SNs and passes it to the BS. RNs can either be static or mobile. Unlike Static RNs (SRNs) that are located once within the network lifetime, Mobile RNs (MRNs) are given certain mobility features such that they can be relocated on demand [16]. The use of this type of RNs helps resolving bottleneck problems during the network lifetime. In fact, MRNs can be seen as a proactive solution to maintain connectivity and fault-tolerance when some communication paths are running out of energy or losing connectivity [17]. Also, one of the unique features of OEM applications is the 3-D space monitoring where the height of a node is as important as its horizontal position [12], which cannot be considered by 2D deployment algorithms. For instance, in monitoring the gigantic redwood trees in California, some experiments required placing the sensors at varying heights ranging from the ground surface up to tens of meters [13]. Moreover, monitoring the intensity of certain gases, like CO₂ [28], [15], [14], requires sensor placement at different heights such that monitoring coverage and accuracy requirements are met. Such a 3-D monitoring can easily be secured if the coverage space is modeled as a 3-D grid, which is a typical coverage model in OEM applications [2], [28]. Hence, network nodes can only be placed at the vertices of this grid. In fact, the grid model limits the search space to a finite number of points. The shape of the grid building units can either be a cube, a hexagonal, an octahedron or any regular shape chosen to meet certain coverage levels [19]. Other advantages of the 3-D grid modeling include exclusion of all positions where node deployment is not possible, and accurate description of the possible routing paths [18]. In spite of the aforementioned grid advantages, placing the WSN nodes on the grid vertices might affect their deployment optimality. Nevertheless, this effect is fortunately controllable by the grid edge length (i.e. the deployment optimality is proportional with the count of vertices, and inversely proportional with the grid edge length). Thus, a more restricted search space (without affecting the deployment optimality) is required. Meanwhile, the overall network cost is proportional to the total number of nodes deployed. Hence, the lower the number of nodes, the lower the overall cost. However, communication reliability and fault tolerance require abundant node deployment. Consequently, a tradeoff exists between the overall cost and the network performance. Hence, the network deployment problem can be best modeled as an optimization problem. The objective is to maximize the network lifetime through reducing the energy consumption, while the constraints are cost efficiency, communication reliability, and fault-tolerance. This problem will be mathematically modeled in subsequent sections.

Recently, there have been several proposals for an energy-efficient, lifetime-maximizing WSN deployment. These proposals differ both in the type of nodes used in the network and their deployment strategies. In terms of the type of nodes used, most of the work in literature has considered the use of only sensor nodes deployed in a target area. Very few researchers have considered the use of RNs to improve the communication range and network lifetime. PEDAP and its power-aware version, PEDAP-PA [31], L-PEDAP [32], EESR [33], AND MLDA [34] are examples of WSNs that consider only sensor nodes deployed in a target area, and implement various schemes to improve the network lifetime. For instance, PEDAP considers minimizing the total energy expended by the network in a round of communication, but it does not consider the issue of balancing the energy consumption among the nodes. It consumes less energy to find a route and is able to achieve a good lifetime for the last node, but does not provide for load balancing among SNs and reliable communication in the network. PEDAP-PA is an improvised version of PEDAP that considers balancing the energy consumption among the SNs by computing their remaining energy using a cost function. However, this cost function considers only the transmitting nodes’ residual energy, and the routing tree is recomputed after a pre-defined number of rounds, which is a major drawback when considering an improvement in the reliability of the system. Moreover, both PEDAP and PEDAP-PA are centralized algorithms that were designed for smaller deployment areas and might be unsuitable for large scale deployments such as OEM applications. EESR and L-PEDAP consider the remaining energy levels at both the transmitter and transceiver SNs to achieve better load balancing. EESR uses Kruskal’s algorithm for the routing tree construction and works best when the SN is in the same transmission range and can communicate directly with the sink. L-PEDAP is capable of automatically re-routing a packet to the destination when it finds that the energy level of a node is less than the threshold value. Thus L-PEDAP achieves both load-balancing and reliable communication as it is capable of identifying node failure and recovering from it, unlike EESR. However, L-PEDAP fails to minimize the energy consumption and communication time. This is true even in the case of MLDA that fails to achieve the desired tradeoff between communication delay and network lifetime in the system. Thus, none of these works in existing literature have been able to satisfactorily address the problem of maximizing network lifetime while reducing the energy consumption, and considering the constraints of cost efficiency, communication reliability, and fault-tolerance.

Moving on to node deployment strategies, they are mainly classified as random and deterministic strategies (see Table 1). In random deployments, nodes’ positions can be chosen in a purely random deployment plan or, based on a weighted random deployment plan, where the distributed nodes’ density is not uniform in the monitored areas. For instance, K. Xu et al. [20], studied random deployment of static RNs in a two dimensional (2-D) plane. The authors proposed an efficient deployment strategy that maximizes the network lifetime when all RNs reach the BS with a single hop only. Motivated by the weakness of the uniform random deployment; the authors proposed a weighted random deployment strategy with a gradually increasing density of nodes as the distance to the BS increases. This strategy compensates the number of RNs for the energy needed to reach the BS. Hence, monitoring reliability can be sustained for longer periods of time, i.e. network lifetime is maximized.

In contrast, deterministic deployments aim at deploying nodes exactly on specific, predefined locations. These deployments can be accomplished through centralized or distributed approaches (see Table 1). In centralized approaches, global information gathering is required to end up with the targeted nodes’ positions. For the most part each node requires a complete knowledge of the whole network topology. However, in distributed approaches deployment decisions are made based solely on some local knowledge per node. For example, a deterministic deployment strategy for mobile data collecting nodes was proposed in [22], assuming centralized knowledge and decisions made at the Base Station (BS). These mobile nodes move along a set of predefined tracks in the sensing field. In the proposed deployment strategy, SNs were able to relay data in addition to their sensing duties. It was shown that using data collectors (mobile relays) extends the network lifetime compared to conventional WSNs using static SNs only. In fact, data collectors were used earlier in [23], [24]. The network lifetime was divided into equal length time intervals, called rounds. The data collectors are relocated at the beginning of each round based on a centralized algorithm running at the Base Station. The objective was to minimize the aggregate consumed energy during one round. It was shown that the optimal locations according to this objective function remain optimal even when the objective becomes to minimize the maximum energy consumed per SN. It should be remarked that these two energy metrics are not suitable for finding the optimal locations of mobile nodes since the optimal solutions will not be functions of time, i.e. time-independent. The reason is that the maximum or aggregate energy consumed per round might not change with time, and hence, locations of the data collector will not change with time. Consequently, the locations calculated may be far from optimal. Despite the advantages of these proposals, the deployed networks were prone to network partitioning and/or communication loss due to lack of fault tolerance. In addition, these proposals are designed for 2D deployment problem, which is not the case in OEM applications. In OEM, a 3D deployment plan for environment sensors is a must to achieve the desired outdoor observations [12], [13]. Moreover, ignoring candidate sensory positions in the 3D space can waste numerous opportunities in reducing energy consumptions based on closer locations and in achieving better connectivity performance. In [25], a fault-tolerant random deployment was proposed. In particular, the authors proposed a distributed deployment algorithm to achieve a desired level of fault tolerance for all-SNs WSN. The transmission power of every node is gradually increased until either the distance between two neighboring nodes exceeds a specific threshold or the maximum transmission power is reached. In this deployment, fault tolerance is achieved at the expense of added cost. Transmission power adaptation requires complex hardware that raises the per-node cost, hence increasing the overall network cost. In addition, power adaptation results in added energy consumption, hence causing lifetime reduction. In [26], another fault-tolerant WSN deployment was proposed. The authors considered the case where at least two disjoint paths exist between each pair of SNs. To achieve a desired level of fault tolerance, deterministic RN placement was used. The problem was formulated as an optimization problem. However, it turned to be NP-hard. Hence, a polynomial time approximating algorithm was proposed instead. This algorithm identifies candidate positions for RNs that cover the maximum number of SNs while assuming a regular communication range shape. Thus, RNs positions may not be accurate since it solely depends on the transmission ranges that are irregular in practice. Alternatively, fault tolerance could be achieved through deploying spare (redundant) node. In fact, faulty RNs are even more harmful to the network than SNs. This behavior was shown in [3] for a deterministic, grid-based deployment. Hence, an efficient fault-tolerance should account for faulty SNs as well as RNs.

In this article, a comprehensive network deployment problem is considered based on decentralized algorithms running not only at the system sink (BS), but also at the core network nodes (SNs/RNs). SNs and RNs are jointly deployed. In addition, some MRNs are used to release the pressure from overloaded paths and fix any connectivity problems. Thus, the targeted problem can be stated as follows: Given a 3D deployment space and a limited number of SNs and static/mobile RNs, find the optimal positions that prolong the network lifetime¹ while maintaining connectivity & certain fault-tolerance constraints. Accordingly, our main contributions towards solving this problem can be summarized as:

• We overcome the huge search space of the candidate RNs positions by finding a subset of the grid vertices for these RNs based on their intersecting communication ranges.

• The optimization problem is divided into initial deployment and periodic redeployment. In the initial deployment, the optimal locations of all nodes are found. In the periodic deployment, MRNs are relocated based on a decentralized decision made by deployed nodes at their present positions.

• Efficient energy metrics, the minimum node residual energy and the total energy consumed, are used to maximize the network lifetime. These two metrics guarantee an influential MRN relocation.

• An upper bound for the maximum network lifetime in ideal operation conditions is derived. This bound is used to show the performance gains achievable by the proposed two-phase solution.

The remainder of this paper is organized as follows. Section 2 describes the system model and the mathematical framework. The proposed deployment strategy is presented and discussed in Section 3. Section 4 presents the numerical results, while conclusions are drawn in Section 5.