1 Introduction
With the development of wireless networks and hardware technologies, especially very large-scale integrated circuit technology, it has become possible to integrate sensing, communication, and computing capabilities into a single sensor node. Wireless sensor networks (hereinafter referred to as sensor networks) are composed of tens of thousands of such nodes.
It integrates sensing and communication functions, aiming to monitor various parameters (such as temperature, humidity, and target location) and collect sensitive data in harsh or inaccessible environments. Sensor networks are now widely used, including analyzing environmental conditions in remote, uninhabited areas (such as monitoring forest fires by collecting temperature data); installing sensing nodes on specific vehicles to analyze local communication traffic and design optimal routes from source to target; in the military, they can be used to monitor, locate, and track target movements; and in highly polluted areas, they can collect relevant information for disaster recovery. Because these sensor nodes are limited in size and have limited energy, recharging a large number of nodes in the network is almost impossible. Therefore, it is necessary to consider how to minimize node energy consumption to extend the continuous operating time of the sensor network under limited energy conditions.
2. Preparations
Sensor nodes can be divided into the following three modules according to their functions.
① Sensing Module. Assuming the data source generation rate is r, the sensing energy consumption per unit time of a node is Psense = a3r, where a3 is a constant (in J/bit). ② Communication Module. Given a sending node u and a receiving node v, with a data transmission rate of r and a distance d between the two nodes, we have Ptrans(u, v) = (a11 + a2dk)r and Prec = a12r. Ptrans(u, v) represents the transmission energy consumption of node u per unit time, and Prec represents the reception energy consumption of node v per unit time. Here, k is the path attenuation exponent (k is typically 2 or 4), while a11, a12, and a2 are wireless communication constants (the value of a2 depends on k). ③ Computation Module. Compared to the sensing and communication modules, the energy consumption of the node's computation module is very small and can usually be ignored.
This paper assumes:
① Sensor networks are applied in target tracking scenarios. ② As shown in Figure 1, E is the data source node at a certain moment in the network. The sensor nodes are mainly used to monitor the specified target point and transmit the collected results to the remote collection node B in a multi-hop manner. Here, we assume that the energy of node B is sufficiently large, that is, the energy consumption of node B is not considered. Assume that there are N nodes in the sensor network. ③ The nodes are randomly distributed in a finite area R. The communication radius of the nodes is rt. Data transmission between nodes is bidirectional. That is, for any two nodes u and v in the network, if node u can communicate directly with node v, then node v can also communicate directly with node u. As shown in Figure 1, the sensing radius of the node is ds. That is, the sensing node can only sense the existence of the data source within a distance of ds. ④ All sensing node antennas in the sensor network are located on the same horizontal line, and the antennas are omnidirectional. ⑤ Using GPS technology, the sensor nodes can obtain their own location information. Currently, the accuracy of GPS can reach about 5 meters.
As can be seen from reference [3], given a two-dimensional space R, the sensing radius ds of the sensor node, the energy consumption parameters a11, a12, a2, a3 and the path decay index k, the number of sensing nodes N, the initial energy E of each node, and assuming that the data source trajectory follows a certain uniform probability distribution function lsource(x, y), then the upper limit of the network lifetime T can be obtained as follows.
3. Algorithm Idea
Sensor networks are essentially data-centric self-organizing networks. However, previous studies on the lifetime of sensor networks assumed that the surrounding environment was reliable, meaning that sensing nodes would only fail when their energy was completely depleted. This did not take into account the impact of the environment on the sensing nodes.
In sensor networks, since all nodes within a distance of the data source ds can collect data and forward it, how to ensure that these nodes can continue to operate for as long as possible is a problem we are concerned about. To address this problem, we have adopted the concept of fault-tolerant redundancy and introduced backup sensing nodes so that the failure of a certain sensing node will not affect the normal operation of the entire network. Therefore, in this paper, we propose a backup-based distributed algorithm to extend the continuous working time of the network. Simulation results show that under the node failure environment, by reasonably controlling the number of nodes, the continuous working time of this algorithm is longer than that of the method in reference [3].
4. Algorithm Description
In the actual operation of the algorithm, it can be divided into three stages: the initialization stage, the data transmission stage, and the task takeover stage.
During the initialization phase, each node in the network needs to determine the minimum hop count to the collection node B. Each node is initially set to the active state. The collection node first broadcasts a HOP message with a communication radius rt. Except for the collection node, the initial hop count of each node is set to infinity. The initial hop count of the HOP message is set to 0. When a node receives a HOP message, it checks if it has already received the message. If it has not received it, the node adds the sender's information (including the sender's ID) to its routing table, sets the sender as its upstream node, increments the hop count of the HOP message by 1 and sets its value as its own hop count, and then forwards the new HOP message to its neighboring nodes with a communication radius rt. Otherwise (i.e., a node has previously received the HOP message), the node will back off for a period of time before sending the HOP message. Furthermore, the node will only consider HOP messages with the minimum hop count received within the backoff time.
After the backoff time, the node compares the hop count of the received HOP message with the hop count previously stored by the node: if the former is less than 1, the node adds the sender's information to its routing table as another different upstream node; if the former is less than the latter and the difference is greater than 1, the sender becomes the node's new upstream node. Accordingly, both the node hop count and message hop count in the routing table need to be updated, and the modified HOP message will continue to be forwarded by the node; if the former is not less than the latter, the node discards the newly received HOP message. Note that in the above description, the node only forwards the message immediately upon receiving the first HOP message. This strategy has the advantage of speeding up the transmission rate of HOP messages and shortening the execution delay of the initialization phase. On the other hand, the benefit of introducing a backoff time is that it allows the node to wait in a waiting state so that it can receive more HOP messages from its neighboring nodes. Although a longer backoff time increases the completion time of this phase, the method is effective because this phase only needs to be executed once for a static sensor network and also avoids the increased energy consumption due to additional message forwarding. After all, energy is the most valuable resource for sensor nodes. As can be seen, after the initialization phase is executed, each node will accurately know its minimum hop count and all its upstream nodes.
Once the designated target for monitoring appears, the data transmission phase begins. Multiple nodes sensing the target use information exchange to select the node with the fewest hops and the highest energy as the target's sensing node. Other nodes sensing the target act as backup nodes, entering a dormant state to reduce energy consumption. Before actual data transmission, the actual sensing node uses the MECN algorithm proposed by Rodoplu et al. and information from its upstream nodes to establish a minimum energy-consuming path from the source node to the collecting node B.
The task takeover phase will begin when one of the following two situations occurs: ① Active task switching during normal operation. At this time, as the sensing node operates, its own energy consumption decreases. When it can no longer complete the data sensing (or transmission) task, it will actively initiate a task takeover command. ② Task switching when a sensing node malfunctions. Since all backup nodes are in a dormant state during normal operation, the heartbeat technology in distributed systems cannot be used to detect the status of the sensing nodes. Here, we use a timeout method. That is, when the collecting node B cannot receive data from the target point within a specified time interval T, there is a possibility that the sensing node has failed. In this case, the collecting node B will send a flooding message to the sensor network, using the principle of directional diffusion to wake up the nodes in the backup node set and re-initialize them.
When none of the sensor nodes within a distance of the data source node ds have sufficient power to forward data to any neighboring node, the entire sensor network will cease operation.
5. Algorithm Message Load Analysis
In this section, we will conduct a simple analysis of the message load of the proposed algorithm from the perspective of the average number of messages processed by a single node.
Initialization phase: The number of HOP messages broadcast by each node depends on the backoff time setting. It can be seen that if the backoff time is large enough, each node will broadcast a maximum of 2 HOP messages: the first HOP message will be broadcast after the node receives the first HOP message from its neighboring nodes; and the second HOP message will be broadcast after the backoff time expires.
Data transmission phase: The sensing node only needs to sense the broadcast messages between multiple nodes of the specified target to determine the target; that is, the node only needs to broadcast one message in this phase.
Task takeover phase: During normal operation, the task takeover node only needs to broadcast one message; similarly, during anomalies, the task takeover node only needs to send one flood message from the collection node (in fact, this phase is the opposite of the initialization phase where nodes broadcast HOP messages; we can expect that if the backoff time for this step is set properly, most nodes will broadcast at most once). Therefore, each node broadcasts at most one message during this phase.
In summary, if the backoff time is set appropriately during the initialization and task takeover phases, the message load of the sensor node in the algorithm can be a maximum of 4 broadcast messages.
6 Simulation Results
The discrete-time simulation tool Ns-2.29, developed by Berkeley University, was used. It was assumed that all nodes in the sensor network were randomly and uniformly distributed within a 1500 m × 1500 m rectangular area, with N nodes (150 ≤ N ≤ 1500). The initial energy of each node was set to 2 J, solely to shorten the simulation time and without altering the simulation's behavior. The wireless transmission radius rt of each node was set to 50 m, the sensing radius ds to 20 m, the path attenuation parameter k to 4, a3 = 50 nJ/bit, a11 = 45 nJ/bit, a12 = 135 nJ/bit, a2 = 0.001 pJ/bit/m⁴, the data source generation rate to 1 bit/s, the routing protocol used on the nodes to adopt the AODV protocol for wireless ad hoc networks, and the application layer traffic generator to adopt CBR (constant bit rate) with a size of 512 bit/s. To obtain more reliable data results, each algorithm was executed 100 times. For each experiment, 10 different random sensor network topologies were generated. The final simulation results were obtained by averaging the results, with the confidence interval set at 95%.
In actual operation, the performance of the algorithm is considered under two conditions: normal and failure (random simulation of the failure of sensing nodes). Given different numbers of network nodes, the new algorithm is compared with the method in reference [3], and the ratio between the simulation results obtained in the experiment and the theoretical value obtained by equation (1) is given, as shown in Figure 2.
As can be seen from Figure 2, under normal circumstances, the network lifetime of the new algorithm is shorter than the theoretical value in reference [3]. This is because, compared to reference [3], some of the energy of the sensor nodes in the new algorithm is consumed in the execution of algorithm message exchange during operation. When the sensing nodes fail, it can be seen that when 150≤N≤550 or N≥750, the network lifetime of the new algorithm is shorter than the theoretical value in reference [3]. This is because when 150≤N≤550, due to the small number of sensing nodes in the network, some nodes act as both sensing points and relay nodes, so the time performance of the entire network will decrease rapidly. When N≥750, as the number of nodes in the network increases, the density of the entire network increases, the number of neighboring nodes of each node increases, and the probability of channel contention and channel collision will also increase accordingly. This will also have an adverse effect on the network lifetime. However, when 550≤N≤750, the network lifetime of this algorithm is longer than the theoretical value in reference [3], that is, the network performance of the new algorithm is better than that of reference [3]. Therefore, the distributed algorithm proposed in this paper is feasible, especially for handling node failures in the network.
Figure 3 illustrates the relationship between the average message load per node and time during the algorithm's operation. As can be seen from Figure 3, the average message load per node does not increase significantly with the increase in deployed nodes, which verifies that the algorithm designed in this paper is applicable to large-scale sensor networks. Furthermore, it is observed that from the start of operation until the node reaches the network's lifetime, the maximum message load per node is only 5, thus confirming the algorithm's effectiveness.
7. Conclusion
To extend network lifetime, a distributed algorithm based on backup-aware nodes is proposed. Experiments demonstrate that this algorithm achieves further improvements in time efficiency, thus guiding us on how to optimize node management under current conditions to enhance network lifetime. Future research directions include:
① In the algorithm proposed in this paper, when the sensing node fails, the collecting node B actively initiates flooding information, which will lead to congestion of the network's wireless channel and a decrease in network performance. Furthermore, some nodes may receive the same flooding information from multiple neighboring nodes, consuming valuable memory resources. Exploring alternative methods to avoid this unnecessary communication overhead and resource consumption is a direction for future research.
② In the algorithm, it is assumed that the number of data source nodes is one and their positions are fixed. In the future, a distributed fault-tolerant algorithm should be considered when there are multiple data source nodes and they are moving.
③ Sensor network algorithms are all application-environment dependent. In some cases, using only a single sensing node can affect the accuracy of data acquisition results due to the different angles from which the node observes the object. This necessitates cooperation among multiple nodes to improve data accuracy. Finding a balance between accuracy and network energy consumption is also a worthwhile research topic.
