| The wireless sensor network (WSN) technology has attracted increasing attention from industry, academia and government because of its great potential as a low-cost solution to a wide spectrum of real-world challenges. A sensor network is a self-organizing network of low-power embedded sensor nodes communicating with each other through wireless communication. Tightly coupled with the physical world, a sensor network is capable of capturing critical information from the environment in real-time. This information enables the users to make better decisions in various sectors of the human society.Event detection is a very compelling class of applications of sensor networks, whose major task is to detect vital physical events such as fire and gas leakage. For event detection applications, the design objectives are twofold. On the one hand, the system should provide high performance of event detection. Performance goals include shorter detection latency and higher detection probability. On the other hand, the system should operate in a very energy-efficient way. Two key reasons account for this. First, the system is expected to be functional for a long lifetime, e.g., several months or even years. Second, sensor nodes are powered by small batteries and thus have a very limited lifetime. It has been a grand challenge to design energy-efficient sensor networks. In this thesis, we present our systematic research on low-power distributed event detection using embedded sensor networks.We propose a novel approach for energy-efficient event detection using low duty-cycled sensor networks. With this advantageous approach, a number of fundamental problems should be addressed. We have performed theoretical analysis, advanced algorithm design, distributed protocol development and comprehensive experiments to solve the problems. First, there is an intrinsic tradeoff between event detection and energy efficiency. We precisely characterize, by mathematical analysis, the intrinsic tradeoff between detection performance and system lifetime. This result is crucial for system designers when developing embedded sensor networks for distributed event detection.Second, a great many practical applications have a clear energy budget of sensor networks for event detection. For these applications, it is highly imperative to maximize the detection performance of the whole sensor system. After formulating the wakeup scheduling problem as a quadratic optimization problem, we show that it is extremely difficult to derive the global optimum because of its NP-hardness. We design CAS, a distributed algorithm that allows each sensor to determine its wakeup time cooperatively with its neighboring sensors. This algorithm significantly improves event detection performance with minimal computation and communication overhead.Third, we thoroughly explore the paramount QoS provision problem for distributed event detection in sensor networks. It is highly desirable for many real-world applications to provision network-wide guaranteed QoS for event detection. In response to this need, we design PAD, a fully distributed protocol which adaptively tunes the active probability of every sensor for energy efficiency. The unique feature is that this algorithm can dynamically maintain guaranteed QoS while consuming minimal energy. The algorithm can work in the absence of costly time synchronization.Keywords: Wireless Sensor Networks, Event Detection, Quality of Service, Statistical Guarantee, QoS Provision, Power Conservation, Duty Cycling and Wakeup Scheduling... |
