Joint Event Detection And Environment Perception In Wireless Sensor Networks

With characters of low power, low cost, distributed and self-organizing, wireless sensor networks is applied to more and more fields. But there’re still some problems to be solved, and, in this thesis we tried to solve a few of them by introducing new algorithms.In the event detection applications, when events occur in the area of a wireless sensor network(WSN), there would be corresponding signals generated from the events. The signals emitted from the events attenuate in the environment and will be measured by the sensor nodes. Then the WSN can estimate both the locations and magnitudes of the events from the received signal strength (RSS). The environment is a non-ignorable factor that has influence on the attenuation of the signals. It is necessary to be estimated, to make the event detection more accurate.In many cases, the monitored area is large, so large-scale wireless sensor network is needed. If we use a centralized network topology, multi-hop communication between sensor nodes and the central node would easily lead to congestion and packet loss. And if the central unit does not work well, the whole network may break down. Collabora-tion through information sharing between neighbors helps to accomplish the detection mission, and also lead to system characteristics of communication load balancing and system robust. So it is worth concentrating on distributed algorithms.(1) Problem Formulation.Considering the impact of environment, we introduce an en-vironmental parameter in the signal attenuation model. Through exploiting the s-parse nature of the events, we propose an-norm regularized least squares formu-lation that automatically estimates the number of the events as well as their locations and magnitudes; the attenuation coefficient of the environment is also an optimiza-tion variable in the formulation.(2) Centralized Algorithm. The centralized algorithm solves the е1-norm regularized least squares optimization with alternating direction method.(3) Distributed Algorithms. We develop a decentralized algorithm to solve the joint event detection and environment perception problem using the alternating direction method of multipliers. Through exploiting the problem structure, the decentralized algorithm in each node boils down to three steps:an event detection step that is convex; an environment perception step that is a non-convex one-dimensional op-timization problem; a multiplier update step that contains only algebraic operations. To improve energy-efficiency of the wireless sensor network, we further develop a heuristic scheme that accelerates the proposed decentralized algorithm.(4) Simulation and Tests. Numerical simulations demonstrate the effectiveness of the decentralized joint event detection and environment perception algorithms and we also test the algorithms with data from the experimental platform.In summary, this thesis we present a sparse optimization model for event detection and environment perception, and develop centralized and decentralized algorithms. We also validate performance and practicality the algorithms with some simulations and experimental tests.