Energy Efficient Distributed Target Tracking And State Detection Algorithms

Estimation and detection theories, which are the two embranchment of statistical signal processing, are widely used to fetch information in electronic signal processing systems like radar, communication, sonar, speech, image processing, biomedicine, environment monitoring and seismology. Detection theory is used to determine if the interested event is happening or which discrete state the environment is now in while estimation theory obtains more details about the interested event.Wireless sensor networks have become increasingly important in a number of civilian and military applications. Sensor nodes can be densely deployed in a large sensing field to acquire information of interest, which will be used in applications like target tracking, environmental monitoring, security and so on. These are all the field of estimation and detection theories. There is often a trade-off between performance and energy consumption for a limit of sensor power. People are interested in minimizing energy consumption while keeping a good performance or maximizing the performance under certain energy or bandwidth constraint. We investigated how to minimize the network communication cost while keeping most of the performance for the application of target tracking and state detection, so as to reduce energy consumption and improve network lifetime. Among these two applications, target tracking is a typical estimation problem. Position and speed of the target need to be estimated each time step. And state detection is a detection problem, by which state of the environment is determined.Consider target tracking problem in wireless sensor networks with limited energy, it is quite necessary to decide which sensors should and how to take part in the tracking process as the quality of data from each sensor is different from each other. Two new tracking algorithms for wireless sensor networks are proposed, one based on leader agent and the other is a consensus based distributed algorithm. In the leader agent based tracking algorithm, communication costs both in data aggregation and that between leader agents are considered. Error matrix from the tracking process is used to select the optimal sensor set to get the best tracking performance under certain communication constraint. As the constraint optimization problem is very complex, Gauss-Seidel method and convex relaxation method are used to make the problem much easier to solve. The experimental results show that our algorithm performs better under the same communication constraint.For the distributed tracking algorithm, consensus method is used to broadcast local information each tracking step, and every sensor will reach a consensus (or approximate) global optimal result. In our newly proposed distributed tracking algorithm, only sensors with informative data are selected to take part in the consensus algorithm, which will reduce communication cost and speed up the convergence rate while the tracking performance is almost unchanged.In distributed state detection, every sensor uses not only its own data, but also information from its neighbors. Yet the exchange of observed data between neighbors will lead to large communication cost. A new distributed detection algorithm based on private data and actions of neighbors is proposed in the light of social learning process and convergence property is analyzed. Decision exchange takes much less communication cost compared with a continuous variable like observed data or likelihood ratio. When the state of nature is binary, only 1 bit is needed. Experimental results show that our algorithm will converge much faster with the same communication cost.