Mobility-assisted, energy-efficient algorithms for distributed inference in Wireless Sensor Networks

In this dissertation, we investigate major problems in a resource limited wireless sensor network for detection and estimation, in signal processing and communication perspectives. Specific problems addressed in the dissertation can be categorized as follows: (1) Power management under correlated observations (2) Distributed node scheduling for sequential inference (3) Impact of node mobility on detection and dynamic state estimation in mobility assisted sensor networks.;The dissertation is started by focusing on efficient power management in distributed detection and estimation with correlated local observations. In most potential applications, once deployed WSNs need to work unattended for a long time. Since sensor nodes are equipped with finite energy supplies, power management is considered to be a core issue in designing a WSN. Also, in a practical densely deployed sensor network, it is more likely that the observation noise is dependent across sensor nodes. The performance analysis under correlated observations is analytically complexco mpared to that with independent observations. We explore optimal power scheduling techniques with correlated observations under different practical correlation models.;Among all possible power consumption modes, the transmission power is the most dominant in a sensor node. A large communication power is consumed in a traditional centralized data/decision fusion since nodes have to communicate with a central fusion center. On the other hand, the reliability of centralized data fusion architecture depends on the robustness of the fusion center for failure. Hence, distributive approaches for data fusion are desirable in power constrained sensor networks. In the second part of the dissertation, a distributed sequential methodology is proposed for parameter estimation in a stationary sensor network in which nodes are assumed to exchange information locally via noisy communication channels. Distributed algorithms for node selection in the sequential estimation process are developed taking the trade-off between information gains and the communication costs into account.;In this dissertation, the cost of deploying mobile nodes is investigated in terms of the required node density to achieve desired performance measures. Several important performance measures in target detection in a mobility assisted sensor network are derived assuming random and independent node mobility models. In situations where random node mobility models are inefficient, it is required to navigate mobile nodes purposefully to compensate for the lack of performance resulted in a all-static network. To that end, a new interactive distributed protocol, collaborating among static and mobile nodes, for node mobility management is proposed to improve the coverage over time in an efficient manner. The worst case performance in detecting mobile targets by hybrid sensor networks is derived in terms of the exposure.;Contrast to continuous movements for target detection, due to energy constraints, it may be desired to keep mobile nodes stationary until a target is detected with a certain confidence level, or useful statistics regarding the target locations are available under stationary configuration. We develop two decision fusion architectures for distributed detection by a hybrid sensor network when nodes are allowed to move only if necessary depending on the application requirements. Moreover, the cost of allowing nodes to be mobile is investigated analytically in terms of the minimum fraction of mobile nodes required to achieve desired performance gains under certain constraints.;Estimation of the state of a moving target is an another important application of wireless sensor networks. In the last part of the dissertation, we address the problem of non-linear/non-Gaussian dynamic state estimation by hybrid sensor networks. The node mobility is exploited to improve the tracking quality dynamically. (Abstract shortened by UMI.)...