Quantized Information Based Target State Estimation And Fusion In Wireless Sensor Networks

Recent advances in micro-electro-mechanical systems (MEMS), sensing technology, wireless communications, and distributed signal processing have enabled the development of low-cost, low-power wireless sensor networks (WSNs). Target tracking is one of the fascinating application scenarios for WSNs, which has been studied and widely applied to national defense, environmental monitoring, intelligent transportation, and industrial automation etc.Target tracking in a WSN has the advantages such as better robustness and higher accuracy, while each node in the network has limited energy supply and communication bandwidth. Therefore, quantized information based state estimation and fusion for target tracking in WSNs is investigated in this dissertation. Specifically, the contributions can be stated as follows.1. Design of Reduced-Order Filters for Complex SystemsBased on a brief overview of state estimation and fusion, the energy-to-peak filtering that has attracted less attention is considered, including full-order and reduced-order ones. Based on linear matrix inequalities (LMIs) technique, both the necessary and sufficient condition and an explicit solution of the filter are given. Besides, a unified information fusion model is set up with the optimal solution given. Then, the robust estimation fusion algorithm for systems with uncertainties is presented.2. Distributed Quantized Track FusionConsidering the limited energy and bandwidth, local covariance matrices are compressed and vector quantized with the state estimates. Then, considering the unknown or incomplete correlation of local estimation, the fusion center (i.e. the cluster head) fuses the local quantized messages through a novel robust approach, i.e. internal ellipsoidal approximation.3. Target-Oriented Sensor Clustering StrategyThe tree-based and static clustering-based target tracking systems have such shortages as expensive routing, high redundancy etc. Hence, a target-oriented dynamic clustering strategy is proposed: Nodes around the target are activated with monitoring reports exchanged; the node with more residue energy and less average communicational distance is competed as cluster head; other activated ones participate this cluster as member nodes and make observation on the target. Simulation results show that comparing with the randomly selection, the target-oriented scheduling strategy save energy more that 42%.4. Adaptive Quantized Measurement based Target TrackingThe local measurements are quantized adaptively and transmitted to the fusion center; the fusion center estimates the target state according to the received messages. Attentions are focused on adaptive bandwidth allocation and adaptive quantization thresholds selection. The posterior Cramer-Rao lower bounds (CRLBs) for target tracking using adaptive quantized measurements are also given.5. Channel-Aware Target Tracking and Cross-Layer OptimizationThere is little result on target tracking in WSNs with uncertain channels. According the typical uncertain model for digital channels, binary symmetric channels (BSC), the problem of channel-aware target tracking is investigated with the posterior CRLBs derived. Based on the posterior CRLBs of channel-aware target tracking, the problem of cross-layer design and optimization for sensor scheduling is considered. Furthermore, a heuristic approach to sensor scheduling is given to surround the optimization complexity.6. Scalable Distributed Target Tracking for Peer-to-Peer (P2P) Sensor NetworksFirst, based on dynamic consensus strategy, two scalable distributed filters are proposed for P2P sensor networks, one is distributed robust filtering, and the other is distributed sigma-point Kalman filtering. Both strategies have the advantage that only information exchanges between neighboring nodes are necessary, and network-wide agreement can be achieved for tareget state estimation. This makes both strategies scalable for large-scale sensor networks. Besides, a novel dynamic consensus is proposed with its convergence and stability discussed. Finally, a scalable hybrid estimation fusion framework is presented, which is robust against link failure and varying topology.