Abstract
With a growing demand for location-aware applications and services, indoor positioning techniques based on wireless sensor networks (WSNs) have received great attention in recent years. Positioning in WSNs can be realized in a centralized or decentralized way, where the decentralized approach generally has a better energy efficiency than the centralized approach in large-scale and/or dense networks. It is therefore worth exploiting decentralized positioning schemes that can provide good location accuracy and tracking performance for WSNs. In this dissertation, we propose some new decentralized positioning algorithms based on recursive optimization procedures for WSNs, including the recursive least-squares, the improved projection onto convex sets (POCS), and the recursive weighted least absolute value method. Each of these algorithms is derived from the minimization of a recursive-in-time cost function and then realized in an iterative decentralized manner. Specifically, the target location can be calculated iteratively by taking a weighted average of the local estimates based on the participating sensor nodes’ reliability information, where a participating sensor node computes the newest location estimate according to its own observation and the most recent local estimate passed from the previous participating sensor node. All the proposed schemes adopt an effective rule for updating the step size at each iteration and guarantee convergence to a stationary point, where every one of them is equivalent to the incremental subgradient method or the POCS method with an appropriate variable step size. Computer simulation results show that the proposed schemes have better location accuracy than previous related methods. To track a target in WSNs, we further propose a decentralized weighted extended Kalman filtering (WEKF) scheme for positioning and tracking, where a message-passing algorithm is adopted for inter-sensor-node communications and for adaptively selecting the participating sensor nodes as the target moves within the area of interest. During each iteration, the current participating sensor node computes a local estimate and passes it on to the next participating sensor node for further processing. The update process is circulated only among the selected participating sensor nodes that surround the target. A convergence analysis is given to show that the proposed WEKF-based method converges. Computer simulation results are also given to demonstrate the effectiveness of the proposed approach.