Algorithms for Position and Data Recovery in Wireless Sensor Networks by Lance Doherty Research Project

Networks of hundreds or thousands of sensor nodes equipped with sensing, computing and communication ability are conceivable with recent technological advancement. Methods are presented in this report to recover and visualize data from wireless sensor networks, as well as to estimate node positions. A communication system is assumed wherein information from sensor nodes can be transferred to a centralized computer for data processing, though suggestions are made for extensions to distributed computation. Specifically, this report presents four topics. First, the notion of using network connectivity to reconstruct node positions via linear or semidefinite programming is explored. Random feasible node placement and bounding methods are both found to increase in precision with the indiviual geographical constraints. Second, the potential effectiveness of two correlation-based sensor data encoding schemes is reported. Blind correlation methods are found to provide meager compression while semi-blind correlation can effectively reduce bandwidth requirements by one-half. Third, trajectory reconstruction through a sparse sensor network is used to track objects with expectation-minimization techniques. Trajectories can be distinguished providing that sufficient spatial or temporal separation exists. Fourth, optical flow algorithms are used to visualize time-varying continuous flow around the network. A qualitative analysis of the reconstructed flow for several case studies suggests a minimal node density as related to flow speeds.