Localization in mobile wireless sensor networks via sequential global optimization

We develop a novel approach to source localization in mobile wireless sensor networks. Standard approaches make explicit assumptions relating to the statistical characteristics of the physical process and propagation environments which result from distributional model assumptions in a likelihood-based inference method. In contrast, we adopt an approach known in statistics as a non-parametric modeling framework which allows one to relax the number of required statistical assumptions, specifically with regard to the distributional properties of the received signal and the physical process. This is achieved via a re-formulation of the problem as a flexible non-parametric regression model via the framework of Gaussian Processes. Coupling this modeling perspective with a Bayesian optimization mechanism, we frame the global optimization objective as a sequential decision problem. We then develop an efficient algorithm to sequentially select the optimal location at which the mobile sensor should obtain observations under communication and mobility constraints. Simulation results demonstrate the efficiency of the algorithm at achieving accurate localization in a wireless sensor network.