VSPSA for Acoustic Source Localization in Wireless Sensor Networks

Acoustic source localization is one of the important applications of wireless sensor networks. Among the energy based acoustic source localization methods, Maximum Likelihood (ML) is known as an accurate algorithm, while its computation cost is high. Expectation Maximization (EM) algorithm reduces the computing complexity, but it is easy to trap into local optimum. In this paper, we propose a Simultaneous Perturbation Stochastic Approximation (SPSA)-based solution that aims at achieving accurate acoustic source localization and fast convergence by computing the approximate gradient of the target function to estimate the position of acoustic source. Furthermore, an island model constructed using Voronoi diagram is presented to significantly reduce the searching space and improve the searching efficiency. Through extensive simulation, we demonstrate that the Voronoi enhanced SPSA (VSPSA) algorithm outperforms EM algorithm significantly with higher localization accuracy, lower computation complexity, and better robustness in noise environment. Testbed experiments also demonstrate the feasibility of this algorithm.