Robust Expectation-Maximization Algorithm for Multiple Wideband Acoustic Source Localization in the Presence of Nonuniform Noise Variances

Abstract

Wideband source localization using acoustic sensor networks has been drawing a lot of research interest recently. The maximum-likelihood is the predominant objective which leads to a variety of source localization approaches. However, the robust and efficient optimization algorithms are still being pursuit by researchers since different aspects about the effectiveness of such algorithms have to be addressed on different circumstances. In this paper, we would like to combat the source localization based on the realistic assumption where the sources are corrupted by the noises with nonuniform variances. We focus on the two popular source localization methods for solving this problem, namely the SC-ML (stepwise-concentrated maximum-likelihood) and AC-ML (approximately-concentrated maximum likelihood) algorithms. We explore the respective limitations of these two methods and design a new expectation maximization (EM) algorithm. Furthermore, we provide the Cramer-Rao lower bound (CRLB) for all these three methods. Through Monte Carlo simulations, we demonstrate that our proposed EM algorithm outperforms the SC-ML and AC-ML methods in terms of the localization accuracy, and the root-mean-square (RMS) error of our EM algorithm is closer to the derived CRLB than both SC-ML and AC-ML methods.