Maximum Likelihood Source Localization using the EM Algorithm to Incorporate Prior Parameter Distributions

Richard Perry and Kevin Buckley¹
ECE Department, Villanova University, Villanova, PA 19085
buckley@ece.vill.edu, (610)-519-5658, FAX: (610)-519-4436

Abstract:

In this paper we introduce a new algorithm for the estimation of source location parameters from array data given prior distributions on unknown nuisance source signal parameters. The conditional maximum-likelihood (CML) formulation is employed, and ML estimation is obtained by marginalizing over the nuisance parameters. In general, direct solution of this marginalization ML problem is intractable. We introduce an expectation-maximization (EM) algorithm solution, which is applicable to any prior distribution.