Experimental study of the EM and SAGE algorithms with application to sonar data

Abstract

We study the expectation maximization (EM) and space alternating generalized EM (SAGE) algorithms by processing real data measured by a towed hydrophone array. The EM and SAGE algorithms are well known numerical methods for locating modes of likelihood functions. The complicated multi-dimensional search for finding ML DOA estimates can be simplified to one dimensional search by EM and SAGE. Computational efficiency can be further improved by the fast EM and SAGE algorithms which use smaller search spaces than those used by EM and SAGE. Experimental results show that the SAGE algorithm converges faster than the EM algorithm in most cases. Without losing estimation accuracy, the fast EM and SAGE algorithms lead to a significant reduction in computational time.