An approach to maximum likelihood identification of autoregressive marine mammal sources by passive sonar

Abstract

In this paper we propose a marine mammal classification method that relies in the assumption that the sources are autoregressive (AR). By incorporating the AR coefficients of each source we make explicit their contribution to the signals at array sensors. A logarithmic likelihood function is introduced in the frequency domain so that all available information from the sources can be incorporated thus letting a proper classification. We show that it is possible to deal with different sources regardless the closeness of their center frequency and their relative location. In the simulations we explore the potential applications of our method in real situations where it is needed to identify sources as they are detected and localized.