A Gibbs sampling approach to matched-field source localization and deconvolution

Abstract

In narrow-band source localization the source spectrum is typically unknown. Matched-field techniques employ normalization schemes to remove the uncertainty for the unknown amplitude at the source, whereas consideration of phase differences between replicas and true field addresses the unknown phase issue. In this work we implement a matched-field technique that simultaneously estimates source location and source spectrum. The estimation is performed using an efficient Gibbs sampling scheme. Results of the Gibbs sampling estimation approach are compared to those of the linear (Bartlett) processor under several noise levels. The comparison reveals that the simultaneous process of deconvolution and localization has superior performance to that of the linear localization process. Moreover, improvement in localization is obtained without a significant increase in computational requirements. In addition to the improved performance, the proposed method has the advantage of providing more information than conventional matched-field localization, since it also estimates the source spectrum which is often of great interest. [Work supported by ONR.]