Stochastic matched-field processing.

Abstract

Listen

Translate article

Matched-field processing (MFP) exploits the acoustic field to estimate either a source position or environmental parameters. Many have noted that the performance of MFP degrades rapidly with uncertainty about the field especially if adaptive methods are used to suppress sidelobes. It is clear that robust MFP methods are needed for working with experimental data. There are three categories of uncertainty that have been examined: (i) observational (array geometries and sensor responses are imprecisely known), (ii) statistical (ambient field covariances have errors), and (iii) environmental (the propagation is uncertain because of oceanographic characterization). The third category is examined in this presentation. The fundamental issue for robust MFP is to match the stochastic propagation with array processing that responds to an ensemble of replicas not just one from a deterministic model. At high SNRs one can search the ensemble for the appropriate sample vector using efficient searching algorithms such as simulated annealing. At modest SNRs the processing must incorporate errors from the noise. The approaches to MFP with environmental uncertainty are reviewed and some of the relevant signal processing literature for the ‘‘detection of random signals in noise’’ are noted. [Work supported by Mathematical Sciences, ONR.]