Rank-N stochastic matched field processing

Abstract

Underwater acoustic source localization can be performed using matched field processing (MFP). Successful localization depends on accurate signal and environmental input parameters to a propagation model. Some parameters, notably the sound velocity profile (SVP), vary both spatially and temporally, and are thus difficult to treat as deterministic quantities. Instead, one can model the SVP and the resultant pressure field as random vectors. This allows one to detect targets using second order statistics of the received and modeled pressure fields. Baggeroer et al. [J. Acoust. Soc. Am. 106, 2126] demonstrated successful target localization using individual singular vectors of a modeled covariance matrix. A new method of MFP is proposed which combines energy from orthogonal singular vectors of the modeled covariance matrix. Each correlation is weighted and summed to yield a detection statistic. A maximum-likelihood nonrandom parameter estimation method is then applied to determine the source location. Through inclusion of signal energy which would normally be lost to environmental mismatch, this method significantly reduces estimation bias in a simulated 200-m shallow water waveguide. Localization results are shown using both simulated and experimental data from the 1998 Santa Barbara Channel Experiment. [Work supported by SAIC, DARPA and ONR.]