Linear and nonlinear Bayesian localization in ocean acoustics

Abstract

Ocean-acoustic localization can be considered an inverse problem that involves estimating model parameters that specify the location of one or more acoustic sources and/or receivers based on fitting measured acoustic data, which can include observable quantities such as travel times, travel-time differences, modal dispersion, or acoustic-field structure. In a Bayesian inversion approach, data and prior information are used to compute the posterior probability density (PPD) of the model parameters, providing uncertainty analysis that quantifies the information content of the problem. The Bayesian formulation provides the generality to treat all uncertain parameters (e.g., both source and receiver locations, environmental properties, clock drifts) as unknowns subject to appropriate levels of prior information. Marginalizing over nuisance parameters can improve localization accuracy or at least account for parameter uncertainties in the localization uncertainty. Some Bayesian localization problems can be solved efficiently using linearization and iteration, with closed-form approximations for the PPD. In other cases, nonlinear (numerical) methods such as Markov-chain Monte Carlo sampling or trans-dimensional inversion are required. This talk will illustrate these concepts with a series of examples including array-element localization for moored and towed arrays, tracking autonomous underwater vehicles at a test range, marine-mammal localization, and multi-source matched-field localization