Sound speed estimation and source localization with particle filtering and a linearization approach

Abstract

In previous work, a particle filtering method was developed that provided estimates of multipath arrival times from short-range data and, subsequently, employed them in geometry, bathymetry, and sound speed inversion. The particle filter provided probability density functions of arrival times, that were then “propagated” backwards through a sound propagation model for inversion. That implies that every particle from the probability density is employed in the inversion scheme, creating a potentially computationally cumbersome process. In this work, we develop a new method for such parameter estimation which relies on linearization. The novel aspect is that the Jacobian matrix now includes derivatives with respect to Empirical Orthogonal Function coefficients. The approach, requiring only a few iterations to converge, is particularly efficient. Results from the application of this technique to synthetic and real (SW06) data are presented and compared to full-field inversion estimates. [Work supported by ONR and the NSF CSUMS program.]