Sequential inversion of self-noise using adaptive particle filter in shallow water.

Abstract

The geoacoustic inversion based on a horizontal towed array sonar receiving tow-ship noise has demonstrated a promising technique for the parameter inversion in shallow water. In order to characterize the evolution of parameters in the time-varying environment, the adaptive particle filter for the sequential inversion is presented in this paper. The inversion problem is formulated as a dynamic and nonlinear process in the Bayesian framework, due to the fact that the self-noise is recorded sequentially in space and time. To deal with the interparameter correlations and time-varying noise process, the adaptive sequential importance sampling is carried out based on the estimated covariance matrix of parameters that is updated on-line. And the particles are proposed with an adaptive shift to handle the rapidly varying parameters. The tonal components at low frequencies of the self-noise are used in the inversion. The sequential inversion method is verified through the processing of both synthetic data and the sea-trial data in the shallow water environment. The results show that the adaptive particle filter method can achieve a more stable and accurate estimate than successively running global optimization algorithms and can do better than particle filter inversion in a rapidly varying environment.