Moving source parameter estimation in an uncertain environment

Abstract

Environmental uncertainty is one of the limiting factors in the matched-field localization. Within a Bayesian framework, environmental focalization has been widely used to solve the augmented localization problem, in which the environmental parameters, source ranges and depths are considered to be the unknown variables. However, the position of the moving source varies with time, which limits the observation interval and the number of acoustic signals. Therefore, it has to estimate lots of unknown parameters with the limited observation information. When the source moves fast or the environment has great uncertainty, the environmental focalization gets worse. Taking the parameter estimation of Kalman filter in the non-stationary process as a reference, the acoustic signals from a series of observations are treated in a simultaneous inversion. This provides the most informative solution, since data from multiple source locations are brought to bear simultaneously on the environmental unknowns, which in turn constrain the source locations better. In this article, the time-unvarying parameters are introduced to describe the source position. The source positions are inverted indirectly by the time-unvarying parameters, which reduces the estimated parameter dimension effectively. At the same time, the current estimated results are treated as the priori information of the next inversion, which establishes the new prior distribution and cost function. It could compensate for some individual abnormal data effectively and realize continuous localization of the moving source. The noise signals radiated from a surface ship target are processed and analyzed. The Bayesian tracking algorithm greatly increases the observation interval and the number of acoustic signals, and enhances the localization accuracy in an uncertain water environment. Tracking results of the ship noise indicate that simultaneous inversion of multiple acoustic observations with constant velocity track model and the Thikhonov regularization provides a better solution than sequential independent inversions. It is indicated that the Bayesian tracking method learns the uncertain environment as more observations become available. It is discovered that the maximum a posteriori solution and the two-dimensional solution have similar results according to the global positioning system value. The reason is that the source locations are treated implicitly by the source speed, which is similar to the marginal probability distribution by reducing the multidimensional posterior probability density to the representative two-dimensional probability distributions.