Non-stationary covariance estimation for adaptive beamforming

Abstract

Standard adaptive beamforming (ABF) assumes a quasi-stationary covariance. However, in low frequency shipping noise the covariance is not stationary, due to the relative motion of ships and receiver. A physical model was previously developed for the behavior of the covariance due to ship motion, and a technique was developed to provide improved estimation. Relatively wide filters were used previously in order to simultaneously estimate the covariance from all moving ships. It has recently been recognized the it is most important in ABF to handle the loudest ship, which is usually the closest and therefore the most non-stationary. Here we will show simulations of the results of using more narrow filters to estimate the covariance due to the loudest ship. The technique will be contrasted to a method of matched field processing with moving ships developed by Lisa Zurk, as well as with derivative-based updating, another non-stationary estimation technique. The potential to improve the estimate by using waveguide invariance across a band of frequencies will also be discussed.