Narrowband Acoustic Mode Estimation Using Performance-Weighted Blended Filtering

Abstract

Estimating acoustic mode coefficients is a common problem in underwater applications such as tomography and source localization. Vertical arrays facilitate the separation of a narrowband signal into its constituent modes. A spatial matched filter provides a correct estimate of the mode coefficients if the array’s sampling of the modeshapes preserves their orthogonality. If the sampled modeshapes are not orthogonal, the matched filter for a desired mode can be overwhelmed by leakage of energy from other modes. A pseudo-inverse (PI) mode filter addresses the leakage problem by placing nulls that eliminate mode interference, assuming that the total number of modes arriving at the array is known. Since the number of modes is not known a priori, it can be difficult to choose the right rank for the PI mode filter. This paper proposes a new approach to mode filtering that blends mode filters of different ranks based on their observed performance. The blend coefficients are obtained from an algorithm devised by Buck and Singer for adaptive beamforming [1]. Simulations demonstrate the ability of the new mode filter to achieve better performance than a fixed-rank mode filter, particularly for non-stationary scenarios.