Passive acoustic monitoring using random matrix theory

Abstract

Cross-correlating ocean noise is a potential alternative to using active sources to monitor and study ocean environments. However, directional sources in the medium (usually ships) often introduce a bias in the cross-correlations, making the travel time estimates unreliable. Here, we use recent results in random matrix theory for the eigenvalue density of isotropic noise sample covariance matrices to separate the directional noise from the diffuse noise field. The eigenvalues obtained from ocean data agree well with the theoretical results. Beamforming on the diffuse noise components reveals a fairly spatially isotropic nature for the noise field, which fits the assumption. The cross-correlations using the diffuse noise field alone converge to the expected travel times (i.e., unbiased estimates) and are stable temporally.