Implementing physical constraints for noise-only mode shape estimation on real data

Abstract

Many underwater acoustic tasks sample a narrowband pressure field with a vertical line array. In the absence of strong local sources, the noise sampled by an array includes both a spatially correlated and a spatially uncorrelated component. The acoustic waveguide's modes form a basis for the spatially correlated noise component generated by distant sources. This basis can be estimated from the eigenvectors of the noise sample covariance matrix. Propagation physics constrain the mode shapes to be real, but the eigenvectors are generally complex. These complex vectors require a phase rotation for each eigenvector prior to taking the real part. Previous work found that the eigenvector mode estimates' phases include spatially correlated noise. This noise correlation degrades the performance of spatial averaging and minimum variance estimates of the unknown phase rotation. The best estimate is the phase of the array element with the maximum magnitude for each mode, which is robust to the noise's spatial correlation. This research evaluates the previously developed phase rotation techniques using a real data set. The data was collected off the north coast of Elba Island by the SACLANT Centre in 1993. [This work was supported by ONR.]