Eigenvalue/eigenvector characterization of slope reflected signals

Abstract

Eigenvector/eigenvalue analysis of experimental acoustic propagation data obtained in the region of the East Australian Continental Slope indicates this method is sensitive to changes in multipath structure and can distinguish multiple signal arrivals. A 16‐element horizontal array at 300 m was towed in deep water, while a 152‐Hz source at 18‐m depth was moved from shallow to deep water. Source‐receiver separation was about 60 nmi with source bearing near array endfire. Shipboard processing consisted of 1024‐Hz time sampling, 4K FFT processing, and production of cross‐spectral matrices for selected frequency bins around 152 Hz. The eigenvalues of a spatially whitened cross‐spectral matrix may be used to count the number of statistically independent arrivals. These may be from multiple sources or from multipath effects of a single source. The eigenvalues for each sample cross‐spectral matrix (five minute averages) were calculated and displayed versus time in order to correlate the number of large eigenvalues with the number of signal arrivals. In addition, conventional, minimum variance (ML), and eigenvector beamforming produced wavenumber‐time displays of signal arrivals. For comparison, a GRASS‐based ray trace model, using actual bottom reflection coefficients and deep water sound‐speed profile, was used to predict expected arrivals. Comparisons of the visible shift in eigenvalues with arrivals in both the beamformed and modeled data show the eigenvalue detection method to be sensitive to changes in the multipath structure and capable of distinguishing multiple signal arrivals that were not detectable with conventional beamforming methods.