Environmentally adaptive signal processing in shallow water

Low-frequency signals propagating in shallow water are conveniently represented as a set of acoustic normal modes individually characterized by different dependencies of pressure on depth. These differences have been exploited to form spatial filters which selectively enhance single modes of propagation. The performance of these filters is degraded by errors in the knowledge of the calculated depth function. This paper discusses an alternative approach to filter construction, in which filter coefficients are adaptively extracted from measurements of naturally-occurring acoustic noises. When the acoustic array densely samples the water column, the modal eigenfunctions can be obtained by diagonalization of the cross-spectral density matrix formed by averaging over an ensemble in which the modal components add incoherently. In the experimentally more practical case, in which the receiving array does not span the water column, useful spatial filters can be constructed from noise cross-spectral density matrices