Multichannel myopic deconvolution using ambient noise sources

Abstract

The ocean ambient noise has been increasingly utilized for ocean passive sensing and monitoring applications. By recording the received signals from the same individual noise source (for example, the shipping noise) at multiple hydrophones simultaneously, we develop a framework, called multichannel myopic deconvolution, which can allow us jointly estimate the source and the channel responses without any assumption about the source, but using some priori knowledge of the channel. Our work on this classical signal processing problem has two novel aspects. First, we recast the corresponding bilinear system of equations as a linear system with a rank constraint. This allows us to apply recently developed algorithms and analytical tools from the field of low-rank recovery to the blind channel estimation problem, yielding insight into the conditions under which accurate channel estimation is possible. Second, we incorporate (continuous-time) parametric uncertainty about the Green's functions as subspace constraints in the low-rank recovery problem. These subspaces are generated in a systematic way using the singular value decomposition, and their dimension can be directly related to the amount of priori knowledge we have about the channel. We will present simulations in shallow water environments of the proposed approach from relatively short observation times.