Detection on a distributed network using information geometry

Abstract

A passive detection scheme based on information geometry is tested on underwater acoustic field data acquired through multiple sensors from the Distributed Network Consensus experiment. These detectors use a signal discovery method wherein a target is detected when the statistics of that signal are sufficiently different from those in a broadband background. The signal excess is viewed here as the non-Euclidean distance between two probability distributions interpreted as points in a Riemannian manifold. The SNR is artificially lowered to create a low SNR scenario and it is shown that in spite of this, the geometric detectors perform quite well.