A bivariate model for Arctic undersea noise

Abstract

A bivariate model is developed for the impulsively contaminated noise of the Arctic undersea environment. It is based on one proposed by S. V. Czarnecki [‘‘Nearly Optimal Detection of Signals in Non-Gaussian Noise,’’ Ph.D. thesis, Dept. of EECS, Princeton Univ. (October 1983)] where impulsive and normal background noise regions are treated as coming from two independent Gaussian sources with differing variance. The noise is considered as the output of a switch that toggles between the two noise sources. In this paper, a statistical analysis is conducted on a 10-min time series of Arctic in-water noise data (Fram II) to provide a stochastic description of the real noise data within the framework of such a model. A median filter-based detector is developed to detect the state of the noise source selection switch and partition the time series into the impulsive and normal background regions. Goodness-of-fit tests are used to check the dual source noise distributions against the Gaussian hypothesis. A point process description of the operation of the source selection switch is also derived by statistically analyzing the switch transition inter-arrival times.