Automatic signal detection in noise using entropy.

Abstract

Automatic detection of signals in noise is a common problem in many areas of acoustics. In the field of passive acoustic monitoring of marine mammals, the signals to be detected are vocalizations. The noise originates from natural (wind, waves, and rain) and man‐made sources (e.g., shipping, construction, and seismic surveys). Signal characteristics vary broadly: frequency ranges from a few Hz to 200 kHz, duration from milliseconds to seconds to hours. Noise characteristics vary by similar orders of magnitude. While specific automatic detectors have been designed to successfully find specific calls in specific environments, the challenge is to find a large variety of calls in a large variety of noise. An exploitable difference between calls and noise is that most noise is a result of stochastic processes (wind, waves, rain, cavitating propellers + seismics generate gas bubbles underwater of varying size + resonance frequency), while many animal signals are a result of deterministic processes (vibrating st...