The blind source separation and classification of diver's noise based on adaptive beamforming and frequency selection

Abstract

This paper introduces an algorithm to separate and classify the diver’s noise in the presence of other noise interferers. The passive detection and classification for divers has been discussed before by using the breathing rate as features. But the performance of this approach significantly diminishes when other noise sources existing simultaneously, especially the other divers or louder shipping noises. In this paper, the wideband MVDR is used as an adaptive beamformer for horizontal linear array to better discriminate the azimuth of diver source of interest and separate the noises from different sources. The frequency content of the diver noise is also estimated adaptively according to azimuth distribution with no prior knowledge to enhance the separation ability from noises of similar azimuth and used to improve the correct rate of the classification. The performance of the proposed algorithm is discussed using both diver noise recorded in tank and at-sea shipping noise recorded in ocean for varying signal to noise ratio.