Binaural Listening based DOA Estimation of Underwater Targets using Unsupervised GMM

Highly accurate estimation of Direction of Arrival (DOA) of underwater targets with less number of hydrophones has always remained a challenging problem among the acoustic signal processing research community. In addition to the recurring overheads in terms of money, the large footprints of the conventionally used arrays with hundreds of sensors make them infeasible to be used along with small platforms. Motivated by the hearing ability of animals with two ears, this paper proposes a two-sensor DOA estimation approach utilizing the binaural cues extracted from the target signals of interest. An unsupervised GMM clustering is performed on the extracted features in order to estimate the DOA of targets within an angular range of 0° to 180°. Evaluation of the proposed system using various performance metrics yields promising results on binaural data synthesized from underwater target records acquired during various expeditions conducted in the Indian Ocean.