Direction-of-Arrival Estimation Using Learned Iterative Shrinkage and Thresholding Algorithm

Abstract

The paper examines direction-of-arrival (DOA) estimation in underwater acoustics using Learned Iterative Shrinkage and Thresholding Algorithm (LISTA). As a learned sparse signal recovery approach based on neural networks, LISTA is investigated in the underwater acoustic multi-source DOA estimation through simulations. Different LISTA models are trained using plane wave signals with different Signal-to-Noise-Ratios (SNRs). The tested multi-channel data received by the Horizontal Line Array (HLA) are generated by an acoustic propagation model with multiple sources in a shallow water environment. The one-snapshot DOA estimation performance of the LISTA models for a single frequency is demonstrated on the noisy data with different SNRs. The results show that the LISTA models can generate sparse solutions for DOA estimation and using low-SNR noisy data as the training set for LISTA is beneficial to the DOA estimation.