Acoustic source localization

Abstract

Many technologies rely on underwater acoustics. Most of these applications are able to denoise Gaussian noise from the surrounding, but have trouble removing impulse like noises. One source of impulse-like noise is the snapping shrimps. The acoustic signals they emit from snapping their claws hinder technologies, but can also be used as a source of ambient noise illumination due to the rough uniformity in their spatial distribution. Understanding the spatial distributions of these acoustic signals can be useful in working to mitigate or amplify their effects. Our collaborators in Singapore use a multi-array sensor to take measurements of the sound pressures of the environment in the ocean. This thesis investigates in solving the signal reconstruction problem -- given the measurements, reconstruct the locations of the original signal sources. A numerical model for the system consisting of the snapping shrimp signals, the environment, and the sensor is formulated. Three methods of reconstruction -- Disciplined Convex Programming, Orthogonal Matching Pursuit, and Compressive Sensing -- are explored, and their robustness to noise, and sparsity are examined in simulation. Results show that Two-Step Iterative Shrinkage Threshold (TwIST) is the most robust to noisy and non-sparse signals. The three methods were then tested on real data set, in which OMP and TwIST showed promising consistency in their results, while CVX was unable to converge. Since there is no available information on ground truth, the consistency is a promising result.