Deconvolution of underwater acoustic marine mammal signals from the Northern Gulf of Mexico

Abstract

Extracting useful information from electronically recorded data measurements often requires the signal to be separated from the noise and the blurring impulse response deconvolved out. In order to denoise LADC-GEMM underwater acoustic data, we apply and compare several techniques from Fourier frequency filtering and wavelet denoising. We also look at methods of estimating the acoustic impulse response (system response, transfer function, etc.) for the northern Gulf of Mexico to be used in implementing several deconvolution methods applied to the same dataset. Deconvolution techniques tested include Fourier deconvolution, Wiener filtering, various blind deconvolution methods, as well as combination algorithms. Comparisons of techniques will be presented, showing that the different methods may yield different results. Metrics for analysis will include both the final quality of results along with computational efficiency. Some recommendations will be made within the framework of marine mammal signal detection; although specifics of denoising and deconvolution will often be data dependent, these ideas can be applied more broadly to a variety of situations.Extracting useful information from electronically recorded data measurements often requires the signal to be separated from the noise and the blurring impulse response deconvolved out. In order to denoise LADC-GEMM underwater acoustic data, we apply and compare several techniques from Fourier frequency filtering and wavelet denoising. We also look at methods of estimating the acoustic impulse response (system response, transfer function, etc.) for the northern Gulf of Mexico to be used in implementing several deconvolution methods applied to the same dataset. Deconvolution techniques tested include Fourier deconvolution, Wiener filtering, various blind deconvolution methods, as well as combination algorithms. Comparisons of techniques will be presented, showing that the different methods may yield different results. Metrics for analysis will include both the final quality of results along with computational efficiency. Some recommendations will be made within the framework of marine mammal signal detectio...