Self-Noise Suppression for AUV without Clean Data: a Noise2Noise Approach

The passive detection performance of an autonomous underwater vehicle (AUV) is affected by various types of noise. One of the main components of the noise is the AUV’s self-noise. Inspired by the statistical reasoning shown in Noise2Noise, we propose a deep learning model for AUV’s selfnoise suppression. It can learn the spectrum features of the noise from single-channel data in a self-supervised learning setting and reconstruct clean signals. The tests on South China Sea experimental dataset show that the model is able to intelligently suppress AUV’s self-noise for single hydrophone signals.