Self-Noise Cancellation in Passive SONAR using Deep Regression Neural Network

Target signal detection in the presence of tow-ship noise is an important requirement in towed array sonar applications. The self-interference is a severe problem especially in the ocean conditions where multipath propagation is supported. In this paper, we propose a novel deep learning-based algorithm using deep regression neural network (DRNN) for effective self-noise suppression with minimal power loss for target signal. In the training phase, the designed DRNN tries to minimize the root mean square error between the received signal and the desired output which is the target alone signal. In the testing phase, the DRNN is fed with the received signal and the neural network learns to suppress the self-noise caused by the tow ship. Due to the self-learning ability and high feature extraction capability, the network performs extremely well. To the best of our knowledge, deep learning-based algorithm is employed for the first time, for tow-ship noise suppression application, as per literature. The efficacy of the proposed method is evaluated over simulated data and the simulations were done as received by a towed linear array of acoustic sensors in the shallow ocean scenario. Experimental results show that the proposed method can improve the accuracy significantly.