Machine Learning Solutions for Improved Channel Estimation in Underwater Acoustic Communication

Scattering and multipath propagation are the most significant distortion encountered in underwater acoustic communications. They make the channel estimation a challenging problem to solve. The multipath propagation implies that signal arrives at the receiver in the form of multiple time-varying scattering by the moving sea surface and diffusion scattering by the ocean floor. This results in frequency dependent fading leading to a channel response which is challenging to estimate. In this paper we propose to use wavelet analysis and support vector machines to achieve this goal of channel estimation. Wavelet transforms are particularly suitable for the task because they are an effective tool for signal representation which is localised both in time and frequency. The localised structures in the delay vs time images are effectively represented by Wavelet transforms. The features obtained by the Wavelet Transform are used to isolate regions representing multipath propagation. The experiments demonstrate the efficacy of the proposed approach.