Reinforcement Learning-Based Underwater Acoustic Channel Tracking for Correlated Time-Varying Channels

Due to the correlation nature of underwater acoustic (UWA) channels, channels can be decomposed into the reduced-rank principal components and the corresponding channel subspace. A suitable rank in the subspace channel tracking is crucial for the performance. It relies on the correlation of the UWA channels, and the correlation of the UWA channels is usually time-varying. However, in conventional subspace channel tracking, it is set as a fixed parameter. In this work, we propose a reinforcement learning method to obtain a time-varying rank, thereby to improve the subspace channel tracking for the time-varying UWA channels. We provide three improved channel tracking approaches through the reinforcement learning. Experimental results show that improved channel tracking approaches with RL have the around 1.5-2.2dB lower prediction error both in calm sea and rough sea with the at-sea data.