Reinforcement Learning-Based Virtual Time Compressed Mirror for Underwater Acoustical Channel Equalization Based on Main Path to Side Path Ratio Criterion

Underwater acoustical (UWA) channels exhibit severe multipath propagation and long delay spreads, resulting in serious intersymbol interference in communications. In this article, we propose a virtual time compressed mirror (VTCM), a sparse adaptive equalizer, and its reinforcement learning (RL)-based version, RLVTCM. VTCM can compress the time-spread channel with a virtual mirror based on sparse estimation, while RLVTCM enhances this approach by exploiting RL to adjust VTCM's parameter-setting according to the UWA environment adaptively. In addition, we introduce a main path to side path ratio (MSR) criterion to evaluate the equalization performance in multipath channels before demodulation. Simulations and experiments demonstrate that both VTCM and RLVTCM significantly improve communication performance. MSR consistently reflects symbol error rate performance.