Underwater Acoustic Channel Prediction in delay-Doppler Domain based on Reinforcement Learning

The underwater acoustic (UWA) channel in the delay-Doppler domain is more sparse and slower time-varying than in the time-delay domain. Hence, the channel dynamics can be tracked more efficiently in the delay-Doppler spreading function (DDSF) representation. In this paper, we propose a data detection scheme with sparse channel prediction in the DDSF representation. Moreover, the sparse channel prediction is further improved by the reinforcement learning in the DDSF representation. Experimental result shows that proposed schemes can obtain a 3.6-7.5 dB bit error rate (BER) performance improvement compared with the traditional channel prediction.