Deep Learning-Aided OFDM Demodulation Scheme for Undersea RF Short-Range Communication

Abstract

The realization of undersea wireless communication using radio frequency (RF) signals is hindered by severe attenuation and signal distortion, particularly due to complex propagation mechanisms and limitations of RF hardware. This paper proposes a deep learning-based orthogonal frequency division multiplexing (OFDM) demodulation scheme that employs a long short-term memory (LSTM) network, thereby eliminating the need for conventional channel estimation and equalization. The proposed method is evaluated under two numerically modeled undersea RF channel scenarios: (i) direct-path propagation and (ii) combined direct and lateral wave propagation. While achieving performance comparable to conventional least squares (LS) demodulation for BPSK, QPSK, and 8-PSK in the direct-path case, the LSTM-based approach significantly outperforms the LS method under combined direct and lateral wave conditions, yielding a 2â 6 dB improvement in bit error rate (BER) across various modulation schemes. Notably, the gain is more pronounced with higher-order modulations such as 8-PSK, demonstrating the potential of deep learning to enhance the robustness of undersea RF communication in challenging environments.