Using LSTM Autoencoder and Dynamic Signal Decomposition for Efficient PAPR Reduction in UWA-OFDM

Underwater Acoustic Communication (UWA) is an emerging wireless technology that enables reliable and long range data transmission in challenging underwater environments. To support high data rates and mitigate Inter Symbol interference (ISI) caused by multipath propagation, Orthogonal Frequency Division Multiplexing (OFDM) is widely adopted in UWA systems. High Peak-to-Average Power Ratio (PAPR) continues to be a significant constraint in OFDM based UWA systems, as it causes nonlinear distortion in power-constrained acoustic transmitters and results in spectrum regrowth. This study presents a data driven strategy for PAPR reduction using a Long Short Term Memory-Autoencoder (LSTM-AE) as well as noise reduction from time-domain OFDM signals using the Local Mean Decomposition technique (LMD). The LSTM-AE is designed to learn the basic components of the OFDM waveform and rebuild a low PAPR variant without sacrificing signal integrity. Simulation results indicate that the suggested strategy significantly decreases PAPR while maintaining Bit Error Rate (BER) performance across different Signal-to-Noise Ratio (SNR) levels. Moreover, Complementary Cumulative Distribution Function (CCDF) analysis demonstrates substantial improvement relative to conventional PAPR reduction methods, including Tone Reservation (TR), Partial Transmit Sequence (PTS), Active Constellation Extension (ACE) and Fully Connected Neural Network (FCNN). Compared to the original OFDM signal, the LSTM-AE approach reduced PAPR by 9.4 dB, reaching 4.8 dB. It also maintain better energy efficiency than ACE, FCNN, TR, and PTS at all SNR levels. The suggested LSTM-AE method is a feasible low complexity alternative for improving the efficiency and resilience of UWA OFDM systems.