Neural OFDM: An ANN-Based Adaptive Encoder-Decoder System for Robust Transmission over Watermark Underwater Acoustic Channel

Abstract

Problem: Underwater wireless communication (UWC) systems face severe performance degradation due to multipath fading, Doppler dispersion, and high transmission losses inherent in acoustic channels. These challenges significantly affect signal reliability, limit data rates, and complicate accurate channel estimation, making conventional modulation techniques such as FFT-OFDM and DWT-OFDM less effective in dynamic underwater environments.

Aim: This study aims to develop a robust and adaptive communication framework that enhances transmission reliability and eliminates the dependency on explicit channel estimation in UWC systems.

Main Contribution: The paper proposes a novel hybrid Orthogonal Frequency Division Multiplexing (OFDM) architecture that integrates a neural network-based encoder–decoder framework with Discrete Tree Complex Wavelet Transform (DTCWT) modulation. Additionally, the system leverages a realistic WATERMARK channel model under NOF1 underwater conditions to enable end-to-end learning of signal representations.

Method: The proposed model employs a deep learning-based encoder–decoder structure to learn optimal signal mappings directly from transmitted to received signals without requiring prior channel state information. DTCWT modulation is incorporated to provide improved time-frequency localization and robustness against channel impairments. The system is trained and evaluated using simulated underwater acoustic conditions modeled by the WATERMARK framework.

Result: Simulation results demonstrate that the proposed DTCWT-OFDM with neural encoder–decoder significantly outperforms conventional FFT-OFDM and DWT-OFDM systems. Notably, the model achieves a reduction in Bit Error Rate (BER) by up to 35–45%, improves Peak Signal-to-Noise Ratio (PSNR) by approximately 3–5 dB, and enhances Structural Similarity Index (SSIM) by 8–12% under varying Doppler and noise conditions.

Conclusion: The integration of deep learning with advanced wavelet-based modulation provides a powerful solution for overcoming key challenges in UWC systems. The proposed approach demonstrates superior robustness, reliability, and signal quality, making it a promising candidate for next-generation secure and high-fidelity underwater communication applications.