Diffusion model-based underwater acoustic communication receivera).

Area of Science:

• Underwater Acoustic (UWA) Communications
• Signal Processing
• Machine Learning

Background:

• Underwater acoustic channels are challenging due to signal distortion.
• Existing systems like deep transfer learning (DTL) require retraining for new environments.
• Robust and adaptive UWA communication is crucial for various applications.

Purpose of the Study:

• To propose a new frequency hopping binary frequency shift keying (FH-BFSK) UWA communication system.
• To leverage denoising diffusion probabilistic models (DDPM) for signal reconstruction and convolutional neural networks (CNN) for demodulation.
• To achieve high performance without requiring additional adaptation in dynamic UWA environments.

Main Methods:

• A FH-BFSK UWA communication system architecture is proposed.
• A DDPM is used to reconstruct distorted Mel-spectrograms from UWA channels.
• A CNN is employed for demodulating the reconstructed spectrograms.

Main Results:

• The proposed system demonstrates superior performance compared to conventional methods.
• The system achieves performance comparable to DTL-based systems.
• Simulations and experiments validate the effectiveness of the DDPM-CNN approach.

Conclusions:

• The DDPM-based approach enables direct deployment in dynamic UWA environments without retraining.
• This method offers enhanced flexibility and suitability for complex and variable UWA propagation channels.
• The proposed system represents a significant advancement in robust UWA communication.