Meta-learning-aided orthogonal frequency division multiplexing for underwater acoustic communications

Area of Science:

• Electrical Engineering
• Computer Science
• Ocean Engineering

Background:

• Underwater acoustic (UWA) communication systems face challenges due to environmental variability, leading to performance degradation.
• Orthogonal frequency division multiplexing (OFDM) is a key technology in UWA systems, but its robustness to channel changes is limited.
• Existing machine learning methods struggle with rapid adaptation to novel UWA environments.

Purpose of the Study:

• To propose a novel meta-learning-based framework for UWA-OFDM systems.
• To address the issue of environmental mismatch in real-world UWA applications.
• To enable rapid adaptation to unknown UWA environments using limited data.

Main Methods:

• A meta-learning strategy is employed, treating different UWA environments as distinct tasks.
• Meta-training is used to learn a robust model from multiple previously encountered UWA tasks.
• The learned model is then fine-tuned for new, unseen UWA environments with a small number of updates.

Main Results:

• The proposed meta-learning UWA-OFDM system demonstrated superior bit error rate (BER) performance compared to traditional UWA-OFDM.
• The method significantly outperformed conventional machine learning frameworks in various UWA scenarios.
• Experimental validation was conducted using the at-sea-measured WATERMARK dataset and a lake trial.

Conclusions:

• Meta-learning provides an effective solution for environmental mismatch in UWA-OFDM systems.
• The proposed approach offers enhanced learning ability and adaptability for UWA communication.
• This framework holds promise for improving the reliability of underwater acoustic communications.