Deep transfer learning-based variable Doppler underwater acoustic communications.

Area of Science:

• Underwater Acoustic Communication
• Signal Processing
• Machine Learning

Background:

• Underwater acoustic communication faces challenges due to Doppler effects and multipath propagation.
• Conventional systems often require complex Doppler estimation, limiting performance in dynamic environments.

Purpose of the Study:

• To propose a deep transfer learning (DTL)-based system for variable Doppler frequency-hopping binary frequency-shift keying underwater acoustic communication.
• To develop a receiver employing a convolutional neural network (CNN) for direct signal demodulation, bypassing explicit Doppler estimation.

Main Methods:

• Utilized DTL by pre-training a CNN on simulated data and fine-tuning it on specific communication scenario data.
• Employed Mel-spectrograms for the CNN to learn frequency-to-symbol mapping within frequency-hopping groups.
• Implemented the CNN as the receiver's demodulation module.

Main Results:

• The proposed DTL-CNN system achieved superior performance compared to conventional methods.
• Performance gains were particularly significant under variable speed motion conditions in shallow water acoustic channels.
• Direct demodulation without Doppler estimation proved effective.

Conclusions:

• The DTL-based CNN approach offers a robust and efficient solution for underwater acoustic communication systems.
• This method enhances communication reliability in dynamic underwater environments with variable Doppler shifts.