Training a U-Net based on a random mode-coupling matrix model to recover acoustic interference striations

Area of Science:

• Ocean Acoustics
• Signal Processing
• Machine Learning

Background:

• Acoustic interference striations (AISs) are crucial for underwater acoustic detection.
• Distorted AISs pose challenges for accurate signal interpretation.
• Existing methods for AIS recovery are often limited in complex environments.

Purpose of the Study:

• To develop and evaluate a U-Net based model for recovering distorted acoustic interference striations (AISs).
• To assess the model's performance in realistic range-dependent waveguides influenced by nonlinear internal waves (NLIWs).

Main Methods:

• A U-Net architecture was trained using synthetically generated data.
• A random mode-coupling matrix model was employed for rapid generation of diverse training datasets.
• The trained U-Net was tested in simulated range-dependent waveguides with varying nonlinear internal wave (NLIW) characteristics.

Main Results:

• The U-Net model demonstrated successful recovery of acoustic interference striations (AISs) even from distorted signals.
• Performance remained effective across various signal-to-noise ratios.
• The model showed robustness against different amplitudes, widths, and shapes of nonlinear internal waves (NLIWs).

Conclusions:

• The U-Net, trained with data from a simplified model, exhibits significant capability in recovering acoustic interference striations (AISs).
• This approach offers a promising method for enhancing underwater acoustic signal processing in the presence of complex environmental factors like nonlinear internal waves (NLIWs).