Predicting transmission loss in underwater acoustics using convolutional recurrent autoencoder network

Area of Science:

• Ocean acoustics
• Deep learning
• Signal processing

Background:

• Underwater acoustic propagation is complex, influenced by variable ocean parameters and scales.
• Generalized physical models struggle to predict transmission loss across diverse scenarios.
• Accurate underwater noise prediction is crucial for marine operations.

Purpose of the Study:

• To develop a data-driven deep learning model for predicting far-field acoustic propagation in the ocean.
• To utilize a convolutional recurrent autoencoder network (CRAN) for learning transmission loss distributions.
• To demonstrate the CRAN model's capability in capturing essential acoustic phenomena.

Main Methods:

• Proposed a convolutional recurrent autoencoder network (CRAN) architecture.
• Trained the CRAN model on data to learn reduced-dimensional representations of physical data.
• Applied the model to a 2D ocean domain with depth-dependent sources to predict transmission loss.

Main Results:

• The CRAN model successfully learned far-field acoustic transmission loss distributions.
• Demonstrated the model's ability to capture physical effects like geometric spreading, refraction, and reflection.
• Validated the CRAN's effectiveness in a complex ocean acoustics environment.

Conclusions:

• The CRAN model provides a powerful, data-driven approach for underwater acoustic transmission loss prediction.
• This deep learning model can learn complex ocean acoustics phenomena without explicit physical modeling.
• Potential applications include real-time underwater noise prediction for marine vessel decision-making and control.