Deep unsupervised adversarial domain adaptation for underwater source range estimation.

Area of Science:

• Acoustics
• Signal Processing
• Machine Learning

Background:

• Accurate underwater source localization is crucial for various applications.
• Traditional methods often require extensive labeled data, limiting practical use.
• Environmental variations and data corruption pose significant challenges for source localization.

Purpose of the Study:

• To propose a novel underwater source range estimation method using unsupervised domain adaptation (UDA).
• To develop a practical approach that bypasses the need for labeled measured data.
• To enhance the robustness of source localization against environmental mismatches and data limitations.

Main Methods:

• Training a deep neural network classifier with labeled simulated acoustic data.
• Employing adversarial learning within UDA to align distributions of simulated and measured data in latent space.
• Adjusting an adaptation module using unlabeled measured data to mitigate distribution divergence.

Main Results:

• The UDA method successfully projects measured and simulated data into a common space, reducing distribution discrepancies.
• Range estimation using UDA demonstrated superior performance compared to match-field-processing.
• The method proved robust across scenarios with limited data (few snapshots, few array elements), low signal-to-noise ratio, and environmental parameter mismatch.

Conclusions:

• Unsupervised domain adaptation offers a practical and robust solution for underwater source range estimation.
• The proposed UDA method effectively handles environmental parameter mismatch and data scarcity.
• This approach significantly advances the capabilities of underwater acoustic localization systems.