Matched field source localization with Gaussian processes.

Area of Science:

• Ocean acoustics
• Signal processing
• Machine learning

Background:

• Acoustic field data is often sparsely observed.
• Accurate source localization in the ocean is crucial for various applications.
• Conventional methods face challenges with sparse data and noise.

Purpose of the Study:

• To apply Gaussian processes for predicting densely sampled acoustic fields from sparse observations.
• To improve ocean source localization using Gaussian processes combined with matched-field processing.
• To evaluate the performance of Gaussian process-based localization against conventional methods.

Main Methods:

• Utilized Gaussian processes to predict a dense acoustic field from sparse measurements.
• Integrated Gaussian process predictions with matched-field processing for source localization.
• Analyzed the impact of kernel choice and hyperparameters on prediction quality.
• Compared the proposed method with conventional processing under varying noise levels.

Main Results:

• Gaussian processes successfully predicted dense acoustic fields from sparse data.
• The denser predicted field reduced ambiguity function sidelobes in source localization.
• The Gaussian process-based processor demonstrated a higher probability of correct localization than conventional processing, particularly at increased noise levels.
• The method showed benefits from both denoising and denser field prediction.

Conclusions:

• Gaussian processes offer a robust approach for enhancing acoustic field prediction in sparse observation scenarios.
• Combining Gaussian processes with matched-field processing significantly improves ocean source localization accuracy and reliability.
• The proposed method provides superior performance over conventional techniques, especially in challenging, noisy underwater environments.