Multifrequency matched-field source localization based on Wasserstein metric for probability measures.

Area of Science:

• Acoustics
• Signal Processing
• Statistical Inference

Background:

• Matched-field processing (MFP) is a generalized beamforming technique for underwater source localization.
• Traditional MFP relies on correlating array data with replica vectors.
• Existing methods face challenges with ambiguity and performance under limited data conditions.

Purpose of the Study:

• To reformulate the Matched-field processing scheme using a statistical metric.
• To develop a novel Wasserstein processor for enhanced underwater source localization.
• To improve robustness and reduce ambiguity in source localization, especially with deficient snapshots.

Main Methods:

• Reformulation of MFP by computing a statistical metric between Gaussian probability measures of cross-spectral density matrices (CSDMs).
• Application of the Wasserstein metric to develop a processor that leverages intrinsic properties of CSDMs.
• Derivation of a multifrequency processor and an approach to derive the averaged Bartlett processor using Wasserstein metric on Dirac measures.

Main Results:

• The Wasserstein processor effectively suppresses ambiguities and distinguishes multiple sources.
• A multifrequency Wasserstein processor improves localization statistics with deficient snapshots.
• Demonstrated effectiveness and robustness through acoustic simulations and the SWellEx-96 experiment data, showing reduced ambiguity.

Conclusions:

• The Wasserstein metric provides an innovative and effective approach for Matched-field processing in underwater acoustics.
• The developed Wasserstein processor offers improved localization accuracy and ambiguity reduction.
• This method presents a new perspective for Matched-field processing, including a novel derivation of the averaged Bartlett processor.