Physics-based source separation using acoustic vector sensors in three dimensions, with comparisons to adaptive beamforming and subspace methods

Previously, we introduced a method leveraging vector sensors—compactdevices that measure both acoustic pressure and particle velocity—for extracting time signatures and bearings from two angularly separated sources overlapping in time and frequency. This physics-based approach demonstrated that a two-dimensional vector sensor could resolve azimuths, amplitudes, and relative phases of two horizontal plane waves using closed-form inversion formulas, even for the case of a single FFT snapshot. In this work, we extend the algorithm to three dimensions and evaluate its performance against conventional and adaptive beamforming approaches, including Minimum Variance Distortionless Response (MVDR) beamforming and the MUltiple SIgnal Classification (MUSIC) algorithm. Through quantitative comparisons, we highlight the strengths and limitations of the physics-based approach in resolving spatially separated sources, particularly in scenarios that are challenging for standard approaches, such as rapidly moving sources. Simulations and experimental data illustrate the refined algorithm's performance relative to conventional beamforming, adaptive beamforming, and subspace methods, as well as diagnostics for identifying calibration issues. [Work sponsored by ONR TFO.]