Source separation and direction-of-arrival estimation for signals overlapping in time and frequency with a single acoustic vector sensor.

Abstract

A single acoustic vector sensor directly measures acoustic field directionality, but separating arrivals that overlap in time and frequency remains challenging. Beamforming, covariance-based direction finding, and statistical source-separation methods can be effective when sufficient spatial, temporal, spectral, or statistical diversity is available. However, coherent or transient components occupying the same time-frequency bins present a more difficult separation problem. Here, a physics-based method is introduced showing that a single vector sensor can recover the arrival angles, amplitudes, and relative phase of two overlapping plane waves at the same frequency. The approach exploits reactive intensity generated by interference between the waves, which is oriented perpendicular to the bisector of their wavevectors. This geometry defines a natural coordinate system leading to closed-form analytic solutions for source separation and direction-of-arrival estimation. Simulations illustrate separation of correlated signals, such as coherent multipath, while sensitivity analyses examine robustness to sensor miscalibration and additive noise. Field data from the Beaufort Sea demonstrate extraction of weak transient airgun and biological signals masked by intense broadband vessel noise. These results suggest that the method can be effective even in circumstances where acoustic arrivals are more accurately characterized as normal modes instead of plane waves.