Sparse DOA Estimation of Wideband Sound Sources Using Circular Harmonics

摘要

Sparse signal models are in the focus of recent developments in narrowbandDOA estimation. Applying these methods to localizing audio sources, however, ischallenging due to the wideband nature of the signals. The common approach ofprocessing all frequency bands separately and fusing the results is costly andcan introduce errors in the solution. We show how these problems can beovercome by decomposing the wavefield of a circular microphone array and usingcircular harmonic coefficients instead of time-frequency data for sparse DOAestimation. As a result, we present the super-resolution localization methodWASCHL (Wideband Audio Sparse Circular Harmonics Localizer) that is inherentlyfrequency-coherent and highly efficient from a computational point of view.