Speaker
Description
Many acoustic direction-of-arrival (DoA) algorithms rely on digital beamforming with omnidirectional receivers, which demands the highest computational load per method when applied to a full-sphere focus grid. Conventional methods often neglect the spatial filter characteristics inherent in passive acoustic gain patterns, which are standard in studio microphones. This work introduces a phase-less DoA estimation method for a full-sphere focus grid, leveraging the comb-like filter characteristics of non-omnidirectional microphones when excited by a noise source. Through cepstral decomposition, informed by domain knowledge of the spatial filter, we derive an analytical relationship between the rotation-ambiguous angle of arrival and the filter parameters observed in the spectrum. For an array of orthogonally positioned non-omnidirectional microphones, superposing relative direction estimators yields one or more sound source directions in 3D space. We first simulate scenarios for far-field noise sources and then experimentally validate them under free-space conditions using real noise sources. Classification metrics and spatial error for source detection are presented, with comparisons across different array geometries. As the proposed method employs cepstral analysis per channel rather than per discrete focus point, it reduces computational load, making it suitable for autonomous low-power sensor networks in noise monitoring applications.