Multiple acoustic source localization using ambiguous phase differences under reverberative conditions

Abstract

“Phase ambiguity” leads to confusion in computational source localization where multiple source locations are introduced from a cross-spectral phase value measured by two sensors at high frequencies, where sound wavelength is shorter than sensor interval. In this paper, a frequency domain algorithm for broadband source localization by two sensors is proposed for solving “phase ambiguity” confusion under actual conditions. Using the overlapped and averaged phase differences of the cross-spectral phases measured over the audible frequency range, multiple source azimuths are identified from each phase difference as much as possible over the azimuth range of ±90°. The frequency-independent azimuths extracted from multiple azimuths by Hough transformation provide the target source azimuths. The azimuths for the two loudspeakers can be identified simultaneously in this way from the phase differences measured over the full audible frequency range within an error of approximately 6° under reverberative conditions. By removing the numerical noise during source azimuth identification, the estimated source distribution corresponds to the diameters of the loudspeakers. When it is necessary to distinguish between near or far sound sources around the microphones, the horizontal azimuths for the sources can be precisely identified from all directions except at approximately ±90° if judgment of the front or back is given.