Blind localization and counting of multiple acoustic sources in a randomly distributed sensor system

We consider a sensor system to localize multiple acoustic sources through the estimation of time difference of arrival. For locally wide sense stationary (WSS) acoustic sources, space-time correlation matrix, although not in Toeplitz (or even block Toeplitz) form, has eigenvalues which asymptotically contain the power spectral density (PSD) of the sources. By exploiting the whole eigen system including all the eigenvalues and eigenvectors, we show that multiple acoustic sources can be separated through high-resolution bandpass eigen filtering vectors. Using cross correlation among each eigenvector's piece with respect to each sensor, the relative time delay of arrivals (TDOA) can be obtained. Simulation results are presented to illustrate the performance of the proposed scheme.