Blind broadband source localization and separation in miniature sensor arrays

Abstract

We study the ability of a sensor array to blindly separate and localize broadband traveling waves impinging on the array, with additive sensor noise. We consider arrays smaller than the shortest wavelength in the sources, such as MEMS acoustic arrays or VLSI arrays of RF receivers. A series expansion about the center of the array of the time-delayed signals emanating from the sources reduces the problem of separating and localizing the delayed sources to that of separating instantaneous signal mixtures using conventional tools of Independent Component Analysis. The covariance of the noise in the estimated sources is expressed in terms of the covariance of the sensor noise and the angular direction of the sources. Physical simulations demonstrate separation and localization of three non-coplanar speech sources using a planar array of four sensors within a 1 mm radius.