ABSTRACT
Blind separation of real-world acoustic sources is generally considered a hard and unsolved problem, with mixed degree of success in practical realizations. This chapter presents an approach, gradient flow, which avoids problem of separating delayed mixtures in the case of arrays of very small aperture, that is, of dimensions significantly smaller than shortest wavelength in the sources. It presents a mixed-signal very-large-scale integration (VLSI) adaptive microsystem implementing gradient flow interfaced with a miniature microphone array that performs real-time acoustic source separation and localization at microwatts of power, for use in intelligent hearing aids and acoustic surveillance. Gradient flow principle is implemented in ultrapower mixed-signal VLSI application specific integrated circuit (ASIC) for 3D localization of a broadband acoustic source. Mismatch across sensor array introduces errors in estimation of spatial gradients, which translates to errors in bearing estimates. To mitigate this effect, common-mode correction of the gain mismatch errors in sensors is applied to the estimated spatial gradients, prior to bearing estimation.
