Micropower adaptive VLSI systems for acoustic source localization and separation

Blind separation of real-world acoustic sources is generally considered a hard and unsolved problem, with mixed degree of success in practical realizations. Closely linked to acoustic source separation is the problem of source localization, or bearing angle estimation. Wave propagation of sound complicates the task of separating multiple coexisting sources using independent component analysis (ICA) [1], which conventionally assumes instantaneous mixture observations. In contrast, convolutive ICA techniques explicitly assume convolutive or delayed mixtures in the source observations. Convolutive ICA techniques [2-4] are usually much more involved and require a large number of parameters and long adaptation time horizons for proper convergence. Inspiration from biology suggests that for very small aperture (spacing between acoustic sensors, i.e., tympanal membranes), small differences (gradients) in sound pressure level are more effective in resolving source direction than actual (microsecond scale) time differences. The remarkable auditory localization capability of certain insects at a small (1%) fraction of the wavelength of the source owes itself to highly sensitive differential processing of sound pressure through intertympanal mechanical coupling [5] or interaural coupled neural circuits [6].