A PACED ANALOG SILICON MODEL OF AUDITORY ATTENTION

Abstract

This paper describes a 1MHz CMOS implementation of an integrate and fire neuron network. The system models the human ability to separate unknown sounds under natural conditions. Sound sources are separated based on binaural time delay of auditory nerve spiking pattern. The acoustical attention is guided by the novelty of a sound, acquired knowledge and interaction with other sensoric systems. Signals are coded into current pulses modelling the response of the Inner Hair Cell-Ganglion complex. Uniform paced analog models of pulse propagating cells and synapses are used at different processing levels. A module based layout generator for the realization of different network size and connectivity is under construction. The system is designed to provide a silicon solution for the interaction of acoustical with other sensoric information at the autonomous robot platform MILVA. 1 Introduction