Sound source localization with an integrate-and-fire neural system

This paper briefly describes a biologically inspired model for sound source localization. In the whole neural system an advanced integrate-and-fire neuron model is used for spike coding, time delay evaluation and in a Winner-Take-All (WTA) network. We discuss a current-mode CMOS synapse circuit transforming the binary action potential (AP) into an analog postsynaptic potential (PSP) with long time constant. The integrate-and-fire neuron compares the sum of incoming excitatory and inhibitory PSP currents to a time-varying threshold and generates AP pulses. A chip architecture is shown which performs two modes of processing, either a binaural time delay estimation in a single frequency channel or a WTA function. It can be combined with a modular neural processing system with adjustable parameters.