An Analog VLSI Model of Periodicity Extraction in the Human Auditory System

This paper presents an electronic system that extracts the periodicity of a sound. It uses three analog VLSI building blocks: a silicon cochlea, two Inner Hair Cell circuits and two spiking neuron chips. The silicon cochlea consists of a cascade of filters which delays (and filters) the input sound as it passes along the cascade. The time delay added by each individual filter in the cochlea increases exponentially with position of the filter along the cochlea. The frequency for which the time delay between two outputs that are n-stages apart corresponds to a phase delay of 2π therefore decreases exponentially along the cochlea. In the system presented in this paper we create spike trains from the output of the cochlear filters and we compare the output of each filter with the output of a filter four sections earlier in the cascade. If the signal frequency corresponds to the inverse of the time delay between these two filters, then the two spikes in the spike trains created at these two outputs will coincide. Detecting these coincidences can yield very selective filters, i.e., filters that respond only to a very narrow range of periodicities, but that at the same time still respond after a few periods of the input signal. This is an advantage over traditional band-pass filters, where an increase in selectivity has to be traded off against a decrease in response time.