Part of Advances in Neural Information Processing Systems 9 (NIPS 1996)
Fernando Pineda, Gert Cauwenberghs, R. Edwards
We propose a neuromorphic architecture for real-time processing of acoustic transients in analog VLSI. We show how judicious normalization of a time-frequency signal allows an elegant and robust implementation of a correlation algorithm. The algorithm uses binary multiplexing instead of analog-analog multiplication. This removes the need for analog storage and analog-multiplication. Simulations show that the resulting algorithm has the same out-of-sample classification performance (-93% correct) as a baseline template-matching algorithm.