An analysis of noise enhanced information transmission in an array of comparators

Abstract

An array of N comparators subject to the same input signal and independent additive noise, with the outputs from each comparator summed, is a useful noise model for a range of systems including flash analog-to-digital converters, Digital Multibeam Steering sonar arrays and parallel neurons. It has previously been shown that for certain threshold configurations the transmitted information through such an array is maximised for non-zero noise. This behaviour has been termed Suprathreshold Stochastic Resonance (SSR) [1] and in this paper we show that SSR occurs for a number of different signal and noise distributions. Also presented is an analysis of the variance of the quantisation error incurred when all thresholds are set equal to the signal mean, for Gaussian and uniform distributions. It is shown that the minimum error variance is given by a non-zero value of noise.