An artificial chaotic spiking neuron inspired by spiral ganglion cell: paralleled spike encoding, theoretical analysis, and electronic circuit implementation.

Abstract

A novel chaotic spiking neuron is presented and its nonlinear dynamics and encoding functions are analyzed. A set of paralleled N neurons accepts a common analog input and outputs a set of N chaotic spike-trains. Three theorems which guarantee that the neurons can encode the analog input into a summation of the N chaotic spike-trains are derived: (1) a spike histogram of the summed spike-train can mimic waveforms of various inputs, (2) the spike-trains do not synchronize to each other and thus the summed spike-train can have N times higher encoding resolution than each single spike-train, and (3) firing rates of the neurons can be adjusted by internal parameters. The theorems are proven by using nonlinear iterative maps and are confirmed by numerical simulations as well. Electronic circuit implementation methods of the paralleled neurons are also presented and typical paralleled encoding functions are confirmed by both experimental measurements and SPICE simulations.