Designing Simple Nonlinear Filters Using Hysteresis of Single Recurrent Neurons for Acoustic Signal Recognition in Robots

摘要

In this article we exploit the discrete-time dynamics of a single neuron with self-connection to systematically design simple signal filters. Due to hysteresis effects and transient dynamics, this single neuron behaves as an adjustable low-pass filter for specific parameter configurations. Extending this neuro-module by two more recurrent neurons leads to versatile high- and band-pass filters. The approach presented here helps to understand how the dynamical properties of recurrent neural networks can be used for filter design. Furthermore, it gives guidance to a new way of implementing sensory preprocessing for acoustic signal recognition in autonomous robots.