Signal Processing by Model Neural Networks

摘要

Voltage controlled oscillator model neurons (VCONs) are electronic circuits that are similar to phase locked loops, but designed to account for certain experimental observations of neurons. They are constructible electronic circuits, and they provide teaching tools that involve (relatively) simple mathematical models based on brilliant circuits designed by engineers. The model makes accessible the study of phase locking, an important physical phenomenon that makes possible stable frequency-encoded information processing even in the presence of noise. VCONs also enable the design of networks of circuits that might be useful as analog control devices in robotics, give interesting examples of rotation vectors in high-order dynamical systems, and can process, store, and recognize frequency-encoded information. Presented here are several VCON networks motivated by observations by physiologists. They fire bursting patterns similar to neural circuits in the thalamus and reticular complex of mammalian brains; they reproduce searchlight behavior that is speculated to be a mechanism by which a brain focuses attention on one among many competing stimuli; they convert a temporal signal into a spatial pattern of phase locked firing, similar to a tonotopic mapping in mammalian auditory systems; they store frequency-encoded information in autocorrelating filters that are similar to neurotransmitter synapses at chemical equilibrium; and they recognize stored signals by cross-correlation with new inputs. These networks and their computer simulations are presented here.