Sigma-delta modulation neural networks

Abstract

It is shown that sigma-delta ( Sigma - Delta ) modulation can be used to model the information coding process of biological neurons. Signal analysis results demonstrate that Sigma - Delta modulation processes a noise shaping property by which signal and noise are separated into low (baseband) and high frequency bands, respectively. Restoring the signal with high S/N ratio can be accomplished with a lowpass filter. This property is used to demonstrate that Sigma - Delta modulation can outperform stochastic logic in terms of coding accuracy. The results of simulation on a Sigma - Delta modulation Hopfield neural network are presented. They demonstrate that Sigma - Delta modulation can significantly improve the performance of the network on the immunity of falling into false states. The addition of noise can help Sigma - Delta modulation neural networks escape from one locally stable state to another. >