Noise-enhanced Information Transmission In A Non-linear Neural Network

In communications and other fields, it is always a traditional way to improve signalprocessing and enhance information transmission by removing or diminishing noises in primitivesignal. However, this results in information loss inevitably. Stochastic resonance (SR) is differentfrom traditional methods, it always adds noises to signal rather than eliminate noises to make thesignal and the noises achieve certain synergy in nonlinear systems, which can improve signalprocessing and enhance information transmission.This paper studies mainly the applications of noises in multiple signal detection andinformation transmission. The main work of this paper is as follows:1. The paper discusses in detail stochastic resonance and supra-threshold stochasticresonance (SSR) of noise-improved nonlinear multiple signal detection under the background ofGaussian mixture noise and gets some new results. The occurrence of SR or SSR is a gradualprocess, which isn’t affected completely by whether the signal is supra-threshold or sub-threshold;SSR can exist when the input signal is supra-threshold.2. Based on Shannon mutual information, this thesis studies SR and SSR of noise-enhancedinformation transmission by numerical computation and computer simulation in a non-linearneural network in detail, which is affected by both the Gaussian additive and multiplicativenoises. Some conclusions are achieved, especially, SSR occurs more easily by changing additivenoise than by changing multiplicative noise, in some degree, which may explain why people tendto only consider the impacts of additive noises and ignore multiplicative noises wheninvestigating SR. Then this part shows a simple application of SR in image processing, whichmakes SR more concrete.3. Consideration for extending the applications of SR, the paper studies in detail SSR ofadditive and multiplicative noises enhanced information transmission in a revised non-linearmultilevel threshold neuron networks. The results demonstrate the effects of noises leading to anenhancement of information transmission are closely connected with additive and multiplicativenoise intensity, the number of threshold units and the threshold level of system.