La simulazione di processi non gaussiani mediante filtri AR non lineari

The present work investigates the possibility to simulate time histories of non Gaussian stochastic processes by means of non linear auto-regressive digital filters. The use of auto-regressive filters in the numerical simulation is based on the request to estimate the most likely value of the future observation of a stochastic process, given p past observations of the same – or of a correlated – process. The present work illustrates a proposal for a methodology to determine the – non-linear – relation among such past observations and the best estimation of the next one. Besides, the statistical characteristics of the difference between the estimation and the actual observation are analysed for increasing order of the digital filter. Eventually, particolarizations and simplifications directly implementable for the analysis and subsequent simulation of experiment results are examined.