Recursive Estimation for Nonlinear Wiener Systems by Online Implementation of Bayes' Rule

摘要

A linearization which permits on-line estimation of multiple random variables when their weighted sum is observed through severe nonlinearity is presented. Analysis to guarantee nice properties of the procedure is not available. For modest numbers of random variables the on-line computing requirements are readily met by microcomputers. The power of the Gaussian approximation in estimating a constant random variable is shown. The observed convergence of computed conditional distributions towards the normal shape is a surprise: it might be a manifestation of the central limit theorem.