Nonstationary smoothing and prediction using network theory concepts

Nonstationary signal and noise are assumed, such that ensembles with the same pertinent statistics may be generated by passing white noise through "physical," time-variable, linear networks composed of a finite number of components. The problem is to find the linear, least-squares smoothing and prediction operators. The point of departure is Bode and Shannon\´s solution for stationary statistics. Analogous results are obtained for nonstationary systems, by examining analogous operations in frequency domain, time domain, and differential equation terms.