Non-linear system identification : an overview

Abstract EN

System identification consists in the characterization of a system from an analysis of observed input and output signals.

In essence, the ultimate aim of system identification is prediction such that given a description of the system transfer parameters and the input, the output can be completely specified for any time.

This paper reviews traditional and contemporary methods used for non-linear system identification.

Their motivations and justifications to characterize their properties are discussed for several classes of non-linear systems.

Comparisons to conventional linear approaches are made.

Parametric and non-parametric models, excited by both deterministic and stochastic signals are presented.

Many contributions in the area of non-linear system identification make the simplifying assumption that the input excitation is white and Gaussian.

This document discusses issues related to non-linear transformations of non-Gaussian data, and gives examples of real-life situations where linearity and Gaussianity assumptions are not valid.