An information theoretic approach to statistical linearization

The paper presents a procedure to derive a linear input/output model which is a statistical equivalent of a nonlinear dynamic stochastic system driven by a Gaussian white-noise input process. The key issue of such a procedure is using a statistical linearization criterion which is the condition of coincidence of the mutual information of the input and output processes of the system and the mutual information of the input and output processes of the system model. The approach provides obtaining explicit relationships which determine the weight coefficients of the linearized model.