Single Minimum Nonlinearity Wiener System Identification by Weighted Principal Component Analysis

Wiener system identification with a finite impulse response (FIR) model is investigated in this paper, focusing on the challenging case of non Gaussian input distribution and non monotonic nonlinearity. The proposed method assumes that the static nonlinear function of the Wiener system has a single minimum (or maximum), but does not assume any parametrization of the nonlinear function. Based on a modified principal component analysis (PCA), referred to as weighted PCA, the FIR coefficients of the Wiener system are estimated without estimating the nonlinear function of the Wiener system. The numerical computation cost is essentially equivalent to those of two standard PCA. Numerical examples in harsh practical conditions, with data generated by Wiener systems involving a discontinuous nonlinear function or an infinite impulse response, are presented to illustrate the robustness and effectiveness of the proposed method.