Randomly Initiated Cyclostationary Excitations for Dimensionality Reduction in Wiener System Identification

The problem of nonparametric estimation of nonlinear characteristics in the Wiener system is considered. In this task, the traditional kernel algorithm suffers from dimensionality resulting from the memory length of the dynamic block. A special class of input sequences has been proposed that allows us to reduce the dimension and, consequently, improve the rate of convergence of the estimator to the true characteristics. Theoretical analysis of the proposed method is presented.