Nonparametric identification of linear (almost) periodically time-varying systems using cyclic-polyspectra

Abstract

Novel algorithms are presented for nonparametric input/output identification of systems using kth-order cyclic-polyspectra at known cycles. Errors-in-variables models with generally cyclostationary inputs are considered. The proposed methods for k>or=3 are insensitive to contamination of both input and output data by even cyclostationary Gaussian noise of unknown covariance. Additional insensitivity to different types of input disturbances is delineated. Consistent and asymptotically normal sample cyclic-polyspectrum estimators are used for implementation, and simulations illustrate the proposed algorithms.<>