IDENTIFICATION OF DYNAMIC ERRORS-IN-VARIABLES MODELS FROM DISCRETE TIME FREQUENCY DOMAIN POWER SPECTRA

This paper considers a problem of identifying errors-in-variables (EIV) models under the assumption that a discrete frequency power spectrum is given. A new formulation of identifying dynamic EIV models is presented based on the prediction error approach, where the upper bound of the noise spectrum is taken into account by frequency weighting. An identification algorithm for EIV models is given via a subspace identification method and a J-spectral factorization technique. Based on the derived algorithm, uncertainty modeling is briefly discussed. Numerical simulation results are also included.