Finite Record Effects of the Errors-in-Variables Estimator for Linear Dynamic Systems

The frequency domain errors-in-variables (EIV) estimator for linear dynamic systems with periodic excitations is generally formulated as a maximum likelihood (ML) estimator, where the estimated input-output noise (co)variances are used. In this paper, we study the finite record influences on the uncertainty of the ML estimator. Furthermore, we discuss when a significant influence can occur and derive a practical method to circumvent this problem without the need of introducing parametric noise models.