"Design of Experiments" for the Identification of Linear Dynamic Systems

One of the most important problems in direct digital controller design is to determine the process model from the sampled data. Goodness of the control depends considerably on the accuracy of the identified model. This paper investigates how to generate optimal input signal series for the identification of linear discrete-time systems in order to improve the accuracy of estimates. The determinant of the covariance matrix or of the inverse of the information matrix is considered as a measure of the error in the parameter estimates. Very simple methods are presented for the minimization of these criteria in case of an amplitude constrained input signal.