SUBSPACE ALGORITHMS FOR THE IDENTIFICATION OF MULTIVARIABLE DYNAMIC ERRORS-IN-VARIABLES MODELS

We consider the problem of identifying multivariable finite dimensiona l linear time-invariant systems from noisy input/output measurements. Apart from the fact that both the measured input and output are corrup ted by additive white noise, the output may also be contaminated by a term which is caused by a white input process noise; furthermore, all these noise processes are allowed to be correlated with each other. We shall develop a solution to this problem in the framework of subspace identification and we shall show that our algorithms give consistent estimates when the system is operating in open- or closed-loop. Two re alistic simulation studies are presented to demonstrate the practical applicability of the proposed algorithms. (C) 1997 Elsevier Science Lt d.