ABSTRACT
In recent years, a number of subspace system identification methods have been developed that involve primarily a singular value decomposition computation (see particularly Larimore [27,28], Verhaegen [46], and Van Overschee and De Moor [45]). Such procedures permit the completely automatic and reliable identification of multivariable system. However only the canonical variate analysis (CVA) procedure has been developed on the basis of optimal statistical inference principles, and as a result only it achieves optimal statistical accuracy while the others can be considerably less accurate. In this paper, the major concepts and results involved in CVA are developed, and a number of important applications are discussed.
