Convergence properties of an extended least squares (ELS) estimator

By factorizing the A- and B-polynomials in an ARMAX model and filtering the input/output data with the appropriate factors, the parameters of the model are estimated in a decoupled fashion. This may improve the robustness of some estimators significantly, e.g., when applied to very stiff systems. Earlier work on these techniques has established both local and global convergence properties of some LS, IV, and PE variants. An ELS variant is considered herein, and its local and global convergence properties are analyzed.