Performance evaluations and robustness issues of discrete time linear quadratic Gaussian control design

It is demonstrated that the linear-quadratic-Gaussian (LQG) design methodology does not provide global system-independent guaranteed robustness properties as that of the linear-quadratic (LQ) design. Therefore, the design problem must be tackled deliberately. It is shown that, under a sufficiently small sampling interval, the performance measured by state covariance can be improved to some extent if the estimator gain has some specific form. It is also shown that, in the case of filtering version estimator-based systems, the robustness properties of discrete-time LQ design can be recovered perfectly.<>