Linear-Quadratic-Gaussian Controllers with Specified Parameter Robustness

Despite the attractions of linear-quadratic-Gaussian (LQG) controllers, they have had limited acceptance in practice due to lack of knowledge about their robustness to uncertainties in the plant model. A suitable measure of robustness, like the gain margin or the phase margin of classical controllers, was lacking. A new robustness measure, based on the expected parameter deviations from their nominal values, is proposed here and it is used to design robust LQG controllers using nonlinear programming software.