Observer-based controllers (e.g., linear quadratic Gaussian [LQG] controllers) are quite interesting for different practical reasons and from the implementation point of view. Probably the key advantage of these controller structures lies in the fact that the controller states are meaningful variables as estimates of the physical plant states. It follows that the controller states can be used to monitor (online or ofine) the performance of the system. Such a meaningful state allows us also to initialize the state of the controller or to update controller state during control mode switching. Note that this simple property does not hold for general controllers with state-space description:

x A x B y

u C x D y

= + = +

⎧ ⎨⎪ ⎩⎪ .