Performance improvement by innovations feedthrough in the linear-quadratic-Gaussian control problem

For a class of practical applications where the control problem may be approximated by the linear-quadratic-Gaussian formulation, it is proposed that performance of the "separation-theorem" controller (measured in terms of physical energy input and expected transient response) may be improved by feed through of the innovations (residual) process of the Kalman filter. Explicit values for the gains on the innovations signal are derived using the Matrix Minimum Principle.