Control of Uncertainty-Affected Discrete Time Linear Systems via Convex Programming

In [1], we have demonstrated that robust optimization of linear finite-horizon control in a discrete time linear dynamical system affected by uncertain disturbances becomes computationally tractable if, rather than using natural parameterization of a linear output-based control law, one passes to a specific re-parameterization of the law, representing it as affine control law based on the so called “purified outputs”. With the traditional parameterization, the states and the controls, being affine in the initial state and the disturbances, are highly nonlinear in the parameters of the control law; with the new parameterization, the states and the controls become bi-affine, that is, affine in the initial state and the disturbances, the parameters of the control law being fixed, and affine in the parameters of the control law, the disturbances and the initial state being fixed. As a result, synthesis of a finite-horizon control law satisfying, in a robust fashion, a given system of linear constraints on the finite-horizon state-control trajectory, reduces to solving an explicit convex program and thus becomes computationally tractable. In this follow up paper we extend the above methodology to optimizing infinite-horizon control in a time-invariant linear system and illustrate our methodology on examples involving discrete time H∞and L1-control.