Structured finite-dimensional controller design by convex optimization (Q1611927)

The problem of centralized design with single-objective specifications for linear time-invariant finite-dimensional systems has admitted an elegant and computationally efficient semidefinite programming (SDP) solution since the mid nineties, which is far from being the case for decentralized (or structured) multi-objective design. The aim of the paper is thus to describe a sufficiently general class of problems for which decentralized multi-objective design also boils down to SDP. Under a one-block assumption (of the kind arising in \(H_{\infty}\) control) on one subsystem in the feedback interconnection, it is shown that the Youla-Kučera parameter (of all stabilizing controllers) has a structurally decomposing effect (in addition to the traditional linearizing effect) on closed-loop system matrices. As a result, bounded-degree controller design boils down to solving algebraic matrix equations and SDP problems, and explicit formulas are obtained for the MacMillan degree of the controller. The approach is then specialized to multi-objective \(H_{\infty}\) control as well as to robust design with exact disturbance decoupling.