Computational complexity reduction in scaled \({\mathcal H}_ \infty\) synthesis (Q1367185)

The authors consider a class of robust control design problems with structured uncertainty which can be formulated as optimization problems with a scaled \(H^\infty\) norm bound constraint on the closed loop transfer matrix. In the case of dynamic output-feedback this problem may be reduced to solution of linear matrix inequalities with nonconvex constraints on the product of scaling matrices and complexity grows exponentially with the number of these constraints. In the paper a method of reducing the number of nonconvex constraints is proposed for special cases . For the so-called rank-one robust control design [see e.g. \textit{M. K. H. Fan} and \textit{A. L. Tits} IEEE Trans. Autom. Control, AC-31, 734-743 (1986; Zbl 0607.93019)] the nonconvex constraints can be removed leading to a convex programming problem. For the problem with higher rank, by reducing the number of nonconvex constraints the computational complexity is improved. The authors treat both discrete- and continuous-time problems given by state-space models or transfer matrices with uncertain coefficients. The numerical examples presented by the authors illustrate the reduction of computational complexity in solving the global optimization problems.