Robust controllers for uncertain linear multivariable systems (Q581345)

Three distinct problems in the design of stabilizing controllers for imprecisely known linear multivariable systems are considered. Suppose that P(s) is a true plant transfer matrix and that P(s) is nomially modeled by \(P_ 0(s)\). The first problem is concerned with additive and multiplicative perturbations. In the case of additive uncertainty, P(s) is assumed to satisfy \(\| P(j\omega)-P_ 0(j\omega)\| <| r(j\omega)|\) for all \(\omega\), where r(s) is a prespecified stable rational function, while in the case of multiplicative uncertainty, P(s) is assumed to satisfy \(P(s)=(I+L(s))P_ 0(s)\) where \(\| L(j\omega)\| <| r(j\omega)|\) for all \(\omega\). The second problem deals with stable factor perturbations. \(P_ 0(s)\) is factored as \(N_ 0(s)D_ 0^{-1}(s)\) where \(N_ 0(s)\) and \(D_ 0(s)\) are right coprime stable rational matrices. This class of perturbations consists of all \(P(s)=N(s)D^{-1}(s)\) for some stable rational matrices N(s) and D(s) such that \[ \| \left[ \begin{matrix} N(s)-N_ 0(s)\\ D(s)- D_ 0(s)\end{matrix} \right]\| <| r(s)|,\text{ for all s with Re\((s)>0\)} \] where r(s) is a prespecified stable rational function. The third problem concerns with graph metric perturbations. In this case, P(s) is assumed to satisfy \(d(P(s),P_ 0(s))<r\) where d denotes the graph metric distance and r is a positive real number. The feedback controller C is called robust controller if C stabilizes all plants P(s) within the specified class stated above. The problem of finding such a controller is regarded as a certain \(H_{\infty}\)-norm minimization problem. From this point of view, the paper gives the conditions for the existence of a robustly stabilizing controller C for each of the above three types of plants with uncertainty.