Finite Global $H_{\infty}$ Optimization of the $Q$-Filter Time Constant for Nominal-Response Recovery in SISO Disturbance Observers under Parametric Plant Uncertainty

Abstract

This paper develops a finite real-algebraic procedure for global H-infinity optimization of the time constant of a fixed-shape Q-filter in a single-input/single-output (SISO) disturbance observer (DOB) under affine parametric plant uncertainty. Robust internal stability defines a possibly disconnected feasible domain, and the objective is the worst-case weighted mismatch between the uncertain DOB reference response and the nominal closed-loop response. A recovery-error factorization preserves affine dependence of the performance numerator and denominator on the same uncertainty vector. Phase lifting and a complex-polynomial Edge theorem reduce the uncertainty box to finitely many polynomial edges, while quadratic dependence of the squared frequency-domain inequality on the edge coordinate permits analytic elimination of that coordinate. The resulting semialgebraic formulation has only the squared frequency as a universally quantified variable. It determines the global infimum, decides whether it is attained, and constructs the complete optimizer set over the full robust-stability domain. A DOB case study with a plant family containing non-minimum-phase members yields a unique optimal time constant of approximately 0.0364075 and a worst-case norm of approximately 2.29975. The optimizer lies strictly inside the first robust-stability component, and every smaller feasible time constant in that component has a larger worst-case norm. Finite algebraic certificates establish robust stability, performance feasibility, and strict global optimality.