H-infinity parameter estimation for state-space models

This paper investigates the nonlinear problem of parameter identification for systems with time-varying state-space models using the idea of an extended filter. To treat uncertainty in the data or the model, the paper follows the approach of H-infinity filtering. It shows that the parameter estimation error satisfies exactly albeit locally the prescribed gain criterion as long as the extended filter meets a modified-tighter-criterion. In the infinite horizon case, the paper also shows that the modified H-infinity criterion leads to an asymptotically stable state estimation error without assuming zero-state observability of the associated error equation.