Optimal input design for Hammerstein (FIR) model identification with unknown but bounded errors

Abstract

The problem of optimal input design for Hammerstein system identification is considered when the linear dynamic part of the model is FIR and the measurement errors are unknown but bounded. Under such a condition the identification of the Hammerstein model parameters can be accomplished by passing through the identification of a linearized augmented Hammerstein model from which overbounds to the Hammerstein model parameter uncertainties can be derived. The presented results refer to optimal parameter identification, in a worst error sense, of the linearized augmented Hammerstein model for which optimal input sequences, minimizing the radius of the parameter uncertainty region, are analytically derived.