OPTIMAL ROBUST SYSTEM IDENTIFICATION: BOUNDED STOCHASTIC DISTURBANCES

摘要

In this contribution we show that log cos(πx/(2C)) is the optimally robust norm for predicition error methods with respect to bounded stochastic disturbances. This norm minimizes the maximum asymptotic covariance matrix of the parameter estimates for the family of prediction errors which are white and amplitude bounded by the constant C. This norm is also optimal within a factor two for unknown but bounded disturbances in a deterministic setting.