Subspace-based identification: Weighting and pre-filtering of instruments (Q5960314)

A subspace-based state-space system identification (4SID) problem is considered for a discrete-time time-invariant linear stochastic system operating in open loop. The asymptotic estimation error of the extended observability matrix, estimated from input-output measurement data, is examined. Novel weighting matrices and pre-filters in the instrumental variable 4SID framework are derived. Their optimality in the particular sense specified in the paper is discussed. Several numerical examples illustrate the benefits of the proposed weighting and pre-filtering strategies.