Realization of stable models with subspace methods (Q5961505)

The paper deals with estimating the dynamics of stable linear state-space systems by stable approximants of the models computed from the least-squares approach. Either asymptotic or marginal stability is taken into account for the approximate model. In the first stage, the shift invariance approach is used where the system matrix is recovered from the extended observability or controllability matrix as a least-squares solution and then respectively modified to assure stability. In the second stage, the state sequence approach is discussed, where a state sequence is first estimated and then the system matrices are obtained by solving the respective least-squares problem. The accuracy of the stabilized solutions obtained by several computational algorithms derived for both the shift invariance and the state sequence approach is investigated by computer simulation.