Spectrum estimation in innovation models by a nuclear norm optimization based algorithm

Abstract

In this paper, identification of multi-input/multioutput (Ml MO) state-space models in the innovation form by a regularized-nuclear norm optimization based subspace algorithm is studied. Parametrization issues arc carefully addressed for MIMO state-space models. The optimization problem formulated in this paper allows one to utilize a variety of norms in the objective function including the nuclear and the quadratic norms without affecting the parametrization results. A numerical example illustrates the results derived in the paper.