Recursive 4SID algorithms using gradient type subspace tracking (Q1614337)

The paper is devoted to system identification algorithms in the case where the order of a MIMO system to be identified is a priori known. The main novelty introduced by the authors is that, although they apply the same minimization problem for the recursive subspace state-space system identification (SSSS-ID, further abbreviated as 4S-ID), the search for the global minimizer is performed with the gradient type subspace tracking method used in array signal processing. The proposed 4S-ID algorithms can directly estimate the subspace spanned by the column vectors of the extended observability matrix of the system to be identified without performing the singular value decomposition. Furthermore, the authors obtain a proof of the local convergence property for the gradient type subspace tracking technique under the assumption that the step size belongs to the interval \((0,1)\). A numerical example illustrates the advantage of the proposed 4S-ID algorithm, showing the 4S-ID algorithm to be more robust with respect to the choice of the initial values than the corresponding PAST (projection approximation subspace tracking) algorithm of T. Gustafsson (1997, 1998).