Subspace model identification (SMI) methods, a new class of black-box algorithms to setting up a linear state space model directly from input-output data, have drawn much research attention recently.