In this paper an approximation method for discrete-time linear systems with the internal structure is proposed. For this, the method of Mullis-Roberts is used. A reduced order model constructed through this method retains the same internal structure as the original one. It is also an optimal approximation in the sense that it minimizes the curvilinear integral of the squared norm of an approximation error along the unit circle of z-plane.
It is shown that the Levinson-Wiggins-Robinson algorithm can be used as a fast recursive algorithm computing the reduced order model.
A feature of this method is that we can compute reduced order ARMA-type models of each subsystem independently. Thus, subsystems to be reduced are arbitary chosen.This approximation method can also be used for identifying each subsystem composing the total system.