A new method is presented for identifying the controllable and observable part of a multiinput multi-output system from measurement of the input-output signals contaminated with noises. The pulse transfer function matrix is estimated using a vector input-output equation of an autoregressive-moving average type. The minimal realisation of the system is given by the Ho-Kalman's algorithm.
The Markov parameters are used to determine the degree of the minimal polynomial of the system matrix and the minimal order of the system.