Abstract
To identify a multivariable stochastic linear system with unknown structure, we suggest the use of a particular vector difference equation model. This model requires only one structure index (minimal order) for its unique representation. Thus it is not necessary for consistent parameter estimation of this model to determine the canonical structure indices. A new method proposed for determining the minimal order of this model from input-output observations is based on the behavior of the estimated Markov parameters. By using this method, we can determine the model order with less computational burdens than the conventional methods. Simulation results are presented to illustrate the model order determination method.
