Input-output identification of stochastic systems

A method is proposed for the identification from input-output records of the Kalman filter representation of a multiple input - multiple output linear discrete stochastic system. The method is based on a particular canonic form for the state space representation of the system that has the advantage of minimality of the number of parameters to be identified, and is a stochastic version of a recently proposed algorithm for the minimal realization problem for deterministic systems.