Stochastic approximation algorithm for the identification of linear multivariable systems

A stochastic approximation algorithm is presented for on-line identification of linear, multivariable, discrete-time systems from noisy data without prior knowledge of the statistics of measurement noise. The algorithm uses a normalized mean-square error criterion which improves the initial convergence of the identification scheme. The results of a simulated example are given which indicate that the proposed algorithm provides good estimates even for large noise-to-signal ratios.