Adaptive IIR identification of stochastic systems with noisy input-output data

This paper is concerned with adaptive IrR filtering for linear systems with noisy input and output measurements. A new and numerically efficient procedure for estimating the variances of the white input and output noises is established so that the adaptive I1R filter based on the bias-eliminated least-squares algorithm can be efficiently implemented. This new adaptive IrR filter can achieve a substantial reduction in the computational effort, and meantime it can retain almost the same parameter estimation accuracy. Numerical results that illustrate the attractive properties of the new adaptive IrR filter are presented.