Iteratively reweighted correlation analysis method for robust parameter identification of multiple-input multiple-output discrete-time systems

In the engineering practices, the distributions of measurements are non-Gaussian as they contain outliers. As some slight deviations from the Gaussian assumption would probably cause the performance of an estimator to degrade significantly, a novel iteratively reweighted correlation analysis method is proposed for robust parameter estimation of multiple-input multiple-output (MIMO) systems, in the presence of Student's t-noises. The iterative method achieves good robustness and high efficiency by the combination of multivariable correlation analysis and t-distribution based M-estimators. The appropriate updating weights are able to enter into the sample cross-correlation function, so that the heavy tails are lowered, and the impact of outliers is weakened to the greatest extent. Based on the robust finite impulse response models, the identification procedure is then to reconstruct the noise-free estimates to identify the parameters of an MIMO system. The theoretical discussions and simulation results demonstrate that the proposed method works well.