This paper deals with the nonparametric identification of linear dynamic systems within an output-error framework. In the system, the input is an arbitrate signal cover a broad enough frequency band, while the output is disturbed by a filtered white noise with unknown variance. Since the full maximum likelihood method used in the frequency domain causes calculation complexity, this paper develops a nonparametric method to cope with the complexity. According to the property that the frequency response function and the system leakage term can be locally approximated very well via a low-order degree polynomial, a frequency domain estimator is developed, which can obtain the estimates for the frequency response function and the output noise variance. Finally, the parameters identification results for one real model can validate the effectiveness of the new proposed nonparametric method.
This paper deals with the nonparametric identification of linear dynamic systems within an output-error framework. In the system, the input is an arbitrate signal cover a broad enough frequency band, while the output is disturbed by a filtered white noise with unknown variance. Since the full maximum likelihood method used in the frequency domain causes calculation complexity, this paper develops a nonparametric method to cope with the complexity. According to the property that the frequency response function and the system leakage term can be locally approximated very well via a low-order degree polynomial, a frequency domain estimator is developed, which can obtain the estimates for the frequency response function and the output noise variance. Finally, the parameters identification results for one real model can validate the effectiveness of the new proposed nonparametric method.