The Local Polynomial Method for nonparametric system identification: Improvements and experimentation

The Local Polynomial Method (LPM) is a recently developed procedure for nonparametric estimation of the Frequency Response Function (FRF) of a linear system. Compared with other nonparametric FRF estimates based on windowing techniques, it has proved to be remarkably efficient in reducing the leakage errors caused by the application of Fourier transform techniques to non periodic data. In this paper we propose a modification of the LPM that takes account explicitly of constraints between the coefficients of the polynomials at neighbouring frequencies. This new variant contributes a new and significant reduction in the Mean Square Error of the FRF estimates. We also discuss the effects of the various design parameters on the accuracy of the estimates.