Parametric System Identification on Logarithmic Frequency Response Data

Abstract

Gradient search methods that fit the parameters of a user-defined transfer function model to experimental Logarithmic frequency response data are presented. The methods match a model based on physically significant parameters, including natural frequencies of poles and zeroes and damping ratios of complex poles and zeroes. The algorithms construct and utilize their own analytical gradient descent functions, based on the desired model. One method attempts to fit both log magnitude and phase, while another identifies a minimum phase transfer function model from only log magnitude frequency response data. The performance of the log magnitude algorithm is shown to be superior to traditional methods using non-logarithmic frequency response data, including those used in commercially available frequency response analyzers. The algorithms are shown to perform well, especially for systems with lightly-damped dynamics.