Improvement of Prony -based least square method of system identification via nonlinear parameter transformation and finite -difference processing of signals.

摘要

This dissertation presents an approach to improve Prony's method of identifying a linear time-invariant system. An algorithm is developed based on the introduction of finite differences. The method is based on a nonlinear transformation of parameters, which leads to data averaging. This method yields better results than a direct application of the least squares approach to Prony's method. In addition, recursive analysis for both the conventional Prony's method and the new finite-difference method has been derived and implemented. A numerical example is studied in detail to demonstrate the improvement attained by the new algorithm. Signals in the study are assumed to be contaminated by zero-mean Gaussian white noise.