Title :
Performance analysis of estimation algorithms of nonstationary ARMA processes
Author :
Ding, Feng ; Shi, Yang ; Chen, Tongwen
Author_Institution :
Control Sci. & Eng. Res. Center, Southern Yangtze Univ., Jiangsu, China
fDate :
3/1/2006 12:00:00 AM
Abstract :
The correlation analysis based methods are not suitable for identifying parameters of nonstationary autoregressive (AR), moving average (MA), and ARMA systems. By using estimation residuals in place of unmeasurable noise terms in information vector or matrix, we develop a least squares based and gradient based algorithms and establish the consistency of the proposed algorithms without assuming noise stationarity, ergodicity, or existence of higher order moments. Furthermore, we derive the conditions for convergence of the parameter estimation. The simulation results validate the convergence theorems proposed.
Keywords :
autoregressive moving average processes; correlation methods; gradient methods; least squares approximations; parameter estimation; convergence theorem; correlation analysis based methods; estimation algorithms; gradient based algorithms; least squares algorithms; moving average systems; nonstationary ARMA process; nonstationary autoregressive systems; parameter estimation; parameter identification; performance analysis; Autoregressive processes; Convergence; Filtering; Least squares approximation; Parameter estimation; Performance analysis; Signal analysis; Signal processing algorithms; Spectral analysis; Time series analysis; Autoregressive (AR) models; autoregressive moving average (ARMA) models; convergence properties; gradient search; least squares filtering; martingale convergence theorem; moving average (MA) models; parameter estimation; recursive identification;
Journal_Title :
Signal Processing, IEEE Transactions on
DOI :
10.1109/TSP.2005.862845