Title :
Rapidly converging adaptive IIR algorithms
Author :
Soni, Robert A. ; Jenkins, W. Kenneth
Author_Institution :
Dept. of Comput. & Electr. Eng., Illinois Univ., Urbana, IL, USA
Abstract :
Several new adaptive infinite-impulse (IIR) filtering algorithms based upon the the algorithm developed by Fan and Jenkins (1986) are proposed. The Fan-Jenkins algorithm was shown to experimentally possess the ability to converge to the global minimum of the mean square error (MSE) even in cases where the MSE surface is ill-conditioned. By incorporating estimates of the Hessian matrix in the adaptive filter coefficient update expressions, the new versions of the algorithm appear to improve the convergence performance in comparison to the traditional least mean square (LMS) type algorithms and to preserve the ability of the algorithm to converge to the global minimum of the MSE. The last mean square (LMS), recursive least square (RLS), Gauss-Newton (GN), and the fast quasi-Newton forms of the algorithm are formulated and compared via simulation
Keywords :
Hessian matrices; IIR filters; Newton method; adaptive filters; convergence of numerical methods; filtering theory; least mean squares methods; recursive estimation; Fan-Jenkins algorithm; Gauss-Newton algorithm; Hessian matrix; LMS algorithms; adaptive IIR algorithms; adaptive filter coefficient update; adaptive infinite-impulse filtering algorithms; convergence performance; fast quasiNewton algorithm; global minimum; mean square error; recursive least square algorithm; simulation; Adaptive filters; Convergence; Equations; Filtering; IIR filters; Least squares approximation; Least squares methods; Mean square error methods; Newton method; Recursive estimation;
Conference_Titel :
Image Analysis and Interpretation, 1996., Proceedings of the IEEE Southwest Symposium on
Conference_Location :
San Antonio, TX
Print_ISBN :
0-7803-3200-8
DOI :
10.1109/IAI.1996.493759
