Title :
A robust general normalised gradient descent algorithm
Author :
Mandic, Danilo P. ; Obradovic, Dragan ; Kuh, Anthony
Author_Institution :
Dept. of Electr. & Electron. Eng., Imperial Coll. London
Abstract :
A modification of the generalised normalised gradient descent (GNGD) algorithm is introduced which caters for the steady state performance of the algorithm. This is achieved by combining the search-then-converge and gradient adaptive step size approaches. This way the proposed algorithm exhibits very fast convergence and small steady state error. Simulations on linear and nonlinear signals in prediction setting support the analysis
Keywords :
adaptive signal processing; convergence of numerical methods; gradient methods; GNGD; general normalised gradient descent algorithm; gradient adaptive step size approach; linear signal; nonlinear signal; search-then-converge approach; steady state performance; Analytical models; Convergence; Educational institutions; Finite impulse response filter; Least squares approximation; Predictive models; Robustness; Signal analysis; Stability; Steady-state;
Conference_Titel :
Statistical Signal Processing, 2005 IEEE/SP 13th Workshop on
Conference_Location :
Novosibirsk
Print_ISBN :
0-7803-9403-8
DOI :
10.1109/SSP.2005.1628578
