Title :
EA crossover schemes for a MLP channel equaliser
Author :
Power, P. ; Sweeney, F. ; Cowan, C.F.N.
Author_Institution :
Dept. of Electr. & Electron. Eng., Queen´´s Univ., Belfast, UK
Abstract :
This paper presents an evolutionary algorithm (EA) in a form similar to the LMS algorithm, which is applied to MLP learning. The gradient-based update term of the LMS is replaced with the EA non-gradient-based random distancing matrix. This matrix is created to share solution gene information between selected parent chromosomes. The channel equalisation problem is used to compare this algorithm against an EA averaging style operator, which has previously been examined in this area. It is shown that there is an improved learning capability in the MLP filter when the EA updating operators are unrestrained in the range of the gene exchange
Keywords :
equalisers; evolutionary computation; learning (artificial intelligence); multilayer perceptrons; EA crossover schemes; MLP channel equaliser; averaging style operator; channel equalisation problem; evolutionary algorithm; gene exchange; gradient-based update term; learning capability; multilayer perceptrons; nongradient-based random distancing matrix; parent chromosomes; solution gene information; Backpropagation algorithms; Evolutionary computation; Filtering; Finite impulse response filter; Iterative algorithms; Least squares approximation; Neural networks; Noise level; Nonlinear filters; Signal to noise ratio;
Conference_Titel :
Electronics, Circuits and Systems, 1999. Proceedings of ICECS '99. The 6th IEEE International Conference on
Conference_Location :
Pafos
Print_ISBN :
0-7803-5682-9
DOI :
10.1109/ICECS.1999.812309
