Adaptive Learning Algorithms with Fast Convergence Rate for Minimum Bit Error Rate Nonlinear Equalizers

ZHU Ren-xiang,WU Le-nan

DOI: https://doi.org/10.3969/j.issn.1007-0249.2012.02.017

2012-01-01

Abstract:Fast converged and adaptive algorithms based on quasi Newton method are proposed in this paper for online training of minimum bit error rate nonlinear equalizers.Bit error rate is estimated by Parzen window.Learning can be switched between sliding window-stochastic gradient algorithm and sliding window-quasi Newton algorithm by a decisive variable.This makes the new algorithms much more stable with a fast convergence rate.Moreover,by modifying the quasi Newton method,the new algorithms can be used both for adaptive learning and for high-dimensional parameters training.Minimum bit error rate nonlinear equalizers are used for interference suppression and channel equalization in communications,and their high efficiency is proved by simulation results.