Abstract
The neural network equalization for a generalized partial response maximum likelihood (GPRML) system in a perpendicular magnetic recording channel with thermal decay is studied. First, a designing method of the simplified neural network equalizer (NNE) with the noise whitening function is proposed. The method is based on thinning out of the connections of NNE which is obtained by the alternately iterative learning using a hybrid genetic algorithm and a Levinson-Durbin algorithm. Then, the bit error rate degradation performances with the elapsed time for the GPR1ML and PR1ML systems using this simplifed NNE (GPR1ML-NNE and PR1ML-NNE) are obtained and compared with that for the GPR1ML system using a conventional transversal filter as an equalizer (GPR1ML-TF). It is clarified that both GPR1ML-NNE and PR1ML-NNE show the excellent performance compared to that of GPR1ML-TF.
