Neural networks equalizers for nonlinear magnetic recording channels

Nonlinear distortion in perpendicular magnetic recording channels is known to degrade the overall system performance. In this work, we propose two nonlinear equalizers based on neural network (NN). One involves symbol decision of received signals using a multilayer perceptronNN equalizer (MLPNNE) only, and the other includes the NN equalizer to shape received signal to a partial-response target followed by a maximum likelihood (ML) sequence detection scheme using Viterbi algorithm (ML-MLPNNE). When applied to nonlinear channels generated by Volterra model (VM), it is shown that these two proposed equalizers give similar BER performances. At the BER of 10−4, they provide about 10-dB SNR gains over the conventional partial-response maximum likelihood (PRML) technique. The MLPNNE with the simple threshold needs simpler implementation than the ML-MLPNNE although noise correlation is a disadvantage.