Improved equalization for digital recording using nonlinear filtering and error confinement

In this work, an extended class-IV partial response (EPR4) equalizer for a digital recording channel is replaced by a nonlinear equalizer which is based on a neural network. It is shown that such a scheme has several decibels of signal-to-noise ratio gain compared to linear equalization, when the channel is corrupted by transition noise and media nonlinearity. An error confinement approach, as opposed to the conventional minimum mean square error approach, is shown to further enhance the performance of the nonlinear equalizer