Application of a neural network for detection at strong nonlinear intersymbol interference

As recording density rises read signals are increasingly distorted by nonlinear intersymbol interference (ISI). Against this background an artificial neural network with a new decision making scheme has been set up and trained to work as a detector. Tests have been performed with experimentally captured read signals from a modified disk drive with magneto-resistive (MR) read heads. In comparison with multi-level decision feedback equalization (MDFE) the detection results show superior performance at extremely high linear recording densities. An error rate of 4.10/sup -6/ has been achieved at a user density D/sub u/=3.5. We describe the architecture and the training procedure of the neural network and present detection results.