R&D: Deep Neural Network, Data Detection Channel for HDDs by Learning

IEEE Transactions on Magnetics has published an article written by Yuwei Qin, and Jian-Gang Zhu, Data Storage Center, Carnegie Mellon University, Pittsburgh, PA, USA.

Abstract: “The application of a deep neural network (DNN) as the detection channel for hard disk drive (HDD) data recovery at high user bit density and the prominent magnetic transition jitter noise are investigated in this article. Directly trained with the un-equalized readback signals without any prior knowledge of the magnetic recording channel, the DNN can automatically learn the signal characteristics, in particular the correlations between the input signals and the impact from the noise. As a result, the DNN read channel not only adapts the inter-symbol interference (ISI) but also demonstrates strong resilience against the colored magnetic noise. Our simulation results also reveal that to fully harness the learning power of the DNN data detection channel, the neural network inputs must cover the ISI spread. In addition, the training data must be sufficiently representative so that the inductive bias learned by the DNN detection channel can be used as good prior knowledge for actual detection.“