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.

中文翻译：

---

深度神经网络：通过学习实现硬盘驱动器的数据检测通道

本文研究了将深度神经网络 (DNN) 作为硬盘驱动器 (HDD) 数据恢复检测通道在高用户位密度和突出的磁跃迁抖动噪声中的应用。直接用未均衡的回读信号训练，无需任何磁记录通道的先验知识，DNN 可以自动学习信号特征，特别是输入信号与噪声影响之间的相关性。因此，DNN 读取通道不仅可以适应符号间干扰 (ISI)，而且还表现出对有色磁噪声的强大弹性。我们的模拟结果还表明，为了充分利用 DNN 数据检测通道的学习能力，神经网络输入必须覆盖 ISI 传播。此外，