About the event
Student: Amirhossein Sayyafan
Advisor: Drs. Benjamin J. Belzer and Krishnamoorthy Sivakumar
Degree: Electrical and Computer Engineering, Ph.D.
Dissertation Title: Deep Learning-based Turbo-detection and Equalization for Two- and Three-dimensional Magnetic Recording
Abstract: This dissertation considers various machine-learning-based signal processing architectures for equalization and detection of two- and three-dimensional magnetic recording signals for hard disk drives (HDDs). Recording in multiple dimensions on magnetic hard drives has been a challenge in the HDD industry. The objective of reading approaches for magnetic recording is to detect the highest density of information possible with an acceptable error rate.
The first part of this dissertation considers a concatenated Bahl-Cocke-Jelinek-Raviv (BCJR) detector, a low-density parity-check (LDPC) decoder, and a deep neural network (DNN) architecture for a turbo-detection system with one- and two-dimensional magnetic recording (1DMR and TDMR) for HDD. Two types of equalizers are investigated: a linear filter equalizer with a 1-D/2-D partial response (PR) target and a convolutional neural network (CNN) PR equalizer.
The second part investigates DNN-based turbo-detection for multilayer magnetic recording (MLMR), an emerging HDD technology that employs vertically stacked magnetic media layers with readers above the top-most layer. The proposed system employs two layers with two upper-layer tracks and one lower-layer track. The reader signals are processed by CNNs to separate the upper- and lower-layer signals and equalize them to 2-D and 1-D PR targets, respectively. In the two first parts, as the baseline, we consider the standard 1-D and 2-D pattern-dependent noise prediction (PDNP) method, which has become standard practice in the HDD industry.
The third part considers a turbo-detection system that includes a CNN- based equalizer, a BCJR trellis detector, a CNN-based media noise predictor (MNP), and an LDPC channel decoder for TDMR in the presence of track-misregistration (TMR). Spatially varying read- and write-TMR models are developed. The write- and read-TMR are modeled as cross-track-independent down-track-correlated random processes. We investigate the performance of the proposed system on simulated TDMR readback waveforms with TMR. The comparison baseline is a 1-D BCJR detector with PDNP and soft intertrack interference (ITI) subtraction, referred to as 1-D PDNP with LLR exchange.
The simulation results show that in all the cases, the proposed DNN-based methods outperform the PDNP, achieve higher areal density, and are more robust against the noise and write- and read-TMR than PDNP.