Multitrack Detection for Magnetic Recording

The thesis develops advanced signal processing algorithms for magnetic recording to increase areal density. The exploding demand for cloud storage is motivating a push for higher areal densities, with narrower track pitches and shorter bit lengths. The resulting increase in interference and media noise requires improvements in read channel signal processing to keep pace. This thesis proposes the multitrack pattern-dependent noise-prediction algorithm as a solution to the joint maximum-likelihood multitrack detection problem in the face of pattern-dependent autoregressive Gaussian noise. The magnetic recording read channel has numerous parameters that must be carefully tuned for best performance; these include not only the equalizer coefficients but also any parameters inside the detector. This thesis proposes two new tuning strategies: one is to minimize the bit-error rate after detection, and the other is to minimize the frame-error rate after error-control decoding. Furthermore, this thesis designs a neural network read channel architecture and compares the performance and complexity with these traditional signal processing techniques.Ph.D