Novel Turbo Equalization Methods for the Magnetic Recording Channel Elizabeth Chesnutt 95 Pages Directed by Dr. John R. Barry The topic of this dissertation is the derivation, development, and evaluation of novel turbo equalization techniques that address the colored noise problem on the magnetic recording channel. One new algorithm presented is the noise-predictive BCJR, which is a soft-output detection strategy that mitigates colored noise in partial-response equalized magnetic recording channels. This algorithm can be viewed as a combination of the traditional BCJR algorithm with the notion of survivors and noise prediction. Additionally, an alternative equalization architecture for magnetic recording is presented that addresses the shortcomings of the PRML approach, which dominates magnetic recording. Specifically, trellis-based equalizers are abandoned in favor of simple equalization strategies based on nonlinear filters whose complexity grows only linearly with their length. This research focuses on the linear-complexity SFE algorithm and on investigating the possibility of lowering the SFE filter calculation complexity. The results indicate that with using the proposed novel SFE method, it is possible to increase the information density on magnetic media without raising the complexity. The most important result presented is that partial-response equalization needs to be reconsidered because of the amount of noise enhancement problems that it adds to the overall system. These results are important for the magnetic recording industry, which is trying to attain a 1 Tb/cm2 information storage goal.

Bit-Patterned Media (BPM) has been recognised as one of the candidate technologies to achieve an areal density beyond 1Tb/in2 by fabricating single-domain islands out of continuous magnetic media. Though much attention has been focused on the fabrication of BPM, existing lithography techniques demonstrate difficulties in producing uniform islands over large areas cost effectively; the resulting fabricated islands often vary in position and size. The primary purpose of the research documented in this thesis is to investigate the issue of island geometry variations on the data recovery process from a perpendicular patterned media with head and media configurations optimised to achieve an areal density of 1Tb/in2. In order to achieve the research aim, a read channel model has been implemented as a platform to evaluate the read channel performance numerically. It can be also altered to investigate new read channel designs. The simulated results demonstrate that island geometry variations have a detrimental effect on read channel performance. It has shown that a BPM system can be tolerant to island position variations, but more effort needs to be paid to the effect that island size variations have on the read channel performance. A new read channel design revolving around the design of a modified trellis has been proposed for use in the Viterbi detector in order to combat the effect of island geometry variations. The modified trellis for island position variations results in extra states and branches compared to the standard trellis, while the modified trellis for island size variations results in only extra branches. The novel read channel designs demonstrate an improved read channel performance in the presence of island geometry variations even with increasing amounts of island position and size variations. There are two ways to obtain the read channel performance in terms of the bit-error-rate (BER): a) by running a numerical Monte-Carlo simulation to count the number of bits in error at the output of the read channel model and b) using an analytical approach to calculate the BER by approximating the noise into a known distribution. It is shown that both ways demonstrate very similar results, which indicates as long as the distribution of the noise present in read channel model is predictable, the analytical approach can evaluate the BER performance more efficiently, especially when the BER is low. However, the Monte-Carlo simulation is still useful for understanding of the correlation of the errors. Novel trellis proposed in this work will contribute to the commercial development of BPM in two ways: a) to improve the data recovery process in BPM systems, b) to allow a tolerance of 10% size variations for the existing fabrication techniques.