Adaptive signal processing and efficient sequence detection for digital magnetic recording channels

This thesis is concerned with developing totally adaptive signal processing approaches for increasing the storage capacity of disk drives. From a signal processing perspective, three immediate concerns must be addressed. The first is intersymbol interference (ISI). The extensive length of ISI in the magnetic recording channel creates a fundamental tradeoff between complexity and performance. At very high densities, ISI can extend over 10--20 symbol periods. Equalization in this environment must balance the conflicting requirements of constraining ISI and Viterbi state complexity with limiting noise enhancement and error rate. For about a decade, partial response equalization targets have served as a good compromise, however recording densities are now exceeding their useful range. New equalization methods must be developed to effectively increase storage capacity.; Other problems posed by increased recording density are higher levels of signal nonlinearity and the prevalence of correlated, sequence-dependent noise due to the magnetic medium. Conveniently applied linear equalization of the nonlinear channel results in misequalization that degrades error rate performance. Also, it has been shown that sequence-dependent, correlated media noise, upon equalization, can seriously degrade the error rate performance of the Viterbi algorithm (VA), whose optimal performance requires uncorrelated noise. Complicating matters is the fact that neither channel nonlinearity nor media noise is easily characterized with a high degree of accuracy until after the manufacture of the entire storage device. This strongly suggests the need for adaptive signal processing approaches.; This thesis describes equalization and detection strategies that address the aforementioned problems, keeping in mind complexity constraints required for feasible hardware implementation. Adaptive equalization approaches are developed to whiten and reduce the distortion at the VA's input to the extent possible, regardless of media noise variations. Viterbi branch metrics are adaptively adjusted to account for misequalization due to channel nonlinearity and sequence-dependent media noise. VA implementation is also addressed. Complexity in the VA, due to the adaptive nature of the detection scheme, is alleviated through the use of piecewise linear branch metrics in lieu of the traditionally applied Euclidean-distance metrics. The linear metrics are also shown to be applicable to sequence-dependent and non-Gaussian noise.