Adaptive Minimum-Bit-Error Rate PDNP Detection for Magnetic Recording

The granular nature of the recording medium in magnetic recording leads to a type of noise known as media noise; it arises because each magnetic grain can take on only one of two polarities, which causes the boundary of each written bit to be random in shape, coinciding with the boundaries of the randomly sized grains. A key feature of media noise is that it depends on the data being written, and in particular is more pronounced in the vicinity of bit transitions. A widely used strategy for mitigating media noise in a trellis-based detector is pattern-dependent noise prediction (PDNP); in this approach, each bit pattern (which determines a trellis branch) will have its own set of branch metric parameters (including the signal levels, noise predictor coefficients, and residual variances). Traditionally these detector parameters are chosen according to some form of a minimum-mean-squared-error (MMSE) criterion. In this paper, we propose the adaptive minimum-bit-error rate (AMBER) algorithm for adapting these pattern-dependent parameters with the aim of minimizing BER. The AMBER algorithm updates the parameters whenever the add-compare-select operation in the Viterbi detector selects an incorrect path instead of the correct path; the parameters are then updated so as to increase the metric of the incorrect path, and reduce the metric of the correct path. Numerical results based on a set of quasi-micromagnetic simulated channel waveforms show that the AMBER PDNP detector provides at least a 20% BER decrease and an 8% increase in areal density over a traditional MMSE PDNP detector.