Dynamic Grain State Estimation for High-Density TDMR: Progress and Future Directions

摘要

Dynamic grain state estimation (DGSE) algorithms for 2-D magnetic recording (TDMR) employ probabilistic message-passing algorithms that jointly estimate magnetic grain configurations and coded data bits, in order to iteratively assist channel decoding. At high densities (e.g., between 1 and 3 magnetic grains per coded bit), occasionally, a bit will not be written on any grain, and hence will effectively be overwritten (or erased) by bits on surrounding grains. DGSE enables the detection of overwritten bits so that their log-likelihood ratios are assigned small magnitudes, effectively making them erasures, which are easily filled in by linear channel codes. Past papers employing Bahl-Cocke-Jelinek-Raviv-based detectors on a simple four-rectangular-grain model have shown that the DGSE is surprisingly resilient to mismatch between the detector’s assumed grain model and the actual model generating the data. This paper presents a novel DGSE–TDMR detector based on the generalized belief propagation (GBP) algorithm. The new detector employs a random discretized-nuclei Voronoi grain model. Simulation results show that the GBP-based TDMR turbo-detector accurately detects the overwritten bits and that it achieves low decoded bit error rates at densities as high as 0.4966 user bits per grain.