Two-dimensional magnetic recording (TDMR) is a new technology to increase the areal density of magnetic recording systems with conventional magnetic disks. We incorporate signal processing and machine learning techniques to dynamically estimate the states of magnetic grains, and recover the information written on hard disk under TDMR channel.

Firstly, we introduce several grain models involved to explain magnetic grain properties. The read and write models in TDMR channel are also discribed and compared against different grain densities.

Then, a generallized belief propagation (GBP) based algorithm is proposed to dynamically estimate grain states interactively and iteratively. The algorithms pass messages of graph region based beliefs to each other region, and combines disparate top level regoins together to aggregate into an overall estimate of grain states and produce a soft decision in form of log likelihood ratio (LLR). And a low density parity check (LDPC) decoder is serially concatenated to give final decision.

Next, a tiny but more efficient approach called local area influence probabilistic (LAIP) algorithm is proposed. LAIP utilizes grain interferance among sorrounding bits to estimate the influence on magnetizations of target bits.

Then, LAIP is extended to work with TDMR channel with two-dimensional inter-symbol interference. A Bahl-CockeJelinek-Raviv (BCJR) equalizer is concatenated to turbo detect coded bits. Moreover, we use specially desired pattern training method to adapt LAIP to grain flipping (GFP) model. The best achieved areal density is 0.18 user bits per grain at 18nm track pitch.