Graph Neural Network (GNN) for Joint Detection–Decoder MAP–LDPC in Bit-Patterned Media Recording Systems

Abstract

With its high area density, bit-patterned media recording (BPMR) is emerging as a leading technology for next-generation storage systems. However, as area density increases, magnetic islands are positioned closer together, causing significant two-dimensional (2D) interference. To address this, detection methods are used to interpret the received signal and mitigate 2D interference. Recently, the maximum a posteriori (MAP) detection algorithm has shown promise in improving BPMR performance, though it requires extrinsic information to effectively reduce interference. In this paper, to solve the 2D interference and improve the performance of BPMR systems, a model using low-density parity-check (LDPC) coding was introduced to supply the MAP detector with the needed extrinsic information, enhancing detection in a joint decoding model we call MAP–LDPC. Additionally, leveraging similarities between LDPC codes and graph neural networks (GNNs), we replace the traditional sum–product algorithm in LDPC decoding with a GNN, creating a new model, MAP–GNN. The simulation results demonstrate that MAP–GNN achieves superior performance, particularly when using the deep learning-based GNN approach over conventional techniques.