Attention-Adaptive Fusion GNN Decoder for LDPC Codes
IEEE Communications Letters · 2026 · 30 · 2347-2351
DOI: 10.1109/lcomm.2026.3705887 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Communications Letters, 2026. Verfasst von Long Jin. Die Arbeit nennt 13 Literaturangaben.
Abstract (Originalsprache)
Graph neural networks (GNNs) can improve LDPC decoding, but their complexity limits practical deployment. We propose an Attention-Adaptive Fusion GNN (AAF-GNN), a lightweight residual decoder that augments a weighted belief propagation (WBP) backbone with learnable neural corrections. Unlike a baseline residual GNN with global scaling, AAF-GNN uses node-wise attention to adaptively combine WBP estimates and neural residuals at the bit level. Simulation results show that AAF-GNN outperforms standalone GNN decoders and conventional WBP baselines, while parameter-count and runtime analyses demonstrate a favorable trade-off between decoding performance and computational cost.
Zitieren
APA
Jin, L. (2026). Attention-Adaptive Fusion GNN Decoder for LDPC Codes. IEEE Communications Letters, 30, 2347-2351. https://doi.org/10.1109/lcomm.2026.3705887
Vancouver
Jin L. Attention-Adaptive Fusion GNN Decoder for LDPC Codes. IEEE Communications Letters. 2026;30:2347-2351. doi:10.1109/lcomm.2026.3705887
BibTeX
@article{Jin2026,
author = {Jin, Long},
title = {Attention-Adaptive Fusion GNN Decoder for LDPC Codes},
journal = {IEEE Communications Letters},
year = {2026},
volume = {30},
pages = {2347--2351},
doi = {10.1109/lcomm.2026.3705887}
}
EUVIMED