Next-Generation Passive Optical Network Based on Sparse Code Multiple Access and Graph Neural Networks

Sparse code multiple access (SCMA) is a promising technology to provide high throughput and overall improved system performance at affordable cost for next-generation passive optical networks (NG-PONs). Message passing algorithm (MPA) based on a factor graph is usually used for low-complexity multi-user detection in SCMA. However, MPA requires accurate and effective channel estimation due to the interference between the user's signals on the same resource block and suffers uncertain convergence caused by the cycles in the factor graph. In this paper, a graph neural networks (GNN)-based detection method is proposed for SCMA-PON, which performs channel impairment compensation and signal detection in a joint manner. 25.5 Gb/s SCMA-OFDM system over 20/60 km single mode fiber link is simulated to demonstrate the feasibility with 8.5-G class optical devices. The simulation results show that GNN-based detection method outperforms MPA and is more robust to the nonlinear distortion for the same level of computational complexity.