Graph Neural Network Assisted Efficient Signal Detection for OTFS Systems
Xufan Zhang, Shengyu Zhang, Lixia Xiao, Shuo Li, Tao Jiang
Abstract
In this letter, an efficient graph neural network (GNN) assisted detector is conceived for the orthogonal time frequency space (OTFS) system. Specifically, the transmit symbols are viewed as the nodes of GNN, obtaining the detection results through aggregation, update, and output modules. Firstly, the aggregation module is employed to weigh the connections between adjacent nodes. Subsequently, the update module amends node features according to the calculated connection value and the node’s information. Finally, after a certain number of iterations, the output module classifies nodes relying on the final features to realize signal detection. Simulation results confirm that the proposed GNN-assisted detector outperforms the latest intelligent detector by around 1 dB.