Cooperative Spectrum Sensing Algorithm Based on Eigenvalue and Graph Convolutional Network
Weijia Huang, Yonghua Wang, Hao Yuan, Yaohua Hu · IEEE Sensors Journal · 2025
In cooperative spectrum sensing (CSS), the hidden node problem can arise due to factors such as hidden terminals or shadowing effects, which hinder secondary users (SUs) from effectively participating in the decision-making process of the primary user (PU). Existing algorithms based on the graph convolutional network (GCN) can mitigate this issue; however, using an energy-based graph construction method may result in poor sensing performance under low signal-to-noise ratio (SNR) conditions and prolonged training time. To address these challenges, this article proposes a novel GCN algorithm based on eigenvalues (EigGCN). First, a different number of SUs is set up to model the hidden node problem. Then, SUs are defined as nodes in the graph, with the eigenvalues of the covariance matrix used to represent the node features. A complete graph is constructed to depict the relationship between SUs. Finally, the graph dataset is trained using a GCN to obtain a graph classifier, which is then utilized for classifying signal data. Compared to the energy-based approach for constructing graphs, the proposed algorithm demonstrates improvements in both detection probability and sensing time. Furthermore, this article introduces varying levels of antenna fading to SUs and validates the exceptional performance of the algorithm in dynamic wireless environments.