Hierarchical Deep Learning for Multi-PU Cooperative Spectrum Sensing: A Sequential 1D-CNN and GCN Framework for Robust Feature Learning

Efficient dynamic spectrum access is vital in cognitive radio networks (CRNs) due to the escalating demand for radio frequency spectrum. This work focuses on solving the cooperative spectrum sensing problem in multi-primary user CRNs. A novel hierarchical deep learning architecture, termed the CNN-GCN model, is proposed to integrate a one-dimensional Convolutional Neural Network (1D-CNN) with a Graph Convolutional Network (GCN) for enhanced feature representation and relational learning. The model explicitly learns intra-features from individual secondary user signals using a shared 1D-CNN layer. The output of the 1D-CNN forms a structured feature matrix, which is then processed by the GCN to learn inter-features across all SUs. Experimental results demonstrate that the proposed model achieves superior performance compared to existing models.