Spectrum Data Graph Structure Learning Based On Dual-View Contrastive Learning For Spectrum Prediction Of ISCC

In the upcoming 6G era, an efficient spectrum prediction method is crucial for meeting the requirements of complex intelligent applications in wireless networks. This paper proposes a spectrum data graph structure learning method and a Multi-Scale Feature Fusion Graph Convolutional Network model (MFGCN) aimed at optimizing Integrated Sensing, Communication, and Computing (ISCC) for spectrum analysis and management. Our method addresses the issue of noisy graph connections commonly found in existing graph-based spectrum prediction methods. By employing a dual-view graph structure contrastive learning approach, it enhances the consistency of the graph structure, thereby supporting the high reliability demands of 6G networks. The MFGCN model combines the learning of local and global signal correlations, using convolutional kernels of various scales to capture signal patterns and adapt to both short-term and longterm environmental changes. Experimental results demonstrate the model’s efficacy in real-world spectrum analysis and prediction, offering novel 6G data utilization strategies.