Short-Term Spectrum Situation Prediction Using 3D Transformer with Geographic Priors

The accurate prediction of electromagnetic spectrum dynamics is essential for enhancing operational efficiency in wireless communication and defense systems. However, conventional prediction techniques reliant on physical modeling or ray tracing are often hindered by high computational load and limited adaptability in complex propagation environments. To overcome these challenges, this chapter introduces a 3D Transformer-based model termed 3D-STP for short-term spectrum situation prediction. The model effectively integrates geographic structural priors through a novel location encoding mechanism, maintains 3D structural integrity to capture fine-grained spatial-spectral features, and employs hierarchical temporal aggregation for efficient multi-step prediction. Experimental evaluations demonstrate that the proposed approach not only achieves high predictive accuracy with naturally smooth spectral transitions, but also significantly reduces inference time compared to conventional numerical methods. These contributions collectively advance the deployment of AI-aided spectrum management in practical dynamic scenarios.