Dynamic Graph Temporal-Frequency Dual-Channel Network for Multi-Band Spectrum Prediction

Abstract

Spectrum prediction is crucial in cognitive radio networks, and previous studies have introduced various spectrum modeling methods. However, these methods typically only capture static patterns from historical spectrum data, overlooking the dynamic nature of the spectrum environment. Therefore, this letter proposes a dynamic graph temporal-frequency dual-channel network (DGTFDN) to capture the time-varying correlations among multiple frequency bands and adaptively update and aggregate features. Additionally, the proposed method employs a dual-channel network to simultaneously model both non-structural and structural features from multiple frequency bands, and then utilizes an adaptive gating mechanism to fuse the two types of features. The experimental results show that the proposed method achieves better prediction results than other comparative methods, with improvements of 0.3638, 0.7828, and 2.7077% in MAE, MRSE, and MAPE respectively.