CCTNet: Clustering-ConvTransformer Network for Wide-band Multi-step Satellite Spectrum Prediction

• Main Authors: Chen, Jindi, Xie, Zhuochen, Yang, Wenxin, Yan, Mubiao, Liu, Huijie
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Computational modeling; LEO satellite; Low earth orbit satellites; Prediction algorithms; Predictive models; Satellites; spectrum clustering; spectrum prediction; spectrum sharing; Transformer algorithm; Transformers
• ISSN: 1938-1883
• DOI: 10.1109/ICC51166.2024.10622376

Satellite spectrum-sharing systems exhibit longer transmission delays and occupy larger bandwidths in comparison to terrestrial spectrum-sharing systems. These extended latencies, coupled with the time needed for devices to process broadband data, often render short-term prediction results ineffective. Hence, the need for rapid and precise Wide-band Multi-Step (WBMS) satellite spectrum prediction arises as an effective solution to address this challenge. To tackle this problem, we propose the Clustering-ConvTransformer Network (CCTNet). CCTNet significantly enhances the computation speed of the system through its clustering module. Additionally, it incorporates an improved Transformer network combined with convolution to ensure reliable multi-step prediction accuracy. Experimental evaluations on real-world satellite spectrum data demonstrate that CCTNet achieves a computational speed of only 0.38 seconds, marking a 55-fold improvement over non-clustered networks. Furthermore, in terms of prediction accuracy, the CCTNet has also illustrated superiority compared to other state-of-the-art (SOTA) baselines.