A Multi-Channel Prediction Method via Separate Extraction of Temporal-Frequency Correlations

The advancement of wireless communication is hindered by inefficient and imbalanced spectrum utilization. Spectrum prediction offers a promising solution to this challenge. Most existing spectrum prediction methods rely on a channel-dependent (CD) strategy, however, their performance often lags behind that of the channel-independent (CI) strategy, due to inherent limitations in handling inter-channel correlations. To address this, we propose a hybrid method that integrates the strengths of both CI and CD strategies. First, the CI strategy is employed to independently extract temporal correlations, thereby effectively reducing inter-channel interference and expanding the dataset. Then, we propose a model-agnostic plugin that leverages the CD strategy to extract inter-channel correlations and seamlessly integrates them into the prediction process. Experimental results demonstrate that our proposed method significantly enhances the performance of the original models, and outperforms existing spectrum prediction algorithms.