TY  - JOUR
AU  - Han, Yu
AU  - Song, Jiayi
AU  - Wang, Zhiqi
AU  - Xiang, Wei
AU  - Mathiopoulos, P. Takis
AU  - Gui, Guan
AU  - Lin, Yun
PY  - 2025
DA  - 2025/11/20
TI  - A Novel Electromagnetic Spectrum Prediction Model Based upon Multi-Dimensional Feature Fusion
JO  - Chinese Journal of Information Fusion
T2  - Chinese Journal of Information Fusion
JF  - Chinese Journal of Information Fusion
VL  - 3
IS  - 1
SP  - 1
EP  - 16
DO  - 10.62762/CJIF.2025.747641
UR  - https://www.icck.org/article/abs/CJIF.2025.747641
KW  - electromagnetic spectrum prediction
KW  - multi-dimensional decoupling
KW  - bidirectional cross-attention mechanism
KW  - temporal convolutional network
AB  - In the era of increasingly scarce spectrum resources, electromagnetic spectrum (EMS) prediction has emerged as a critical means for enhancing spectrum utilization efficiency. However, most of the existing EMS methods primarily exploit low-dimensional features such as temporal, frequency, or spatial characteristics in an individual fashion, which limits their ability to fully capture the inherent complexity of spectrum dynamics. To improve the performance, this paper proposes a novel EMS prediction model, which involving three operations, namely multi-dimensional decoupling, feature fusion and temporal prediction. Firstly, for multi-dimensional decoupling operation, we propose a Multi-dimensional Feature Extraction (MFE) module, which characterizes the complex temporal-frequency-spatial variations of EMS data by leveraging both single-dimensional features and cross-dimension dependencies (i.e., temporal-frequency, temporal-spatial, and frequency-spatial relationships). By explicitly modeling these correlations, the MFE module enhances the prediction performance of the proposed model. Secondly, to reduce redundancy between these decoupled multi-path features, we introduce a Tensor-Feature-Fused (TF) module. Through a bidirectional cross-attention mechanism, the proposed TF module enables symmetric information exchange between multi-path features and the original spectrum data, by selectively integrating both inter-path features and intra-feature information. Finally, by employing a Temporal Convolutional Network (TCN), the data obtained by the TF module are processed to capture multi-scale dependencies so that the accuracy of spectrum prediction is enhanced. The performance of the proposed model, termed as MFE-TFTCN, has been extensively evaluated by means of computer simulations. Various experimental results obtained through the use of a publicly European multi-location dataset have demonstrated that, compared to state-of-the-art EMS prediction methods, the proposed model achieves superior prediction performance by effectively capturing temporal-frequency-spatial interdependencies.
SN  - 2998-3371
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Han2025A,
  author = {Yu Han and Jiayi Song and Zhiqi Wang and Wei Xiang and P. Takis Mathiopoulos and Guan Gui and Yun Lin},
  title = {A Novel Electromagnetic Spectrum Prediction Model Based upon Multi-Dimensional Feature Fusion},
  journal = {Chinese Journal of Information Fusion},
  year = {2025},
  volume = {3},
  number = {1},
  pages = {1-16},
  doi = {10.62762/CJIF.2025.747641},
  url = {https://www.icck.org/article/abs/CJIF.2025.747641},
  abstract = {In the era of increasingly scarce spectrum resources, electromagnetic spectrum (EMS) prediction has emerged as a critical means for enhancing spectrum utilization efficiency. However, most of the existing EMS methods primarily exploit low-dimensional features such as temporal, frequency, or spatial characteristics in an individual fashion, which limits their ability to fully capture the inherent complexity of spectrum dynamics. To improve the performance, this paper proposes a novel EMS prediction model, which involving three operations, namely multi-dimensional decoupling, feature fusion and temporal prediction. Firstly, for multi-dimensional decoupling operation, we propose a Multi-dimensional Feature Extraction (MFE) module, which characterizes the complex temporal-frequency-spatial variations of EMS data by leveraging both single-dimensional features and cross-dimension dependencies (i.e., temporal-frequency, temporal-spatial, and frequency-spatial relationships). By explicitly modeling these correlations, the MFE module enhances the prediction performance of the proposed model. Secondly, to reduce redundancy between these decoupled multi-path features, we introduce a Tensor-Feature-Fused (TF) module. Through a bidirectional cross-attention mechanism, the proposed TF module enables symmetric information exchange between multi-path features and the original spectrum data, by selectively integrating both inter-path features and intra-feature information. Finally, by employing a Temporal Convolutional Network (TCN), the data obtained by the TF module are processed to capture multi-scale dependencies so that the accuracy of spectrum prediction is enhanced. The performance of the proposed model, termed as MFE-TFTCN, has been extensively evaluated by means of computer simulations. Various experimental results obtained through the use of a publicly European multi-location dataset have demonstrated that, compared to state-of-the-art EMS prediction methods, the proposed model achieves superior prediction performance by effectively capturing temporal-frequency-spatial interdependencies.},
  keywords = {electromagnetic spectrum prediction, multi-dimensional decoupling, bidirectional cross-attention mechanism, temporal convolutional network},
  issn = {2998-3371},
  publisher = {Institute of Central Computation and Knowledge}
}