A Deep Long-Term Joint Temporal-Spectral Network for Spectrum Prediction.

Area of Science:

• Wireless communication
• Signal processing
• Machine learning

Background:

• Spectrum prediction is crucial for efficient spectrum resource management in cognitive radio networks.
• Traditional methods struggle with complex environments and lack real-time prediction capabilities.
• Existing models often fail to capture temporal-spectral dynamics effectively.

Purpose of the Study:

• To develop a deep learning model for simultaneous temporal-spectral and multi-slot spectrum prediction.
• To enhance the accuracy and timeliness of spectrum prediction in dynamic environments.
• To overcome limitations of traditional methods in perceiving complex spectrum states.

Main Methods:

• A hierarchical spectrum prediction system utilizing Bi-ConvLSTM and a seq2seq framework.
• Bi-ConvLSTM for capturing time-frequency characteristics.
• Attention mechanism integrated to prevent information loss in seq2seq framework.

Main Results:

• The proposed model demonstrates significant advantages over benchmark schemes.
• Achieved 6.15% MAPE, 0.7749 MAE, 1.0978 RMSE, and 0.9628 R2.
• Outperformed all baseline deep learning models in spectrum prediction accuracy.

Conclusions:

• The developed deep learning model offers superior performance for temporal-spectral and multi-slot spectrum prediction.
• The integration of Bi-ConvLSTM, seq2seq, and attention mechanisms effectively addresses limitations of prior approaches.
• The model provides a more robust and timely solution for spectrum resource management.