Deep Learning for Spectrum Prediction in Cognitive Radio Networks: State-of-the-Art, New Opportunities, and Challenges

TL;DR

This paper tackles spectrum prediction for cognitive radio networks to enable dynamic spectrum access by modeling nonlinear, spatiotemporal usage patterns. It reviews DL-based intra-band and cross-band approaches, and introduces ViTransLSTM, a ViT-LSTM framework that combines visual self-attention with LSTM to capture local and global dependencies in spectrum maps, validated on real-world data. It also explores cross-band data enhancement via DGAN and cross-band transfer learning (DTL) to address data scarcity and distribution shifts. The study highlights key challenges and opportunities, including incomplete data, multi-modal fusion, and multi-domain information fusion, shaping future DL research for spectrum prediction.

Abstract

Spectrum prediction is considered to be a promising technology that enhances spectrum efficiency by assisting dynamic spectrum access (DSA) in cognitive radio networks (CRN). Nonetheless, the highly nonlinear nature of spectrum data across time, frequency, and space domains, coupled with the intricate spectrum usage patterns, poses challenges for accurate spectrum prediction. Deep learning (DL), recognized for its capacity to extract nonlinear features, has been applied to solve these challenges. This paper first shows the advantages of applying DL by comparing with traditional prediction methods. Then, the current state-of-the-art DL-based spectrum prediction techniques are reviewed and summarized in terms of intra-band and crossband prediction. Notably, this paper uses a real-world spectrum dataset to prove the advancements of DL-based methods. Then, this paper proposes a novel intra-band spatiotemporal spectrum prediction framework named ViTransLSTM. This framework integrates visual self-attention and long short-term memory to capture both local and global long-term spatiotemporal dependencies of spectrum usage patterns. Similarly, the effectiveness of the proposed framework is validated on the aforementioned real-world dataset. Finally, the paper presents new related challenges and potential opportunities for future research.

Figures (5)

• Figure 1: System model.
• Figure 2: RMSE comparison of various spectrum prediction schemes. All schemes are executed on a PC featuring an Intel(R) Xeon(R) E5-2698 v4 CPU @ 2.20GHz, NVIDIA Tesla V100 GPU 32GB graphics card, and 256GB RAM, utilizing PyTorch 2.1.0 with the Python programming language. The DL-based schemes use mean squared error (MSE) as the loss function, with 20 training epochs, a batch size of 32, and an early stopping patience of 6.
• Figure 3: The overall structure of the ViT-LSTM.
• Figure 4: Comparison results of the average MSE and PSNR between ViTransLSTM and baselines under the prediction of 10 frames.
• Figure 5: Summary of research challenges for DL-based spectrum prediction.