CNN-Transformer-Based Cooperative Spectrum Sensing in Cognitive Radio Networks

Recently, deep learning (DL)-based data-driven spectrum sensing has attracted much attention due to its powerful capabilities to exploit underlying signal features. Existing DL-based cooperative spectrum sensing (CSS) detectors usually extract the signal features from each secondary user (SU) individually, followed by feature fusion at a central fusion center (FC). However, this cannot extract correlation features across SUs. To address this limitation, this letter proposes a CNN-Transformer neural network to exploit the correlations among the received signals across and within SUs. Leveraging the CNN-Transformer, we propose a posteriori probability ratio detector, which is optimal for the binary hypothesis testing problem under the Neyman-Pearson criterion. Moreover, unlike most DL-based CSS detectors, the proposed detector considers imperfect and independent reporting channels between SUs and the FC, and exploits the temporal correlation of signals among neighboring samples. Numerical results are provided to demonstrate the superior performance of the proposed detector, showing an improvement of approximately 6 dB over traditional model-driven energy detectors. Moreover, the proposed detector improves the detection probability to 0.99 from 0.88 of data-driven detectors under Rician fading channels.