Robust DNN-Enabled Cooperative Spectrum Sensing

In this contribution, we investigate cooperative spectrum sensing in cognitive radio networks using a deep neural network approach. Our proposed architecture leverages the strengths of both a one dimentional convolutional neural network and a bidirectional long short-term memory network. This combined approach aims to effectively learn the activity patterns of primary users for accurate spectrum sensing. The signal generation process incorporates diverse modulation techniques commonly used in communication systems. Additionally, the channel model considers complexities encountered in real-world deployments. These complexities mimic the challenges of multipath propagation and relative motion, making the training data more realistic. The performance of the proposed deep neural network is compared to benchmark methods using only CNN or LSTM at the secondary user level. We evaluate the models using receiver operating characteristic curves and probability of detection for various signal-to-noise ratio levels. The simulation results showcase the effectiveness of the proposed DNN, achieving superior probability of detection compared to the benchmarks and potentially surpassing state-of-the-art methods, particularly demonstrating very good performance at very low SNR values.