Primary User Detection in Cognitive Radio using Spectral-Correlation Features and Stacked Denoising Autoencoder

This paper proposes a primary user detection method in cognitive radio system based on the features of cyclic spectral correlation and the stacked denoising autoencoder. The cyclic spectral correlation, which is sensitive to OFDM signal, is utilized as the feature-extraction tool. Then, we use the SDAE as a classifier due to its high classification accuracy. Concretely, we suppose that OFDM is applied by PU because OFDM is more commonly used than other single carrier modulations. Therefore, our mission turns to the classification of OFDM signal. In addition, a long symbols sequence is not required for this method simplifying the detection procedure and rendering rapid detection more achievable. The results of the probability of successful detection, the change in the probability of successful detection corresponding to different corruption levels are presented, demonstrating significant performance advantages over energy detection.