Spectrum Sensing for Cognitive Radio Using FLOM and CNN in Alpha Noise

The existing techniques of spectrum sensing mostly make decisions mostly using model-driven or data-driven detection approach. These methods often suffer performance degradation under non-Gaussian noise conditions, especially, alpha noise, which has no energy or other second order statistics for detection. In view of the powerful capability of fractional lower order statistics (FLOS) in solving the sensing performance degradation under non-Gaussian noise, a novel spectrum sensing algorithm based on FLOS and convolutional neural networks (CNN) is proposed. In contrast to the existing deep learning-based spectrum sensing techniques which use second order statistics such as energy and covariance matrices as their inputs, in this work, the observed data with fractional low-order moments (FLOM) is pre-processed and then the covariance matrix of FLOM is obtained. Finally, the extracted feature of the covariance matrix can be obtained by CNN and used as a basis for decision making. Simulation results show that a reasonable CNN model is feasible and the proposed algorithm has a higher detection probability than that of traditional spectrum sensing methods. In particular, when the fractional low-order coefficient p is 0.7, the probability of detection of the proposed method outperforms the CNN-based method by a factor close to 2 at a GSNR=-5dB in alpha noise.