Deep Learning Based Spectrum Sensing Method for Cognitive Radio System

Abstract

Cognitive radio (CR) network is the promised paradigm to resolve the spectrum shortage and to enable the cooperation in heterogeneous wireless networks in 5G and beyond. CR mainly relays on Spectrum Sensing (SS) strategy by which the vacant spectrum portion is identified. Therefore, the sensing mechanism should be accurate as much as possible, as long as the subsequent cognition steps are mainly depended on it. In this paper, an efficient and blind SS algorithm called Deep Learning Based Spectrum Sensing (DBSS) is proposed. This algorithm utilizes the deep learning approach in SS by using Convolutional Neural Network (CNN) as a detector instead of energy thresholding. In this algorithm, the computed energies of the received samples are used as dataset to feed the optimized CNN model in both training and testing phases. The proposed algorithm is simulated by MATLAB, the simulation scenarios divided into: CNN optimization (training) and SS. The last scenario shows the detection ability of the proposed algorithm for PU under noisy environment. The simulation results show that the proposed algorithm reached high detection probability (Pd) with low sensing errors at low SNR. In addition, high recognition ability to identifying Primary User (PU) signal form noise only signal is achieved as well. Finally, the proposed algorithm is validated with respect to real spectrum data that supported by SDR in an experimental signal transmission and reception scenario.