A Cognitive Radio Spectrum Sensing Implementation Based on Deep Learning and Real Signals

In a cognitive radio environment, spectrum sensing is an essential phase for improving spectrum resources management. Based on a deep learning method and real signals, a new spectrum sensing implementation is proposed in this work. The real signals are artificially generated, using an ARDUINO UNO card and a 433 MHz wireless transmitter, in ASK and FSK modulation types. The reception interface is constructed using an RTL-SDR receiver connected to MATLAB software. The signals classification is carried out by a convolutional neural network (CNN) classifier. Our proposed model’s main objective is to identify the spectrum state (free or occupied) by classifying the received signals into a licensed user (primary user) signals or noise signals. Our proposed model’s performance evaluation is evaluated by two metrics: the probability of detection (Pd) and the false alarm probability (PFA). Finally, the proposed sensing method is compared with other used techniques for signal classification, such as energy detection, artificial neural network, and support vector machine. The experimental results show that CNN could classify the real signals better than traditional methods and machine learning methods