Deep Learning Approach for Automatic ModulationRecognition in Cognitive Radio

Abstract

Intelligent radios collect information by sensing signals within the radio spectrum, and the automatic modulation recognition (AMR) of signals is one of their most challenging tasks. Although the result of a modulation classification based on a deep neural network is better, the training of the neural network requires complicated calculations and expensive hardware. The proposed system consists of Convolutional Neural Network (CNN) combined with a deep learning-based method, to achieve higher-accuracy AMR that are trained on different datasets. The proposed system adopt dropout instead of pooling operation to achieve higher recognition accuracy. The proposed system is combined with CNNs and is designed for recognition of eight modulation modes of BPSK, QPSK, 8PSK, GFSK, CPFSK, PAM4, 16QAM, and 64QAM.These modulation modes are widely used in modern communication systems, including optical communications and satellite communications. When unknown signals are detected, the initial CNN trained on IQ samples is employed to recognize easily distinguishable modulation modes except 16QAM and 64QAM. This CNN does not have the capacity to distinguish between them, but it can separate them from other modulation modes. Therefore, they are categorized into the same class (QAMs), from which the other CNN trained on constellation diagrams can distinguish 16QAM and 64QAM. The experimental results show the efficiency of each of the modulation modes. The system reveals that BPSK modulation has 97% accuracy, PAM4 has 91% accuracy, PAM8 has 88%, PSK4 and PSK8 both have 93% accuracy, QAM4 has 94%, QAM16 has 81% and the final one QAM64 has 87%. The overall system undergoes 90% accuracy which the best value to be measured.