Modulation Classification on Radio Communication Using Convolutional Neural Network

Abstract

Modulation type classification for radio communications signals is purposed and examined the effect of deep learning on the classification performance in this paper. In this work, search how utilize a convolutional neural network (CNN) for classification of modulation is searched. Firstly, artificial, impaired channel waveforms are created. A CNN is trained that used the generated waveforms as data, then, modulation classification is realized. In the end, Software-defined radio (SDR) hardware and over-the-air signals are utilized to test the CNN. Different effects of offset of carrier frequency, rate of symbol, multipath fading, delay spread, and thermal noise in the simulation are considered. All of these impacts can be fully modeled decently and also, there are many additional dispersion influences that can be shaped artificially except within the scope of our work here. The CNN that is trained in this paper identifies such as; Binary phase-shift keying (BPSK), Quadrature phase-shift keying (QPSK), (8-PSK), 16-quadrature amplitude modulation (QAM), 64 QAM, PAM4 (4-pulse amplitude modulation), GFSK (Gaussian frequency-shift keying), CPFSK (Continuous phase frequency shift keying), B-FM (Broadcast FM), DSB-AM (Double sideband amplitude modulation), SSB-AM (Single sideband amplitude modulation). The trained CNN predicts and makes comparisons about the modulation sort of each frame for 1024 and 512 channel-impaired samples in this paper. Trade-offs of the performance shown in this paper consider principal parameters in data symbols and training symbols generation, enhance sensing and focus future workings on the optimization of such systems. © 2022 IEEE.