Research on Modulation Signal Detection Method based on Deep Learning

Abstract

Facing the complex electromagnetic environment, the modulation mode of communication signal is becoming increasingly complex. The existing detection methods of modulation mode of communication signal can not detect the modulation mode of communication signal accurately and quickly. In order to facilitate the presentation, we represent the digital signal on the complex plane, form the constellation map according to the mapping formula, analyze the difference of the characteristics of the constellation map, and train and test the constellation map. It can be found that when the signal-to-noise ratio is lower than 20dB, the classification accuracy of the characteristics of the constellation map is greatly affected for the 64QAM signal with the largest number of points and the smallest radius. To solve this problem, A method of signal constellation de-noising using VMD is proposed. Compared with the pre-de-noising method, the average accuracy of VGGNet-16 classification is increased by 7.76%; The average accuracy rate of ResNet-18 classification increased by 9.77%; The average accuracy rate of ResNet-50 classification increased by 7.57%. This method improves the accuracy of constellation classification detection, which is difficult to improve, and lays a good foundation for the research of modulation signal detection methods.