Classification of RF Signals Based on Image and Sequence Inputs

Radio frequency (RF) signals are widely used in commercial and military wireless communications, and the accurate classification of such signals is of great theoretical significance and practical application value. This study aims to solve the practical problem of RF signal classification based on a 5-classified signal dataset (abbreviated as 2021 dataset) provided by a company. In this paper, we design ResNet18 based on constellation map input, ResNet18 based on hybrid map input and CNN network based on sequence input, and compare and analyze the performance of the deep learning algorithms by exploring the deep learning algorithms under two input modes: image and sequence. It is shown that the classification accuracy of ResNet18 using hybrid graph input reaches 95.79%, while the CNN model with sequence input performs better in terms of classification accuracy and real-time performance, with an accuracy of 98.22%, and the number of parameters is only about 1/8 of that of ResNet18.