Research on Signal Modulation Classification under Low SNR Based on ResNext Network

Abstract

To address the shortcomings of existing methods such as low recognition accuracy and poor anti-interference performance under low signal-to-noise ratios, this paper proposes the RFSE-ResNeXt (Residual-fusion squeeze–excitation aggregated residual for networks, RFSE-ResNeXt) network. In this paper, we improve the residual structure of the network based on the ResNeXt network and then introduce the compressed excitation structure to improve the generalization ability of the network. The improvement of the residual structure of the network leads to a good improvement in the overall recognition accuracy of the network; meanwhile, the compressed excitation structure improves the confusion phenomenon when the network faces complex signals with low signal-to-noise ratios. The experimental results show that the proposed network improves the recognition accuracy by 4% on average at a very low SNR of -10dB and reduces the misclassification of AM-DSB into CPFSK by about 27%.