Modulation Recognition based on One-Dimensional Complex Convolutional Networks under Low SNR

With the increasing complexity of electromagnetic environments, improving the accuracy of signal modulation recognition under low signal-to-noise ratio (SNR) has become a challenge. To address this issue, this paper proposes a one-dimensional complex convolutional network (1DCCN) for processing raw complex signals and recognizing their modulation types. Firstly, a mapping relationship from real convolution to complex convolution is sought. Secondly, multi-scale complex convolution is used to enhance the perception ability of the network. Subsequently, a sub-sampling complex convolution layer is designed to replace the pooling layer for data down-sampling. Finally, the network model is verified by using real measured signals. Experimental results show that FSK, BPSK, QPSK, 8QAM and 16QAM modulation signals can be effectively recognized under low SNR. When the SNR is -5dB, the average recognition rate of the five modulation signals can reach 91%.