Modulation Recognition Algorithm Based on ResNet50 Multi-feature Fusion
Xianghui Liu, Zhongdong Wu, Chunyang Tang · 2021
In order to solve the problem of low accuracy of modulation pattern recognition under the condition of low signal-to-noise ratio in traditional modulation pattern recognition methods, this paper proposes a modulation recognition algorithm based on ResNet50 multi-feature fusion. This method uses the ResNet50 network that introduces an attention mechanism, and on this basis, through the extraction of three data characteristics of IQ signal, signal spectrogram and constellation diagram, and the multi-feature fusion of the three characteristic data as the input of the network. By comparing the improved ResNet5 and ResNet50, VGG-19, VGG-16 under the conditions of -10dB, -5dB, 0dB, 5dB, and 10dB in different signal-to-noise ratios. The experimental result shows that the improved method can effectively identify 7 types of digital signal modulation modes, namely 2FSK, 4FSK, BPSK, QPSK, 8PSK, 16QAM, and 64QAM under the condition of low signal-to-noise ratio. Especially when the signal-to-noise ratio is -5dB, the average recognition rate of the ReSNet50 network based on multi-feature fusion can reach 93.41%.