ResNet Based Modulation Pattern Recognition Method in Low SNR

Modulation pattern recognition is an important research direction in communication systems, and plays an important role in electromagnetic spectrum management, communication signal interception and electronic countermeasures. In order to solve the problem of low modulation pattern recognition performance under low signal-to-noise ratio(SNR), a wavelet fusion noise reduction algorithm is proposed. The algorithm uses the Optuna hyperparameter optimization framework to dynamically construct a wavelet basis function and a wavelet threshold search space, and converts the optimal wavelet basis function. and the optimal wavelet threshold range as a parameter, the original IQ signal is denoised and then fused with the IQ signal to form a new data input neural network, which retains more details of the original IQ signal while denoising the signal. The residual neural network(Resnet) obtains a squeeze and excitation transpose one-dimensional ResNet6, and the squeeze and excitation transpose block is added to the one-dimensional residual block to perform secondary noise reduction on the signal, which improves the recognition accuracy of the modulation pattern under low SNR. The effectiveness of the method is verified by extensive experiments under the open source dataset RML2016.10a.