Satellite Communication Signal Recognition Based on Multi-Scale Recurrent Convolutional Neural Network
Addressing the current oversight of feature fusion across different scales in satellite commu-nication modulation classification algorithms,we propose a Multi-Scale Recurrent Convolutional Neu-ral Network(MSRC)model.This network integrates dual-branch design,compression and excita-tion strategy,multi-scale residual networks,and long short-term memory networks,aiming to com-prehensively capture the multi-scale features of signals and effectively model time sequences.Experi-mental results demonstrate that the proposed model achieves a recognition accuracy of 97.1％above 0 dB,further increasing to 99％at 13 dB.In comparison to classical CNN2 and LSTM2 models,our model exhibits significant advantages in recognition accuracy.Moreover,compared to the closely-performing CLDNN2 model,our model reduces parameter quantity by 47.7％and shortens training time by 68％.Notably,the recognition accuracy of QAM 16 and QAM64 modulation styles notably improves and maintains a high level,further confirming the effectiveness of the model's multi-scale feature fusion strategy.
万方数据
维普
