Automatic Modulation Recognition Based on Dual-Channel Fusion Hybrid Neural Network

Aiming at the shortcomings of low recognition accuracy, few recognition types, and poor anti-interference ability in existing automatic modulation recognition methods, a new modulation recognition method based on dual-channel fusion hybrid neural network (DCFHNN) was proposed. By utilizing In-phase/Quadrature (I/Q) data and Amplitude/Phase (A/P) data as inputs to the DCFHNN, the complementary information in different domains can enrich the feature dimension. The spatial features of the dual-channel data are extracted by the deep residual shrinkage network (DRSN), then fuse the extracted features and sent to the bidirectional gated recurrent unit (BiGRU) to extract temporal features. Furthermore, adding the attention mechanism helps to improve the effectiveness of feature extraction by enhancing the focus on key features. Experiments are evaluated on open source RadioML2016.10a dataset, the experimental results demonstrate that the proposed method outperforms other deep learning methods. When the signal-to-noise ratio (SNR) is above OdB, the classification accuracy can reach 92.3%, effectively improving automatic modulation recognition's accuracy and robustness.