Open Set Modulation Recognition for Robust Channel Based on Multimodal Information Fusion
Maomao Zhang, Guofeng Wei, Peng Tang, Guoru Ding, Huali Wang · 2024
In the field of wireless communication, modulation recognition is crucial for ensuring the accurate demodulation of signals. However, current open-set modulation recognition (OSMR) algorithms typically assume that the training and testing data originate from the same or similar channel distribution, which is often not the case in practical scenarios. When the test data encounters a significantly different channel environment, such as varying noise levels, multipath effects, or interference types, the shift in channel distribution leads to differences in data distribution, thus impacting modulation recognition performance. We propose an OSMR algorithm based on multimodal information fusion for adversarial training (MFAT-OSMR). Firstly, we employ multi-modal information and a convolutional block attention module (CBAM) for feature extraction and representation learning. By integrating information from multiple transform domains, our model can gain complementary advantages and enhance its ability to represent key features of the data through the attention mechanism. To reduce the disparity between the source domain and the target domain, adversarial training is employed. By incorporating an adversarial classifier in the feature space, it prompts the feature extraction network to acquire a more universal feature representation. Enhancement of the model’s ability to adapt to unfamiliar channels. We validate the effectiveness of the proposed algorithm through experiments. The experimental results demonstrate that our algorithm showcases superior recognition performance and increased robustness across various channel environments.