EMD and VMD Empowered Deep Learning for Radio Modulation Recognition

Tao Chen, Shuncheng Gao, Shilian Zheng, Shanqing Yu, Qi Xuan, Caiyi Lou, Xiaoniu Yang

Abstract: Deep learning has been widely exploited in radio modulation recognition in recent years. In this paper, we exploit empirical mode decomposition (EMD) and variational mode decomposition (VMD) in deep learning-based radio modulation recognition. The received IQ sequences are decomposed by EMD or VMD and the decomposed components are spliced and fed into the designed deep neural network for classification. In order to reduce the computational complexity, we further propose to dowansample the decomposed components and input these downsampled components into the network for classification. Simulation results show that the proposed methods perform far better than other transform-based methods in terms of recognition accuracy. There is also performance gain of our proposed methods over IQ-based method and the performance gain is larger when using immature or shallow network architecture for classification or the recognition is in a few-shot scenario where only small number of training samples is available. Results also show that the proposed downsampling scheme can further improve the accuracy and reduce the computational complexity at the same time with a properly chosen downsampling factor.