Power of Deep Learning for Amplitude-phase Signal Modulation Recognition
Xiong Zha, Xin Qin, Yumei Zhou, Hua Feng Peng · 2019
This article presents our initial results in deep learning for signal modulation recognition in satellite communication system. In this article, we propose a signal classification model based on the recurrent neural network. Different from existing methods that need estimating the signal parameters or design the features by the experienced experts, we use LSTM (Long Short-Term Memory) to extract deep features of signal sequence from time domain. Using this architecture, we realized the modulation recognition for target signals end-to-end finally. And the method is insensitive to frequency offset and timing deviation, which has a strong application prospect in engineering. From our simulation results, the deep learning based approach can achieve considerable performances that the signals recognition rate is close to 98% in 6dB. Furthermore, the deep learning based approach is more robust than conventional methods when there is frequency offset, timing deviation and low Signal-to-noise ratio.