Signal modulation waveform recognition method based on STF-Net

Signal modulation waveform recognition is one of the key technologies in the field of spectrum cognition and an important means to achieve monitoring and control of spectrum resources for low-orbit satellites. To address the issues of high parameter count and computational complexity in existing deep learning-based modulation waveform recognition methods, a lightweight signal modulation waveform recognition method based on space-time fusion network (STF-Net) is proposed. The method first preprocesses the signals into dual-channel data in the time-frequency domain. It then utilizes convolutional neural network (CNN) to extract signal spatial features and reduce feature redundancy. Long short-term memory (LSTM) is employed to capture temporal information and output recognition results. Experimental results show that the proposed method achieves an average recognition accuracy of 91.79% for modulation waveforms when the signal-to-noise ratio is greater than 0dB. Compared with equivalent methods, the proposed method reduces the parameter count by 96% and improves efficiency by 2.7 times.