A hybrid VGG16-LSTM architecture for automatic signal modulation classification

Abstract

Traditional modulation recognition methods often show limited robustness in complex electromagnetic environments, especially under multipath fading, Doppler shifts, and non-stationary noise. To address this issue, this paper investigates a hybrid VGG16-LSTM architecture for automatic signal modulation classification. In the proposed framework, Rayleigh fading is introduced to simulate complex channel conditions, VGG16 is employed to extract spatial features, and an LSTM module is used to capture temporal dependencies in signal evolution. Experiments are conducted on five modulation schemes, namely BPSK, QPSK, 8PSK, 16QAM, and 64QAM, under different signal-to-noise ratio (SNR) conditions. The results show that all compared models achieve better classification accuracy as the SNR increases. Under the evaluated experimental setting, the proposed VGG16-LSTM model achieves stable performance and obtains slightly higher average accuracy than the compared baseline models. These results indicate that combining convolutional feature extraction with temporal modeling is a feasible approach for modulation recognition in fading and noisy channel conditions.