Multimodal attention-based deep learning for automatic modulation classification

Wireless Internet of Things (IoT) is widely accepted in data collection and transmission power system, with the prerequisite that base station wireless IoT be compatible a variety digital modulation types to meet requirements terminals different modes. As key technology communication, Automatic Modulation Classification (AMC) manages resource shortage improves spectrum utilization efficiency. And for better accuracy efficiency classification signal modulation, Deep learning (DL) frequently exploited. It found real cases signal-to-noise ratio (SNR) signals received by remains low due complex electromagnetic interference from equipment, increasing difficulties accurate AMC. Therefore, inspired attention mechanism multi-layer perceptron (MLP), AMC-MLP introduced herein as novel AMC method SNR signals. Firstly, sampled I/Q converted constellation diagram, smoothed pseudo Wigner-Ville distribution (SPWVD), contour diagram spectral correlation function (SCF). Secondly, convolution auto-encoder (Conv-AE) used denoise extract image feature vectors. Finally, MLP employed fuse multimodal features classify model utilizes characterization advantages images modes boosts Results simulations on RadioML 2016.10A public dataset prove well provides significantly range than other latest deep-learning methods.