Spatial-Temporal Hybrid Feature Extraction Network for Few-Shot Automatic Modulation Classification

Abstract

Automatic modulation classification (AMC) plays an important role in wireless spectrum monitoring. Motivated by the success of deep learning in various informatics domains, many AMC methods based on deep learning have been proposed. However, they usually require a large amount of labeled training samples for each category of modulation, which is hardly applicable in real-world AMC tasks. To tackle this issue, this paper proposes a novel few-shot learning framework, namely the spatial-temporal hybrid feature extraction network (STHFEN). In STHFEN, two feature extraction networks are designed to map the wireless communication signals into the spatial feature space and the temporal feature space, respectively. Then, a hybrid inference classifier is designed to combine the classification results in the two feature spaces. To train STHFEN more effectively, a hybrid loss function which promotes better inter-class separability of signals in the two feature spaces is proposed. Experimental results have demonstrated the effectiveness and robustness of STHFEN in few-shot AMC tasks.