Deep-Learning Hopping Capture Model for Automatic Modulation Classification of Wireless Communication Signals

Recent years have witnessed a surge of developments in deep learning (DL) motivated by a variety of contemporary applications. The conventional DL-based automatic modulation classification (AMC) methods are always relying on a great quantity of data. In this article, we propose a DL-based AMC model with short data for the spectrum sensing of wireless communication signals. First, a hopping transform unit is proposed to represent the transient variation occurred either by frequency, amplitude, or phase modulations. Second, a bidirectional long short-term memory-based hopping feature perception model, namely deep-learning hopping capture model (DHCM), is built for the AMC. A comprehensive comparison of the DHCM with other existing methods is then provided under various signal-to-noise ratios. The experimental results demonstrate the superiority of the proposed method under short data.