Automatic Modulation Classification Based on Complex-Valued Convolutional Neural Network and Semi-Supervised Learning

TL;DR: The proposed CCNN model is capable of extracting features from the real and imaginary parts of complex-valued signals, thereby improving the classification accuracy, and the CCNN-SSL is validated on the RML2016.10A and RML2016.10B datasets, demonstrating the superiority of the proposed approach.

Abstract: In recent years, with the rapid development of 5G and intelligent communication, the proliferation of wireless terminal devices has led to the scarcity of spectrum resources. Cognitive radio, as a key technology for spectrum regulation and management, plays a crucial role in automatic modulation classification, which is an essential part of cognitive radio. Therefore, automatic modulation classification techniques have become increasingly important. This paper proposes an automatic modulation classification method based on Complex-Valued Convolutional Neural Network and Semi-Supervised Learning (CCNN-SSL). Deep learning has been a hot research topic in recent years, and its remarkable performance in image recognition has been widely recognized. It has also been found that applying deep learning to signal modulation classification achieves promising results. However, existing deep learning methods for modulation classification are primarily based on fully supervised learning, relying on a large amount of annotated data for neural network training. In practical signal recognition tasks, annotating a significant amount of data is costly. The proposed method effectively addresses this issue, and the CCNN model is capable of extracting features from the real and imaginary parts of complex-valued signals, thereby improving the classification accuracy. The performance of CCNN-SSL is validated on the RML2016.10A and RML2016.10B datasets, demonstrating the superiority of the proposed approach.