Quadruplet Convolution Neural Network With C3 Loss (C3-QCNN) for Signal Classification
Chen Yang, Shuyuan Yang, Zhixi Feng, Min Wang · IEEE Transactions on Instrumentation and Measurement · 2022
In this letter, a new Quadruplet Convolution Neural Network (QCNN) is proposed to cooperatively and contrastively learn deep features of raw radio signals for accurate Automatic Modulation Classification (AMC), when very few labeled samples are available. Four convolutional channels with shared weights are constructed to deal with the In-phase (I) path and Quadrature (Q) path of signals. Moreover, a new C3-loss function that contains thecooperativeloss,contrastiveloss andclassificationloss, is designed for QCNN to extract discriminative and concise features of signals. Some experiments are taken on the public dataset generated with GNU Radio, which has eleven types of modulation signals with different signal-to-noise ratios. The experimental results show that C3-QCNN achieves high classification accuracy and outperform its counterparts.