Modulation Classification of MQAM Signals Based on Gradient Color Constellation and Deep Learning

Modulation classification is a key issue in noncooperative communication systems, and signal constellation images can be used as input features of deep learning (DL) networks for classification. However, the conventional gray constellation image cannot exactly reflect density and location information of constellation points. To solve this problem, this paper proposes a gradient color constellation (GCC) algorithm based on the density of constellation points, which converts the density of constellation points into color data to realize its visualization, and uses two deep learning network models, i.e., the modified convolution neural network (M-CNN) and the residual network (ResNet), as classifiers. The experimental results show that, compared with the scheme based on gray constellation, the overall classification accuracy of the seven multilevel quadrature amplitude modulation (MQAM) signals under low signal-to-noise ratios (SNRs) is improved by 3%-4%.