Modulation recognition algorithm using innovative CNN and cyclic-spectrum graph

In view of the defects that the existing modulation recognition algorithm of communication signals has low recognition rate and poor adaptability. A scheme for automatic modulation recognition using innovative convolutional neural network (CNN) and cyclic-spectrum graph was proposed. In our proposed scheme, cyclic spectrum theory is utilized to establish a concise features representation of a unified graph which can accurately express the statistical features of modulated signals. Then, an innovative CNN model using small convolution kernels is designed to enhance features extraction, so as to deeply analyze the features of modulated signals in graph domain. When the signal- to- noise ratio (SNR) is -2dB, the average recognition rate of proposed algorithm can reach 90.48% with limited computing resources. The final test results indicate the graph features are highly discriminant, and CNN has good performance in extracting high- dimensional features. The algorithms can be effectively applied in communication network to strengthen the adaptability of communication switching equipment.