Classification Model of Wireless Signals Based on Higher-order Statistics

Automatic modulation classification technology is an indispensable step in cognitive radio, and its recognition accuracy is related to the orderly progress of subsequent communications. In this paper, we introduce the higher-order statistics into residual neural network for the precise classification of different modulation types. The classification technology can recognize 24 digital and analog modulation types under both synthetic simulated channel effects and over-the-air recordings. We also consider a rigorous baseline method using residual neural network and compare performance between two approaches under a wide range of signal-to-noise ratio. Experimental results show that our proposed method achieves an average accuracy of 96.4% and obtains better performance in correct classification probability than the baseline method, especially in lower signal-to-noise ratio.