Method for the Automatic Modulation Classification for the Sums of Phase- Shift-Keying Signals Based on Polynomial Features

This paper considers the task of automatic modulation classification for the sums of phase-shift keying (PSK) signals which occupy the same frequency bandwidth and have the same modulation. This problem is an important step which precedes the subsequent processing of such signals. For the solution of this classification problem we implement the method based on two machine learning tools: logistic regression or two-layer feedforward neural network. It allows to determine the class to which the sum of two PSK signals belongs. As the features used for the input of these classifications tools, we use second-order polynomial features calculated based on higher order cumulants of the received signal. The results of numerical experiments using different channel parameters and noise levels show the advantage of implementing second-order features over the raw features both in the case of logistic regression and two-layer feedforward neural network.