Complicated Interference Identification via Machine Learning Methods

Under complicated electromagnetic environments, useful signals often interfere with other electromagnetic systems nowadays. Then how to suppress the interferences is a critical problem. Since there is no universal suppression method for each kind of interferences, we need to identify the interferences, and then perform the specific suppression method to tackle with the interferences. In this paper, we simulate 15 types of possible interference signals, some are mixed by two different kinds of interferences, and extract seven features in the time-frequency domain under different interference-to-noise ratios (INRs). By fully analyzing these features, several common classifiers, such as random forest, gradient boosting, and a neural network, are designed to identify these complicated interferences. Numerical experiments are provided to demonstrate the effectiveness of the proposed methods. The final accuracy can exceed 95.20% at the condition INR=20dB.