Robustness enhancement of distribution based binary discriminative features for modulation classification

In this paper, we propose distribution based binary discriminative features and a novel feature enhancement process for automatic modulation classification. The new features exploit the signal distribution mismatch between two modulations. Signal distributions on I-Q segments, amplitude and phase, are considered to produce a comprehensive feature set for improved robustness. Logistic regression is used to reduce feature dimension and enhance classification robustness. To accomplish multi-class classification, a class oriented feature space is created for the K-nearest neighbours classifier. The test results show that the proposed method is able to achieve excellent performance in simulated environments.