This paper discusses the automatic modulation classification of weak communication signals using distributed low-cost sensors. The concept of a secondary user in sensor networks is presented and very high-order statistics are used as modulation features. Two feature-based methods, single-variable and multivariable modulation classifiers, are proposed for estimating unknown modulation schemes through single-input multiple-output signal sensing channels.
The new approaches acquire multiple signal observations collected from distributed sensors and leverage the channel diversity to enhance signal power and reduce bias in estimation. The experiment demonstrates that the network centric modulation classifier achieves significantly improved performance in terms of probability of correct classification than the current state-of-the-art single sensor modulation classifier, and the multivariable modulation classifier is more robust to the channel parameter variations than the single-variable classifier.