Automatic modulation classification, the procedure between signal detection and signal demodulation, is a major task for communication engineers to study of the modulation format of detected signals, with multiple civilian and military utilization. The technique of modulation classification has helped the military detect and decode hostile signals from the other side, and improved the efficiency of the Software Defined Radio during the transmission of the message. At the same time, without the information of the signal power, frequency offset, and channel deviation, this job is full of challenge. Thus, the investment in automatic modulation classification is worth the price.
In this graduate project, decision tree and deep learning have been applied to the automatic modulation classification.
For the decision tree method, the classification accuracy is above 90% when SNR is beyond 11 dB, while in deep learning method, the classification is above 90% when SNR is beyond 2dB. With applying the methods on stimulation signals and an open dataset RML2016.10a, the reliability and validity of the two algorithms have been satisfied.