Modulation Recognition: An Unsupervised Learning Approach

Modulation recognition is an integral part of many communication systems. For example, it is widely used in cooperative communication, and stealthy decoding a signal in non-cooperative communication. This work proposes a modulation recognition method using unsupervised learning aided clustering analysis of constellations in single-user (SU) and multi-user (MU) scenarios. The idea is to pose modulation classification as a pattern recognition problem in the I/Q plane, and then design modulation classifiers by exploiting patterns formed by the constellations. The proposed algorithm classifies SU versus MU mode of communication in the first step, followed by recognising modulation used by the users. Simulation results show that the designed method achieves more than 90% average accuracy of correct classification above 8 dB SNR, and significantly outperforms the existing methods.