Robust classification of quadrature amplitude modulation constellations based on GMM

The performance of a maximum likelihood (ML) based modulation classifier is highly sensitive to whether the model matches the real situation. In this paper, the quadrature amplitude modulation (QAM) signals are concerned and a ML modulation classifier which is robust to frequence offset, phase jitter, amplitude fluctuation and time delay is proposed. A new model is proposed to represent constellation as a Gaussian Mixture Model (GMM). The Gaussian Discrimination Analysis (GDA) algorithm is used to estimate the parameters of the GMM by offline data training. Maximum Likelihood criterion is used to classify the modulation of real intercepted data. Numerical results show the superiority with respect to robustness of this new model and reasonably good performance under additive white gauss noise (AWGN).