Modulation Classification for Overlapped Signals via Clustering Analysis of Super-Constellations

Modulation recognition (MR) plays an important role in military and civilian applications of cooperative and non-cooperative communications. The existing literature has introduced several MR methods for single-user scenarios but few papers have studied multi-user MR. This work proposes an MR method that employs a clustering analysis of super-constellation in a completely overlapped MU scenario. A super-constellation refers to the mapping of superposed symbols in the I/Q plane. A blind MR for stealthy decoding conversation between two users is considered with parameters like user gain, noise variance etc. being unknown in practical impairments such as carrier frequency offset, timing and phase offsets. The proposed algorithm utilizes agglomerative hierarchical clustering along with various cluster validation techniques to determine the optimal number of clusters and their respective centroids within the super-constellation. Subsequently, amplitude and phase-based features are extracted from these centroids to enable accurate MR. The simulation results demonstrate that the classification accuracy of the proposed method is i) significantly better than the features-based methods like cumulants and higher-order statistics; and ii) comparable with the deep learning-based methods that crucially rely on the availability of training data. Furthermore, our analysis reveals that the proposed method has significantly lower complexity than the existing techniques.