MODULATION CLASSIFICATION BASED ON AUGMENTED CONSTELLATION DIAGRAMS

Modulation Classification (MC) is an increasingly relevant design feature in wireless communications and plays an essential part in cognitive radio networks. Deep learning methods are decisive in performing MC. MC methods based on constellation diagrams usually achieve excellent performance because of the constellation diagrams' discriminative characteristics. This paper uses deep learning models to classify generated signals' constellation diagrams by their modulation type. We propose a constellation diagram-based MC architecture that uses different training and testing resolutions to classify the modulations of the RadioML dataset. The observed improvement of the classification accuracy relies on the fact that a lower training resolution improves the classification at test time. We also perform a comparative analysis of the model by examining the impact on the classification accuracy when different levels of resolution are applied.