Deep Learning Scheme for Modulation Classification under Fading Channels

The field of automatic modulation classification using deep neural networks has undergone significant development in recent years. In this research, M-QAM and N-PSK modulation formats have been considered for classification. The deep neural network architecture is directly fed with raw signals that are often corrupted with channel noise and fading effects such as Rayleigh and Nakagami. The proposed classifier structure also incorporates normalized cumulants of higher order from the received signal. The proposed DNN architecture’s performance is evaluated by comparing its efficiency for raw data and extracted features-based classification. We have validated the classifier efficiency through three different cases. Our proposed classifier demonstrates remarkable capability in efficiently classifying modulation formats.