A software-defined radio testbed for deep learning-based automatic modulation classification

摘要

Automatic modulation classification (AMC) is the demodulation process on the receiver side, which is a crucial protocol for current and next-generation intelligent communication systems. This method becomes complicated, in the presence of channel noise, to identify the modulation of the transmitted signal, that is, the transmitter and receiver with its ambiguous parameters like timing information, signal strength, phase offset, and carrier frequency. Two fundamental approaches are used for the AMC, namely, the signal statistical feature-based approach and the maximum likelihood approach. Current Feature-Based AMC approaches typically built for a limited set of modulation; a comprehensive AMC approach utilizing convolutional neural networks (CNN) is suggested in this article to overcome this obstacle. Altogether, 11 different types of modulations considered. In this method, without an extraction function, the transmitted signal can be identified directly. Also, the features of the received signal are known directly by using this method. The classification accuracy using CNN seems to be remarkable, especially for low SNRs. In this article, a realistic AMC framework that can be quickly applied to provide reliable efficiency in numerous commercial real-time scenarios has developed and tested. Therefore, to prove the functional viability of our proposed model, it was applied to the software-defined radio test-bed.