Deep Learning Based Modulation Classification in Radio Access Network

The growing demand for traffic shaping and manipulation for efficient last-mile coverage has driven extensive research using emerging artificial intelligence to overcome the capacity hurdle in next-generation wireless systems. In the radio access network (RAN), modulation format can be critical information to determine the channel condition and implement resource allocation and data compression so as to improve the quality of service and spectral efficiency of data transmission. This paper proposes a novel deep learning-based automatic modulation classification algorithm that trains neural networks as classifiers to determine the modulation format of signals transmitted in the RAN. To conquer the problem of limited training data with the received dataset, the proposed algorithm exploits a signal generation method that generates massive simulated data for training instead of relying on the received signal stream. To eliminate the issue of inconsistent data distribution between the training and inferring stages, data augmentation is applied to increase the diversity of training data and enrich the training dataset. The experiment results demonstrate that the proposed algorithm accurately identifies the modulation formats of signals received from the RAN.