Hardware Implementation of Automatic Modulation Classification with Deep Learning

Satish Kumar, Anurag Singh, Rajarshi Mahapatra

Abstract: Very recently, neural network based deep learning (DL) models are being applied to solve the problems of communication system. In this work, we design a deep neural network (DNN) based automatic modulation classification (AMC) scheme for wireless communication scenario. In particulars, we have implemented a DL network (ModNet) to identify the modulation scheme of received signal in real time using universal software radio peripheral (USRP) hardware. To identify the correct modulation scheme, the constellation diagram of the received complex symbols is plotted in image format. Thereafter, this image is send to pre-trained ModNet classifier for classification. Based on the constellation image and intelligence gained during training, ModNet determines the modulation scheme of the received signal. The proposed ModNet architecture is based on well-known image classification network AlexNet. Simulation result shows, the proposed modulation detection scheme provides higher classification accuracy than other existing modulation classification methods. The same has been verified through hardware implementation.

External IDs:dblp:conf/IEEEants/KumarSM19a