Radio Signal Classification using Deep Learning

Abstract

Radio signal classification is a modulation recognition that are used by a wide collection of applications in radio communications and electromagnetic spectrum management. It is the process of deciding, based on observations of the received signal, what modulation is being used at the transmitter. Significant progress has been made in this area using Deep Learning (DL). In recent years, DL has shown success in solving radio signal classification problems. There are two important types of Neural Networks (NN) in DL, Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). Long Short-Term Memory (LSTM) is one of the most popular RNN architectures that perform well in classifying signals. The aim of this work is to determine the appropriate NN model architecture that achieves good performance and high accuracy of modulation classification of the signals. Thus, in this document we tried two different approaches. The first one uses CNN, while in the second we combine CNN with LSTM in order to perform classification. The dataset DeepSig:RadioML2016, is used for the performance analysis. Experiment results shows that the use of the second approach of LSTM-CNN achieved better performance compared to the first one that use only CNN.