Deep-Learning Based Modulation Identification in Wireless Communication System

Abstract

This paper studies the performance of deep learning (DL) for the identification of analog and digital modulation in orthogonal frequency-division multiplexing (OFDM) based wireless communication systems. OFDM is the key enabling technology for 4G and going to be part of 5G communication system implementation. The conventional methods for signal modulation identification (SMI) are discrete wavelet transform (DWT), adaptive wavelet transform (AWT) and, mixed parameters. These are based on statistical models and therefore encounter the bottleneck of accuracy. To handle this problem, machine learning (ML) algorithms were proposed which uses support vector machine (SVM), k nearest neighbors (KNN), decision tree (DT), etc. However, the feature extraction has to be done manually which makes these methods very difficult to implement in practical OFDM systems. Therefore many DL based SMI methods are proposed which handles the limitations of the mathematical models by making use of the availability of large datasets. In this paper, we have used a Convolutional Neural Network (CNN) for identification of both analog and digital modulation in 4G/5G wireless system. This proposed deep learning based method achieves accuracy higher than 95% which is better than the traditional methods.