Deep Learning Approach for Wireless Signal and Modulation Classification

Abstract

This paper aims to classify signal and modulation classes of a given wireless signal with high accuracy using a model having a low number of parameters. We propose an end-to-end method to classify a wireless signal based on its signal and modulation type using a CNN-based architecture. The proposed architecture is similar to that of the LeNet-5. Firstly, we implement signal and modulation classification using a decision tree, followed by a random forest algorithm, classic examples of machine learning(ML) based algorithms. Since our dataset is a time series, we also implement using RNN-LSTM based model for the classification. The proposed model has fewer parameters than that of the CNN-based, RNN-LSTM based architectures. Moreover, it achieves better accuracy for a wide range of signal-to-noise ratios than a decision tree, random forest, RNN-LSTM based classifiers. © 2021 IEEE.