Classification of advanced radio modulation techniques using deep learning networks

Abstract

This work focuses on the automatic classification of analog and digital modulation schemes in modern wireless communication systems using both deep learning and machine learning techniques. A thorough experimental study was conducted using the RadioModRec-1 dataset, which includes a variety of real-world modulation types under different signal conditions. Three models were implemented and compared: a Convolutional Neural Network (CNN), Random Forest, and eXtreme Gradient Boosting (XGBoost). The results demonstrate that XGBoost achieved the highest classification accuracy of 98.45%, followed closely by Random Forest with 97.32%, while the CNN reached an accuracy of 70.64%. These outcomes confirm the strong performance of ensemble learning methods in structured signal environments, while also highlighting the adaptability of DL models in handling raw input data. Overall, this study emphasizes the potential of Artificial Intelligence driven approaches to improve the efficiency, accuracy, and robustness of modulation recognition, contributing to the advancement of intelligent and reliable wireless communication systems.