An autoencoder-based learning method for wireless communication protocol identification

As protocols play respective roles to fulfill different communication services, it is important to identify protocols before analyzing and managing the system. In the past decade, there have been a lot of researches on protocol identification using machine learning methods, which achieve promising results. However, the features of protocol used for identification mainly rely on engineering skill and domain expertise, which may not be available for the complicated wireless communication systems, such as encryption-based systems. In this paper, we propose an unsupervised-based learning method to make the feature extraction more intelligently and automatically. We first review the limitation of the traditional identification methods, especially the part of feature extraction. After that, an unsupervised deep learning based method, autoencoder, is proposed for automatically extracting the features of the original protocol data. Then, we construct the identification model based on the extracted features and a Support Vector Machine based classifier. Finally, experimental results show the effectiveness of the proposed method.