Deep Neural Networks for Jamming and Interference Classification in 5G Physical Layer
Romppanen, Viljami (2023)
2023
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-11-17
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202310168840
https://urn.fi/URN:NBN:fi:tuni-202310168840
Tiivistelmä
The fifth generation (5G) of cellular networks is bringing major performance improvements and connecting new industries with wider application areas than the previous generations. The exposed nature of every wireless technology, including 5G, makes them vulnerable to being interfered. Interference can be intentional by jamming attacks or unintentional by other devices in the network. These physical layer threats can cause denial-of-service problems in the network. To ensure the security and availability of 5G communications, it is important to develop identification methods for jamming and interference.
In this thesis, three deep learning (DL) approaches are proposed for classifying different jamming and interference models. To train and evaluate those DL approaches, a 5G spectrogram-related jamming and interference dataset is generated. The proposed DL architectures are convolutional neural network (CNN), long short-term memory (LSTM), and combined CNN-LSTM. The goal is to find out which DL model works the best and if there is a significant difference between two different classification tasks. Those tasks are binary and multi-class classifications. Binary classification denotes whether the input is jammed or not, whereas multi-class classification recognizes the type of the jamming model.
CNN and CNN-LSTM results are almost identical with binary classification accuracies around 97% and multi-class classification accuracies around 90%. LSTM is notably worse with accuracies around 90% and 70% respectively. Overall, all models show sufficient performance, but CNN performs the best in terms of results and efficiency. Also, performance differences between the two classification tasks are not alarming, so either task is suitable for the proposed approaches.
In this thesis, three deep learning (DL) approaches are proposed for classifying different jamming and interference models. To train and evaluate those DL approaches, a 5G spectrogram-related jamming and interference dataset is generated. The proposed DL architectures are convolutional neural network (CNN), long short-term memory (LSTM), and combined CNN-LSTM. The goal is to find out which DL model works the best and if there is a significant difference between two different classification tasks. Those tasks are binary and multi-class classifications. Binary classification denotes whether the input is jammed or not, whereas multi-class classification recognizes the type of the jamming model.
CNN and CNN-LSTM results are almost identical with binary classification accuracies around 97% and multi-class classification accuracies around 90%. LSTM is notably worse with accuracies around 90% and 70% respectively. Overall, all models show sufficient performance, but CNN performs the best in terms of results and efficiency. Also, performance differences between the two classification tasks are not alarming, so either task is suitable for the proposed approaches.