A novel anomaly detection mechanism for Open radio access networks with Peer-to-Peer Federated Learning
Attanayaka, Dinaj (2022-12-14)
© 2022 Dinaj Attanayaka. Ellei toisin mainita, uudelleenkäyttö on sallittu Creative Commons Attribution 4.0 International (CC-BY 4.0) -lisenssillä (https://creativecommons.org/licenses/by/4.0/). Uudelleenkäyttö on sallittua edellyttäen, että lähde mainitaan asianmukaisesti ja mahdolliset muutokset merkitään. Sellaisten osien käyttö tai jäljentäminen, jotka eivät ole tekijän tai tekijöiden omaisuutta, saattaa edellyttää lupaa suoraan asianomaisilta oikeudenhaltijoilta.
Julkaisun pysyvä osoite on
Open radio access network (O-RAN) has been recognized as a revolutionary architecture to support the different classes of wireless services needed in fifth-generation (5G) and beyond 5G networks, which have various reliability, bandwidth, and latency requirements. It provides significant advantages based on the disaggregation and cloudification of the components, the standardized open interfaces, and the introduction of intelligence. However, these new features including the openness and the distributed nature of the O-RAN architecture have created new forms of threat surfaces than the conventional RAN architecture and require complex anomaly detection mechanisms. With the introduction of RAN intelligent controllers (RICs) in the O-RAN architecture, it is possible to utilize advanced artificial intelligence (AI) and machine learning (ML) algorithms based on closed control loops to perform automated security management in a data-driven manner, including detecting anomalies. In this thesis, the use of Federated Learning (FL) for anomaly detection in the O-RAN architecture is investigated, which can further preserve data privacy in a sensitive data processing system such as RAN. A Peer-to-Peer (P2P) FL-based anomaly detection mechanism is proposed for the O-RAN architecture and provides comprehensive analysis of four variants of P2P FL techniques. Three of the models are based on secure multiparty average computing, and the other is a homomorphic averaging-based model that provide protection against semi-honest local trainers. Moreover, the proposed models are simulated using the UNSW-NB15 dataset in a Python environment and the performance is tested using the same dataset. The simulation results indicated that all the proposed models have improved accuracy and F1-score values.
- Avoin saatavuus