Federated Learning for Distributed NWDAF Architecture
Parsa Rajabzadeh, Abdelkader Outtagarts · 2023
Machine Learning (ML) has been considered to play a key role in processing collected data from the network Functions(NFs) in 5G. Network Data Analytic Function(NWDAF) is a new 5G component that is designed to provide analytics for any Network Functions(NFs). However, sending all data to a central NWDAF instance is extremely time-consuming and it raises concerns about security vulnerabilities and data overload. To tackle these issues, a distributed architecture for NWDAFs is proposed to perform parallel processing such that there are multiple NWDAFs co-located on the edges near the NFs in 5G. Therefore, predictive analytics are generated with considerably less latency service in the 5G. However, existing multi-node ML frameworks for distributed networks are not suitable for this scenario due to security and robustness issues. In order to address this issue, we developed a demonstration that showcases our novel centralized Federated Learning framework. The proposed centralized Federated Learning Framework is specially designed to meet the challenges that are roused in the distributed NWDAF architecture. Eventually, the performance of this framework is compared with current solution and potential candidates in our distributed NWDAF demonstration.