FLADxG: Federated Learning Based Anomaly Detection Framework for xG Systems

Anomaly detection (AD) is a critical component of closed-loop automation, designed to monitor desired system resiliency. A centralized AD implementation can introduce an excessive communication overhead for distributed access and xHaul networks. Thanks to the growing adoption of cloud-native infrastructure and network function disaggregation, infrastructure monitoring time-series can be distributed and processed near the data sources. In this work, we propose the FLADxG framework to use federated learning to distribute the AD process, with as a reference use-case the cellular mobile access infrastructure. We build it using LSTM AutoEncoders, implemented as AI functions, showing how we can drastically reduce training times without severe detriment to performance.