An emerging modality, increasingly utilized by edge devices for training machine learning models in a distributed and collaborative manner, is Federated Learning (FL). FL offers a unique blend of improved learning quality and the imperative need for data privacy. Despite the numerous advantages of this emerging paradigm, there exists a critical factor that could significantly impact its
... [Show full abstract] efficiency in future 5G and 6G application scenarios. This factor relates to potential delays arising from the limited communication resources available for connecting clients to servers, which could excessively slow down the process and diminish its effectiveness, especially in light of the new real-time applications that characterize 5G and 6G scenarios. To address this challenge, the thesis introduces a novel approach to client selection. Unlike the various communication streamlining methods proposed thus far, this approach begins with a networking research perspective and leverages the capabilities of Software-Defined Networking (SDN). It focuses on the dynamic selection and continuous updating of clients participating in the Federated Learning (FL) process. This innovative approach ensures that the distributed learning process remains highly effective and efficient, resulting in an overall reduction in the FL process time under varying network traffic loads. This effectiveness is demonstrated through a performance evaluation campaign conducted using an implemented testbed platform.