Performance Evaluation of Federated Learning in Intelligent Wireless Receivers

In the next generation of mobile networks, the task of Machine Learning (ML) model training can be allocated to the end user devices taking into account their increasing processing capabilities. One of the approaches which is foreseen for the 6th Generation (6G) of mobile networks is Federated Learning (FL). It has several benefits in terms of overhead reduction, and offloading of cloud servers. However, the performance of trained ML models through FL and their convergence time need to be studied. In this paper, FL performance is evaluated and compared to a conventional Centralized Learning (CL) approach. The ML is applied in intelligent wireless receivers for equalization of received signals. The results show that FL is capable of converging to the same accuracy as CL but needs a longer time. Besides, FL shows significant reduction of transmission overhead.