NOMA-Empowered Federated Learning for Enhancing Uplink Efficiency in Wireless Networks.

This paper proposes an uplink non-orthogonal multiple access (NOMA)-assisted federated learning (FL) framework to enhance the stable and accurate convergence in wireless networks. In this system, clients upload their local models securely to the base station through uplink NOMA to improve the uplink transmissions efficiency. We mathematically formulate the total latency optimization problem, taking into account the quality-of-service (QoS) requirements for clients, the power allocation at the base station (BS), and the frequency considerations. Simulation results show the that of the proposed FL framework achieves an 11% higher test accuracy and reduces latency by 25% compared to FedAvg and FedProx.