Multi-Server Federated Edge Learning for Low Power Consumption Wireless Resource Allocation Based on User QoE

Tianyi Zhou, Xuehua Li, Chunyu Pan, Mingyu Zhou, and Yuanyuan Yao

10.23919/JCN.2021.000040

Abstract : Federated edge learning (FEL) deploys a machinelearning algorithm by using devices distributed on the edge ofa network, trains massive local data, uploads the local modelto update the parameters after training, and performs alternateupdating with global model parameters to reduce the pressurefor uplink data transmission, prevent systematic time delay andensure data security. This paper proposes that an optimal balancebetween time delay and energy consumption be achieved by optimizingthe transmission power and bandwidth allocation basedon user quality of experience (QoE) in a multi-server intelligentedge network. Given the limited computing capability of devicesinvolved in FEL local training, the transmission power is modeledas a quasi-convex uplink power allocation (UPA) problem, anda lower energy consumption bandwidth allocation algorithm isproposed for solution-seeking. The proposed algorithm allocatesappropriate power to the device by adapting the computingpower and channel state of the device, thereby reducing energyconsumption. As the theoretical deduction result suggests thatadditional bandwidth should be allocated to those devices withweak computing capabilities and poor channel conditions torealize minimal energy consumption within the restraint time.The simulation result indicates that, the maximum gain of theproposed algorithm can be optimized by 31% compared with thebaseline.​ 

Index terms : Bandwidth optimization, federated edge learning, QoE, uplink power allocation