FAS-assisted federated learning over wireless communication systems

FAS-assisted federated learning over wireless communication systems

연구 분야: Networking

학회: Science China Information Sciences

This paper examines the energy efficiency of a multi-user system where federated learning (FL) is implemented in a distributed manner across all nodes. Each user employs a fluid antenna system (FAS) to improve the channel condition, while the base station (BS) is equipped with multiple traditional fixed-position antennas (FPAs). When performing the FL algorithm, each user first trains a local model and transmits it to the BS over shared time-frequency resources. Then, the BS aggregates the received models and broadcasts the combined model back to all users. These steps are repeated until the FL model achieves a desired accuracy level. The system energy is mainly consumed in the computation and transmission processes at the user side. To save energy, we develop an optimization framework that minimizes the total energy consumption by jointly optimizing the learning accuracy, transmit power, antenna positions, and the BS receivers. Since the optimization variables are highly coupled, the problem is non-convex and quite complex. To address the issue, we propose an iterative algorithm to obtain a suboptimal solution of the problem. Simulation results have verified the effectiveness of the algorithm and also the advantages of FAS over the conventional FPA technology.