Energy Minimization for IRS-Aided Wireless Powered Federated Learning Networks With NOMA
MohammadHossein Alishahi, Paul Fortier, Ming Zeng, Quoc‐Viet Pham, Xingwang Li
Abstract
This paper considers the scenario where multiple Internet-of-Things (IoT) devices collaborate to train a distributed model using federated learning. Wireless power transfer (WPT) is employed to address the issue of limited battery life of IoT devices, while non-orthogonal multiple access (NOMA) is utilized to facilitate data transmission. Besides, an intelligent reflecting surface (IRS) is applied to assist both energy transfer and data transmission. On this basis, a joint resource allocation problem is formulated to minimize the total energy consumption for the considered IRS-aided FL-WPT networks with NOMA. The non-convex problem is first solved by developing a combination of semi-definite programming relaxation (SDR) with a two-dimensional search algorithm. To lower the computational complexity, SDR with a bisection algorithm is further employed by exploiting the inherent structure of the formulated problem. Numerical results not only validate the equivalence of these two algorithms in performance but also unequivocally establish the superior efficiency of the proposed method over benchmark schemes in terms of energy consumption.