Adaptive Retransmission for Efficient Wireless Federated Learning

Federated learning (FL) facilitates collaborative training of a shared machine learning (ML) model across distributed devices while ensuring data privacy. However, the limited radio resources of mobile devices and the presence of unreliable wireless channels pose significant challenges to the robustness of FL deployment over wireless networks. To address these issues, this paper proposes an adaptive FL retransmission framework. First, the impact of unreliable transmission on the FL convergence rate is analyzed. Then, a resource block (RB) allocation problem is formulated from a long-term perspective to minimize the convergence rate, subject to a total RB consumption constraint. To solve this problem, it is reformulated as a variational calculus optimization problem, and the Euler-Lagrange equation is employed to derive the optimal solution. Experimental results demonstrate that the proposed scheme outperforms benchmark methods.