Wireless Federated Learning with Retransmission

Federated learning (FL) enables distributed devices to collaboratively train a shared machine learning (ML) model while protecting their data privacy. However, the limited radio resources of mobile devices and unreliable channels pose per-formance bottlenecks for deploying FL over wireless networks. In this paper, we propose a novel federated learning framework with frequency-diverse retransmission to address these issues. First, we analyze the impact of unreliable transmission on the convergence rate. Next, we formulate an optimal resource block (RB) allocation problem to minimize the model convergence error subject to the energy budget. To solve this problem, we transform it into a multiple-choice knapsack problem and use dynamic programming to solve it. Experimental results demonstrate the superiority of the proposed scheme compared to the benchmark schemes.