Optimizing Uplink MIMO Transmission for Model Uploadand Aggregation in Federated Learning

This paper focuses on optimizing multi-antenna transmission to enhance the efficiency of model upload in digital federated learning systems. Unlike previous research that assumes perfect channel state information (CSI) for local model upload in federated learning, this work takes into consideration the imperfection of CSI and computes the achievable data rates accordingly. The problem of minimizing the mean squared error (MSE) of the global aggregated model is formulated, which is found to be non-convex. To address this non-convexity and obtain an efficient suboptimal solution, we propose an iterative algorithm based on Majorization Minimization. The advantages of the proposed algorithm are validated through numerical results.