A Cell-Free Millimeter-Wave Channel Estimation Algorithm Based on Federated Learning

Although machine learning is powerful in physical layer design like channel estimation, its centralized training approach generates significant communication overhead and privacy risks due to the need for processing large amounts of data centrally. We leverage the distributed deployment characteristics of access points (APs) in a cellular-free network architecture in a cell-free massive MIMO system and combine federal learning for channel estimation, this solution ensures accurate channel estimation, effectively addressing the potential issues mentioned above. A novel model aggregation approach is proposed in this paper, which incorporates a feedback mechanism to bring the training model closer to the convergence upper bound. Simulation results demonstrate that the proposed algorithm yields a neural network model with superior channel estimation performance and requires fewer training rounds compared to the benchmark algorithm.