Random Orthogonalization for Federated Learning in Massive MIMO Systems

We propose a novel communication design, termed random orthogonalization , for federated learning (FL) in massive multiple-input and multiple-output (MIMO) wireless system. The key novelty of random orthogonalization comes from the tight coupling FL two unique characteristics MIMO - channel hardening favorable propagation. As result, can achieve natural over-the-air model aggregation without requiring transmitter side state information (CSI) uplink phase FL, while significantly reducing estimation overhead at receiver. extend this principle to downlink develop simple but highly effective broadcast method FL. also relax assumption by proposing an enhanced design both communications, that does not rely on or Theoretical analyses with respect machine performance are carried out. In particular, explicit relationship among convergence rate, number clients, antennas is established. Experimental results validate effectiveness efficiency MIMO.