Multiple Parallel Federated Learning via Over-the-Air Computation

This paper investigates multiple parallel federated learning in cellular networks, where a base station schedules several FL tasks and each task has group of devices involved. To reduce the communication overhead, over-the-air computation is introduced by utilizing superposition property access channels (MAC) to accomplish aggregation step. Since all use same radio resource transfer their local updates BS, order separate received signals different tasks, we zero-forcing receiver combiner mitigate mutual interference across groups. Besides, analyze impact device selection on convergence our framework. Also, formulate an optimization problem that jointly considers vector design for improving performance. We address decoupling it into two sub-problems solve them alternatively, adopting successive convex approximation (SCA) derive vector, then scheduling with greedy algorithm. Simulation results demonstrate proposed framework can effectively straggler issue achieve near-optimal performance tasks.