Federated Deep Deterministic Policy Gradient for Power Control in 6G in-X Subnetworks

This paper introduces a federated deep deterministic policy gradient (FDDPG) method for power control optimization in 6G in-X subnetworks. The method enables efficient collaborative training of a DDPG agent for distributed wireless power control (WPC) without sharing raw sensing data, ensuring data privacy for all subnetworks. The algorithm's performance was evaluated in an in-factory mobile 6G in-X subnetworks deployment and benchmarked its performance against state-of-the-art DDPG solutions with centralized and distributed training, WMMSE, maximum power, and Sequential linear-quadratic pro-gramming (SLQP). The FDDPG solution achieved comparable performance to complex iterative optimization benchmarks.