Adaptive Federated Learning-based Joint Pilot Design and Active User Detection in Scalable Cell-free Massive MIMO Systems
Lei Diao, Jiamin Li, Pengcheng Zhu, Dongming Wang, Xiaohu You · 2023
Massive ultra-high reliability ultra-low latency communication (mURLLC), a promising core service of 6G, is likely to adopt grant-free random access (GFRA) as its access scheme for low access latency. However, massive user access brings many challenges to the process, especially the huge computing overhead and delay will hinder the scalability of the system. Based on a distributed cell-free massive multiple input multiple output (CF-mMIMO) edge computing system, we propose a scalable active user detection (AUD) scheme in the first step of GFRA. First, we divide the remote radio units (RRUs) into several clusters and allocate pilots and power to users with deep deterministic policy gradient (DDPG) algorithm. Then, within each RRU cluster, a noise-learning-based convolutional structure is applied on the edge distributed unit (EDU) to train local detection network. Furthermore, by introducing an adaptive federated learning algorithm, the globally optimal AUD performance is achieved under the delay constraint. The simulation results show that the proposed scalable AUD scheme is suitable for massive users scenario.