Spatial-Separable NOMA-Based Intelligent Hierarchical Fast Uplink Grant for mURLLC Over Cell-Free Networks

Massive ultra-reliable and low latency communications (mURLLC) has emerged as a dominating 6G-standard service. Fast uplink grant is an effective means to solve the uplink access in mURLLC due to the advantages of low signaling cost and collision-free. However, as two key challenges in fast uplink grant, active set prediction and optimal scheduling still lead to high resource wastage and low access success probability. Based on the special spatial sparsity of cell-free networks, we propose the spatial-separable non-orthogonal multiple access (SSNOMA). Compared with NOMA, SSNOMA allows multiple users to share same three-dimensional (3D) resources composed of time-frequency resources and pilots, so as to reduce the resource wastage caused by prediction errors. Futhermore, we design an intelligent hierarchical fast uplink grant framework. In this framework, the upper controller is responsible for scheduling users from the predicted active user set to ensure the optimal quality of service (QoS) and active probability, while the lower controller strictly controls the allocation of uplink grants among the scheduled users to maximize spectral efficiency. In addition, considering the limited ability to collect information in massive user access, the upper confidence bound (UCB)-based multi-armed bandit (MAB) algorithm is used in the upper layer to schedule fast grant users, while the multi-agent deep deterministic policy gradient (MADDPG) is used in the lower layer to perform specific grant allocation. Simulation results show that the proposed SSNOMA-based intelligent hierarchical framework can significantly improve the utilization of limited resources, and track long-term scheduling experience as well as QoS, effectively supporting mURLLC.