Deterministic Flow Scheduling in 5G-TSN for IIoT: A Learning-Based Adaptive MCG-GF-NOMA Approach

The Industrial Internet of Things (IIoT) demands ultra-reliable low-latency communication (URLLC) to support massive time-sensitive devices with stringent requirements on deterministic transmission, ultra-dense connectivity, and bounded latency. To address this challenge, the convergence of 5G and Time-Sensitive Networking (TSN), termed 5G-TSN, has emerged as a key enabler for deterministic industrial networks. In this paper, we propose a joint 5G-TSN flow scheduling framework that guarantees deterministic transmission for massive time-sensitive data flows. Our approach integrates Grant-Free Non-Orthogonal Multiple Access (GF-NOMA) in 5G with Cyclic Queuing and Forwarding (CQF) in TSN. Specifically, we introduce a learning-based adaptive Multi-Configured Grant (MCG) subframe structure to optimize spectral efficiency in GF-NOMA. We then formulate a deadline-aware flow scheduling problem to maximize the number of successfully transmitted flows under strict latency constraints. To address this non-stationary and non-convex problem, we develop a Proximal Policy Optimization (PPO)-based algorithm that dynamically adjusts MCG configurations based on real-time ACK feedback and deadline requirements. Extensive simulations demonstrate that our solution outperforms existing benchmarks, achieving up to $\mathbf{2 3 . 8 1 \%}$ improvement in the number of successfully transmitted flows under heavy flow IIoT scenarios compared with the MCG subframe structure.