Two-Stage Delay-Aware Intelligent Scheduling for QoS Guarantee in Cell-Free Networks

In Ultra-Reliable Low-Latency Communication (URLLC) scenarios under cell-free architectures, packet scheduling must meet stringent delay constraints. To address challenges such as heterogeneous delay sensitivity among tasks and the lack of timeout risk prediction in traditional methods, this paper proposes a two-stage delay-aware intelligent scheduling algorithm, referred to as DAIS-PPO. In the first stage, a Mixture Density Network (MDN) is employed to model the distribution of the remaining delay of packets in the queue and to predict their timeout risk. In the second stage, the predicted timeout risk (PTR) serves as input to a scheduling model formulated as a Constrained Markov Decision Process (CMDP), where the Proximal Policy Optimization (PPO) algorithm is employed to learn a scheduling policy for packets with heterogeneous delay constraints. To reflect the impact of limited frequency-domain resources on scheduling capacity, a dynamic subband allocation mechanism is designed based on scheduling utility, which adaptively determines the number of packets that can be scheduled in each slot. Additionally, Lagrange multipliers are introduced to dynamically constrain the system’s delay violation rate, achieving a balance between scheduling utility and QoS guarantees. Simulation results demonstrate that DAIS-PPO strictly satisfies delay requirements under low traffic loads and maintains robust, stable performance under high loads. It consistently outperforms several comparative algorithms, demonstrating its potential for deployment in delay-sensitive scenarios.