Context-Aware Multi-DQN-Driven Preemptive Scheduling for Network Slicing in Next-Generation Cellular Networks

The coexistence of Ultra-Reliable Low-Latency Communications (URLLC) and enhanced Mobile Broadband (eMBB) services in 5G and beyond networks introduces a challenging trade-off between stringent latency requirements and high throughput demands. In this work, we propose a context-aware multi-Deep Q-Network (CA-M-DQN) framework for preemptive scheduling, where a set of specialized DQN agents is indexed by Channel Quality Indicator (CQI) and URLLC packet size, and a lightweight dispatcher selects the appropriate policy at runtime. The proposed approach is implemented in a 3GPP-compliant system-level simulator and evaluated across a range of operating conditions, including scale-out, mobility, bandwidth variation, and high-load scenarios. The results show that CA-M-DQN maintains reliable URLLC performance while preserving eMBB throughput under nominal and moderately stressed conditions, and achieves consistent behavior across varying channel regimes without requiring retraining. At the same time, the evaluation highlights limitations under sustained extreme load, where a monolithic DQN baseline can offer improved generalization. This observation reflects the trade-off inherent in context-specialized policies and helps clarify the operating regime of the proposed approach. Overall, the results suggest that context-aware modularization provides a practical and effective framework for balancing URLLC and eMBB requirements, while leaving room for extensions that improve robustness under heavy traffic conditions.