DPAS: Demand-Pressure Adaptive Slicing for Real-Time 6G Network Resource Management

Abstract

Next-generation 6G base stations must jointly serve eMBB, URLLC, and mMTC traffic within sub-millisecond scheduling windows under strict and conflicting latency constraints. Existing methods either neglect URLLC urgency or fail to enforce per-slice bandwidth floors. We propose DPAS (Demand-Pressure Adaptive Slicing), a closed-form, zero-training scheduler that represents each slice as a weighted demandpressure ratio $P_{i}^{w}=w_{i} D_{i} / B_{i}$ and reallocates bandwidth via a lightweight gradient update with a floor-preserving two-stage projection. The algorithm runs in $\mathcal{O}(N)$ per slot, requires no offline training, and-under our linearised model and simulationstypically settles within 6-8 slots. On 3.67 million real-world CDRs across seven spatial zones, DPAS achieves above 99% queueingdelay SLA for all three slice types.