Adaptive Resource Allocation for eMBB and URLLC Under Dynamic Backhaul Traffic: Puncturing or Slicing?

To satisfy the heterogeneous QoS requirements of eMBB and URLLC, network slicing and puncturing have emerged as two most widely-used resource allocation methods. However, under dynamic backhaul traffic (e.g., bursty eMBB and URLLC traffic), relying solely on either slicing or puncturing may cause performance degradation. This paper investigates adaptive resource allocation in integrated access and backhaul (IAB) networks to tackle the performance degradation resulting from eMBBandURLLCtraffic bursts in the backhaul. To release more backhaul communications resources, we first leverage edge caching to serve eMBB traffic at the edge as much as possible. Based on the caching decisions, we then allocate the backhaul resources according to the URLLC arrival patterns. Specifically, for sporadic URLLC arrivals, a greedy puncturing algorithm is proposed to minimize the delay. For bursty URLLC arrivals, we propose a hybrid algorithm that switches between puncturing and slicing methods to minimize the delay. To determine the timing for switching, we employ a deep reinforcement learning framework to learn the switching probability based on backhaul traffic states. Simulation results validate the effectiveness of our proposed algorithms in reducing URLLC delay while guaranteeing eMBB throughput. In particular, the proposed HPS algorithm achieves lower delay than other algorithms, especially under intensive URLLC traffic bursts.