Puncturing-Based Resource Allocation for URLLC and eMBB Services via Matching Theory and Unsupervised Deep Learning

The coexistence of Ultra-Reliable and Low-Latency Communications (URLLC) and enhanced Mobile BroadBand (eMBB) brings significant challenges for service pairing and resource allocation in beyond fifth-generation (B5G) wireless networks. To meet the reliability requirement of URLLC services and improve the fairness of eMBB services, we first develop a supervised learning-based resource allocation policy for eMBB services. Then, a two-phase resource allocation framework is proposed for URLLC services: 1) eMBB/URLLC service pairing and 2) URLLC resource allocation. In the first phase, matching theory pairs eMBB and URLLC services for better fairness. In the second phase, URLLC resource allocation policy is optimized by a constrained unsupervised learning algorithm. Simulation results show that our proposed framework can achieve better trade-offs among fairness, throughput, and reliability compared with two existing baselines. For example, a dynamic proportional fairness algorithm can meet the reliability requirement of URLLC when its average packet arrival rate is below 0.7 packets/mini-slot. The proposed algorithm can support URLLC services with an average packet arrival rate of 1.6 packets/mini-slot.