Deep Reinforcement Learning-Based Joint Scheduling of eMBB and URLLC in 5G Networks

Jing Li, Xi Zhang

Abstract: To satisfy tight latency constraints, ultra-reliable low latency communications (URLLC) traffic is scheduled by overlapping the on-going enhanced mobile broad band (eMBB) transmissions (i.e., puncturing approach), which causes eMBB users unprecedented rate loss and hence degraded quality-of-service (QoS). To tackle this issue, this letter proposes to achieve QoS tradeoff between eMBB and URLLC in 5G networks. We jointly optimize bandwidth allocation and overlapping positions of URLLC users' traffic with deep deterministic policy gradient algorithm observing channel variations and URLLC traffic arrivals. Simulation results show that the proposed system-wide tradeoff method achieves the best tradeoff performance.