An Approach to Transmitting URLLC Data with Different Latency Requirements over eMBB Services Based on Deep Reinforcement Learning

Abstract

For different quality of service (QoS), the coexistence of enhancement of mobile broadband (eMBB) and ultra-reliable low-latency communications (URLLC) on the same radio spectrum leads to a challenging scheduling optimization problem. For the moment, puncturing is a practical alternative. In this paper, we propose a deep reinforcement learning (DRL) approach to dynamically schedule resources to meet the stringent latency requirements of URLLC while minimizing the impact on average eMBB throughput. In addition, we model a novel realistic application scenario model for transmitting URLLC data with different latency requirement over eMBB services. Compared to traditional scheduling algorithms, the simulation results show that this method never violates the latency requirements of URLLC traffic and successfully ensures that the number of codewords in outage is at a minimum.