RL-Based Time Division Duplex Optimization for eMBB and URLLC Coexistence

Enhanced mobile broadband (eMBB), characterized by its high data rates, and ultra-reliable low-latency communications (URLLC), known for its dependability, are transforming the digital landscape. Nevertheless, the eMBB-URLLC coexistence poses significant challenges in balancing high-speed data transmissions with real-time reliability, particularly considering the scarce wireless resources. In this paper, we explore dynamic time division duplex (TDD) pattern selection to enhance coexistence performance. Thus, we formulate an optimization problem to maximize eMBB throughput while ensuring fairness among them and meeting URLLC reliability requirements. To solve this problem, we develop a deep reinforcement learning framework using a double deep Q-network (DDQN). Results from our near-product 5G simulator show that our solution significantly improves eMBB throughput and reduces the throughput interquartile range by over 40% compared to competing baselines while meeting URLLC reliability requirements.