The coexistence of Enhanced Mobile Broad-band (eMBB) and Ultra-Reliable Low-Latency Communications (URLLC) services is a common scenario in 5G. While eMBB strives for high data rates using slots as transmission time intervals (TTIs), URLLC emphasizes reliability and low latency using mini-slots. Puncturing scheme is introduced by 3GPP, which means puncturing a set of Resource Blocks (RBs) from ongoing eMBB transmissions and reallocating them for URLLC traffic. In this paper, a DRL-based dynamic resource slicing scheme for eMBB and URLLC traffic considering puncturing is proposed, where the data rate, Quality of Service (QoS) satisfaction and rate stability for eMBB users are simultaneously optimized on mini-slot-level timescale, by employing an improved Deep Q-learning (DQN) algorithm. Simulation results demonstrate that the proposed algorithm outperforms the baseline algorithms while ensuring the latency and reliability requirements of URLLC and successfully protects eMBB users under adverse conditions.
Deep Reinforcement Learning Based Dynamic Resource Slicing for eMBB and URLLC Traffic Considering Puncturing
2024-06-24
1751921 byte
Conference paper
Electronic Resource
English