Dynamic SDN-Based Radio Access Network Slicing With Deep Reinforcement Learning for URLLC and eMBB Services

Radio access network (RAN) slicing is a key technology that enables 5G to support heterogeneous requirements of generic services, namely ultra-reliable low-latency communication (URLLC) and enhanced mobile broadband (eMBB). In this paper, we propose two time-scales RAN mechanism optimize the performance URLLC eMBB services. large time-scale, an SDN controller allocates radio resources gNodeBs according short each gNodeB its available end-users requests, if needed, additional from adjacent gNodeBs. We formulate problem as non-linear binary program prove NP-hardness. Next, for model Markov decision process (MDP), where large-time scale modeled single agent MDP whereas shorter time-scale multi-agent MDP. leverage exponential-weight algorithm exploration exploitation (EXP3) solve single-agent deep Q-learning (DQL) resource allocation. Extensive simulations show our approach efficient under different parameters configuration it outperforms recent benchmark solutions.