Cost-Aware Deep Reinforcement Learning for eMBB-URLLC Multiplexing Resource Allocation

The multiplexing of Enhanced Mobile Broadband (eMBB) and Ultra-Reliable Low-Latency Communications (URLLC) results in a loss of the eMBB transmission rate. Existing multiplexing resource allocation schemes focus on enhancing the eMBB data rate, ignoring the Quality of Experience (QoE) of URLLC users and degrading the profit of the operator. Considering all users' QoE, we define the operator profit as the weighted difference between the average QoE and the operator cost which integrates the eMBB transmission rate loss and the successful gain of the URLLC service. This paper focuses on the cost aware eMBB-URLLC multiplexing resource allocation (CAMRA) to maximize the operator profit. With our proposed event-driven Actor-Critic (EDAC) Deep Reinforcement Learning (DRL) algorithm, CAMRA scheme can enhance the operator profit by more than 30% compared to the Rate- Aware strategy,