Multi-objective Optimization of Energy and Latency in URLLC-enabled Wireless VR Networks

Abstract

Energy and latency are important metrics for performance evaluation in ultra-reliable and low-latency communication-enabled wireless virtual reality networks. However, these two metrics often conflict with each other. Therefore, in order to strike a balance between energy efficiency and latency, a novel model is proposed for the energy and latency optimization of reconfigurable intelligent surface-assisted networks. To investigate the tradeoff between energy and latency, the meta-learning-based multi-objective soft actor-critic (MO-SAC) algorithm is proposed. The algorithm assigns dynamic weights to the objectives during training and the trained model is able to achieve a fast adaptation to the new tasks. The numerical results verify the efficiency of meta-learning-based MO-SAC, where the trained model is able to quickly adapt to new tasks.