A Deep Reinforcement Learning Strategy for MEC Enabled Virtual Reality in Telecommunication Networks

Abstract

One of the most anticipated features of 5G and subsequent networks is wireless virtual reality (VR), which promises to transform human interaction via its immersive experiences and game-changing capabilities. Wireless virtual reality systems, and VR games in particular, are notoriously slow due to rendering issues. But most academics don't care about data correlation or real-time rendering. Using mobile edge computing (MEC) and mmWave-enabled wireless networks, we provide an adaptive VR system that enables high-quality wireless VR. By using this architecture, VR rendering operations may be adaptively offloaded to MEC servers in real-time, resulting in even greater performance advantages via caching.The limited processing power of VR devices, the need for a high quality of experience (QoE), and the small latency in VR activities make it difficult to connect wireless VR consumers to high-quality VR content in real-time. To solve these problems, we provide a wireless VR network that is enabled by MEC. This network makes use of recurrent neural networks (RNNs) to provide real-time predictions about each user's field of vision (FoV). It is feasible to simultaneously move the rendering of virtual reality material to the memory of the MEC server. To improve the long-term VR users' quality of experience (QoE) while staying within the VR interaction latency limitation, we provide decoupling deep reinforcement learning algorithms that are both centrally and distributedly run, taking into consideration the connection between requests' fields of vision and their locations. When compared with rendering on VR headsets, our proposed MEC rendering techniques and DRL algorithms considerably improve VR users' long-term experience quality and reduce VR interaction latency, according to the simulation results.