Rendered Tile Reuse Scheme Based on FoV Prediction for MEC-Assisted Wireless VR Service

Abstract

Interactive real-time wireless Virtual Reality (VR) has become increasingly popular because it provides an immersive experience for VR users anytime, anywhere. Introducing multi-access edge computing (MEC) to wireless VR service can solve the problem that VR devices having insufficient rendering power. However, the problem that some tiles in the overlapping field of views (FoVs) may be rendered repeatedly usually be ignored. Repeated rendering would waste the computing resources of edge nodes and damage users' quality of experience (QoE). This paper proposes a rendered tile reuse scheme based on FoV prediction and 3 C (caching, computing, and communication) optimization for the MEC-assisted VR service. Firstly, we model the above scheme as an optimization problem that aims to maximize the total users' QoE value under the motion to photons (MTP) delay constraints. Secondly, we use the recurrent neural network model with gated recurrent unit (GRU) architecture to dynamically predict the users' FoV in the next time slot. Thirdly, we use the proximal policy optimization (PPO)to learn the question's solution iteratively based on the results of FoV prediction. The simulation results show that our proposed algorithm is superior to other algorithms in improving the value of total users' QoE and reducing the MTP delay.