Learning-Based Prediction, Rendering and Transmission for Interactive Virtual Reality in RIS-Assisted Terahertz Networks

The quality of experience (QoE) requirements wireless virtual reality (VR) can only be satisfied with high data rate, reliability, and low VR interaction latency. This rate over short transmission distances may achieved via the abundant bandwidth in terahertz (THz) band. However, THz waves severe signal attenuation, which compensated by reconfigurable intelligent surface (RIS) technology programmable reflecting elements. Meanwhile, latency mobile edge computing (MEC) network architecture due to its computation capabilities. Motivated these considerations, this paper, we propose an MEC-enabled RIS-assisted indoor scenario, taking into account uplink viewpoint prediction position transmission, MEC rendering, downlink transmission. We two methods, are referred as centralized online gated recurrent unit (GRU) distributed federated averaging (FedAvg), predict viewpoints users. In uplink, algorithm that integrates long-short term memory (LSTM) convolutional neural networks (CNN) is deployed locations line-of-sight non-line-of-sight statuses users time. downlink, develop a constrained deep reinforcement learning select optimal phase shifts RIS under constraints. Simulation results show our proposed achieves near-optimal QoE genie-aided benchmark algorithm, about times improvement compared random shift selection scheme.