QoE-Aware 3D Video Streaming via Deep Reinforcement Learning in Software Defined Networking Enabled Mobile Edge Computing

Abstract

With the advancements of wireless network transmission technology, 2D video is hard to satisfy people's requirement for multimedia. Therefore, the high-definition 3D video that can bring a whole new viewing experience is starting to enter people's vision. However, when a tremendously large number of users play 3D video, it puts enormous computational pressure on the cloud server, which incurs high transmission latency. To release the tension, in this articl we consider a promising computing and networking architecture by incorporating Mobile Edge Computing (MEC) and Software-defined Networking (SDN) and propose a novel resource allocation model (RAM) to allocate resources and reduce delay. At the same time, we introduce the Quality of Experience (QoE) Model (QoEM), which uses information collected during 3D video playback to adaptively allocate the rate of future tiles. The model addresses the problem of assigning the best transmission speed to the block in the case of time-varying characteristicfactors during transmission. We propose an Actor-Critic-based deep reinforcement learning algorithm for viewport prediction and QoE optimization, called QoE-AC. For the differential transmission in the playback phase, we use the LSTM network for bandwidth and viewport prediction, while combining the historical information of the blocks into the Actor-Critic network as observations. The network can be adaptively assigned the best transmission speed for future tiles based on observations to maximize QoE. Finally, the experimental results show that the actual performance of the model is much better than other existing 3D video network models. Under different QoE targets, our proposed system can be adapted to all situations and has a 10%-15% performance improvement.