Tile-based proactive virtual reality streaming via online hierarchial learning

Wireless virtual reality (VR) can provide unconstrained immersive experience, which however is resource-demanding to satisfy the unique user experience, i.e., motion-to-photon latency and degree of overlap (DoO). To improve the quality of experience with constrained resource, in this paper we propose tile-based proactive VR streaming, which selects, transmits and computes the tiles in a segment most likely requested in future before playback. To select the tiles to be delivered under limited communication and computing resources, it is necessary to learn the user behaviour in requesting tiles in an online manner. We formulate a joint communication and computing duration allocation and tile selection problem to maximize the average DoO for a VR video under the communication and computing resource constraints. To reduce computational complexity and implicitly predict the tile request information, we decouple the original problem into two subproblems, which are respectively solved via convex optimization and hierarchial online learning. Simulation results on a real dataset demonstrate evident gain of the proposed method over the first-predict-then-optimize scheme.