Hybrid Policy Learning for Energy-Latency Tradeoff in MEC-Assisted VR Video Service

Virtual reality (VR) is promising to fundamentally transform a broad spectrum of industry sectors and the way humans interact with virtual content. However, despite unprecedented progress, current networking computing infrastructures are incompetent unlock VR's full potential. In this paper, we consider delivering wireless multi-tile VR video service over mobile edge (MEC) network. The primary goal minimize system latency/energy consumption arrive at tradeoff thereof. To end, first cast time-varying view popularity as model-free Markov chain effectively capture its dynamic characteristics. After jointly assessing caching capacities on both MEC server playback device, hybrid policy then implemented coordinate replacement deterministic offloading, so fully utilize resources. underlying multi-objective problem reformulated partially observable decision process, deep gradient algorithm proposed iteratively learn solution, where long short-term memory neural network embedded continuously predict dynamics unobservable popularity. Simulation results demonstrate superiority scheme in achieving trade-off between energy efficiency latency reduction baseline methods.