An Online Caching Scheme for 360-Degree Videos at the Edge

360-degree videos have gained considerable popularity by offering immersive experiences to viewers. However, they consume significantly high bandwidth and demand for specialized caching schemes at the edge network. Previous caching schemes are limited as they either rely on complete historical data or neglect the viewer-varying characteristics in 360-degree videos. In this paper, we present an online caching scheme for 360-degree videos that leverages feedback from sequentially arriving viewers at the network edge. Our scheme consists of two components: an online tile popularity prediction component that accurately predicts the popularity of the tiles with the PopPred algorithm, and an online tile-bitrate caching optimization component that optimizes caching decisions to enhance viewers’ quality of experience (QoE) with the CacheOpt algorithm. We prove that both algorithms have sublinear regret, i.e., $O(\sqrt K ),$ where K is the number of viewers. We also conduct comprehensive experiments using real-world data to show that our caching scheme achieves better performance with lower regrets, higher QoE, and higher hit ratios compared with existing algorithms.