Adaptive Streaming Method based on Stochastic Prediction of Throughput on the Move

Abstract

A video streaming service is getting available at all times and all places. DASH is a common protocol for the video streaming and various technologies to control it are proposed. Most of them use the past throughput to when and what bitrate to select. However, the throughput varies dynamically especially on the move taking the train or the bus and predicting the throughput variance in advance is difficult. So relying on only the past throughput leads to improper control for streaming. We investigate the throughput variance caused by the moving state and propose a novel technique considering the moving context. Experimental results shows that our method reduces the video stalling time by 40%.