EnDASH - A Mobility Adapted Energy Efficient ABR Video Streaming for Cellular Networks

User experience of watching videos in smartphones while travelling is often limited by fast battery drainage. Existing client video players use adaptive bitrate (ABR) streaming through Dynamic Adaptive Streaming over HTTP (DASH) to improve user's Quality of Experience (QoE) while ignoring the energy savings aspect, which has been addressed in our work. In this paper, we propose EnDASH - an energy aware wrapper over DASH which minimizes energy consumption without compromising on QoE of users, under mobility. First, we undertake an extensive measurement study using two phones and three service providers to understand the dynamics between energy consumption of smartphones and radio related network parameters. Equipped with this study, the proposed system predicts cellular network throughput from the radio parameters within a finite future time window. The prediction engine captures the effect of associated technology and vertical handovers on throughput, unlike existing works. EnDASH then uses deep reinforcement learning based neural networks to first tune the playback buffer length to the average predicted cellular network throughput and then to select an optimal video chunk bitrate. It achieves a near 30% decrease in the maximum energy consumption than state-of-the-art ABR Pensieve algorithm while performing almost at par in QoE.