On the trajectory of video quality transition in HTTP adaptive video streaming

Abstract

browsers, TVs. Each of these platforms has specific requirements with respect to transmission and video quality. Moreover, the environment within which most of the video streaming clients operate is both unreliable and varies over time. However, regardless of the access device, users want the best viewing experience possible. HTTP adaptive streaming (HAS) is the most successful technology so far that allows content providers to cater for the requirements of the multitude of devices and contexts. The process through which a HAS client chooses a video rate is called adaptive bitrate selection (ABR). The first generation of ABRs relied on throughput estimation and selected the highest video rate lower than the measured throughput [14]. This is based on the work of Wang et al. [22] that showed if the available TCP throughput is twice the bitrate of the video plus a few seconds of start-up delay TCP can ensure an acceptable video streaming experience. It later became clear that throughput estimation alone is not a sufficient parameter for designing efficient ABR since an accurate bandwidth estimation above the HTTP layer is difficult to achieve [6]. Consequently, any video rate selection algorithm that solely depends on such a relatively inaccurate estimate results in unnecessary rebuffering events [7], an undesirable variability of video rate [6] and sub-optimal video quality [6].Various attempts have been made to improve some of the identified issues of throughput-based ABRs by supplementing throughput measurements with information about the playback buffer [1,10,21]. Using buffer occupancy as a factor in video rate selection has developed from regarding buffer state changes as a complementary factor in making a rate selection decision [10,21] to employing it as the sole metric [7,8]. Though, whatever factor an ABR primarily relies on, it is difficult to build an ABR that maximises Quality of Experience (QoE) without taking buffer state changes into consideration.Abstract HTTP adaptive video streaming matches video quality to the capacity of a changing context. A variety of schemes that rely on buffer state dynamics for video rate selection have been proposed. However, these schemes are predominantly based on heuristics, and appropriate models describing the relationship between video rate and buffer levels have not received sufficient attention. In this paper, we present a QoE-aware video rate evolution model based on buffer state changes. The scheme is evaluated within a real-world Internet environment. The results of an extensive evaluation show an improvement in the stability, average video rate and system utilisation, while at the same time a reduction in the start-up delay and convergence time is achieved by the modified players.