An adaptive-predictive architecture for streaming Scalable Encoded Video

Abstract

The steadily growth of multimedia demands in the Internet can lead to an impending collapse, since there are little efforts in such applications to control their sending rates. In addition to that, it is well known that providing perceptually good quality video streaming is a complex task, in view of the fact that in today’s best effort Internet the available bandwidth can fluctuate strongly and the encoded video can exhibit significant rate variability at several time-scales. On the other hand, one important requirement for streaming multimedia flows is that they must exhibit fairness with competing flows. Therefore, the main research problem addressed in this thesis is to bridge the gap between the available bandwidth variability and the encoded video rate variability, taking into account the requirement of the minimization of the quality variability and the maximization of the overall quality of the video rendered to the user. The main contribution of this thesis is the definition and realization of a novel architecture for video streaming applications in best effort networking environments. We focus on a scenario where the network provides explicit feedback information throughout the network path, which implies that such multimedia streaming must be able to adapt to network conditions efficiently, i.e., it is capable to cope with variations in bandwidth at several time scales. Towards this end, within our scalable architecture we propose several deployable server-side based solutions, which combine most beneficial properties of some innovative congestion control mechanisms, signal processing techniques, and time series analysis. We devise and investigate the mechanisms that implement the proposed solutions, and reveal the efficiency of each approach through simulation. Specifically, we firstly present a comprehensive investigation of the performance of video streaming in the best-effort Internet when using some selected network friendly protocols. As our experiments show, congestion control mechanisms that rely on precise explicit feedback information from the network provide a significantly better quality (i.e., with low intensity variation in quality) to the end-user than those that rely on rate-based slowly responsive ones. Second, using MPEG-4 Fine Granular Scalable pre-encoded video as our target application, we ensure that we meet the most important requirement from the network point of view that is transporting multimedia flows efficiently while exhibiting fairness with competing flows. Our architecture extracts the most precise information from the network level and then provides video source application with consistent and stable information. In summary, we build our solution using appropriate techniques in the networking, signal processing and statistical fields. By merging ideas from several areas, we propose a scalable architecture for video streaming over best effort networks with explicit rate notification. Such novel architecture is flexible to extend, subtract, or change functionalities.