1.1 Problem Statement
The Scalable Video Coding (SVC) Extension of H.264/AVC [1] specifies a multilayer predictive encoding scheme that enables different video coding layers to be extracted and decoded selectively by different user devices according to their playback capability and transport network throughput. This desirable property causes SVC to be regarded as the ideal technology for providing multimedia multicasting to heterogeneous viewing devices.
Nevertheless, a close examination of SVC data format reveals two potential handicaps of its use in heterogeneous multicasting. First, the cumulative bit rates of the SVC layers extracted by individual user devices routinely exceed the bit rates of their corresponding AVC bitstreams due to the loss of coding efficiency in multilayer encoding. When the SVC bitstreams are transported over bandwidth stringent networks such as the cellular telephone networks, this bit rate inflation often implies cost inflation or performance degradation.
Second, the inter-layer dependence relations introduced by the multilayer predictive encoding process requires a viewing device to extract and decode all the reference coding layers on which the target layer of the device depends before the device can decode its target layer. Any un-recoverable loss of a reference layer will reduce the decodable layers to those below the lost layer due to inter-layer dependency and thus degrade the playback quality on the viewing devices. In a best-effort delivery network such as the Internet, a congested along a transport rate can thus a degradation of the video playback quality. Thus, a bandwidth efficient and loss
2
resilient transport scheme must be devised in order to ensure SVC to be the preferred encoding format for video streaming.
In this thesis, we attempted to improve the bandwidth efficiency of heterogeneous multi-casting of SVC bitstreams by devising a distributed bandwidth allocation scheme that can be implemented by every intermediate node along the transport paths. The objective of our scheme is to maximize the overall video playback quality of the viewing devices enrolled in the heterogeneous multicasting session while minimizing the total bandwidth utilization along the transport paths.
Our bandwidth allocation scheme assumes that the SVC bitstream is transported to different types of viewing devices scattered across multiple stub networks over the Internet via several multicast trees each of which is made up of tiers of intermediate nodes (known as the media gateways). Every media gateway (MG) receives specific SVC layers from a finite set of upstream nodes along multiple inbound network connections and dispatches these layers to specific downstream nodes through multiple outbound network connections. Each MG decides how to allocate its outbound bandwidth for transporting specific SVC layers to specific downstream nodes based only on the local information it gathers from its upstream and downstream nodes. No information exchange among the media gateways in the same tier is allowed. To simplify the preliminary design of our scheme, we ignore the asynchronous and lossy nature of the transport network with respect to information exchanges regarding bandwidth allocation. Hence, the decisions of individual MGs are unaffected by the order and the potential loss of control packets.
3
1.2 Research Approach
We will first introduce transport techniques for efficient purpose of SVC multicasting.
¾ Broadcasting over stub networks is to compensate the bit rate inflation introduced by SVC multilayer encoding process.
¾ En-route Bitstream Aggregation over long haul network is to minimize upstream bandwidth while maximize serviced device numbers.
¾ Information Dispersal and Multipath Diversity is to cut layers into equal-rated flows to fit into media gateway more easily and to add error protections on layers unequally by their importance shown in the inter-layer dependency.
There are a lot of different video streams in the network and to we will arrange bitstream allocation of them in a bandwidth efficient and loss resilient transport under these three techniques. Then we will formulate our goal in an optimization models about both playback quality and network bandwidth consumption. And we will design algorithms under this model.
There are some considerations of SVC we should introduce. Both theoretic and empirical work point to some important factors that are highly relative to the performance of bandwidth allocation of the SVC multicasting as the followings:
¾ Inter-layer Dependency: To avoid wastes of bandwidth, viewing devices must not subscribe to any layer that can’t be decoded. On the other hand, they will make subscription based on the extraction path of SVC.
¾ Bitstream Characteristics: SVC layers have their own bandwidth consumption and improvement of playback quality. The Rate Distortion Ratio of a single SVC layer
4
influences the efficiency of transporting it directly.
¾ User Device Types and Distribution over the Internet: Since we want to arrange the bandwidth allocation among large amount of devices, user device types and their distribution should be considered carefully. The largest amount of viewing devices will dominate the performance of bandwidth allocation.
¾ Network Topology: Since we want to arrange the bandwidth allocation over large scale networks, to make a better bandwidth allocation under a specific sub-network highly depends on its topology.
We will consider these factors and design different decision algorithms to approach our goal.
1.3 Thesis Outline
In chapter 2, we will briefly introduce the related work of SVC multicasting. In chapter 3, we will introduce efficient transport techniques. In chapter 4, we describe the optimization model and decision algorithms for bandwidth allocation. In chapter 5, we will explain our message exchange mechanism in detail. In chapter 6, we will show how we implement the scheme and the how we design the experiment models. In chapter 7, we will define the measurements and give the results and its analysis. In chapter 8, we will make a conclusion and tell accomplishments in our works and our future works. And the end of paper, we add a glossary about definitions of symbols we used in this thesis.
5