Adaptive Streaming of Layer-Encoded Multimedia Presentations

Abstract

In the context of a communication network with no QoS guarantees, we describe the problem of adaptive streaming of layer encoded multimedia presentation data as a two-phase decision problem. In phase one the application transfers only base layer data that comprise a presentation of minimum quality, which is stored at the client. When the application determines that commencing play out will result in an uninterrupted presentation of at least minimum quality, it does so, and then transitions into phase two. The application then loops on the decision on which data to send next: another base layer, or an enhancement layer. It does this by considering the deliverability and play-out quality within a sliding window of media units. We present two different algorithms for making this decision, based on two different presentation quality metrics: the total quality metric, which yields an optimization problem that can be solved with dynamic programming, and the refined max-min metric, which yields a computationally inexpensive algorithm for computing an optimal decision. We also consider the problem of progressively rendering static objects after their start times as a means of improving presentation quality. We compare the various approaches with a slide show presentation with a randomly generated sequence of layer-encoded JPEG images, and an Ogg Vorbis audio stream.