| dc.description.abstract | The topic of this dissertation is bitrate adaptive media streaming to receivers in mobile wireless networks. This work was motivated by the recent explosion in popularity of media streaming to mobile devices. Wireless networks will always be bandwidth limited compared to fixed networks due to background noise, limited frequency spectrum, and varying degrees of network coverage and signal strength. Consequently, applications that need to move large amounts of data in a timely manner cannot simply assume that future networks will have sufficient bandwidth at all times. It is therefore important to make the applications themselves able to cope with varying degrees of connectivity.<br>
In order to understand the requirements of streaming in 3G mobile networks, we perform a large number of measurements in Telenor’s 3G network in and around Oslo. Using bandwidth traces from these field experiments, we compare commercial adaptive media streaming clients by Adobe, Apple, and Microsoft in challenging vehicular (bus, ferry, tram and metro) streaming scenarios.<br>
In this comparison, we reveal problems with buffer underruns and unstable video playouts. We therefore develop our own adaptive bitrate media client, and design a new quality adaptation scheme that targets the requirements of mobile wireless networks, reducing the number of buffer underruns and improving stability. We also observe that network conditions are highly predictable as a function of geographical location. Simulations on bandwidth traces from field experiments indicate that the video playout can be made even more stable: A media player that knows its future (bandwidth availability and the duration of the streaming session) can use its buffer more intelligently. Fluctuations in bandwidth can be smoothed out through sophisticated buffering algorithms, resulting in a higher quality video playout with fewer interruptions due to buffer underrun.<br>
Again using our collection of bandwidth traces, we develop a bandwidth lookup service and a new algorithm for quality scheduling that uses historic bandwidth traces to plan ahead, thus avoiding most underruns and offering a far more stable playout with fewer visually disturbing fluctuations in quality. We show that this prediction-based approach greatly improves the performance compared to our best results with non-predictive quality schedulers. Finally, we show how multi-link streaming can be employed to increase the network capacity available to the video receiver, thus improving perceived video quality even further.<br>
All algorithms are developed and tested using custom made simulation tools, and are later verified in real world environments using a fully functional prototype implementation. We demonstrate that our proposed algorithms greatly improve performance in vehicular mobile streaming scenarios. | en_US |