大规模视频网络的感知优化.ppt

Note that the number of layers is selected based on the average SNR in the previous interval. * In a typical video streaming client application, this represents the time difference between the playback pointer and the loaded content pointer It is desirable to keep this value small to avoid unnecessary bandwidth utilization if the user quits early but not too small to avoid frame freezes. A large startup delay causes the playback margin to grow initially but stabilize over time as the buffer contents are consumed. Yes. Thats right. But it is a 10 sec clip so the 10 seconds worth of video were transmitted in 13 seconds. Also note that this is measured from the perspective of the server. To measure it from the client perspective, you need to add the y-axis value to the x-axis value, i.e., the client consumes the last bit at 13+1.7 = 14.7 * Perceptual Optimization ofLarge Scale Wireless Video Networks Robert Heath Joint work with Profs. Alan Bovik, Jeff Andrews, Gustavo de Veciana, and Constantine Caramanis and several graduate students Huawei University Day Chicago November 2011 Outline Introduction UT’s video aware wireless networks project What we are learning Results on video-aware link adaptation Conclusions * Mobile Video Explosion * Figure from Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2010–2015 65x increase in video traffic Mobile video is 66% of mobile data 65% of mobile data will be at home or office, not that mobile No clue where a 65x revenue increase will come from Mobile video is coming and will consume the network Video Aware Wireless Networking Source is a real-time or stored video stream Network designed for efficient video transport Wireless link matches source and channel rates * Source Network Video Streaming * Frame 1 Frame 2 Frame 8 33ms … Group of pictures: 266ms … Channel coherence time: 20ms Videos are long Startup delay to fill buffer (e.g. 2 seconds) Avoid buffer starvation Can be buffered (stored