Perceptually Driven Algorithms for Stored Video Delivery
27 May 2014
Wireless, and to a lesser extent wired, networks exhibit substantial variability in available capacity and/or load variations. As a result stored video delivery protocols have been designed to adapt compression rate to congestion in the network and/or the state of the video clients playback buffer.
WNCG Ph.D. alumnus Vinay Joseph and Prof. Gustavo de Veciana have devised a new class of Dynamic Adaptive Streaming over HTTP (DASH)-based video delivery protocols and complementary network resource management algorithms which are driven by perceptually relevant video quality metrics and user preferences including: mean and variability of video quality, playback rebuffering, and average cost per unit time. Our approach is comprehensive in that it realizes a joint optimization of users' video quality of experience that can in principle account for heterogenous content and device characteristics as well as user preferences.
The proposed family of algorithms is provably optimal in a stationary regime, and easily implementable in the DASH setting with some additional exchange of meta-data amongst clients and servers and possibly state updates at network resources. Overall we believe this framework could in principle be the starting point for a new generation of video delivery protocols.
This research, which received joint support from Intel and Cisco, represents the outcome of discussions involving WNCG colleagues and students, as well as Intel and Cisco engineers and researchers.
- Presentation slides: NOVA