Faculty Recruiting Support CICS

Video Adaptation for High-Quality Content Delivery

29 Oct
Thursday, 10/29/2020 3:00pm to 5:00pm
Zoom Meeting
PhD Thesis Defense
Speaker: Kevin Spiteri

Zoom Meeting:

Meeting ID: 988 8595 1200
Passcode: 801193


Modern video players employ complex algorithms to adapt the bitrate of the video that is shown to the user. Bitrate adaptation requires a tradeoff between reducing the probability that the video freezes (rebuffers) and enhancing the quality of the video. A bitrate that is too high leads to frequent rebuffering, while a bitrate that is too low leads to poor video quality. In this dissertation we propose video-adaptation algorithms to deliver content and maximize the viewer's quality of experience (QoE).

Video providers partition videos into short segments and encode each segment at multiple bitrates. The video player adaptively chooses the bitrate of each segment to download, possibly choosing different bitrates for successive segments. We formulate bitrate adaptation as a utility-maximization problem, and design algorithms to provide provably near-optimal time-average utility.

Real-world systems are generally too complex to be fully represented in a theoretical model and thus present a new set of challenges. We design algorithms that deliver video on production systems, maintaining the strengths of the theoretical algorithms while also tackling challenges faced in production. Our algorithms are now part of the official DASH reference player dash.js and are being used by video providers in production environments.

Most online video is streamed via HTTP over TCP. TCP provides reliable delivery at the expense of additional latency incurred when retransmitting lost packets and head-of-line blocking. Using QUIC allows the video player to tolerate some packet loss without incurring the performance penalties. We design and implement algorithms that exploit this added flexibility to provide higher overall QoE by reducing latency and rebuffering while allowing some packet loss.

 Recently virtual reality content is increasing in popularity, and delivering 360deg video comes with new challenges and opportunities. The viewing space is often partitioned in tiles, and a viewer using a head-mounted display only sees a subset of the tiles at any time. We develop an open source simulation environment for fast and reproducible testing of 360deg algorithms. We develop adaptation algorithms that provide high QoE by allocating more bandwidth resources to deliver the tiles that the viewer is more likely to see, while ensuring that the video player reacts in a timely manner when the viewer changes their head pose.


Ramesh Sitaraman