Improving Video Transportation over Dynamic Wireless Networks
Abstract
This thesis explores video transfer over dynamic wireless communication links in an attempt to understand current limitations and propose new methods to push performance beyond the existing envelope. To this end, I spent considerable time researching and experimenting with two wireless communication methods, 4G LTE and Wi-Fi. I also looked at different problem sets that currently limit these specific technologies. This thesis is therefore broken down into three main sections.
The first major section explores the performance of TCP over live 4G LTE networks in order to gain an understanding of current congestion control protocols and help guide the design of a new congestion control algorithm. Specifically, this section of the thesis attempts to determine the relationship between throughput
and delay over LTE using different existing TCP congestion control algorithms. It then explores the possibility of designing a new delay based congestion control algorithm based on the best features of current congestion control algorithms observed in practice. This sections details the proposed algorithm which is built on top of the existing TCP Cubic congestion control algorithm in the Linux Kernel and tested over a commercial LTE network.
The second major section focuses on improving the performance of Virtual Reality (VR) systems. Streaming VR videos is extremely bandwidth intensive, and also highly sensitive to delay. This highlights the key features of throughput and delay and illustrates the necessity of understanding how to achieve the best performance. This section details the current performance of VR systems and proposes innovative ways to improve performance. These methods focus on two possibilities; reducing the data to be sent through tiling the video and use of DASH, and shifting some of the core functions from a distant server to the very edge of the wireless network - the wireless access point (AP).
Permanent Link
http://digital.library.wisc.edu/1793/76499Type
Thesis
Description
Thesis Advisor: Professor Xinyu Zhang