Quality of service and dependability of cellular vehicular communication networks

authored by
Mark Akselrod
supervised by
Markus Fidler
Abstract

Improving the dependability of mobile network applications is a complicated task for many reasons: Especially in Germany, the development of cellular infrastructure has not always been fast enough to keep up with the growing demand, resulting in many blind spots that cause communication outages. However, even when the infrastructure is available, the mobility of the users still poses a major challenge when it comes to the dependability of applications: As the user moves, the capacity of the channel can experience major changes. This can mean that applications like adjustable bitrate video streaming cannot infer future performance by analyzing past download rates, as it will only have old information about the data rate at a different location. In this work, we explore the use of 4G LTE for dependable communication in mobile vehicular scenarios. For this, we first look at the performance of LTE, especially in mobile environments, and how it has developed over time. We compare measurements performed several years apart and look at performance differences in urban and rural areas. We find that even though the continued development of the 4G standard has enabled better performance in theory, this has not always been reflected in real-life performance due to the slow development of infrastructure, especially along highways. We also explore the possibility of performance prediction in LTE networks without the need to perform active measurements. For this, we look at the relationship between the measured signal quality and the achievable data rates and latencies. We find that while there is a strong correlation between some of the signal quality indicators and the achievable data rates, the relationship between them is stochastic, i.e., a higher signal quality makes better performance more probable but does not guarantee it. We then use our empirical measurement results as a basis for a model that uses signal quality measurements to predict a throughput distribution. The resulting estimate of the obtainable throughput can then be used in adjustable bitrate applications like video streaming to improve their dependability. Mobile networks also task TCP congestion control algorithms with a new challenge: Usually, senders use TCP congestion control to avoid causing congestion in the network by sending too many packets and so that the network bandwidth is divided fairly. This can be a challenging task since it is not known how many senders are in the network, and the network load can change at any time. In mobile vehicular networks, TCP congestion control is confronted with the additional problem of a constantly changing capacity: As users change their location, the quality of the channel also changes, and the capacity of the channel can experience drastic reductions even when the difference of location is very small. Additionally, in our measurements, we have observed that packet losses only rarely occur (and instead, packets are delayed and retransmitted), meaning that loss-based algorithms like Reno or CUBIC can be at a significant disadvantage. In this thesis, we compare several popular congestion control algorithms in both stationary and mobile scenarios. We find that many loss-based algorithms tend to cause bufferbloat and thus overly increase delays. At the same time, many delay-based algorithms tend to underestimate the network capacity and thus achieve data rates that are too low. The algorithm that performed the best in our measurements was TCP BBR, as it was able to utilize the full capacity of the channel without causing bufferbloat and also react to changes in capacity by adjusting its window. However, since TCP BBR can be unfair towards other algorithms in wired networks, its use could be problematic. Finally, we also propose how our model for data rate prediction can be used to improve the dependability of mobile video streaming. For this, we develop an algorithm for adaptive bitrate streaming that provides a guarantee that the video freeze probability does not exceed a certain pre-selected upper threshold. For the algorithm to work, it needs to know the distribution of obtainable throughput. We use a simulation to verify the function of this algorithm using a distribution obtained through the previously proposed data rate prediction algorithm. In our simulation, the algorithm limited the video freeze probability as intended. However, it did so at the cost of frequent switches of video bitrate, which can diminish the quality of user experience. In future work, we want to explore the possibility of different algorithms that offer a trade-off between the video freeze probability and the frequency of bitrate switches.

Organisation(s)
Communication Network Section
Type
Doctoral thesis
No. of pages
99
Publication date
2023
Publication status
Published
Sustainable Development Goals
SDG 11 - Sustainable Cities and Communities
Electronic version(s)
https://doi.org/10.15488/13570 (Access: Open)