Cross-layer optimisation of quality of experience for video traffic

Realtime video traffic is currently the dominant network traffic and is set to increase in volume for the foreseeable future. As this traffic is bursty, providing perceptually good video quality is a challenging task. Bursty traffic refers to inconsistency of the video traffic level. It is at high level sometimes while is at low level at some other times. Many video traffic measurement algorithms have been proposed for measurement-based admission control. Despite all of this effort, there is no entirely satisfactory admission algorithm for variable rate flows. Furthermore, video frames are subjected to loss and delay which cause quality degradation when sent without reacting to network congestion. The perceived Quality of Experience (QoE)-number of sessions trade-off can be optimised by exploiting the bursty nature of video traffic. This study introduces a cross-layer QoE-aware optimisation architecture for video traffic. QoE is a measure of the user's perception of the quality of a network service. The architecture addresses the problem of QoE degradation in a bottleneck network. It proposes that video sources at the application layer adapt their rate to the network environment by dynamically controlling their transmitted bit rate. Whereas the edge of the network protects the quality of active video sessions by controlling the acceptance of new sessions through a QoE-aware admission control. In particular, it seeks the most efficient way of accepting new video sessions and adapts sending rates to free up resources for more sessions whilst maintaining the QoE of the current sessions. As a pathway to the objective, the performance of the video flows that react to the network load by adapting the sending rate was investigated. Although dynamic rate adaptation enhances the video quality, accepting more sessions than a link can accommodate will degrade the QoE. The video's instantaneous aggregate rate was compared to the average aggregate rate which is a calculated rate over a measurement time window. It was found that there is no substantial difference between the two rates except for a small number of video flows, long measurement window, or fast moving contents (such as sport), in which the average is smaller than the instantaneous rate. These scenarios do not always represent the reality. The finding discussed above was the main motivation for proposing a novel video traffic measurement algorithm that is QoE-aware. The algorithm finds the upper limit of the video total rate that can exceed a specific link capacity without the QoE degradation of ongoing video sessions. When implemented in a QoE-aware admission control, the algorithm managed to maintain the QoE for a higher number of video session compared to the calculated rate-based admission controls such as the Internet Engineering Task Force (IETF) standard Pre-Congestion Notification (PCN)-based admission control. Subjective tests were conducted to involve human subjects in rating of the quality of videos delivered with the proposed measurement algorithm. Mechanisms proposed for optimising the QoE of video traffic were surveyed in detail in this dissertation and the challenges of achieving this objective were discussed. Finally, the current rate adaptation capability of video applications was combined with the proposed QoE-aware admission control in a QoE-aware cross-layer architecture. The performance of the proposed architecture was evaluated against the architecture in which video applications perform rate adaptation without being managed by the admission control component. The results showed that our architecture optimises the mean Mean Opinion Score (MOS) and number of successful decoded video sessions without compromising the delay. The algorithms proposed in this study were implemented and evaluated using Network Simulator-version 2 (NS-2), MATLAB, Evalvid and Evalvid-RA. These software tools were selected based on their use in similar studies and availability at the university. Data obtained from the simulations was analysed with analysis of variance (ANOVA) and the Cumulative Distribution Functions (CDF) for the performance metrics were calculated. The proposed architecture will contribute to the preparation for the massive growth of video traffic. The mathematical models of the proposed algorithms contribute to the research community