Monitoring Network Congestion in Wi-Fi, based on QoE in HTTP Video Steaming Services

Abstract: Improvements in Internet technology, development of multimedia applications, protocols and improvement in user devices have led to the popularity of multimedia applications, among which video streaming applications are most popular. Streaming video services are sensitive to network conditions, thus making Quality of Experience (QoE) of end users sensitive to network conditions. QoE is affected by small disturbances in network conditions and end users observe this as blurred video or lost scenes. This may lead to end users giving up the service or switching to another network operator. To avoid this, network operators and service providers need to maintain QoE at a satisfactory level. The purpose of this study is to develop a monitoring method, which can monitor network congestion in Wi-Fi, based on QoE in HTTP video streaming services. This study proposes a QoE assessment method based on Machine Learning (ML), which allows network operators and service providers to predict QoE from network level measurements. This study was conducted in four steps. Initially, network monitoring probes were designed to measure key metrics that affect QoE, which involved development of QoE assessment model based on relationship between Quality of Service (QoS) and QoE, and implementation of an active measurement protocol called Two-Way Active Measurement Protocol (TWAMP) for network level measurements. Subsequently, a direct link was established between subjective QoE and objective network measurements by designing various test cases. Data was collected by performing network measurements on a Wi-Fi testbed to study the impact of wireless rate adaptation and link utilization on QoE by loading WLAN with cross traffic on downlink or bi-directional paths along with YouTube video. A ML approach was then used to classify network level measurements into QoE levels. A set of ML algorithms: SVM, KNN and Logistic Regression were tested and evaluated to build a classification model to be used in network monitoring system module within network management system. Ultimately, the performance of the proposed QoE assessment method was evaluated using five test cases. The results show that this method performed well and give high classification accuracy in all cases. Outputs from this work may be used by network operators and service providers to modify their network management system by developing effective congestion management solutions to bring back QoE to satisfactory level.