A predictive DASH QoE approach based on machine learning at multi-access edge computing
Md Tariqul Islam · 2021
Multimedia video services traffic is rapidly growing in mobile networks in recent years. Video services using Dynamic Adaptive Streaming over HTTP (DASH) techniques have dominated the total internet traffic to carry video traffic. Mobile Network Operators (MNOs) are expected to run on with this growing demand for DASH-supported video traffic while providing a high Quality of Experience (QoE) to the end-users. Besides, operators need to have a crystal notion of video quality perceived by the end-users and correlate them with network-level monitoring or telemetry information for problem identification, root cause analysis, and pattern prediction. To ensure QoE-aware network traffic management, a prerequisite for the MNOs is to monitor the network traffic passively and measure objective QoE Key Performance Indicators (KPIs) (such as resolutions and stalling events) effectively that directly influence end-user subjective feedback. Many literature approaches have been proposed to measure the KPIs aimed to deliver acceptable video service quality. Most of the solutions require end-user awareness, which is not viable from the MNOs' perspective. However, Deep Packet Inspection (DPI), another most widely used solution to estimate the KPIs directly from network traffic, is not a convenient solution anymore for the operators due to the adoption of end-to-end video streaming encryption over TCP (HTTPs) and QUIC transport protocol. Hence, in recent, Machine Learning (ML) has been accepted as a well-recognized solution for estimating QoE KPIs by analyzing the encrypted traffic patterns and statistics as Quality of Service (QoS).