Estimating Video Streaming QoE in the 5G Architecture Using Machine Learning

Abstract

Compared to earlier mobile network generations, the 5G system architecture has been significantly enhanced by the introduction of network analytics functionalities and extended capabilities of interacting with third party Application Functions (AFs). Combining these capabilities, new features for Quality of Experience (QoE) estimation can be designed and introduced in next generation networks. It is, however, unclear how 5G networks can collect monitoring data and application metrics, how they correlate to each other, and which techniques can be used in 5G systems for QoE estimation. This paper studies the feasibility of Machine Learning (ML) techniques for QoE estimation and evaluates their performance for a mobile video streaming use-case. A simulator has been implemented with OMNeT++ for generating traces to (i) examine the relevance of features generated from 5G monitoring data and (ii) to study the QoE estimation accuracy (iii) for a variable number of used features.

Accuracy vs. Cost Trade-off for Machine Learning Based QoE Estimation in 5G Networks

“…In our previous work [2], we proposed a three step approach for estimating QoE in 5G networks. Firstly, ground-truth QoE is received via the AF and made available to the NWDAF.…”

“…In our previous work [2] we proposed a possible integration of an ML-based QoE estimation workflow into the 5G architecture, making use of the newly introduced network functions NWDAF and AF. We evaluated the performance of two basic regression methods for an increasing number of features, following a greedy approach.…”

“…It was proven that the solution of the problem (3) is piece-wise linear to α, and various efficient solvers have been developed and are available online. 2 The number ∑ D d=1 δ(w d ̸ = 0) of used features is monotonically decreasing with respect to α, providing the best feature set for each number of used features. When the total cost budget λ 0 is given, we evaluate the total cost ∑ D d=1 c d δ(w d ̸ = 0) for each of those best feature sets, and choose the one giving the lowest regression error under the budget.…”

“…This paper aims at filling this gap by analyzing this trade-off for QoE estimation based on a video streaming use-case. As an enhancement to our prior work [2], we rely on a mechanism which tunes both, the selected features and the QoE estimation model, for a limited number of features. This allows to evaluate the performance in terms of estimation accuracy versus costs in terms of required input data.…”

Joint optimization strategy for QoE-aware encrypted video caching and content distributing in multi-edge collaborative computing environment

“…In 5G networks, video services will become the mainstream, and the contradiction between explosive growth of data volume and QoE is becoming increasingly prominent [1]. In other words, in a 5G network, when users initiate a request for a certain bit-rate video resource, remote storage devices must respond to users by codec operations within the shortest possible time [2]. Therefore, research on QoE-aware video caching and content distribution technology has attracted extensive attentions from academia and industry.…”