Annotation of the results obtained in 2021
The goal of this project is to develop new cross-layer algorithms and methods for improving the quality of service for augmented and virtual reality (AR/VR) application traffic. In the first year of the project, AR/VR application models and algorithms were developed to identify AR/VR application network traffic flows and improve their quality of service. In particular, the following results were obtained. The traffic of virtual reality applications has been studied and a simulation model of a VR application has been developed, which allows generating traffic for the transmission of streams with specified properties (stream duration, video resolution, parameters of the video encoding algorithm used) in the NS-3 simulation environment. Using the developed model, it was demonstrated that by changing the parameters of the video encoding algorithm, it is possible to significantly reduce quality of service requirements of VR streams and increase the capacity of the wireless network up to two times compared to the parameters used in actual VR applications. The team has developed an algorithm for the automatic collection and labeling of traces of network traffic flows. We used the developed algorithm to collect and label a database of encrypted traffic of various types of data. We have shown that traffic encrypted using modern versions of the Transport Layer Security (TLS) protocol can be accurately classified by the public SNI (Server Name Indication) field of the first TLS message containing the domain name of the server generating the flow. It is expected that SNI field will be encrypted in future versions of the TLS protocol because it reveals sensitive information about the connection. The team has developed an algorithm for encrypted traffic classification in the scenario with hidden SNI field. The algorithm predicts traffic class based on metadata that will not be protected in future versions of the TLS protocol, e.g., cipher suites and key length. It is shown that the proposed algorithm has a 5 times lower error rate and is 3 times faster than state-of-the-art solutions. To meet the high quality of service requirements for AR/VR traffic it is planned to use 5G cellular networks with base stations operating in significantly different frequency bands using the Multi-Connectivity feature. The team analyzed existing traffic balancing algorithms for cellular networks and, using simulation, shown that considered algorithms cannot satisfy the strict requirements for the quality of service of AR/VR traffic without excessive consumption of channel resources. The team developed a new algorithm that explicitly takes into account the quality of service requirements of AR/VR traffic and minimizes the share of consumed channel resources. Developed algorithm provides a 2-fold increase in the network capacity and reduces the use of low-frequency channel resources 10 times compared to existing traffic balancing algorithms. Deployment of the developed solutions will improve the quality of service for AR/VR applications and reduce the consumption of channel resources by this type of traffic. Thus, it is possible both to improve the quality of experience of AR/VR users, and the network capacity. In the first year of the project 7 publications were published, out of which three papers are indexed by the WoS/Scopus databases.

 

Publications

---

Annotation of the results obtained in 2022
An important feature of wireless data transmission technologies is that the user’s movements cause changes in the state of the wireless channel and, as a consequence, in the network bandwidth. Thus to provide the highest quality virtual reality experience for users, cloud-based AR/VR applications must adapt the bitrate of video streams to varying network bandwidth. In the Project, we extended the cloud VR application simulation model developed in the previous phase of the project. We added new features, such as feedback latency, encoding, and playback pipelines, and asynchronous adaptive bitrate selection. The new VR QoS model takes into account the VR update delays between the user motion capture to the VR frame display, as well as the encoding and decoding delays affecting the VR frame generation time. Using the developed model, we investigated the influence of the playback buffer depth on the effective capacity of the WLANs for VR traffic. Furthermore, we developed a method for selecting the optimal buffer depth to satisfy the VR update delay requirements and maximize the capacity network. Numerical results showed that the optimal depth buffer allows up to two times higher effective network capacity compared to a system without a buffer. In addition, we used the developed model to study the performance of the known from the literature VR video bitrate selection algorithms in wireless local area networks. It was shown that the existing algorithms have significant potential to improve the quality of service and the effective capacity of wireless networks for virtual reality traffic. As part of the work on the Project, we developed an architecture for DeSlice virtualization of wireless channel resources of 5G networks (RAN slicing). The developed architecture decomposes the wireless channel resource allocation problem into long-term and short-term problems within and between subnets. This approach provides high quality of service (QoS) for different types of traffic, increases resource utilization efficiency and spectral efficiency, reduces computing load on network devices, and guarantees isolation between operators of virtual subnets. Further, the DeSlice wireless resource virtualization architecture was adapted for use in the 5G networks with MU-MIMO technology. We have developed a method for estimating the amount of resources sufficient to meet the QoS requirements for VR traffic and algorithms to ensure the allocation of a given share of resources to VR streams. The results of simulations have shown that the proposed solution based on the DeSlice architecture can increase the capacity of 5G MU-MIMO systems for VR cloud applications traffic by up to 50% and simultaneously provide at least a 25% increase in the download speed of web pages over the entire considered range of network load. Different traffic categories (video traffic, audio traffic, telephony) require different packet delays, packet loss, and jitter. The provisioning of characteristics needed for a specific traffic category is called Quality of Service (QoS) provisioning. QoS provisioning requires classifying traffic categories first. However, the traffic classification problem became more complicated with the release of the new cryptographic protocol TLS Encrypted ClientHello (ECH). The protocol significantly increases confidentiality and does not transmit in plaintext any parameters that allow accurately classifying traffic categories. As part of the work on the Project, we developed two classification algorithms working in ECH scenarios. The first one is a hybrid algorithm based on the analysis of unencrypted payload and statistical flow data (packet lengths and packet time intervals). It decreases the classification error rate by 11 times compared to algorithms based on payload analysis. The second one considers the correlation between the flows. This algorithm is 1.5 times more effective than the hybrid algorithm in terms of classification error rate. All the results were achieved on the dataset remotely collected in five countries (Germany, Russia, the USA, Spain, and Turkey). The dataset consists of 600 000 download traces of 5 traffic categories: uplink video streaming traffic, buffered video traffic, buffered audio traffic, short buffered traffic, and web traffic. We also developed a cloud traffic classification platform for QoS provisioning. The platform is based on the classifier trained on the collected dataset. With a set of experiments, we showed that the platform increases the mean YouTube bitrate by two times through accurate classification and the use of the proper traffic policy. To improve the quality of service (QoS) of augmented and virtual reality (AR/VR) traffic in the fifth-generation cellular systems (5G), network operators may enable the Multi-Connectivity feature which allows to simultaneously connect a user equipment (UE) to several base stations operating in low-frequency and millimeter-wave frequency bands. In addition to Dual Connectivity where a UE connects to a single millimeter-wave base station, the scenario of connecting a UE to two millimeter-wave base stations simultaneously was considered. The team studied various algorithms for traffic balancing and link management in terms of achieved network capacity and channel resource consumption. As shown in numerical results, the Multi-Connectivity feature allows us to meet the strict QoS requirements of AR/VR traffic and reduce the usage of channel resources. However, the increase in network capacity in the case of Multi-Connectivity is marginal. Therefore, taking into account the increase in cost and power consumption of the mobile device due to support for additional connection, it is preferable to use the Dual Connectivity feature with correctly selected traffic balancing and link management algorithms.

 

Publications

---