E2E Service Performance Enhancement for Tile-based Adaptive 360° VR Video Streaming

Abstract

Network slicing technique enables a flexible and programmable network architecture. Multiple virtual networks (also known as network slices) are created over a shared physical network for customized and fine-grained service provisioning. Due to the immersive viewing experience and enormous vertical markets, 360° virtual reality (VR) video streaming service has attracted significant attention. Users wear a head-mounted display (HMD) to enjoy spherical video content and freely change their viewing orientations during the streaming. Delivering smooth and high-quality VR video content requires high end-to-end (E2E) transmission rate and low latency, posing significant challenges to current mobile networks. To reduce the requirement for a high transmission rate, a 360° video can be partitioned into multiple non-overlapping (temporal-spatial) video tiles, and only the video tiles covered by the predicted field-of-view (FoV) are transmitted with high bitrate levels. Nevertheless, tile-based 360° video streaming suffers from transmission rate fluctuations and inevitable FoV prediction errors driven by viewing behavior dynamics (i.e., head movements). Besides, encoded video tiles exhibit various properties in terms of transmission priority, deadline, and reliability requirement. To support enhanced service provisioning for tile-based 360° video streaming, requested video content needs to be efficiently delivered by considering various video tile properties under network and viewing behavior dynamics. Meanwhile, users must adaptively request video tiles and select bitrate levels based on FoV prediction results and available transmission rate. In this thesis, we focus on E2E enhancement for tile-based 360° video streaming service with core network slicing, where VR video streaming slices are deployed in the core network to facilitate enhanced VR video packet delivery.

First, we propose a customized transmission protocol based on Quick UDP Internet Connections (QUIC) which operates over a VR video slice in the core network. The QUIC protocol is tailored to accommodate the characteristics of tile-based VR video streaming where explicit mapping relations between requested video tiles and QUIC streams are established. Two customized in-network protocol functionalities including packet filtering and caching-based packet retransmission are proposed, to filter out outdated video data due to FoV prediction errors and to achieve efficient packet retransmissions with disparate transmission reliability requirements. A slice-level packet header is designed to support enhanced slice-based VR video packet transmission with the proposed protocol functionalities.

Second, we develop a performance analytical model for layer-encoded video packet services (i.e., processing or transmission) over a video streaming slice in the core network. The disparate service reliability requirements of base layer (BL) and enhancement layer (EL) packets are considered in the analytical model for E2E packet delays, deadline violation probabilities, and throughputs of BL and EL packets. Specifically, a network function virtualization (NFV) node along the routing path of the video streaming slice is split into two consecutive logical nodes, one for packet processing and the other for transmission, based on which a segment-based analysis framework is proposed for E2E service performance modeling. A two-stage queuing model is established to obtain the approximate steady-state probability distribution of queue length at the first node in the first segment, upon which the BL/EL packet delay, deadline violation probability, and throughput at the segment are derived. In addition, the inter-departure time of successive packets departing from the first segment is analyzed based on an approximate M/D/1 system, and the packet departure process at the first segment is approximated as a Poisson process under the assumption of a large packet service rate of the first node. The independence between two consecutive segments is then achieved for analysis tractability, based on which the E2E performance measures are derived.

Lastly, we propose a two-step adaptive scheme for tile-based 360° video streaming to support multi-user enhanced viewing quality-of-experience (QoE) in a time-slotted system. Each video chunk is first prefetched based on the predictive FoV, and the FoV quality of the chunk is then enhanced at a closer-to-playback time instant based on the updated FoV prediction with improved accuracy. Both transmission-driven and device video super-resolution (VSR)-driven methods are adaptively selected to enable device energy-efficient chunk enhancement. At each time slot, the incremental QoE gain of each user is characterized via a time-difference approach, based on which the best candidate chunk is determined. Then, a single-slot problem is formulated for maximizing the total incremental QoE gain while minimizing the total device energy consumption. A particle swarm optimization (PSO)-based iterative solution is proposed to obtain optimal bandwidth allocation, bitrate level selection, and enhancement method selection for multiple users.

This research provides some insights in enhancing tile-based 360° VR video streaming service provisioning, from both network-side and end device-side perspectives.