Adaptive Partially Reliable Delivery of Immersive Media Over QUIC-HTTP/3

Abstract

The increasing popularity of head-mounted displays (HMD) and depth cameras has encouraged content providers to offer interactive immersive media content over the internet. Traditionally, dynamic adaptive streaming over HTTP (DASH) is the go-to standard for video streaming. However, HTTP is built on top of protocols such as the transmission control protocol (TCP), which prioritize reliability over latency, thereby, inducing additional delay due to acknowledgments and retransmissions, especially on lossy networks. In addition, such reliable protocols suffer from head-of-line (HOL) blocking problem at various levels, leading to playout interruptions of the video streaming application. The third generation of HTTP, i.e., HTTP/3 was recently introduced to deal with the issues posed by TCP. As a major change, HTTP/3 replaces TCP with QUIC at the transport layer, which solves the HOL problem at the transport layer. Moreover, the datagram extension of QUIC allows for unreliable data delivery, just like UDP, which could substantially reduce the latency. Combining this feature of QUIC with the quality adaptation capabilities of DASH-based streaming could bring interactive immersive media delivery to the next level. This work proposes the integration of DASH with the concept of partial reliability of QUIC to reduce playout interruptions and increase the quality of the delivered immersive content on lossy networks. Here, the DASH scheme takes quality and prioritization decisions based on the changing network conditions and the user's viewport, respectively. Then, the part of the content with the highest priority, i.e., within the viewport, is delivered reliably and the rest unreliably. To the best of our knowledge, this is the first work to combine adaptive streaming with partial reliability. Herein, we provide an implementation of a headless player which supports HTTP/3 over partially reliable QUIC as well as state-of-the-art protocols like HTTP/3 over reliable QUIC and HTTP/2 over TCP. We performed an extensive evaluation of our proposed solution using real-world 5G throughput traces and bursty packet loss conditions, using point cloud streaming as the use case. Firstly, our evaluation shows that HTTP/2 is highly intolerant to loss and not suitable for streaming immersive media. Furthermore, even at a loss as high as 5%, the partially reliable framework achieves 46% higher throughput and delivers the content with 33% fewer playout interruptions compared to the reliable counterpart. Since current point cloud decoders are sensitive to loss, we applied the forward error correction mechanism to the data sent unreliably to ensure that the client decodes the content at a probability of 99.9%. Applying this overhead to our solution provides a significant gain of 25% in the throughput compared to the state of the art.