Enhancing QoE for Multi-Device Video Delivery: A Novel Dataset and Model Perspective

Multi-device video streaming applications enable seamless playback across various devices, including large-screen TVs, tablets, and smartphones, revolutionizing digital content consumption and enhancing user experience. However, ensuring consistently high quality of experience (QoE) across these heterogeneous devices remains a substantial challenge due to intrinsic differences in screen sizes and viewing conditions. In this paper, we first build an open-source, multi-device, and time-continuous QoE dataset named MCQoE by conducting a large-scale subjective experiment to analyze QoE variations among different screen-size devices. Then, we thoroughly investigate the dataset and observe that video quality and rebuffering impact on TVs is more significant than on other devices, such as middle-size PC monitors and small-screen smartphones, emphasizing the importance of building specific QoE models for different devices. Furthermore, we propose a novel low-complexity but effective QoE model denoted as LiteDC, integrating a temporal dilated convolution network with a targeted pruning technique to align with the computational constraints of embedded platforms. Extensive results show that compared to a state-of-the-art baseline algorithm, LiteDC achieves a remarkable 20.9-fold improvement in execution speed while increasing prediction accuracy by 6.4%. The MCQoE dataset is available for download at https://github.com/yanghaocuc/mcqoe.

I. Introduction

The advent of multi-device video streaming applications, which provide the flexibility to access videos on a variety of devices such as large-screen televisions (TVs), medium-sized tablets and small-screen smartphones, has revolutionized the way users consume digital content. As shown in Figure 1, this technological advancement has significantly enriched the user experience by enabling seamless video playback across different device types, catering to varying user preferences and contexts. According to Conviva’s global statistical analysis in 2022, large-screen TVs, including Smart TVs, Connected TV devices, and Gaming Consoles, constituted 76.7% of total viewing hours, whereas smartphones accounted for 11.3% [1]. However, ensuring a consistently high Quality of Experience (QoE) across these heterogeneous devices presents substantial challenges due to the intrinsic differences in screen sizes and viewing conditions [2]. Understanding the intricacies of user satisfaction and engagement across these diverse devices is essential for enhancing the QoE and the effectiveness of Adaptive Bitrate (ABR) algorithms [3], [4], [5], thereby driving the development of more efficient and robust streaming technologies. Fig. 1.

Framework of multi-device video streaming system.