TL;DR
This paper tackles the challenge of delivering a high-quality video streaming experience on mobile devices by introducing an "environment-aware" approach to HTTP Adaptive Streaming (HAS). By leveraging mobile sensors to understand the user's environment (e.g., noisy bus vs. quiet library), the system dynamically adjusts streaming strategies to prioritize either video fluency or clarity, leading to demonstrably improved Quality of Experience (QoE).
What Problem Does This Paper Address?
HTTP Adaptive Streaming (HAS) has become the de facto standard for video delivery over the internet, dynamically adjusting video quality based on network conditions. However, existing HAS mechanisms largely overlook a crucial factor influencing user experience: the user's environment. Mobile users watch videos in diverse settings – a bustling bus, a quiet library, a busy street – each with unique characteristics like noise levels and motion. These environmental factors significantly impact what users prioritize in their viewing experience. For instance, in a noisy, moving environment, users might tolerate lower video clarity for smoother playback, while in a quiet, static setting, they might prefer higher visual fidelity. Current HAS systems, by failing to account for these environmental nuances, often fail to optimize for the user's actual priorities, leading to suboptimal Quality of Experience (QoE).
Methods
The core of this research lies in developing an "environment-aware QoE model" and integrating it into a HAS rate adaptation algorithm.
- Environment Classification: The system utilizes mobile phone sensors, specifically the accelerometer (for motion detection) and microphone (for noise level), to classify the user's environment into three categories: "dynamic and noisy," "quiet and static," and "normal." This classification is achieved through statistical analysis of sensor data, with defined thresholds for acceleration standard deviation and noise levels.
- Environment-Aware QoE Model: A novel QoE model is proposed that quantifies the impact of video clarity and fluency on user experience, weighted differently based on the identified environment. The model incorporates an "environmental impact adjustment function" that dynamically balances the importance of clarity and fluency. This function is derived from fitting sensor data to observed Mean Opinion Scores (MOS) from user studies.
- Environment-Aware HAS Rate Adaptation Algorithm: The proposed QoE model is then used to modify a benchmark HAS rate adaptation algorithm (specifically, BOLA - Buffer-based Optimal Rate Adaptation). The algorithm's objective function is adjusted to prioritize either fluency (in dynamic/noisy environments) or clarity (in quiet/static environments) based on the output of the environment-aware QoE model. This leads to differentiated rate adaptation strategies tailored to specific environmental conditions.
Datasets
The paper utilizes two main sources of data:
- Sensor Data: Sensor data (accelerometer and microphone) was collected from 871 Android users over six months as part of a real mobile learning system. This data was used for training and validating the environment classification and the QoE model.
- Subjective MOS Scores: MOS scores were collected from 381 volunteer participants who watched videos in various real-world environments (campus, bus, home, library). These scores were crucial for validating the proposed QoE model against existing ones.
The paper does not introduce a new benchmark dataset but rather uses existing video content (from dash.js development community) and collects new MOS scores in diverse environments to validate their model.
Key Insights and Ideas
- Environment Matters for QoE: The most significant insight is that user environment is not a trivial factor but a critical determinant of what constitutes a good video streaming experience. Prioritizing fluency over clarity in a noisy bus is fundamentally different from prioritizing clarity in a quiet room.
- Sensors as QoE Enablers: Mobile phone sensors, often overlooked for streaming optimization, are powerful tools for inferring user context and thus enabling more personalized and effective HAS strategies.
- Differentiated Adaptation is Key: A one-size-fits-all HAS approach is insufficient for mobile users. The paper demonstrates that a differentiated rate adaptation strategy, informed by environmental context, leads to superior QoE.
- Bridging the Gap Between Objective Metrics and Subjective Experience: The research effectively bridges the gap by creating a QoE model that translates objective sensor data into subjective user preferences, guiding the HAS algorithm.
Key Results
- Improved QoE: Experimental results show that the environment-aware HAS scheme provides higher QoE compared to the benchmark BOLA algorithm. In static and quiet environments, the proposed scheme achieved a higher MOS score (4.05 vs. 3.79) and a higher calculated QoE (4.25 vs. 3.61) by increasing the average bit rate. In dynamic and noisy environments, it significantly reduced stalling time (4 vs. 13 occurrences) and improved MOS (3.79 vs. 3.09) and calculated QoE (3.92 vs. 2.97), even with a slightly reduced average bit rate.
- Accurate Environment Classification: The environment classification achieved high accuracy rates, with the acceleration sensor reaching 96% and the noise sensor reaching 97%.
- Model Validity: The proposed environment-aware QoE model demonstrated better alignment with real MOS scores compared to models that did not consider environmental factors.
Limitations and Future Work
- Limited Environmental Factors: The current study primarily focuses on two environmental factors: motion (via accelerometer) and noise (via microphone). The authors acknowledge that other context information, such as device type, network type, and user behavior, also influences QoE.
- Future Work: The authors propose to further explore the relationship between these additional context information types and QoE evaluation methods. They also aim to improve and enrich the environment-aware QoE model for actual video-on-demand applications to provide even higher QoE.
Summary updated