assessing effect sizes of influence factors towards a qoe model for http adaptive streaming
DESCRIPTION
Tobias Hoßfeld, Michael Seufert, Christian Sieber, Thomas Zinner Assessing Effect Sizes of Influence Factors Towards a QoE Model for HTTP Adaptive Streaming. 6th International Workshop on Quality of Multimedia Experience (QoMEX), Singapore, September 2014. Abstract: HTTP Adaptive Streaming (HAS) is employed by more and more video streaming services in the Internet. It allows to adapt the downloaded video quality to the current network conditions, and thus, avoids stalling (i.e., playback interruptions) to the greatest possible extend. The adaptation of video streams is done by switching between different quality representation levels, which influences the user perceived quality of the video stream. In this work, the influence of several adaptation parameters, namely, switch amplitude (i.e., quality level difference), switching frequency, and recency effects, on Quality of Experience (QoE) is investigated. Therefore, crowdsourcing experiments were conducted in order to collect subjective ratings for different adaptation-related test conditions. The results of these subjective studies indicate the influence of the adaptation parameters, and based on these findings a simplified QoE model for HAS is presented, which only relies on the switch amplitude and the playback time of each layer.TRANSCRIPT
www3.informatik.uni-wuerzburg.de
Institute of Computer ScienceChair of Communication Networks
Prof. Dr.-Ing. P. Tran-Gia
Assessing Effect Sizes of Influence Factors Towards a QoE Model for
HTTP Adaptive Streaming
Tobias Hoßfeld, Michael Seufert, Christian Sieber, Thomas Zinner
Michael Seufert2
2
Agenda
1. HTTP Adaptive Streaming
2. QoE Influence Factors of HAS
3. Goal of the Study
4. Study Description
5. Key Influence Factors of HAS QoE
6. Simple QoE Model and Applications
7. Conclusion
Michael Seufert3
3
HTTP Adaptive Streaming
HTTP Adaptive Streaming (HAS) was introduced by MOVE Networks in 2007
Many proprietary streaming technologies exist Apple HTTP Live Streaming Adobe HTTP Dynamic Streaming Microsoft Silverlight Smooth Streaming
Recently also standardized in MPEG: Dynamic Streaming over HTTP (DASH)
Video is available in several encodings and split into small segments/chunks
Description file provides information about video chunks
Michael Seufert4
4
HTTP Adaptive Streaming
Client requests small chunks via HTTP Based on how fast the current (and previous) segments are
downloaded, the bit rate of the next segment is selected. Dimensions for adaptation:
Image quality (codec, quantization) Spatial (resolution) Temporal (frame rate, initial delay, stalling)
Bit rate 1
Bit rate 2
Bit rate 3
Bits
Time
Bit rate 1
Bit rate 2
Bit rate 3
Bits
Time
Bit rate
Time
TCP throughput
Requested chunks
1
2
3
Time
Quality Level
Michael Seufert5
HTTP Adaptive Streaming
Quality of Experience model for HAS is missing
Multiple dimensions have to be taken into account Network Parameters
– Initial Delay– Stalling
Video Parameters– Content– Codec– Resolution– Frame Rate– Image Quality
Key influence parameters have to be identified
Adaptation Parameters– Available representations– Frequency (segment length, number of
switches)– Amplitude (quality representation levels)
Context Psychological Effects
– Recency– Noticeable difference
Michael Seufert6
Survey of Subjective Studies
Many aspects are already researched in different works Subjective studies with varying video quality have been conducted
in related fields More than 150 works were surveyed
Survey paper currently under submission, technical report with preliminary version is available online (University of Würzburg, Institute of Computer Science, Tech. Rep. Nr. 490, online)
HTTP Adaptive Streamin
g
Video Quality
Human Computer Interaction
Networking
etc.
Michael Seufert7
7
Goal of the Study
Subjective investigation of adaptation parameters Switch amplitude (i.e., quality level difference) Switching frequency Recency effects
FP7 Project SmartenIT 10 European partners (academia, network
and cloud operators) Goals:
– Incentive-compatible cross-layer traffic management in the Internet
– Incorporate social awareness and energy efficiency
– Considered cloud applications: Video streaming, file sharing
HAS QoE model can facilitate monitoring and traffic management decisions
http://www.smartenit.eu
Michael Seufert8
8
Study Description
Crowdsourced video quality survey based on QualityCrowd2 framework
14 sec sequence from „Tears of Steel“ in 3 quality layers (resolution reduction, 640x360, 320x180, 160x90)
710 unique participants, 7-9 ratings per user Filtering of unreliable user ratings
Test subjects have to watch whole video sequence User have to rate and answer video content question 11% of users had to be filtered out
Different quality profiles (2 layers) to identify influence of 1 parameter 40 test conditions, minimum 82 (avg. 106) ratings per condition All study details are available in a technical report (University of
Würzburg, Institute of Computer Science, Tech. Rep. Nr. 491, online)
Michael Seufert9
9
Switching Frequency vs Time on Layer
In several works, switching frequency is reported to influence QoE
Often parameters „number/frequency of switches“ and „time on layer“ are correlated and change simultaneously
Keeping „time on layer“ constant, no influence of switching frequency could be found
Michael Seufert10
10
Key Influence Factors of HAS QoE
Determining key influence factors by: Spearman rank-order correlation coefficient One-way analysis of variance (ANOVA) Effect sizes (partial Eta-squared and Cohen‘s f²)
No effect can be observed for last quality level, recency time, and number of switches
Amplitude and time on highest layer are key influence factors
Michael Seufert11
11
Simple QoE Model and Applications
Simple QoE model (2 layers) based on two key influence factors Exponential model provides a very good fit to the data points
(R²=0.98)
Applications of simple QoE model: Monitoring becomes easier New HAS algorithms can be developed QoE-based performance evaluation of HAS algorithms is possible
(e.g., comparison to theoretical QoE-optimum for given conditions)
Michael Seufert12
12
Conclusion
Adaptation strategy related parameters of HAS were investigated and evaluated using subjective tests conducted in a crowdsourcing platform
Main results: Amplitude: high impact of the switching amplitude Frequency: time on highest video quality layer has a significant
impact on the QoE, number of quality switches can be neglected Recency effects: can be neglected
Based on these results, a simplified QoE model for adaptive streaming was derived
Future work: Fortification of findings and investigation of other HAS parameters Extension of simple QoE model Creation of a generally accepted QoE model for HAS Development of new HAS algorithms
Michael Seufert13
Thank you for your
attention!
Bit rate
Time
TCP throughput
Requested chunks
http://www.smartenit.eu