英文学术报告.ppt

Modeling the QoE of Rate Changes in Skype/SILK VoIP CallsLayoutnBACKGROUNDnPRELIMINARY EXPERIMENTnPROPOSED MODELFixed-Rate ModelVariable-Rate ModelnLARGE-SCALE EXPERIMENTnEVALUATIONnCONCLUSION AND FUTURE WORKBackgroundnTHE EFFECTIVE end-to-end transport of delay-sensitive voice data has long been a subject of study in multimedia networking.user-centric approachrate adaptation mechanismsWeberFechners LawnEarly rate-control mechanisms attempted to improve QoS for the data delivered.An increasing trend in this area is to measure network services using the quality of experience(QoE).QoSQoE Synergyrate adaptation mechanismsQoE-Based AdaptationPsychophysicsPRELIMINARY EXPERIMENTMOSbit-rate plot of fixed rate testsThe purpose of the preliminary experiment in this study was to examine,through user tests,whether changes in sending rates decreases the user experience,and at what scale the MOS is influenced by the magnitude and frequency of changes.PROPOSED MODELA.Fixed-Rate Modelwhere br is the bit rate,and,and are coefficients to be determined by SILK characteristics.The bit-rate shift()is caused by the limit of human perception.According to the formula,the term br-,being inside the operator,can only be positive.PROPOSED MODELB.Variable-Rate Modelwhere the effect of change frequency is distributed to the logarithm term,and the effect of hr and lr is distributed to the two subroutines,SCALE()and SHIFT().LARGE-SCALE EXPERIMENTA.ANOVA TestsNonparallel MOST relationships indicate interactionIn that,and are the mean and standard deviation of all scores obtained from tracks of the same(hr,lr)pair.The z-score was then used for ANOVA tests.LARGE-SCALE EXPERIMENTB.Model Specifics1)Coefficients of Fixed-Rate Formula2)Subroutine of Variable-Rate Formula3)Subroutine of Variable-Rate FormulaEVALUATIONA.Prediction AccuracyGoodness of fit of training data The scatter plot in Fig shows the fitted score(as the-axis)and the user score(as the-axis)for each rate pair and time interval test setEVALUATIONA.Prediction AccuracyAccuracy of prediction Fig shows the accuracy of the proposed model.In this figure,the predictions of both data sets closely track the measured data,regardless of the difference in audio content and speakerEVALUATIONB.Comparison to PESQ1)Fixed RatePrediction error:SILK versus PESQ for fixed-rate tracksFor the fixed-rate comparison,we included tracks in both the preliminary and large-scale experimentEVALUATIONB.Comparison to PESQ2)Variable RatePrediction error:SILK versus PESQ for fixed-rate tracksA similar comparison was also conducted for variable tracks,and the accuracy of prediction of both the proposed model and PESQ are indicated in FigCONCLUSION AND FUTURE WORKA.QoE MeasurementIn this study,we quantify the relationship between user experience and bit rate changes for Skype/SILK calls.B.Network DesignA solid understanding of multimedia content is fundamental to better engineering of the network applications.C.Multimedia ContentTaking Skype as the first research target,this study investigates the influence of different bit rates and rate changes of audio streams produced by the SILK codec.Thank You!