www.dsp-ip.com Fast Forward Your Development Video Quality Measurements
Jan 15, 2015
www.dsp-ip.comFast Forward Your Development
Video Quality Measurements
www.dsp-ip.comFast Forward Your Development
Overview
•QoS and QoE•Measurement Methods
▫Objective, Subjective, payload/codec based•Measurement and Monitoring Goals
▫Lab, pre-deployment, Monitoring, failure analysis
•Monitoring Levels▫Transport, Transaction, content
•Monitoring Domains ▫Head-end, network, Home
www.dsp-ip.comFast Forward Your Development
User Quality of Experience factors (QoE)
•What does subscribers want?▫Content
Content quality Content pricing Content availability
▫Viewing experience Video Quality Channel Zapping delay
▫Integration of services Usability One bill New services
▫Customer services
Technical Factors
www.dsp-ip.comFast Forward Your Development
MEASUREMENT TAXONOMYSubjectiveObjectivePayload based, codec aware, codec anaware
www.dsp-ip.comFast Forward Your Development
Measurement methods review
•Subjective▫Human assessment of quality▫Expensive, not for monitoring▫Not repeatable
•Objective▫Measurement devices▫Repeatable▫For both testing and monitoring, failure
analysis
ObjectiveSubjective
Measurements
www.dsp-ip.comFast Forward Your Development
Multimedia monitoring methods
Subjective
MOS(Voice)
BT500(Video)
Measurements
•Subjective measurement arte done by humans according to pre-defined protocols
•Voice – MOS▫Grade voice quality between 1-5▫4+ - is very good quality
•Video – BT500
www.dsp-ip.comFast Forward Your Development
Objective methods
Measurements
•Objective methods are divided into:•Payload based•Packet based
▫CODEC Aware ▫CODEC Independent
•Network Monitoring
www.dsp-ip.comFast Forward Your Development
Payload based Measurements•Payload base methods assess the video
quality based on the video itself•Reference based methods compares
Original video quality to distorted video.•Used mostly in lab equipment for codec
performance analysis and comparison
www.dsp-ip.comFast Forward Your Development
Full Reference Methods
PSNR
FullReference
Objective Payload based
MSE
SSIM
J.144
•Full reference methods compares each frame of the original video to frame of the distorted video and provide distortion measurement
•MSE & PSNR are pixel based similarity measurements
•Most video quality measurements are done on Luma (Y)
www.dsp-ip.comFast Forward Your Development
PSNR & MSE •PSNR and MSE is a pixel base video
quality comparison tools
MSE
LPSNR
2
10log10
www.dsp-ip.comFast Forward Your Development
MSE=0, MSSIM=1
MSE=225, MSSIM=0.949 MSE=225, MSSIM=0.688 MSE=225, MSSIM=0.723
original Image
PSNR/MSE Problems•Quality degradation is not
reflected by MSE/PSNR index
www.dsp-ip.comFast Forward Your Development
SSIM – “Solves” PSNR Problems•SSIM is Structural Similarity index.
Unlike PNR and MSE it does not compare images “pixel by pixel” but as small NxN “environments”
• are average variance and covariance
• the dynamic range of the pixel-values (typically )
www.dsp-ip.comFast Forward Your Development
ITU-T J.144 and ITU-R BT.1683 Full-reference perceptual models•Digital TV•Rec. 601 image resolution (PAL/NTSC)•Bit rates: 768 kbps ~ 5 Mbps•Compression errors
Full Reference testing standards
www.dsp-ip.comFast Forward Your Development
Back to Objective measurement
Measurements
•What happens when we don’t have the original (Reference video) or when we don’t have the processing power to do an extensive comparison?
•We could use network measurements and codec based degradation info to asses video quality
www.dsp-ip.comFast Forward Your Development
Packet – Codec Aware•Monitoring technique•Degrades video based on codec type by
incorporating network parameters data with codec behavior data
•Scales- could monitor thousands of channels
•Examples:▫ VQS (Telchemy)▫ VQI(Brix)▫ V-Factor (QoSMetrics)
0
5
10
15
20
25
30
35
0 5 10 15 20
Packet Loss (%)
PS
NR
(d
B)
Problem area
Robust codec
“Raw”codec
The need a codec aware metrics
www.dsp-ip.comFast Forward Your Development
Codec Aware Methods
Codec awarePacket based
VQI
V-Factor
VQSTelchemy
Objective methods
www.dsp-ip.comFast Forward Your Development
Example V-Factor
•Based on MPQM (Moving Picture Quality Metrics) – high quality video measurement standard
•V = f(QER, PLR, R)▫QER – relative video codec quality▫PLR – Packet loss ratio (based on actual
packet loss, jitter data and jitter buffer model)
▫R – Image complexity factor (2-3)•Adopted by Spirnet
www.dsp-ip.comFast Forward Your Development
Packet – Codec Independent
•Monitoring only•Codec independent•Based on network parameters data only•Scales - could monitor thousands of
channels•Examples:
▫MDI IneoQuest standardized by IETF
www.dsp-ip.comFast Forward Your Development
MEASUREMENT & MONITORINGIn the Lab & In The FieldsPre-Deployment/monitoring/Failure Analysis
www.dsp-ip.comFast Forward Your Development
Measurement & Monitoring phases
AnalysisProblem Solving
Tuning
PreDeployment
Testing
LabTesting
Design
24/7Monitoring
DeploymentDeploymentPhasePhase
Pre-Pre-DeploymentDeploymentPhasePhase
www.dsp-ip.comFast Forward Your Development
Measurement & Monitoring phases
•Design & Lab testing▫Simulation and Emulation of the network▫Lab and testing tools
•Pre Deployment Stage▫Work on actual network▫Load testing▫Lab, testing, diagnosing and monitoring tools
•Deployment (production) Phase▫Mostly monitoring (probes) equipment,
management systems, data filtering and diagnostics equipment
www.dsp-ip.comFast Forward Your Development
MONITORING LEVELS
www.dsp-ip.comFast Forward Your Development
Measurement Levels
•Transport Level•Service (transaction) Level•Media Quality Level
Video Quality
Transport Quality
Transaction Quality
www.dsp-ip.comFast Forward Your Development
Transaction Level•Examples
▫Post-dial delay in PSTN/mobile networks▫Video start time for channel zapping &
Video conf•Requires understanding in both network
monitoring and signaling (IGMP, SIP) and in media coding (analysis of the media to discover dial tone or I frame)
www.dsp-ip.comFast Forward Your Development
Channel zapping delay
•Multicast saves bandwidth but creates signaling delays:▫Multicast Leave +▫Multicast Join + ▫First I Frame +▫Up to 2 seconds buffering time
Leave latency
Join latency
Signaling Latency
First I Frame
Media Latency
Total Channel zapping Latency
Buffering latency
First frame viewed
www.dsp-ip.comFast Forward Your Development
Transport Level
•Example: ▫Packet Loss▫Loss Patterns▫Jitter▫Delay
•Well understood•Defined by ITU and IETF
www.dsp-ip.comFast Forward Your Development
Content Level• Content quality is a payload based measurement.
▫ Requires decoding of the video stream▫ Understanding of the buffering and error concealment
algorithms of the decoder • CPU intensive – Does not scale• Accurate• Used mostly is Lab equipment and diagnostic
equipment• Examples:
▫ PSNR ▫ ITU-T J.144
• Usually requires the reference (original) stream• Tests:
▫ Source artifacts▫ Source quality
www.dsp-ip.comFast Forward Your Development
Standardization landscape
DSL
DSL Forum
TR-64, TR-69LAN and WAN monitoring standards
ITU
Study Group 12
Algorithms for end-to-end transmission performance
ITU
VQEG – Video Quality Expert Group
Video performance measurement based on Subjective tests Database
Used in Diagnostics / Lab
ATIS
IIF – IPTV Interoperability Forum
QoS Metrics Standardization
Used for Monitoring
www.dsp-ip.comFast Forward Your Development
Example: ATIS IIF Quality Metrics
VSTQ - Video Service Transmission Quality ▫ Transmission Quality - codec/ content independent▫ Based on the rate and distribution of effective packet loss and
discardVSPQ - Video Service Picture Quality
▫ Estimated viewing quality▫ Considers the impact of VSTQ, video codec type and rate,
resolutionVSAQ - Video Service Audio Quality
▫ audio listening quality▫ Considers the impact of VSTQ, audio codec type, sample rate, …..
VSMQ - Video Service Multimedia Quality ▫ overall user experience▫ Combined effect of VSPQ, VSAQ, audio-video synchronization..
VSCQ - Video Service Control (Plane) Quality ▫ Considers responsiveness and reliability of control plane (trick
play)
www.dsp-ip.comFast Forward Your Development
Monitoring levels• J.144 and PSNR examines the video content only (payload
measurements)• TR101290 examines only transport stream data and
coherence without examining the video content• V -Factor and VQS looks at packet loss, jitter and loss patterns
data and incorporate it with codec information and video header information
• MDI – Examines only packet loss and packet loss patterns without considering the codec or video information
TR101290MPEG2TS Headers
V-Factor, VQSMDI
J.144, PSNRVideo payload