IPStatRate_Final.pdf

16
Powering IP Video Convergence From Mobile to HD White Paper Envivio IP-based Statistical Rate Control December 2006 Authors: Boris Felts, Jean Kypreos, Thomas Guionnet www.envivio.com

Transcript of IPStatRate_Final.pdf

Page 1: IPStatRate_Final.pdf

Powering IP Video Convergence From Mobile to HD

White Paper

Envivio IP-basedStatistical Rate Control

December 2006Authors: Boris Felts, Jean Kypreos, Thomas Guionnet

www.envivio.com

Page 2: IPStatRate_Final.pdf

Table of Contents

1. INTRODUCTION 2

2. VIDEO QUALITY ASSESSMENT 2

2.1 Subjective Evaluation 2

2.2 Objective Evaluation 3

3. ENVIVIO STATISTICAL RATE CONTROL 6

3.1 Principle 6

3.2 Test Scenario 6

3.3 Results 6

4. OPEN ARCHITECTURE 11

4.1 Architecture and Principle 11

4.2 Open vs. Closed Systems 12

5. CONCLUSION 14

Envivio IP-based Statistical Rate Control

1

Page 3: IPStatRate_Final.pdf

EnvivioIP-basedStatisticalRateControl

1 Introduction Formanyyearsthebroadcastindustryhasusedexpensiveandproprietary

hardwaretoprovidestatisticalmultiplexingfeaturesinMPEG-2transportstream-basednetworks.WiththeadventofIP-basednetworks,thenotionofmultiplexingandde-multiplexingsourcesthatarebundledtogetherisbecominglessimportant,whiletherequirementtofitmultiplechannelsinabandwidth-constrainednetworkremainsessential.

DrawingfromitsexpertiseinMPEG-4andIP-basedvideonetworks,Envivio®hasdevelopeda

nextgeneration,IP-based,statisticalratecontrolmoduleforbroadcastmobileTVnetworksthat

willprovidethehighestqualityvideoservicesinabandwidthconstrainedenvironment.This

technologyisnotlimitedtoDVB-H.Itcanalsobeappliedtootherbroadcastmobilenetworkssuch

asDVB-SH,DMB,3GPPMBMSandMediaFloaswellasbroadbandIPTV,broadcastcontribution

anddistributionapplications.Byaddingstatisticalratecontrolcapabilitieswithinthecoreofthe

4Caster™M2MobileSeriesencoder,Envivioenablesserviceproviderstobothoptimizetheuseof

theirbandwidth,andalsoavoidthecostofpurchasingproprietarystatisticalmultiplexerorstat

muxtechnology.

Thispaperdescribeshowtomeasurethevideoqualityofmultiplecompressedsourcesandthe

qualityandbandwidthgainedbyusingEnviviostatisticalratecontroltechnology.Thispaperalso

describestheEnvivioopenapproachtosystemarchitectureandhowthisapproachprovides

significantbenefitsandadvantagesformobilebroadcastdeployments.

2 VideoQualityAssessment Thissectioncoversseveralmethodsformeasuringvideoqualityandhowitis

possibletomeasurethecumulativequalityofmultiplestreamssimultaneously.

2.1 Subjective Evaluation Subjectiveevaluationreliesonthecomparisonoftheoriginalandcompressedsourcesbymany

Powering IP Video Convergence From Mobile to HD

2

Page 4: IPStatRate_Final.pdf

observersinaverywell-definedcontext.TestprocedureshavebeendefinedbytheVideoQuality

ExpertGroup(www.VQEG.org)tomeasuresubjectivequalityoftelevisionormultimediasignals.

Whilesubjectiveevaluationleadstothemostaccurateresults,itiscomplextoimplement.

Consequently,forthepurposesofthispaper,whilecheckingforaccuracyvisuallyusingsubjective

methodswasemployed,objectivequalitymeasurementsweretheprimarymethodsusedfor

evaluation.

2.2 Objective Evaluation Amongthevariousobjectivecriteriontoevaluatethevideoquality,PSNR(PeakSignaltoNoise

Ratio),PQR(PictureQualityRatio)andSSIM(StructuralSIMilarity),EnviviohasselectedtheSSIM

methodforthisevaluation.Althoughnotperfect,thismeasurementgivesresultsclosertohuman

observationsthanthetwoothersandalsocanbeadaptedtomobileresolutions.

SSIMisamethodformeasuringthesimilaritybetweentwoimages.Itcomparestheoriginaland

thecompressedsequenceandgivesaqualityscorerangingfrom0to100.Avalueof100indicates

thatthereisabsolutelynodifference,whileavalueof0indicatesextremedifferences.

•SingleSource ThefigurebelowshowstheevolutionoftheSSIMscoreonaparticularvideosequencefor

eachframe.TheaverageSSIMgivesanoverallindicationofthequality.

Figure 1 SSIM score for a CIF resolution video encoded at 380 kbps in Constant Bit Rate (CBR) mode.

Envivio IP-based Statistical Rate Control

3

Page 5: IPStatRate_Final.pdf

•MultipleSources Inordertostudytheefficiencyofthecompressionofvarioussourcessimultaneously,theminimum

scoreforallsourcescombinediscomputed(seeFigure2andFigure3).

Thiswayofassessingthequalityonmultiplesourcescorrespondsquitewelltohuman

observation.Indeed,observersareparticularlysensitivetotheworstartifactsandlesssensitiveto

smoothvideodegradations.

TheaverageMinimumSSIMscoregivesthenafairlygoodindicationofthequalityofa

groupofcompressedsources.Thisscorewillbeusedtoassesstheperformancesofvarious

compressionmethods.

Figure 2 SSIM scores for 10 CIF resolution video streams encoded at 380 kbps in CBR. (Lowpeaksrevealframedroppingondifficultsequences.)

Powering IP Video Convergence From Mobile to HD

4

Page 6: IPStatRate_Final.pdf

Figure 3 Minimum SSIM score for 10 CIF resolution video streams encoded at 380 kbps in CBR (average = 64).

AhistogramoftheSSIMscores(numberofsamplesforagivenSSIMscore)isalsoaninteresting

analysisasitsresultsreflectthevideoqualityhomogeneity.

Figure 4 Histogram of the SSIM values for 10 CIF resolution video streams in CBR at 240 kbps. (Frequenciesgiveanindicationofthequalityhomogeneity.Cumulatedfrequenciesindicatetheproportionofvideosunderagiven

qualitythreshold.)

Envivio IP-based Statistical Rate Control

5

Page 7: IPStatRate_Final.pdf

3. EnvivioStatisticalRateControl

3.1 Principle InthecaseofDVB-H,multipleIPservicesaretransmittedinburstsonafixedbandwidth

channel.Inordertorespecttheoverallbandwidthconstraint,therearegenerallytwowaysof

transmittingtheIPtraffic:

-Fixedbandwidthallocation:eachserviceutilizesaconstantbandwidthandthestreams

mustbeconstantbitrate(CBR).Theburstsizeandperiodarefixed.

-Dynamic,variableallocation:eachserviceutilizesavariablebandwidth.Theburstsize

andperiodarevariable.

Thelattercaseisbetterforvideotransmission.Giventheheterogeneousnatureofvideosignals,

itismoreappropriatetousemorebandwidthonpartsofthevideothatarehardtoencodeand

lessontheeasierparts.Statistically,thereisahighchancethatdifferentsourcesdonotneedhigh

bandwidthatthesametime,sothereisaninterestindynamicallyassigningthebitrateforthe

transmissionofeachsource.ThismethodiscalledStatisticalVariableBitRatemode,orStatistical

VBR,andtheallocationmechanismStatisticalRateControl.

Therefore,testingwillcomparetheefficiencyofthefirstapproachbasedonCBR,andtheStatistical

VBRapproach.

3.2 Test Scenario ThegoalofthesetestsistomeasuretheefficiencyofStatisticalVBRagainsttheapproachmost

commonlyusedtodayinDVB-H,CBR.

•BitRateModel Inthediscussionbelow,instantaneousbitratesareindicatedrwhileaveragebitratesare

indicatedR.

ThesimulationshavebeendonewithNvideosourcesonachannelwithatotalavailable

bandwidthRtotal.

IntheCBRcase,theNvideosourcesareseparatelyencodedataconstantbitrate,eachencodedat

Rservice=Rservice=Rtotal/N.

InthestatisticalVBR,theNvideosourcesareencodedinVBRmode,andatanytimetheirbitrates

satisfy:

∑Rservice =Rtotal

TobeclosetoDVB-Hconditions,thefollowingcodecparameterswereused:

•Format:CIF(352x288)at25Hz

•H.264Baselinewith1sI-frameinterval

•Measuremadeon375frames(15s)

Powering IP Video Convergence From Mobile to HD

6

Page 8: IPStatRate_Final.pdf

•Noaudio

•Sources Thebenefitsofstatisticalratecontrolareparticularlyvisiblewithheterogeneoussources.Therefore,

asetofvideosourceswereselectedwithvariouscodingcomplexities,representingwhatisusually

availableonstandardTVchannels:sports,concerts,movies,newsandanimatedmovies.

3.3 Results ForthesimulationsweusedN = 3, 6or10 andRservicevaryingfrom200to400kbps.Thecurvesin

Figure5throughFigure7showtheaverageMinimumSSIMofallsequencesaccordingtoRservicein

CBRandStatisticalVBRmodes.

Figure 5 Average Minimum SSIM of CBR vs Statistical VBR with 3 video sources.

Envivio IP-based Statistical Rate Control

7

Page 9: IPStatRate_Final.pdf

Figure 6 Average Minimum SSIM of CBR vs Statistical VBR with 6 video sources

Figure 7 Average Minimum of CBR vs Statistical VBR with 10 video sources

Powering IP Video Convergence From Mobile to HD

8

Page 10: IPStatRate_Final.pdf

Figure 8 Cumulated SSIM frequencies of 10 video sources encoded in CBR at 380 kbps, Statistical VBR at 240 kbps, CBR at 240 kbps and Statistical VBR at 380 kbps

•Observations Asexpected,theresultsdemonstratethatstatisticalVBRoffersfarsuperiorvideoqualitythanCBR,

andthepositiveeffectofthismodebecomesmoreandmoreimportantasthenumberofvideo

sourcesincreases.Thebitrateallocationalgorithmhasmoreoptionstoefficientlydistributethe

bitswheretheyareneededmost.

Sincethestatisticalratecontrolalgorithmfocusesonimprovingthe“worst”video,theresults

becomesmorehomogenous,andthenumberofsampleswithpoorqualityisreduced.Asshown

onthehistogramofFigure8,thenumberofsampleswithpoorerqualityissignificantlysmaller

inthestatisticalVBRcasecomparedtoCBR.Visually,thishasextremelygoodconsequences–the

heavycompressionartifactsdisappearandallserviceshavethesamehomogenousquality.

•StatisticalVBRGains Allthecurvesshowsomesignificantvideoqualityimprovementsatanygivenbitrate.More

interestingly,thepotentialbitrategainforagivenqualitylevelcanbeestimated.Lookingat

Figures5to7(measuredhereatMinimumSSIM=65)thisgainisalsoconfirmedbythe

VideoSourceS cBr StatiSticalVBrBitrate BandwidthSaVingS

3 310 220 29%

6 340 220 35%

10 380 240 37%

Table 1 Comparative bit rates for an equivalent quality and potential bandwidth savings

Envivio IP-based Statistical Rate Control

9

Page 11: IPStatRate_Final.pdf

histogramonFigure8.ForbothCBRat380kbpsandStatisticalVBRat240kbps,95%ofthevideo

isaboveSSIM=65.Furthermore,thevideoqualityismorehomogeneousasseenonFigure9.On

thisfigure,theblueportionindicatesareaswheresamplesinStatisticalVBR240kbpsdominate,

whiletheredindicatesareaswheresamplesinCBR380kbpsdominate.Theblueportionismore

compactthantheredindicatingthevideoqualityismorehomogeneousinStatisticalVBRthan

CBRmode,evenatlowbitrates.ThequalityscoresaremorewidespreadinthecaseofCBRat380

kbpscomparedtotheonesinStatisticalVBRat240kbps.Theseresultsarealsoconfirmedbyvisual

observations.ThevariationinqualityismuchlessnoticeableinStatisticalVBRmodesandthus

leadstoabetteroverallsubjectivequality.

Figure 9 Histogram of SSIM repartition for 10 video sources encoded in CBR at 380 kbps and in Statistical VBR at 240 kbps

Inconclusion,StatisticalVBRmodebringssignificantimprovementsoverCBR.Evenwith

alimitedsetofsources,thebenefitsofthismethodarequitedramaticintermsofartifact

reduction.Theefficiencyincreaseswiththenumberofheterogeneoussources,anditcanbe

estimatedthat30%to40%performanceimprovementcanbeachievedinaverageoverCBR,

whichwasalreadygreatlyoptimized.

Powering IP Video Convergence From Mobile to HD

10

Page 12: IPStatRate_Final.pdf

4. OpenArchitecture

4.1 Architecture and Principle TheEnviviostatisticalratecontrolimplementationisnotdependantonanexternalmodule.

Thisfunctionalityisavailableasanoptiondirectlyfromtheencoderanddoesnotrequireany

extracomponents.

ForanEnvivioimplementation,anencoderworkinginStatisticalVBRmodeisassignedtoagroup.

Thisgrouprepresentsalltheencoderprofileswhichareactingunderthesamestatisticalrate

controlandwilllaterbesharingthesamebandwidth.

•Master/SlaveRoles Withinagroup,aparticularencoderactsas“master,”anotheroneasbackupofthemaster,andthe

otherencodersare“slaves.”Themastergetstheoverallbandwidthassignmentfromtheoperator

orthemanagementsystem,listenstotheslaves’requestsandallocatesthebitratesdynamically.

AllcommunicationisdoneviaIPprotocols.

Figure 10 DVB-H head-end with master and slave encoders

•Multi-OutputandStatisticalRateControl

MostEnvivio4Caster™encoders,andnotablytheEnvivio4CasterM2,havetheabilitytooutput

multipleprofilessimultaneouslyfromthesameinput.Eachoftheseprofilescanobeydifferent

statisticalratecontrolparameters.Forexample,4CasterM2canoutput3CIFDVB-Hprofiles,each

belongingtoastatisticalratecontrolgroup.

Envivio IP-based Statistical Rate Control

11

Page 13: IPStatRate_Final.pdf

Thisflexibilitycanhavegreatcostadvantageswhenitisnecessarytocomposevariousstatistically

ratecontrolledDVB-Hgroupings,or“bouquets”,sharingcommonchannels.Insteadofusing

dedicatedencodersforeachbouquet,thesameencodercanbereusedtooutputthesame

channeltovariousbouquetsunderdifferentratecontrols.Thussignificantlydecreasingthehead-

endcostcomparedtoencoderstightlylinkedtoaDVB-IPE(IPEncapsulator)whereequipment

needstobereplicatedforeachbouquet.

Figure 10 National/Regional combined head-ends with multiple stat mux groups, each with various master encoders

•FutureExtensions The“master”functionalityhasbeenplacedononeencoder,andanswerstheneedforsmallto

mediumsizedeploymentswhichcantypicallybefoundwithDVB-H.Inthefuture,thisfunctionality

willalsobeavailablefromthecentralmanagementsystem,4Manager,inordertocontrolmultiple

groupssimultaneouslyandaddressscalabilityandredundancyissuesinmuchlargerdeployments.

4.2 Open vs. Closed Systems

•ClosedSystems Quiteoften,statisticalmultiplexerimplementationsrelyonstrongtiesbetweentheencodersand

amultiplexer.Similarly,inthecaseofDVB-H,mostvendorsrelyonproprietarycommunication

mechanismsbetweentheirencodersandtheirDVB-IPEtoinsurethestatisticalratecontrol

function.

8

Powering IP Video Convergence From Mobile to HD

12

Page 14: IPStatRate_Final.pdf

Intheseclosedsystems,theDVB-IPEisthemain“brain”fortheservicebitrateallocations.It

dynamicallysplitsanddistributesthebandwidthbetweenallencodersaccordingtotheneedsof

eachsource.Itthenperformstheproperburstallocationbeforetransmission.

Whilethiskindofsystemisefficientinperformingstatisticalmultiplexing,itisextremelydifficult

todecoupletheencodersfromthemultiplexer,andthereisnostandardwayofnegotiatingrate

allocationbetweenthesetwokindsofelements.Practically,thismeansthatitisnotpossibleto

mixequipment(encodersormultiplexers)fromdifferentvendorswithoutlosingthestatistical

multiplexingfunctionality.Untilnow,thishashelpedbroadcastequipmentvendorslockcustomers

intoasinglebrandsolution.

•OpenSystems TheEnvivioapproachisradicallydifferentfromthisclosedloopimplementation.TheEnvivio

advancedstatisticalratecontroldoesnotrelyonanyproprietarymechanismbetween

encodersandtheIPE.Instead,theratecontrolallocationisdonebetweenencoders,while

theencapsulatoronlytakescareofburstallocationandisnotobligedtosendanycontrolor

indicationtotheencoders.

Thestatisticalratecontrolfeatureisprovidedwithinamoduleoftheencoder.Thismodule

communicateswithalltheotherencodersbelongingtothesamegroupandsimulatesthe

emissionbufferoftheDVB-IPE.Itcanbesettooperateatfixedbandwidth,aswasdonehereto

performthecomparisons,orsetatdynamicbandwidth.Inthislattercase,anexternalmechanism

canincreaseordecreasetheoverallbandwidthattributedtothestatmuxgroup.Thiscanbe

interestingifotherservicesneedtotakepriorityoverthevideoservicesorifextrabandwidthis

availablefortransmission.

theopenarchitectureprovidessignificantadvantages: -encodersandIPEscancomefromvariousvendors

-encodersandIPEsdonothavetobecollocated

-nolock-infactormeansbest-of-breedapproach

-higherreliability(elementsaremoreautonomous,sotheentiresystemislessproneto

bottleneckfailures)

•Summary

AdvAntAges drAwbAcks

ClosedSystems Efficient Nopossibleequipmentmix

Potentiallylowerdelay Morebottlenecksinthesystem

Novendorinteroperabilityrequired Lockedintoonebrandsolution

Equipmentneedstobecollocated

OpenSystems Efficient

Bestofbreedapproach

Morearchitecturesolutionsavailable

Higherreliability

Vendorinteroperability

Highernumberofvendors

Envivio IP-based Statistical Rate Control

13

Page 15: IPStatRate_Final.pdf

5. Conclusion Enviviostatisticalratecontroltechnologyanditsinstantiationinthe4CasterMobileM2encoder

bringtwosignificantadvantagestothemarket:

PictureQualityimprovementorBandwidthSavings -AveryefficientStatisticalVariableBitRatemode,whichbringssignificantadvantages

overConstantBitRate.Thisleadstoeitherdramaticpicturequalityimprovementforthe

sameoverallbitrate,orareductioninthebandwidthrequiredperservice,allowingmore

servicestobecarried.

openStandards-Basedimplementation -AnopenandIP-orientedarchitecture,allowingadvancedservicesandthecreationof

bestofbreedsolutionswithvariouscomponentvendors.

Thoughthebenefitsofstatisticalratecontrolformobile,andparticularlyDVB-Hwerecoveredin

thispaper,thesametechnologyandprinciplescanbeappliedtootherkindsofenvironments,

whichtraditionallyuseCBR,suchas:

contributionofMultipleFeeds -Statisticalratecontroldeliverssomequalityorcompressionimprovementsevenwitha

smallnumberofencoders,fromwhichsatelliteandIPcontributioncoulddrawbenefits.

iPtVorenterprisetVbackbonecostreduction -InordertosavebandwidthontheIPdistributionnetworks(backboneinthecaseofIPTV,

orprivatenetworksinthecaseofEnterprisevideodistribution)thistechnologycouldbe

applied

Thankstoitsefficiencyandopenness,Enviviostatisticalratecontrolenablessomesignificant

qualityandserviceimprovementstotheoperator,formobile,IPTVandcontributionnetworks.

All statements, specifications, recommendations, and technical and other information contained in this document are current as of the date

of publication and presented without warranty of any kind, whether expressed or implied. Envivio reserves the right to revise the information

in the document at its discretion without notice. Envivio shall not be liable for errors or omissions in this document. Envivio shall not be held

liable for damages of any kind resulting from the furnishing or use of this document. Envivio product specifications are subject to change at any

time without advanced notice. All products or services mentioned are the trademark, service marks, registered trademarks or registered service

marks of their respective owners. ©2007 Envivio, Inc. All rights reserved.

Envivio IP-based Statistical Rate Control

14

Powering IP Video Convergence From Mobile to HD

White Paper

Envivio IP-basedStatistical Rate Control

December 2006Authors: Boris Felts, Jean Kypreos, Thomas Guionnet

www.envivio.com

Page 16: IPStatRate_Final.pdf

8

Powering IP Video Convergence From Mobile to HD

3