BASED VIDEO TELEPHONY
P. Cherriman, E.L. Kuan, L. Hanzo
Dept. of Electronics and Computer Science, Univ. of Southampton, SO17 1BJ, UK.
Tel: +44-1703-593125,Fax: +44-1703-594508
Email: [email protected], [email protected]
http://www-mobile.ecs.sot on.a c.u k
ABSTRACT
A multi-mode joint-detection CDMA-based video transceiver is proposed for wireless video telephony,
whichsubstantiallyoutperformsthematched-lteringbasedbench-markervideotransceiver. Forchannel
SNRsin excessofabout5dB near-unimpaired videoquality ismaintained bythe proposed scheme.
1. VIDEOTRANSCEIVER
Inthisstudywetransmitted176x144pixelQuarterCommonIntermediateFormat(QCIF)and128x96pixelSub-QCIF
(SQCIF)videosequencesat10frames/susingarecongurableTimeDivisionMultipleAccess/CodeDivisionMultiple
Access(TDMA/CDMA)transceiver,whichcanbeconguredasa1,2or4bit/symbolschemeshowninFigure1.The
H.263videocodec[1 ]exhibitsanimpressivecompressionratio,althoughthisisachievedatthecostofahighvulnerability
totransmissionerrors,sincearun-lengthcodedstreamisrenderedundecodablebyasinglebiterror. Inordertomitigate
thisproblem,whenthechannelcodecprotectingthevideostreamisoverwhelmedbythetransmissionerrors,werefrain
fromdecodingthecorruptedvideopacketinorder topreventerrorpropagationthroughthereconstructedvideoframe
buer[4]. Wefoundthat itwasmorebenecial invideoqualityterms,if thesecorruptedvideopacketsweredropped
andthe reconstructedframe buerwasnot updated,untilthenextvideopacketreplenishing thespecicvideoframe
area was received. The associated videoperformance degradation was found perceptually unobjectionablefor packet
dropping-or transmission frame error rates (FER) below about 5%. These packetdropping eventsweresignalled to
theremotedecoderbysuperimposingastronglyprotectedone-bitpacketacknowledgementagonthereverse-direction
packet,asoutlinedin[4 ]. Bose-Chaudhuri-Hocquenghem(BCH)[5] andturboerrorcorrectioncodes[6 ]wereusedand
again,the CDMAtransceiverwas capable oftransmitting 1, 2and4 bits persymbol, where each symbolwas spread
usingalowspreadingfactor(SF)of16, asseen inTable1. Theassociatedparameterswillbeaddressedinmoredepth
during our furtherdiscourse. Employing alow spreadingfactor of 16 allowed us to improve the system'smulti-user
performancewith theaid of joint-detection techniques[7]. We notefurthermore thatthe implementationof the joint
detectionreceiversisindependentofthenumberofbitspersymbolassociatedwiththemodulationmodeused,sincethe
receiversimplyinvertstheassociatedsystemmatrixandinvokesadecisionconcerningthereceivedsymbol,irrespective
ofhowmanybits persymbolwereused. Therefore, joint detection receivers are amenableto amalgamation
withthe above 1, 2and 4 bit/symbol modem,since theydo nothaveto be recongured eachtime the
modulation mode isswitched.
InthisperformancestudyweusedthePan-EuropeanFRAMESproposal[2]asthebasisforourCDMAsystem.The
associatedtransmissionframestructureisshowninFigure2,whilearangeofgenericsystemparametersaresummarised
inTable 1. In ourperformance studieswe usedthe COST207 [3 ]seven-path bad urban (BU)channel model, whose
impulseresponseisportrayedinFigure3.
Ourinitialexperimentscomparedtheperformanceofawhiteningmatchedlter(WMF)forsingleuserdetectionand
theMinimummeansquareerrorblockdecisionfeedbackequalizer(MMSE-BDFE)forjointmulti-userdetection. These
simulationswereperformed using 4-level Quadrature Amplitude Modulation (4QAM), invoking bothbinary BCH [5]
and turbo coded[6] videopackets. Theassociated bitrates are summarised inTable 2. The transmission bitrateof
the4QAMmodemmode was 29.5Kbps,which was reducedduetothe approximately half-rateBCH orturbocoding,
plusthe associated video packetacknowledgement feedback ag error control [1 ]and video packetisation overheadto
produceeectivevideobitratesof13.7Kbpsand11.1Kbps,respectively. Amoredetaileddiscussiononthevideopacket
acknowledgement feedback errorcontroland videopacketisation overhead willbe providedinSection2withreference
totheconvolutionallycodedmulti-modeinvestigations.
Figure4 portrays thebit errorratio (BER) performance ofthe BCHcodedvideotransceiverusingbothmatched
lteringandjointdetectionfor2{8users. Thebiterrorratioisshowntoincrease,asthenumberofusersincreases,even
ThenancialsupportoftheMobileVCE,UKandthatoftheEuropeanCommissionisgratefullyacknowledged.
Feedback Information
n
n
Video
Encoder
Assembly
Packet
Mapper
Encoder
FEC
n-class
Mapper
&
TDMA MPX
bits/symbol
QAM
1,2,4,6
Channel
QAM
Demodulator
TDMA DEM
Mapper
&
Decoder
FEC
n-class
Mapper
Disassembly
Packet
Video
Decoder
Figure1: Recongurabletransceiverschematic
Training
sequence
Data
Data
577 microseconds
34 symbols =
34 symbols =
544 chips
107 chips
544 chips
Guard
55
CP
Spread speech/data burst 2
Figure2: TransmissionburststructureoftheFMA1spreadspeech/datamode2oftheFRAMESproposal[2]
Parameter
Multipleaccess TDMA/CDMA
Channeltype C0ST207BadUrban
Numberofpathsinchannel 7
NormalisedDopplerfrequency 3:710 5
Hz
CDMAspreadingfactor 16
Spreadingsequence Random
Frameduration 4.615ms
Burstduration 577s
JointdetectionCDMAreceiver Whitening matched lter (WMF)
or Minimum mean square error
block decision feedback equalizer
(MMSE-BDFE)
No. ofSlots/Frame 8
TDMAframelength 4.615ms
TDMAslotlength 577s
TDMAslots/Videopacket 3
ChipPeriods/TDMAslot 1250
DataSymbols/TDMAslot 68
UserDataSymbolRate(kBd) 14.7
SystemDataSymbolRate(kBd) 117.9
Table1: GenericsystemparametersusingtheFramesspreadspeech/datamode2proposal[2 ]
Features BCHcoding Turbocoding
Modulation 4QAM
Transmission bitrate
(kbit/s)
29.5
Video-rate(kbit/s) 13.7 11.1
Videoframerate(Hz) 10
0
1
2
3
4
5
6
7
8
Path delay ( s)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Normalized
magnitude
Figure3: NormalizedchannelimpulseresponsefortheCOST207[3]seven-pathBadUrbanchannel.
$RCSfile: ber-vs-csnr-bch-mf-vs-jd.gle,v $
8
10
12
14
16
18
20
Channel SNR (dB)
5
10
-4
2
5
10
-3
2
5
10
-2
2
5
10
-1
2
5
10
0
Bit
Error
Ratio
Joint Detection
Matched Filtering
BCH coded 4QAM CDMA
2 Users
4 Users
6 Users
8 Users
Figure4: BERversuschannelSNR4QAMperformanceusingBCHcoded13.7Kbpsvideo,comparingtheperformance
ofmatchedlteringandjointdetectionfor 2{8users.
uponemployingtheMMSE-BDFEmulti-userdetector(MUD).However,whilethematchedlteringreceiverexhibitsan
unacceptablyhighBERforsupportingperceptuallyunimpairedvideocommunications,theMUDexhibitsafarsuperior
BERperformance.
Whenthe BCH codec was replaced by the turbo-codec, the biterror ratio performance of both matchedltering
andtheMUDreceiverimproved,asshowninFigure5. However,asexpected, matchedlteringwasstilloutperformed
bythejoint detectionschemeforthesamenumberofusers. Furthermore,thematchedlteringperformance degraded
rapidlyformorethantwousers.
Figure6showsthevideopacketloss ratio(PLR)fortheturbocodedvideostreamusingmatchedlteringandjoint
detectionfor2{8users. Thegureclearlyshowsthatthematchedlter wasonlycapableofmeetingthetargetpacket
lossratioof5%foruptofourusers,whenthechannelSNRwasinexcessof11dB.However,thejointdetectionalgorithm
guaranteedthe requiredvideopacketloss ratio performancefor 2{8users intheentirerange ofchannelSNRsshown.
Furthermore,the2-usermatched-lteredPLRperformancewasclosetothe8-userMUDPLR.
2. MULTI-MODE VIDEOSYSTEM PERFORMANCE
Havingshownthatjointdetectioncansubstantiallyimproveoursystem'sperformance,weinvestigatedtheperformance
ofamulti-modeconvolutionallycodedvideosystememployingjointdetection,whilesupportingtwousers. Theassociated
convolutionalcodecparametersaresummarisedinTable3.
Belowwenowdetailthevideopacketisationmethodemployed. Thereaderisremindedthatthenumberofsymbols
perTDMAframewas68accordingtoTable1.Inthe4QAMmodethiswouldgive136bitsperTDMAframe. However,
ifwetransmittedonevideopacketperTDMAframe,thenthepacketisationoverheadwouldabsorbalargepercentage
$RCSfile: ber-vs-csnr-turbo-mf-vs-jd.gle,v $
8
10
12
14
16
18
20
Channel SNR (dB)
5
10
-4
2
5
10
-3
2
5
10
-2
2
5
10
-1
2
5
10
0
Bit
Error
Ratio
Joint Detection
Matched Filtering
Turbo coded 4QAM CDMA
2 Users
4 Users
6 Users
8 Users
Figure5: BERversuschannelSNR4QAMperformanceusingturbo-coded11.1Kbpsvideo,comparingtheperformance
ofmatchedlteringandjointdetectionfor 2{8users.
$RCSfile: fer-vs-csnr-turbo-mf-vs-jd.gle,v $
8
10
12
14
16
18
20
Channel SNR (dB)
5
10
-3
2
5
10
-2
2
5
10
-1
2
5
10
0
V
ideo
packet
loss
Ratio
Joint Detection
Matched Filtering
Turbo coded 4QAM CDMA
2 Users
4 Users
6 Users
8 Users
5% PLR target
8 user JD
2 user MF
2 user JD
Figure 6: Video packet loss ratio versus channel SNR for the turbo-coded 11.1 Kbps video stream, comparing the
performanceofmatchedlteringandjointdetectionfor2{8users.
Therefore each protected video packet consists of 683 = 204 modulationsymbols, yielding a transmission bitrate
ofbetween 14.7 and 38.9 Kbpsfor BPSKand 16QAM, respectively. However, inorderto protect the videodata we
employed half-rate convolution coding, using actally represented generator polynomials of 5 and 7. The useful video
bitratewas further reduceddueto the 16bit Cyclic Redundancy Checking (CRC) used for error detections and the
nine-bitrepetition-codedfeedbackerroragforthereverselink. Thisleavesvideopacketsizesof77,179and383bitsfor
eachofthethreemodulationmodes. Theusefulvideocapacitywasnallyfurtherreducedby thevideopacketheader
ofbetween8and 10bits,resulting inusefulor eective videobitratesranging from5 to26.9 KbpsintheBPSKand
16QAMmodes,respectively.
Theproposedmulti-modesystemcanswitchamongstthe1,2and4bit/symbolmodulationschemesundernetwork
control,basedupontheprevailingchannelconditions. AsseeninTable 3,whenthechannelis benign,theunprotected
videobitratewillbeapproximately26.9Kbpsinthe16QAMmode.However,asthechannelqualitydegrades,themodem
willswitchtotheBPSKmodeofoperation, wherethevideobitratedropsto 5Kbps,andfor maintainingareasonable
videoquality,thevideoresolutionhastobereducedtoSQCIF(128x96pels).
Figure7 portraysthe packetloss ratio for the multi-modesystem, ineachof itsmodulationmodes fora rangeof
channel SNRs. It canbe seen that above 14dB the 16QAMmode oers anacceptable packetloss ratio of less than
5%, while providing anunprotected videorateof about 26.9Kbps. If the channelSNR dropsbelow 14dB,the
multi-modesystemisswitchedto4QAMandeventuallytoBPSK,whenthechannelSNRisbelow9dB,inordertomaintain
therequiredquality ofservice, whichisdictated by thepacketloss ratio. Thegurealsoshows theacknowledgement
feedbackerrorratio(FBER)forarangeofchannelSNRs,whichhastobesubstantiallylower,thanthevideoPLRitself.
Thisrequirementis satisedinthegure,since thefeedbackerrorsonlyoccur atextremely lowchannelSNRs,where
Mode BPSK 4QAM 16QAM
Bits/Symbol 1 2 4
FEC ConvolutionalCoding
Transmittedbits/packet 204 408 816
Totalbitrate(kbit/s) 14.7 29.5 58.9
FEC-codedbits/packet 102 204 408
Assigned to FEC-coding
(kbit/s)
7.4 14.7 29.5
Errordetectionperpacket 16bitCRC
Feedbackbits/packet 9
Videopacketsize 77 179 383
Packetheaderbits 8 9 10
Videobits/packet 69 170 373
Unprotectedvideo-rate(kbit/s) 5.0 12.3 26.9
Videoframerate(Hz) 10
Table3: Operational-modespecictransceiverparametersfortheproposedmulti-modesystem
$RCSfile: fer+fber-vs-csnr-fixed-conv.gle,v $ $Date: 1999/04/08 15:04:27 $
0
2
4
6
8
10
12
14
16
18
20
Channel SNR (dB)
5
10
-3
2
5
10
-2
2
5
10
-1
2
5
PLR
/
FBER
16QAM 26.9Kbps
4QAM 12.3Kbps
BPSK 5Kbps
FBER
PLR
10% PLR
5% PLR
Figure7: Videopacketlossratio(PLR)andfeedbackerrorratio(FBER)versuschannelSNRforthethreemodulation
schemesofthe2-usermulti-modesystemusingjointdetection.
$RCSfile: psnr-vs-csnr-missa-sqcif-fixed.gle,v $
0
5
10
15
20
25
Channel SNR (dB)
30
32
34
36
38
40
42
A
verage
PSNR
(dB)
BPSK 5Kbps SQCIF Miss-America
4QAM 12.3Kbps SQCIF Miss-America
16QAM 26.9Kbps SQCIF Miss-America
Figure8: Decodedvideoquality(PSNR)versuschannelSNR forthemodulationmodesofBPSK,4QAMand16QAM
supporting2-userswiththeaidofjoint detection. Theseresultswererecorded forthe Miss-Americavideosequenceat
$RCSfile: psnr-vs-fer-missa-sqcif-fixed.gle,v $
0.0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Video packet loss ratio
30
32
34
36
38
40
42
A
verage
PSNR
(dB)
BPSK 5Kbps SQCIF Miss-America
4QAM 12.3Kbps SQCIF Miss-America
16QAM 26.9Kbps SQCIF Miss-America
Figure 9: Decoded video quality (PSNR)versus video packetloss ratio for the modulation modes of BPSK, 4QAM
and16QAM,supporting2-userswiththeaid ofjoint detection. Theresultswererecordedfor theMiss-America video
sequenceatSQCIFresolution(128x96pels).
Thevideoqualityiscommonlymeasuredintermsofthepeak-signal-to-noise-ratio(PSNR).Figure8showsthevideo
quality in terms of the PSNR versus the channel SNRsfor eachof the modulation modes. As expected, the higher
throughputbitrateofthe16QAMmodeprovidesabettervideoquality. However,asthechannelqualitydegrades,the
videoqualityofthe16QAMmodeisreducedandhenceitbecomesbenecialtoswitchfromthe16QAMmodeto4QAM
atan SNRof about14dB, as it was suggestedby thepacketloss ratio performance of Figure7. Althoughthe video
quality expressed interms of PSNRis superior for the 16QAM modein comparison to the 4QAM mode at channel
SNRsinexcessof12dB,however,duetotheexcessive PLRtheperceivedvideoquality appearsinferiorincomparison
tothatofthe4QAMmode,eventhoughthe16QAMPSNRishigherforchannelSNRsintherangeof12{14dB.More
specically,wefoundthatitwasbenecialtoswitchtoamorerobustmodulationscheme,whenthePSNRwasreduced
byabout1dBwithrespecttoitsunimpairedPSNRvalue. Thisensuredthatthepacketlossesdidnotbecomeobvious,
resultinginahigherperceivedvideoquality,andsmootherdegradationasthechannelqualitydeteriorated.
Theeect ofpacketlosses onthe videoquality quantiedinterms ofPSNRis portrayedinFigure 9. The gure
shows, how the video quality degrades, as the PLR increases. It has beenfound that inorder to ensure aseamless
degradationofvideoqualityasthechannelSNRreduced,itwasthebestpolicytoswitchtoamorerobustmodulation
scheme, whenthePLR exceeded 5%. Thegureclearly shows that a5% packetloss ratio results inaloss of PSNR,
whenswitching to a more robustmodulation scheme. However, if the system did not switch until the PSNRof the
morerobustmodulationmodewassimilar, theperceivedvideoquality associatedwith theoriginally higherrate, but
channel-impairedstreambecameinferior.
3. CONCLUSIONS
Inconclusion,theproposedjoint-detectionassistedmultimodeCDMA-basedvideotransceiversubstantiallyoutperformed
thematched-lteringbasedtransceiver. Thetransceiverguaranteedanear-unimpairedvideoquality forchannelSNRs
inexcess of about 5 dB over the COST207 dispersive Rayleigh-faded channel. The benets of the multimode video
transceiver clearly manifest themselves in terms of supporting un-impaired video quality under time-variant channel
conditions,whereasingle-modetransceiver'squalitywouldbecomeseverelydegradedbychanneleects.
4. REFERENCES
[1] L.Hanzo,P.Cherriman,J.Streit:VideoCompressionandCommunicationsoverWirelessChannels:FromSecondtoThirdGeneration
Systems,WLANsandBeyond,JohnWileyandIEEEPress,inpreparation 1
[2] A.Klein,R.Pirhonen,J.Skoeld,andR.Suoranta,\FRAMESmultipleaccessmode1-widebandTDMAwithandwithoutspreading,"
inProceedingsof IEEEInternational Symposium on Personal, Indoorand MobileRadioCommunications, PIMRC'97, vol.1,
(MarinaCongressCentre,Helsinki,Finland),pp.37{41,IEEE,1{4Sept1997.
[3] \COST207: Digitallandmobileradiocommunications,nalreport."OÆceforOÆcialPublicationsoftheEuropeanCommunities,
1989.Luxembourg.
[4] P. Cherriman,L. Hanzo: ProgramableH.263-based wirelessvideo transceivers forinterference-limitedenvironments, IEEETr. on
CSVT,June1997,Vol.8,No.3,pp275-286
[5] R.Steele,L.Hanzo(Ed.),MobileRadioCommunications,2ndedition,IEEEPress-JohnWileyandSonsLtd.,1999
[6] C.Berrou, A. Glavieux andP. Thitimajshima, \NearShannon LimitError-Correcting Coding andDecoding: Turbo Codes," in
Proceedings,IEEEInternationalConfrenceonCommunications,pp.1064{1070,1993.
[7] EEL.Kuan, L.Hanzo: JointDetectionCDMA TechniquesforThird-generationTransceivers, Proceedingsof ACTS'98, Rhodos,
Greece,June1998,pp727-732