TDSL 1 !
!
TerabitDSL(TDSL)(Useofacopperpair’ssub-millimeter
Waveguidemodes)
G.fastConferenceParis,May10,2017
J.CioffiProfessor
Emeritus,StanfordEE
(CEO/COBASSIA)
K.KerpezSr.Director,StandardsASSIAInc.
ChanSooHwangSr.DirectorR&D
ASSIAInc.
IoannisKanellakopoulosConsulIngCTO
ASSIAInc.
plasmonpolaritonsCouirtesy:BilkentU
TDSL 2 !
DriversforHigher-SpeedDSLs
• MULTITUDEof5Gsmallercells– high-speedlow-latencywiredsupport– New5G-fibercost=400Beuros(foreurope,DTCTO,2016)
• FibertheoreWcalcapacity~500Tbps– Todaysupports1Gbpsto100Gbps(access-network)
• BUT– INSTALLcosts$3000-$4000/home(average)
• $4trillionglobally(insteadpaynaWonaldebts?)– Successfulbusinesscaseneeds<1/10ofthiscost
• Thecoppertwistedpairsarethere(1.3B)– Runfiberpartoftheway($3000/10homesisabe`erbusinesscase)– ConWnuesxinxDSL,socanx=T?
Ovum 2015
Yes,wethinkso
ADSL (G.992.1, G.992.3)
Bandwidth (bps)
Reach
Vectored VDSL G.993.5
VDSL2 (G.993.2)
ADSL2+ (G.992.5) 2001
2003
2007
2014
6Km 3Km 1500m 500m 200m
1G
500M
100M
50M
20M
8M
1M
FTTEx FTTC/B FTTdp
Telco CO
ONU/LT (ends fiber Starts copper)
xDSL fiber
xDSL
G.MGFast 10G
50m
G.Fast G.9902
2018+
CurrentxDSLprogression
TDSL?
TDSL 4 !
Vectoring=1stMassive(MU-)MIMO(2001invenWon/introtostandards)
• MassiveantennasàvecDSLAM• Purple“ChannelHardening”àcopperpairs
– Thosearethewireswithcrosstalkcanceled– ThenMu-MIMO/vectoringagainWi-Fitodevicefromgateway!
• Itisindeedthesamesignalprocessing– DiagonaldominanceofDSL~channelhardening
• Useoflinearprecoderinsteadofnon-linear(uptoapoint)• MaybewecanborrowbackabitfromMassiveMIMO’smmW?
TDSL 5 !
MIMO?àIt’sreallyMISO(orSIMO)
• IninformaWontheory–single-sidedcoordinaWon– Downstream=vectorbroadcastchannel(MISO)– Upstream=vectormulWple-accesschannel(SIMO)– It’swhytheterm“vectoringwasused”(notMISO/SIMO/MIMO)
• MIMOcoordinatesBOTHends– Solotsofantennas/wiresatreceivesideinsameplace– Someearly“H”DSL(Voyan)didthis– Butnotphysicallypossiblewhenthehomesareindifferentplaces
TDSL 6 !
TwoTDSLPaths:mulWpair&singlepair
6
TDSL
MulW
easier/convenWonal
BondmanyPairs
forcellsite,maybe
Single
harder-useTM/TEmodes
enormousbandwith
orrangeatlower
bandwidth
Terabit/son100pairs(tocellordistribuWonpt)
Terabit/son1pair(fromdistribuWonpttohome)
Transmission-linemode Waveguidemode
TDSL 7 !
7
PhantomsDSL2-pair(4x4)
30R31R
20R10R
32R
21R
3V
2V
1V
0
3I
2I
1I
32~I
321 III ++
21~I
20~I
10~I
31~I
30~I
Ingeneral,Npairà2N-1channels
N=100,andR=5Gbps(G.MGfast)à1Tbps
TDSL 8 !
Issueswith“Phantom”TDSL
• Onlyback-haulbecauseofreceivermatching– Ormoregenerallyreceivercoordinatedprocessing
• Mayneedtoomany“repeater”points– EmissionscouldbeproblemaWc
• Limiteduse-• BUT,itisaTerabit• HowaboutaTerabit/lineàWaveguideMode
5/8/17 8
TDSL 9 !
r = 0.5 mm
rcopper = 0.25 mm
plasIcinsulator
copperwire
Aircoresofwaveguide
SinglePair:CableCrossSecWon
• Today’sxDSLonthecopper(differenWalTEMmode)– And/ortheairgaps?(green)
TDSL 10 !
SurfaceWaveTransmission(1909Sommerfeldwave)
• SurfaceMode(orTM10)– WavesusesinglewireinTMmodeasguide
• E.g.Goubauantennaor“G-line”• SeealsoAT&T“AirGig”
– EffecWvelywirelesstransmission• Worksreasonablywell(noatmosphereinsidecable)• Dielectric(plasWc)canhelp(see[Wiltske]),p.971)
keepenergyclose
– Tubewithnon-uniformdilectricconstant• Conformalmappingof1/rdimension
• EnergysWllleaksoffwireifbent Mathfaculty.com
SWEnergy
TDSL 11 !
Surface-WaveMeasurements2006
11
• SinglewireTM01– Wiltse’ssurface-wavemeasurementsare2mm
wirecore,not0.5mm)– Measuresa`enuaWon/m
• WiltseExtrapolaWon– .8dB/[email protected]– Fa`erwires
• Grischkowskyhas.5db/m– For.52mmdiameterCuwire– 2ndwirewouldprobablyimprovetransfer– Likeintwistedpair
• 100mshouldsee50-80dB
• Bendingislessofaproblem – EachwirehasaTMmode– BetweenwiresisaTEMplasmonpolaritonmode– 2ndTEM“plasmonic”(weaker?)tootherpairs–
somewhatlikephantoms/split-pairs– TIRmode– Surfacemode(maybesameasTM…)– 3--4modesperpair
Wiltse
Grischkowsky
TDSL 12 !
3/6differenceisareaofthefree-spaceinatriangular“waveguide”=.16r2
AlWtudeoftriangular“waveguide”=.53r
(600GHzisλ/2)
Ifplas:cused200GHz
r = 0.5 mm
rcopper = 0.25 mm
plasWcinsulator
copperwire
With2L(say100+)wires,therecanbep(2L)transmi`ers(m=0,1,…)orantennasintowaveguidegaps(triangles)ManyreflecWonstoanyspaWalpoint:(waveguideisreally“SwissCheese”withmanyinterconnecWngspaces).BigMIMOopportunity.MIMOmatrixwithGaussianentriesatanypointinspace,whichissameasRayleighfadinginwireless,exceptstaWcsampleEquivalentof“tunable”laserwithvector-coordinatedexcitaWon?(muchlessa`enuaWon–likefiber?)
Aircoresofwaveguide
MIMOProcessor
CrossSecWonGeometry
difference = .32r2
Circle Area=πr2
Outer Hexagon= 63 r
2
TDSL 13 !
Vectoring=MassiveMIMO• Letstrym=1withTMantenna(s)wirelesslyexciWngeachwireend
– PhotoconducWveantennasperhaps?– BothpolarizaWons(TMandTE)foreachwire
• Orpossiblyforpairsofwires
• ThereisalsoaTEMplasmonpolaritonmode– Atleastone,reallytwo– CouldthinkofthisasdualpolarizaWon,butnotquitereally– ThereisalsoatleastoneTIRmode(totalinternalreflecWon)withsheath
• NominallyintersecWonswouldintroducecrosstalkbetweentheTMsandTEMs
– UseMIMOorMISO(justlikeinmmWwireless5G,exceptmmW/10)– Willtendtowardlognormal
• “SwissCheese”Waveguide– ULTRArichsca`ering(exactlywhatmassiveMIMOneeds)
• Coupling(splicing)isopentoinnovaWons,butphotoconducWveandothertypesofantennas/lasers/detectorsdoexistinthesefrequencyrangestoday.
TDSL 14 !
Vectoring~ChannelHardening
• Sayfrom100GHzto300GHz– Use4096tones,soroughly50MHzwideeach– Twowiresinapair,andtwopolarizaWons
• Itsconceivablethateven2.5bits/toneaverage,so1Tbps
CPE1
CPE2
0
OldDSLAMPort1
OldDSLAMPort2
crosstalk
“wireless”
Customer1
Customer2
VectoredDSLAM
TDSL 15 !
Model
15
• Channel(Grischkowsky)
• Xtalk(thispaper)– Lognormal
• 20dBmtotaltransmitpower,flattransmitPSD
• 4096subcarriersfrom100GHzto300GHz,48.8MHzsubcarrierspacing– Bitloadingfrom1to12bits/Hz– 10%phy-layeroverheadremovedbeforepresenWngresults– 4.5dBcodinggain,1.5dBimplementaWonloss– Carriersfrom50GHzto150GHzwereusedforthe10Gbpsresults
• 50pairs,vectorprecodedwitheitherzero-forcinglinearprecoderorNon-LinearPrecoder(NLP)usingGeneralizedDecisionFeedbackEqualizaWon(GDFE);idealchannelesWmaWonassumed.
• -160dBm/HzbackgroundAWGN.
• WealsoaddinaTM2andTEM2modefor400GHzto500GHz(sameparameters)
Meank=0dBVar=6dB
TDSL 16 !
ResultsinTbps[down+up]/pair
5/8/1716
1Tbpsat100m
• CananyPONget1Tbpstoeachcustomer?
0
1
2
3
4
5
6
7
8
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200
DataRate(Tbp
s)perhom
e
LoopLength(m)
100-300GHzTDSL,perHomedatarates
NLP
Linearprecoded
TDSL 17 !
BitLoading(eachpolarizaWonof1wire)
5/8/17 17
0
1
2
3
4
5
6
7
8
9
100 110 120 129 139 149 159 168 178 188 198 207 217 227 237 246 256 266 276 286 295
BitLoa
ding(b
its/H
z)
Frequency(GHz)
50m
100m
150m
TDSL 18 !
LongerRange,LowerSpeed?
100Gbps>300m
10Gbps>500m(~0.5km)
0
0.5
1
1.5
2
2.5
3
10 30 50 70 90 110130150170190210230250270290310330
DataRate(Tbp
s)perHom
e
LoopLength(m)
60-120GHzTDSL,perHomedatarates
NLP
Linearprecoded
TDSL 19 !
Very-highspeedTDSL
5/8/17 19
• AddingTE2andTEM2modesfrom400-500GHz
0
2
4
6
8
10
12
14
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200
DataRate(Tbp
s)perHom
e
LoopLength(m)
100-500GHzTDSL,perHomedatarates
NLP
Linearprecoded
100meters 300meters 500meters
2Tbps 100Gbps 10Gbps
TDSL 20 !
Antennas(analogprocessing)?
5/8/17 20
• Whatwouldantennaslooklike?– Annularringsaroundeachwireend– AlsoatCPEside
– PossiblymulWpleco-centricringsatCPEside– CombinaWons– Catchasmuchdri{ingenergyonCPEreceiveraspossible(dualforupstreamtransmi`er)
• Whatwouldcouplingtowaveguideslooklike?(photoconducWve,photodetect)– Itmaybefeasibletohaveondieacouplerinthis200-400GHzfrequencyrange.
• Couplinglosses?
• Havenotincluded“nestedMIMO”overthe4(ormore)antennasperhomeinresults– Thiswillbealargeimprovement(likevector-bondinginmulW-lineDSL,butperhapsbe`er)
• Currentplotsignorethisimprovement– However,wewereopWmisWcontheenergylossa{erthesheath-breakonthesurfacewaves– Thetwoeffectsmayoffset
TDSL 21 !
DigitalSignalProcessing?
5/8/17 21
• Conversiondevices?– Mightnotuseall200GHz,butsWll…– ADC’srunningat120Gsamples/secondexist(Jari`ech)– UseMulWcarrier(AMTinsteadofDMT)?
• EachtonecouldhaveitsownADC/DAC(soeasilyavailable,butmanyinparallel)
• ProcessingCapabiliWes– VectorEngine,evenatpertoneof50MHz– .1-.25Giga-opspertone– Tera-opsforafullsystem
• CurrentNvidiaTeslaGP100has5Teraflopsin16nmCMOS• 4-7nmonimmediatehorizonandshouldallowcostreducWon
– WithinemergingcapabiliWes– Startat100Gbpsinstead(1/10thecost)?
TDSL 22 !
OpportuniWes–Measurements
5/8/17 22
• Howgoodisthelog-normalmodelforwaveguidemodes’xtalk?– Mightthisxtalkbelarger?
• Even10%ofthesenumbersis>>“G.mgfast”
• Arealcablemeasurementorafewwouldhelp– Thesemodescertainlyexist,butwhatisa`enuaWon?– Best/reasonableantenna/interfacedesigns?
TDSL 23 !
Whataboutyet-higher-ordermodes?
5/8/17 23
• Theyexist!• Higherbandwidths,buta`enuaWon?
– Unknownfornow– Likelyneedevenmoreantennas/wire(MIMO)– 10meters(insteadof100m)mightwork
• NotclearifwavescouldbefocusedlikesurfacewavesbyMIMOprocessingto“hugthewires”astheyseparateandgotoindividualhomes
• Groundedshieldwouldcontainthemthough– PDSL?(P=Petabitor1015)
• TDSLwillprobablybeenoughfornow
TDSL 24 !
Conclusions
• TDSListechnicallyfeasiblewith100pairsandphantomsusedforbackhaul– Alsoroughly1Tbps@100m,100Gbpsat300m,10Gbpsat500m
• ButofcourseonALL100pairsusedtogether• SWllcouldbeveryusefulfor5GcellmulWtude
• Terabit/sDSLperhome(orsmallcell)alsoappearsfeasible– Usingwaveguidemodesandvectoring–SINGLEpair– Measurementsofa`enuaWonwouldhelprefinerate/range
• ProbablywithMIMO-channelcharacterizaWonused• Couldbeexpensivetoprototype
– Becauseofprocessing/converterspeedsinvolved
• Isitworthit?(orshouldwespend$4Trilliontoreplaceallthecopperwithfiberinstead,sayinthenext3-5years…..Orcentury?)– Would5Gsmallcellsbeacceleratedsincethiswouldreduce
deploymentcost?
5/8/17 24
TDSL 25 !
AcknowledgementsandRefs
• Prof.JoeKahn,StanfordU.• Dr.RickyHo,Apple• Dr.S.Galli,Huawei• Prof.LeonidKazovsky,StanfordU.
• J.C.Wiltse,“Surface-WavePropagaWononaSingleMetalWireorRodatMillimeter-WaveandTerahertzFrequencies,”MicrowaveSymposiumDigest,2006.IEEEMTT-SInternaWonal,11-16June2006.
• R.E.Collin,“HertzianDipoleRadiaWngOveraLossyEarthorSea:SomeEarlyandLate20th-CenturyControversies”IEEEAntandPrp.Magazine,Vol46,No.2,April2004.
• T.I.Jeon,J.Zhang,andD.Grischkowsky,“THzSommerfeldwavepropagaWononasinglemetalwire,”Appl.Phys.Le`ers,Vol86,2005.
5/8/17 25
TDSL 26 !
BackUP
5/8/17 26
TDSL 27 !
Baremetalwirewaveguide
• Surfaceplasmonpolariton– EMsurfacewavethattravelsalonganinterfacebetweenmetal(negaWvepermi�vity)and
dielectric(posiWvepermi�vity)basedonsurfaceelectrondensitychangesbelowmetal’splasmafrequency
• Phasevelocityandgroupvelocityissame(likefreespace)ànodispersioniffrequencyiswaybelowplasmafrequency
– E-fielddecaysexponenWallyverWcaltothewireà energyisconfiedneartheconductorsono1/rtypeofpathloss.Onlysmallohmiclossduetoelectron
sca`eringàsmallinmaterialswithhighconducWvityandhighfrequencyà Lossabout0.1%~0.25%offieldstrengthin1cmà0.86dB/m~2.1dB/[email protected]
– Problems• TMmodeàHardtogenerateradiallypolarizedEMwave&lowcouplingcoefficient• Needtobestraightàloseenergyduetobending• ConnecWngtwometalwiresarenoteasy
5/8/17h`ps://nanohub.org/resources/1852/download/2006.10.05-ece695s-l09.pdf
27
TDSL 28 !
Splicer
5/8/17 28